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/net_namespace.h>
63#include <linux/nospec.h>
64#include <linux/btf_ids.h>
65
66#include <net/net_namespace.h>
67#include <net/netns/generic.h>
68#include <net/sock.h>
69#include <net/scm.h>
70#include <net/netlink.h>
71#define CREATE_TRACE_POINTS
72#include <trace/events/netlink.h>
73
74#include "af_netlink.h"
75#include "genetlink.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, bool lock_taken);
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 skb_queue_purge(&sk->sk_receive_queue);
397
398 if (!sock_flag(sk, SOCK_DEAD)) {
399 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
400 return;
401 }
402
403 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
404 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
405 WARN_ON(nlk_sk(sk)->groups);
406}
407
408/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
409 * SMP. Look, when several writers sleep and reader wakes them up, all but one
410 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
411 * this, _but_ remember, it adds useless work on UP machines.
412 */
413
414void netlink_table_grab(void)
415 __acquires(nl_table_lock)
416{
417 might_sleep();
418
419 write_lock_irq(&nl_table_lock);
420
421 if (atomic_read(&nl_table_users)) {
422 DECLARE_WAITQUEUE(wait, current);
423
424 add_wait_queue_exclusive(&nl_table_wait, &wait);
425 for (;;) {
426 set_current_state(TASK_UNINTERRUPTIBLE);
427 if (atomic_read(&nl_table_users) == 0)
428 break;
429 write_unlock_irq(&nl_table_lock);
430 schedule();
431 write_lock_irq(&nl_table_lock);
432 }
433
434 __set_current_state(TASK_RUNNING);
435 remove_wait_queue(&nl_table_wait, &wait);
436 }
437}
438
439void netlink_table_ungrab(void)
440 __releases(nl_table_lock)
441{
442 write_unlock_irq(&nl_table_lock);
443 wake_up(&nl_table_wait);
444}
445
446static inline void
447netlink_lock_table(void)
448{
449 unsigned long flags;
450
451 /* read_lock() synchronizes us to netlink_table_grab */
452
453 read_lock_irqsave(&nl_table_lock, flags);
454 atomic_inc(&nl_table_users);
455 read_unlock_irqrestore(&nl_table_lock, flags);
456}
457
458static inline void
459netlink_unlock_table(void)
460{
461 if (atomic_dec_and_test(&nl_table_users))
462 wake_up(&nl_table_wait);
463}
464
465struct netlink_compare_arg
466{
467 possible_net_t pnet;
468 u32 portid;
469};
470
471/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
472#define netlink_compare_arg_len \
473 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
474
475static inline int netlink_compare(struct rhashtable_compare_arg *arg,
476 const void *ptr)
477{
478 const struct netlink_compare_arg *x = arg->key;
479 const struct netlink_sock *nlk = ptr;
480
481 return nlk->portid != x->portid ||
482 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
483}
484
485static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
486 struct net *net, u32 portid)
487{
488 memset(arg, 0, sizeof(*arg));
489 write_pnet(&arg->pnet, net);
490 arg->portid = portid;
491}
492
493static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
494 struct net *net)
495{
496 struct netlink_compare_arg arg;
497
498 netlink_compare_arg_init(&arg, net, portid);
499 return rhashtable_lookup_fast(&table->hash, &arg,
500 netlink_rhashtable_params);
501}
502
503static int __netlink_insert(struct netlink_table *table, struct sock *sk)
504{
505 struct netlink_compare_arg arg;
506
507 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
508 return rhashtable_lookup_insert_key(&table->hash, &arg,
509 &nlk_sk(sk)->node,
510 netlink_rhashtable_params);
511}
512
513static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
514{
515 struct netlink_table *table = &nl_table[protocol];
516 struct sock *sk;
517
518 rcu_read_lock();
519 sk = __netlink_lookup(table, portid, net);
520 if (sk)
521 sock_hold(sk);
522 rcu_read_unlock();
523
524 return sk;
525}
526
527static const struct proto_ops netlink_ops;
528
529static void
530netlink_update_listeners(struct sock *sk)
531{
532 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
533 unsigned long mask;
534 unsigned int i;
535 struct listeners *listeners;
536
537 listeners = nl_deref_protected(tbl->listeners);
538 if (!listeners)
539 return;
540
541 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
542 mask = 0;
543 sk_for_each_bound(sk, &tbl->mc_list) {
544 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
545 mask |= nlk_sk(sk)->groups[i];
546 }
547 listeners->masks[i] = mask;
548 }
549 /* this function is only called with the netlink table "grabbed", which
550 * makes sure updates are visible before bind or setsockopt return. */
551}
552
553static int netlink_insert(struct sock *sk, u32 portid)
554{
555 struct netlink_table *table = &nl_table[sk->sk_protocol];
556 int err;
557
558 lock_sock(sk);
559
560 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
561 if (nlk_sk(sk)->bound)
562 goto err;
563
564 /* portid can be read locklessly from netlink_getname(). */
565 WRITE_ONCE(nlk_sk(sk)->portid, portid);
566
567 sock_hold(sk);
568
569 err = __netlink_insert(table, sk);
570 if (err) {
571 /* In case the hashtable backend returns with -EBUSY
572 * from here, it must not escape to the caller.
573 */
574 if (unlikely(err == -EBUSY))
575 err = -EOVERFLOW;
576 if (err == -EEXIST)
577 err = -EADDRINUSE;
578 sock_put(sk);
579 goto err;
580 }
581
582 /* We need to ensure that the socket is hashed and visible. */
583 smp_wmb();
584 /* Paired with lockless reads from netlink_bind(),
585 * netlink_connect() and netlink_sendmsg().
586 */
587 WRITE_ONCE(nlk_sk(sk)->bound, portid);
588
589err:
590 release_sock(sk);
591 return err;
592}
593
594static void netlink_remove(struct sock *sk)
595{
596 struct netlink_table *table;
597
598 table = &nl_table[sk->sk_protocol];
599 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
600 netlink_rhashtable_params)) {
601 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
602 __sock_put(sk);
603 }
604
605 netlink_table_grab();
606 if (nlk_sk(sk)->subscriptions) {
607 __sk_del_bind_node(sk);
608 netlink_update_listeners(sk);
609 }
610 if (sk->sk_protocol == NETLINK_GENERIC)
611 atomic_inc(&genl_sk_destructing_cnt);
612 netlink_table_ungrab();
613}
614
615static struct proto netlink_proto = {
616 .name = "NETLINK",
617 .owner = THIS_MODULE,
618 .obj_size = sizeof(struct netlink_sock),
619};
620
621static int __netlink_create(struct net *net, struct socket *sock,
622 int protocol, int kern)
623{
624 struct sock *sk;
625 struct netlink_sock *nlk;
626
627 sock->ops = &netlink_ops;
628
629 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
630 if (!sk)
631 return -ENOMEM;
632
633 sock_init_data(sock, sk);
634
635 nlk = nlk_sk(sk);
636 mutex_init(&nlk->nl_cb_mutex);
637 lockdep_set_class_and_name(&nlk->nl_cb_mutex,
638 nlk_cb_mutex_keys + protocol,
639 nlk_cb_mutex_key_strings[protocol]);
640 init_waitqueue_head(&nlk->wait);
641
642 sk->sk_destruct = netlink_sock_destruct;
643 sk->sk_protocol = protocol;
644 return 0;
645}
646
647static int netlink_create(struct net *net, struct socket *sock, int protocol,
648 int kern)
649{
650 struct module *module = NULL;
651 struct netlink_sock *nlk;
652 int (*bind)(struct net *net, int group);
653 void (*unbind)(struct net *net, int group);
654 void (*release)(struct sock *sock, unsigned long *groups);
655 int err = 0;
656
657 sock->state = SS_UNCONNECTED;
658
659 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
660 return -ESOCKTNOSUPPORT;
661
662 if (protocol < 0 || protocol >= MAX_LINKS)
663 return -EPROTONOSUPPORT;
664 protocol = array_index_nospec(protocol, MAX_LINKS);
665
666 netlink_lock_table();
667#ifdef CONFIG_MODULES
668 if (!nl_table[protocol].registered) {
669 netlink_unlock_table();
670 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
671 netlink_lock_table();
672 }
673#endif
674 if (nl_table[protocol].registered &&
675 try_module_get(nl_table[protocol].module))
676 module = nl_table[protocol].module;
677 else
678 err = -EPROTONOSUPPORT;
679 bind = nl_table[protocol].bind;
680 unbind = nl_table[protocol].unbind;
681 release = nl_table[protocol].release;
682 netlink_unlock_table();
683
684 if (err < 0)
685 goto out;
686
687 err = __netlink_create(net, sock, protocol, kern);
688 if (err < 0)
689 goto out_module;
690
691 sock_prot_inuse_add(net, &netlink_proto, 1);
692
693 nlk = nlk_sk(sock->sk);
694 nlk->module = module;
695 nlk->netlink_bind = bind;
696 nlk->netlink_unbind = unbind;
697 nlk->netlink_release = release;
698out:
699 return err;
700
701out_module:
702 module_put(module);
703 goto out;
704}
705
706static void deferred_put_nlk_sk(struct rcu_head *head)
707{
708 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
709 struct sock *sk = &nlk->sk;
710
711 kfree(nlk->groups);
712 nlk->groups = NULL;
713
714 if (!refcount_dec_and_test(&sk->sk_refcnt))
715 return;
716
717 sk_free(sk);
718}
719
720static int netlink_release(struct socket *sock)
721{
722 struct sock *sk = sock->sk;
723 struct netlink_sock *nlk;
724
725 if (!sk)
726 return 0;
727
728 netlink_remove(sk);
729 sock_orphan(sk);
730 nlk = nlk_sk(sk);
731
732 /*
733 * OK. Socket is unlinked, any packets that arrive now
734 * will be purged.
735 */
736 if (nlk->netlink_release)
737 nlk->netlink_release(sk, nlk->groups);
738
739 /* must not acquire netlink_table_lock in any way again before unbind
740 * and notifying genetlink is done as otherwise it might deadlock
741 */
742 if (nlk->netlink_unbind) {
743 int i;
744
745 for (i = 0; i < nlk->ngroups; i++)
746 if (test_bit(i, nlk->groups))
747 nlk->netlink_unbind(sock_net(sk), i + 1);
748 }
749 if (sk->sk_protocol == NETLINK_GENERIC &&
750 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
751 wake_up(&genl_sk_destructing_waitq);
752
753 sock->sk = NULL;
754 wake_up_interruptible_all(&nlk->wait);
755
756 skb_queue_purge(&sk->sk_write_queue);
757
758 if (nlk->portid && nlk->bound) {
759 struct netlink_notify n = {
760 .net = sock_net(sk),
761 .protocol = sk->sk_protocol,
762 .portid = nlk->portid,
763 };
764 blocking_notifier_call_chain(&netlink_chain,
765 NETLINK_URELEASE, &n);
766 }
767
768 /* Terminate any outstanding dump */
769 if (nlk->cb_running) {
770 if (nlk->cb.done)
771 nlk->cb.done(&nlk->cb);
772 module_put(nlk->cb.module);
773 kfree_skb(nlk->cb.skb);
774 }
775
776 module_put(nlk->module);
777
778 if (netlink_is_kernel(sk)) {
779 netlink_table_grab();
780 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
781 if (--nl_table[sk->sk_protocol].registered == 0) {
782 struct listeners *old;
783
784 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
785 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
786 kfree_rcu(old, rcu);
787 nl_table[sk->sk_protocol].module = NULL;
788 nl_table[sk->sk_protocol].bind = NULL;
789 nl_table[sk->sk_protocol].unbind = NULL;
790 nl_table[sk->sk_protocol].flags = 0;
791 nl_table[sk->sk_protocol].registered = 0;
792 }
793 netlink_table_ungrab();
794 }
795
796 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
797
798 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
799 return 0;
800}
801
802static int netlink_autobind(struct socket *sock)
803{
804 struct sock *sk = sock->sk;
805 struct net *net = sock_net(sk);
806 struct netlink_table *table = &nl_table[sk->sk_protocol];
807 s32 portid = task_tgid_vnr(current);
808 int err;
809 s32 rover = -4096;
810 bool ok;
811
812retry:
813 cond_resched();
814 rcu_read_lock();
815 ok = !__netlink_lookup(table, portid, net);
816 rcu_read_unlock();
817 if (!ok) {
818 /* Bind collision, search negative portid values. */
819 if (rover == -4096)
820 /* rover will be in range [S32_MIN, -4097] */
821 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
822 else if (rover >= -4096)
823 rover = -4097;
824 portid = rover--;
825 goto retry;
826 }
827
828 err = netlink_insert(sk, portid);
829 if (err == -EADDRINUSE)
830 goto retry;
831
832 /* If 2 threads race to autobind, that is fine. */
833 if (err == -EBUSY)
834 err = 0;
835
836 return err;
837}
838
839/**
840 * __netlink_ns_capable - General netlink message capability test
841 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
842 * @user_ns: The user namespace of the capability to use
843 * @cap: The capability to use
844 *
845 * Test to see if the opener of the socket we received the message
846 * from had when the netlink socket was created and the sender of the
847 * message has the capability @cap in the user namespace @user_ns.
848 */
849bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
850 struct user_namespace *user_ns, int cap)
851{
852 return ((nsp->flags & NETLINK_SKB_DST) ||
853 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
854 ns_capable(user_ns, cap);
855}
856EXPORT_SYMBOL(__netlink_ns_capable);
857
858/**
859 * netlink_ns_capable - General netlink message capability test
860 * @skb: socket buffer holding a netlink command from userspace
861 * @user_ns: The user namespace of the capability to use
862 * @cap: The capability to use
863 *
864 * Test to see if the opener of the socket we received the message
865 * from had when the netlink socket was created and the sender of the
866 * message has the capability @cap in the user namespace @user_ns.
867 */
868bool netlink_ns_capable(const struct sk_buff *skb,
869 struct user_namespace *user_ns, int cap)
870{
871 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
872}
873EXPORT_SYMBOL(netlink_ns_capable);
874
875/**
876 * netlink_capable - Netlink global message capability test
877 * @skb: socket buffer holding a netlink command from userspace
878 * @cap: The capability to use
879 *
880 * Test to see if the opener of the socket we received the message
881 * from had when the netlink socket was created and the sender of the
882 * message has the capability @cap in all user namespaces.
883 */
884bool netlink_capable(const struct sk_buff *skb, int cap)
885{
886 return netlink_ns_capable(skb, &init_user_ns, cap);
887}
888EXPORT_SYMBOL(netlink_capable);
889
890/**
891 * netlink_net_capable - Netlink network namespace message capability test
892 * @skb: socket buffer holding a netlink command from userspace
893 * @cap: The capability to use
894 *
895 * Test to see if the opener of the socket we received the message
896 * from had when the netlink socket was created and the sender of the
897 * message has the capability @cap over the network namespace of
898 * the socket we received the message from.
899 */
900bool netlink_net_capable(const struct sk_buff *skb, int cap)
901{
902 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
903}
904EXPORT_SYMBOL(netlink_net_capable);
905
906static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
907{
908 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
909 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
910}
911
912static void
913netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
914{
915 struct netlink_sock *nlk = nlk_sk(sk);
916
917 if (nlk->subscriptions && !subscriptions)
918 __sk_del_bind_node(sk);
919 else if (!nlk->subscriptions && subscriptions)
920 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
921 nlk->subscriptions = subscriptions;
922}
923
924static int netlink_realloc_groups(struct sock *sk)
925{
926 struct netlink_sock *nlk = nlk_sk(sk);
927 unsigned int groups;
928 unsigned long *new_groups;
929 int err = 0;
930
931 netlink_table_grab();
932
933 groups = nl_table[sk->sk_protocol].groups;
934 if (!nl_table[sk->sk_protocol].registered) {
935 err = -ENOENT;
936 goto out_unlock;
937 }
938
939 if (nlk->ngroups >= groups)
940 goto out_unlock;
941
942 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
943 if (new_groups == NULL) {
944 err = -ENOMEM;
945 goto out_unlock;
946 }
947 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
948 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
949
950 nlk->groups = new_groups;
951 nlk->ngroups = groups;
952 out_unlock:
953 netlink_table_ungrab();
954 return err;
955}
956
957static void netlink_undo_bind(int group, long unsigned int groups,
958 struct sock *sk)
959{
960 struct netlink_sock *nlk = nlk_sk(sk);
961 int undo;
962
963 if (!nlk->netlink_unbind)
964 return;
965
966 for (undo = 0; undo < group; undo++)
967 if (test_bit(undo, &groups))
968 nlk->netlink_unbind(sock_net(sk), undo + 1);
969}
970
971static int netlink_bind(struct socket *sock, struct sockaddr *addr,
972 int addr_len)
973{
974 struct sock *sk = sock->sk;
975 struct net *net = sock_net(sk);
976 struct netlink_sock *nlk = nlk_sk(sk);
977 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
978 int err = 0;
979 unsigned long groups;
980 bool bound;
981
982 if (addr_len < sizeof(struct sockaddr_nl))
983 return -EINVAL;
984
985 if (nladdr->nl_family != AF_NETLINK)
986 return -EINVAL;
987 groups = nladdr->nl_groups;
988
989 /* Only superuser is allowed to listen multicasts */
990 if (groups) {
991 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
992 return -EPERM;
993 err = netlink_realloc_groups(sk);
994 if (err)
995 return err;
996 }
997
998 if (nlk->ngroups < BITS_PER_LONG)
999 groups &= (1UL << nlk->ngroups) - 1;
1000
1001 /* Paired with WRITE_ONCE() in netlink_insert() */
1002 bound = READ_ONCE(nlk->bound);
1003 if (bound) {
1004 /* Ensure nlk->portid is up-to-date. */
1005 smp_rmb();
1006
1007 if (nladdr->nl_pid != nlk->portid)
1008 return -EINVAL;
1009 }
1010
1011 if (nlk->netlink_bind && groups) {
1012 int group;
1013
1014 /* nl_groups is a u32, so cap the maximum groups we can bind */
1015 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1016 if (!test_bit(group, &groups))
1017 continue;
1018 err = nlk->netlink_bind(net, group + 1);
1019 if (!err)
1020 continue;
1021 netlink_undo_bind(group, groups, sk);
1022 return err;
1023 }
1024 }
1025
1026 /* No need for barriers here as we return to user-space without
1027 * using any of the bound attributes.
1028 */
1029 netlink_lock_table();
1030 if (!bound) {
1031 err = nladdr->nl_pid ?
1032 netlink_insert(sk, nladdr->nl_pid) :
1033 netlink_autobind(sock);
1034 if (err) {
1035 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1036 goto unlock;
1037 }
1038 }
1039
1040 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1041 goto unlock;
1042 netlink_unlock_table();
1043
1044 netlink_table_grab();
1045 netlink_update_subscriptions(sk, nlk->subscriptions +
1046 hweight32(groups) -
1047 hweight32(nlk->groups[0]));
1048 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1049 netlink_update_listeners(sk);
1050 netlink_table_ungrab();
1051
1052 return 0;
1053
1054unlock:
1055 netlink_unlock_table();
1056 return err;
1057}
1058
1059static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1060 int alen, int flags)
1061{
1062 int err = 0;
1063 struct sock *sk = sock->sk;
1064 struct netlink_sock *nlk = nlk_sk(sk);
1065 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1066
1067 if (alen < sizeof(addr->sa_family))
1068 return -EINVAL;
1069
1070 if (addr->sa_family == AF_UNSPEC) {
1071 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1072 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1073 /* dst_portid and dst_group can be read locklessly */
1074 WRITE_ONCE(nlk->dst_portid, 0);
1075 WRITE_ONCE(nlk->dst_group, 0);
1076 return 0;
1077 }
1078 if (addr->sa_family != AF_NETLINK)
1079 return -EINVAL;
1080
1081 if (alen < sizeof(struct sockaddr_nl))
1082 return -EINVAL;
1083
1084 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1085 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1086 return -EPERM;
1087
1088 /* No need for barriers here as we return to user-space without
1089 * using any of the bound attributes.
1090 * Paired with WRITE_ONCE() in netlink_insert().
1091 */
1092 if (!READ_ONCE(nlk->bound))
1093 err = netlink_autobind(sock);
1094
1095 if (err == 0) {
1096 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1097 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1098 /* dst_portid and dst_group can be read locklessly */
1099 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1100 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1101 }
1102
1103 return err;
1104}
1105
1106static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1107 int peer)
1108{
1109 struct sock *sk = sock->sk;
1110 struct netlink_sock *nlk = nlk_sk(sk);
1111 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1112
1113 nladdr->nl_family = AF_NETLINK;
1114 nladdr->nl_pad = 0;
1115
1116 if (peer) {
1117 /* Paired with WRITE_ONCE() in netlink_connect() */
1118 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1119 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1120 } else {
1121 /* Paired with WRITE_ONCE() in netlink_insert() */
1122 nladdr->nl_pid = READ_ONCE(nlk->portid);
1123 netlink_lock_table();
1124 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1125 netlink_unlock_table();
1126 }
1127 return sizeof(*nladdr);
1128}
1129
1130static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1131 unsigned long arg)
1132{
1133 /* try to hand this ioctl down to the NIC drivers.
1134 */
1135 return -ENOIOCTLCMD;
1136}
1137
1138static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1139{
1140 struct sock *sock;
1141 struct netlink_sock *nlk;
1142
1143 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1144 if (!sock)
1145 return ERR_PTR(-ECONNREFUSED);
1146
1147 /* Don't bother queuing skb if kernel socket has no input function */
1148 nlk = nlk_sk(sock);
1149 /* dst_portid and sk_state can be changed in netlink_connect() */
1150 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1151 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1152 sock_put(sock);
1153 return ERR_PTR(-ECONNREFUSED);
1154 }
1155 return sock;
1156}
1157
1158struct sock *netlink_getsockbyfd(int fd)
1159{
1160 CLASS(fd, f)(fd);
1161 struct inode *inode;
1162 struct sock *sock;
1163
1164 if (fd_empty(f))
1165 return ERR_PTR(-EBADF);
1166
1167 inode = file_inode(fd_file(f));
1168 if (!S_ISSOCK(inode->i_mode))
1169 return ERR_PTR(-ENOTSOCK);
1170
1171 sock = SOCKET_I(inode)->sk;
1172 if (sock->sk_family != AF_NETLINK)
1173 return ERR_PTR(-EINVAL);
1174
1175 sock_hold(sock);
1176 return sock;
1177}
1178
1179struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1180{
1181 size_t head_size = SKB_HEAD_ALIGN(size);
1182 struct sk_buff *skb;
1183 void *data;
1184
1185 if (head_size <= PAGE_SIZE || broadcast)
1186 return alloc_skb(size, GFP_KERNEL);
1187
1188 data = kvmalloc(head_size, GFP_KERNEL);
1189 if (!data)
1190 return NULL;
1191
1192 skb = __build_skb(data, head_size);
1193 if (!skb)
1194 kvfree(data);
1195 else if (is_vmalloc_addr(data))
1196 skb->destructor = netlink_skb_destructor;
1197
1198 return skb;
1199}
1200
1201/*
1202 * Attach a skb to a netlink socket.
1203 * The caller must hold a reference to the destination socket. On error, the
1204 * reference is dropped. The skb is not send to the destination, just all
1205 * all error checks are performed and memory in the queue is reserved.
1206 * Return values:
1207 * < 0: error. skb freed, reference to sock dropped.
1208 * 0: continue
1209 * 1: repeat lookup - reference dropped while waiting for socket memory.
1210 */
1211int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1212 long *timeo, struct sock *ssk)
1213{
1214 struct netlink_sock *nlk;
1215
1216 nlk = nlk_sk(sk);
1217
1218 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1219 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1220 DECLARE_WAITQUEUE(wait, current);
1221 if (!*timeo) {
1222 if (!ssk || netlink_is_kernel(ssk))
1223 netlink_overrun(sk);
1224 sock_put(sk);
1225 kfree_skb(skb);
1226 return -EAGAIN;
1227 }
1228
1229 __set_current_state(TASK_INTERRUPTIBLE);
1230 add_wait_queue(&nlk->wait, &wait);
1231
1232 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1233 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1234 !sock_flag(sk, SOCK_DEAD))
1235 *timeo = schedule_timeout(*timeo);
1236
1237 __set_current_state(TASK_RUNNING);
1238 remove_wait_queue(&nlk->wait, &wait);
1239 sock_put(sk);
1240
1241 if (signal_pending(current)) {
1242 kfree_skb(skb);
1243 return sock_intr_errno(*timeo);
1244 }
1245 return 1;
1246 }
1247 netlink_skb_set_owner_r(skb, sk);
1248 return 0;
1249}
1250
1251static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1252{
1253 int len = skb->len;
1254
1255 netlink_deliver_tap(sock_net(sk), skb);
1256
1257 skb_queue_tail(&sk->sk_receive_queue, skb);
1258 sk->sk_data_ready(sk);
1259 return len;
1260}
1261
1262int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1263{
1264 int len = __netlink_sendskb(sk, skb);
1265
1266 sock_put(sk);
1267 return len;
1268}
1269
1270void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1271{
1272 kfree_skb(skb);
1273 sock_put(sk);
1274}
1275
1276static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1277{
1278 int delta;
1279
1280 WARN_ON(skb->sk != NULL);
1281 delta = skb->end - skb->tail;
1282 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1283 return skb;
1284
1285 if (skb_shared(skb)) {
1286 struct sk_buff *nskb = skb_clone(skb, allocation);
1287 if (!nskb)
1288 return skb;
1289 consume_skb(skb);
1290 skb = nskb;
1291 }
1292
1293 pskb_expand_head(skb, 0, -delta,
1294 (allocation & ~__GFP_DIRECT_RECLAIM) |
1295 __GFP_NOWARN | __GFP_NORETRY);
1296 return skb;
1297}
1298
1299static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1300 struct sock *ssk)
1301{
1302 int ret;
1303 struct netlink_sock *nlk = nlk_sk(sk);
1304
1305 ret = -ECONNREFUSED;
1306 if (nlk->netlink_rcv != NULL) {
1307 ret = skb->len;
1308 netlink_skb_set_owner_r(skb, sk);
1309 NETLINK_CB(skb).sk = ssk;
1310 netlink_deliver_tap_kernel(sk, ssk, skb);
1311 nlk->netlink_rcv(skb);
1312 consume_skb(skb);
1313 } else {
1314 kfree_skb(skb);
1315 }
1316 sock_put(sk);
1317 return ret;
1318}
1319
1320int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1321 u32 portid, int nonblock)
1322{
1323 struct sock *sk;
1324 int err;
1325 long timeo;
1326
1327 skb = netlink_trim(skb, gfp_any());
1328
1329 timeo = sock_sndtimeo(ssk, nonblock);
1330retry:
1331 sk = netlink_getsockbyportid(ssk, portid);
1332 if (IS_ERR(sk)) {
1333 kfree_skb(skb);
1334 return PTR_ERR(sk);
1335 }
1336 if (netlink_is_kernel(sk))
1337 return netlink_unicast_kernel(sk, skb, ssk);
1338
1339 if (sk_filter(sk, skb)) {
1340 err = skb->len;
1341 kfree_skb(skb);
1342 sock_put(sk);
1343 return err;
1344 }
1345
1346 err = netlink_attachskb(sk, skb, &timeo, ssk);
1347 if (err == 1)
1348 goto retry;
1349 if (err)
1350 return err;
1351
1352 return netlink_sendskb(sk, skb);
1353}
1354EXPORT_SYMBOL(netlink_unicast);
1355
1356int netlink_has_listeners(struct sock *sk, unsigned int group)
1357{
1358 int res = 0;
1359 struct listeners *listeners;
1360
1361 BUG_ON(!netlink_is_kernel(sk));
1362
1363 rcu_read_lock();
1364 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1365
1366 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1367 res = test_bit(group - 1, listeners->masks);
1368
1369 rcu_read_unlock();
1370
1371 return res;
1372}
1373EXPORT_SYMBOL_GPL(netlink_has_listeners);
1374
1375bool netlink_strict_get_check(struct sk_buff *skb)
1376{
1377 return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1378}
1379EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1380
1381static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1382{
1383 struct netlink_sock *nlk = nlk_sk(sk);
1384
1385 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1386 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1387 netlink_skb_set_owner_r(skb, sk);
1388 __netlink_sendskb(sk, skb);
1389 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1390 }
1391 return -1;
1392}
1393
1394struct netlink_broadcast_data {
1395 struct sock *exclude_sk;
1396 struct net *net;
1397 u32 portid;
1398 u32 group;
1399 int failure;
1400 int delivery_failure;
1401 int congested;
1402 int delivered;
1403 gfp_t allocation;
1404 struct sk_buff *skb, *skb2;
1405 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1406 void *tx_data;
1407};
1408
1409static void do_one_broadcast(struct sock *sk,
1410 struct netlink_broadcast_data *p)
1411{
1412 struct netlink_sock *nlk = nlk_sk(sk);
1413 int val;
1414
1415 if (p->exclude_sk == sk)
1416 return;
1417
1418 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1419 !test_bit(p->group - 1, nlk->groups))
1420 return;
1421
1422 if (!net_eq(sock_net(sk), p->net)) {
1423 if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1424 return;
1425
1426 if (!peernet_has_id(sock_net(sk), p->net))
1427 return;
1428
1429 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1430 CAP_NET_BROADCAST))
1431 return;
1432 }
1433
1434 if (p->failure) {
1435 netlink_overrun(sk);
1436 return;
1437 }
1438
1439 sock_hold(sk);
1440 if (p->skb2 == NULL) {
1441 if (skb_shared(p->skb)) {
1442 p->skb2 = skb_clone(p->skb, p->allocation);
1443 } else {
1444 p->skb2 = skb_get(p->skb);
1445 /*
1446 * skb ownership may have been set when
1447 * delivered to a previous socket.
1448 */
1449 skb_orphan(p->skb2);
1450 }
1451 }
1452 if (p->skb2 == NULL) {
1453 netlink_overrun(sk);
1454 /* Clone failed. Notify ALL listeners. */
1455 p->failure = 1;
1456 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1457 p->delivery_failure = 1;
1458 goto out;
1459 }
1460
1461 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1462 kfree_skb(p->skb2);
1463 p->skb2 = NULL;
1464 goto out;
1465 }
1466
1467 if (sk_filter(sk, p->skb2)) {
1468 kfree_skb(p->skb2);
1469 p->skb2 = NULL;
1470 goto out;
1471 }
1472 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1473 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1474 NETLINK_CB(p->skb2).nsid_is_set = true;
1475 val = netlink_broadcast_deliver(sk, p->skb2);
1476 if (val < 0) {
1477 netlink_overrun(sk);
1478 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1479 p->delivery_failure = 1;
1480 } else {
1481 p->congested |= val;
1482 p->delivered = 1;
1483 p->skb2 = NULL;
1484 }
1485out:
1486 sock_put(sk);
1487}
1488
1489int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1490 u32 portid,
1491 u32 group, gfp_t allocation,
1492 netlink_filter_fn filter,
1493 void *filter_data)
1494{
1495 struct net *net = sock_net(ssk);
1496 struct netlink_broadcast_data info;
1497 struct sock *sk;
1498
1499 skb = netlink_trim(skb, allocation);
1500
1501 info.exclude_sk = ssk;
1502 info.net = net;
1503 info.portid = portid;
1504 info.group = group;
1505 info.failure = 0;
1506 info.delivery_failure = 0;
1507 info.congested = 0;
1508 info.delivered = 0;
1509 info.allocation = allocation;
1510 info.skb = skb;
1511 info.skb2 = NULL;
1512 info.tx_filter = filter;
1513 info.tx_data = filter_data;
1514
1515 /* While we sleep in clone, do not allow to change socket list */
1516
1517 netlink_lock_table();
1518
1519 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1520 do_one_broadcast(sk, &info);
1521
1522 consume_skb(skb);
1523
1524 netlink_unlock_table();
1525
1526 if (info.delivery_failure) {
1527 kfree_skb(info.skb2);
1528 return -ENOBUFS;
1529 }
1530 consume_skb(info.skb2);
1531
1532 if (info.delivered) {
1533 if (info.congested && gfpflags_allow_blocking(allocation))
1534 yield();
1535 return 0;
1536 }
1537 return -ESRCH;
1538}
1539EXPORT_SYMBOL(netlink_broadcast_filtered);
1540
1541int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1542 u32 group, gfp_t allocation)
1543{
1544 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1545 NULL, NULL);
1546}
1547EXPORT_SYMBOL(netlink_broadcast);
1548
1549struct netlink_set_err_data {
1550 struct sock *exclude_sk;
1551 u32 portid;
1552 u32 group;
1553 int code;
1554};
1555
1556static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1557{
1558 struct netlink_sock *nlk = nlk_sk(sk);
1559 int ret = 0;
1560
1561 if (sk == p->exclude_sk)
1562 goto out;
1563
1564 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1565 goto out;
1566
1567 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1568 !test_bit(p->group - 1, nlk->groups))
1569 goto out;
1570
1571 if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1572 ret = 1;
1573 goto out;
1574 }
1575
1576 WRITE_ONCE(sk->sk_err, p->code);
1577 sk_error_report(sk);
1578out:
1579 return ret;
1580}
1581
1582/**
1583 * netlink_set_err - report error to broadcast listeners
1584 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1585 * @portid: the PORTID of a process that we want to skip (if any)
1586 * @group: the broadcast group that will notice the error
1587 * @code: error code, must be negative (as usual in kernelspace)
1588 *
1589 * This function returns the number of broadcast listeners that have set the
1590 * NETLINK_NO_ENOBUFS socket option.
1591 */
1592int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1593{
1594 struct netlink_set_err_data info;
1595 unsigned long flags;
1596 struct sock *sk;
1597 int ret = 0;
1598
1599 info.exclude_sk = ssk;
1600 info.portid = portid;
1601 info.group = group;
1602 /* sk->sk_err wants a positive error value */
1603 info.code = -code;
1604
1605 read_lock_irqsave(&nl_table_lock, flags);
1606
1607 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1608 ret += do_one_set_err(sk, &info);
1609
1610 read_unlock_irqrestore(&nl_table_lock, flags);
1611 return ret;
1612}
1613EXPORT_SYMBOL(netlink_set_err);
1614
1615/* must be called with netlink table grabbed */
1616static void netlink_update_socket_mc(struct netlink_sock *nlk,
1617 unsigned int group,
1618 int is_new)
1619{
1620 int old, new = !!is_new, subscriptions;
1621
1622 old = test_bit(group - 1, nlk->groups);
1623 subscriptions = nlk->subscriptions - old + new;
1624 __assign_bit(group - 1, nlk->groups, new);
1625 netlink_update_subscriptions(&nlk->sk, subscriptions);
1626 netlink_update_listeners(&nlk->sk);
1627}
1628
1629static int netlink_setsockopt(struct socket *sock, int level, int optname,
1630 sockptr_t optval, unsigned int optlen)
1631{
1632 struct sock *sk = sock->sk;
1633 struct netlink_sock *nlk = nlk_sk(sk);
1634 unsigned int val = 0;
1635 int nr = -1;
1636
1637 if (level != SOL_NETLINK)
1638 return -ENOPROTOOPT;
1639
1640 if (optlen >= sizeof(int) &&
1641 copy_from_sockptr(&val, optval, sizeof(val)))
1642 return -EFAULT;
1643
1644 switch (optname) {
1645 case NETLINK_PKTINFO:
1646 nr = NETLINK_F_RECV_PKTINFO;
1647 break;
1648 case NETLINK_ADD_MEMBERSHIP:
1649 case NETLINK_DROP_MEMBERSHIP: {
1650 int err;
1651
1652 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1653 return -EPERM;
1654 err = netlink_realloc_groups(sk);
1655 if (err)
1656 return err;
1657 if (!val || val - 1 >= nlk->ngroups)
1658 return -EINVAL;
1659 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1660 err = nlk->netlink_bind(sock_net(sk), val);
1661 if (err)
1662 return err;
1663 }
1664 netlink_table_grab();
1665 netlink_update_socket_mc(nlk, val,
1666 optname == NETLINK_ADD_MEMBERSHIP);
1667 netlink_table_ungrab();
1668 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1669 nlk->netlink_unbind(sock_net(sk), val);
1670
1671 break;
1672 }
1673 case NETLINK_BROADCAST_ERROR:
1674 nr = NETLINK_F_BROADCAST_SEND_ERROR;
1675 break;
1676 case NETLINK_NO_ENOBUFS:
1677 assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1678 if (val) {
1679 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1680 wake_up_interruptible(&nlk->wait);
1681 }
1682 break;
1683 case NETLINK_LISTEN_ALL_NSID:
1684 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1685 return -EPERM;
1686 nr = NETLINK_F_LISTEN_ALL_NSID;
1687 break;
1688 case NETLINK_CAP_ACK:
1689 nr = NETLINK_F_CAP_ACK;
1690 break;
1691 case NETLINK_EXT_ACK:
1692 nr = NETLINK_F_EXT_ACK;
1693 break;
1694 case NETLINK_GET_STRICT_CHK:
1695 nr = NETLINK_F_STRICT_CHK;
1696 break;
1697 default:
1698 return -ENOPROTOOPT;
1699 }
1700 if (nr >= 0)
1701 assign_bit(nr, &nlk->flags, val);
1702 return 0;
1703}
1704
1705static int netlink_getsockopt(struct socket *sock, int level, int optname,
1706 char __user *optval, int __user *optlen)
1707{
1708 struct sock *sk = sock->sk;
1709 struct netlink_sock *nlk = nlk_sk(sk);
1710 unsigned int flag;
1711 int len, val;
1712
1713 if (level != SOL_NETLINK)
1714 return -ENOPROTOOPT;
1715
1716 if (get_user(len, optlen))
1717 return -EFAULT;
1718 if (len < 0)
1719 return -EINVAL;
1720
1721 switch (optname) {
1722 case NETLINK_PKTINFO:
1723 flag = NETLINK_F_RECV_PKTINFO;
1724 break;
1725 case NETLINK_BROADCAST_ERROR:
1726 flag = NETLINK_F_BROADCAST_SEND_ERROR;
1727 break;
1728 case NETLINK_NO_ENOBUFS:
1729 flag = NETLINK_F_RECV_NO_ENOBUFS;
1730 break;
1731 case NETLINK_LIST_MEMBERSHIPS: {
1732 int pos, idx, shift, err = 0;
1733
1734 netlink_lock_table();
1735 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1736 if (len - pos < sizeof(u32))
1737 break;
1738
1739 idx = pos / sizeof(unsigned long);
1740 shift = (pos % sizeof(unsigned long)) * 8;
1741 if (put_user((u32)(nlk->groups[idx] >> shift),
1742 (u32 __user *)(optval + pos))) {
1743 err = -EFAULT;
1744 break;
1745 }
1746 }
1747 if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1748 err = -EFAULT;
1749 netlink_unlock_table();
1750 return err;
1751 }
1752 case NETLINK_LISTEN_ALL_NSID:
1753 flag = NETLINK_F_LISTEN_ALL_NSID;
1754 break;
1755 case NETLINK_CAP_ACK:
1756 flag = NETLINK_F_CAP_ACK;
1757 break;
1758 case NETLINK_EXT_ACK:
1759 flag = NETLINK_F_EXT_ACK;
1760 break;
1761 case NETLINK_GET_STRICT_CHK:
1762 flag = NETLINK_F_STRICT_CHK;
1763 break;
1764 default:
1765 return -ENOPROTOOPT;
1766 }
1767
1768 if (len < sizeof(int))
1769 return -EINVAL;
1770
1771 len = sizeof(int);
1772 val = test_bit(flag, &nlk->flags);
1773
1774 if (put_user(len, optlen) ||
1775 copy_to_user(optval, &val, len))
1776 return -EFAULT;
1777
1778 return 0;
1779}
1780
1781static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1782{
1783 struct nl_pktinfo info;
1784
1785 info.group = NETLINK_CB(skb).dst_group;
1786 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1787}
1788
1789static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1790 struct sk_buff *skb)
1791{
1792 if (!NETLINK_CB(skb).nsid_is_set)
1793 return;
1794
1795 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1796 &NETLINK_CB(skb).nsid);
1797}
1798
1799static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1800{
1801 struct sock *sk = sock->sk;
1802 struct netlink_sock *nlk = nlk_sk(sk);
1803 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1804 u32 dst_portid;
1805 u32 dst_group;
1806 struct sk_buff *skb;
1807 int err;
1808 struct scm_cookie scm;
1809 u32 netlink_skb_flags = 0;
1810
1811 if (msg->msg_flags & MSG_OOB)
1812 return -EOPNOTSUPP;
1813
1814 if (len == 0) {
1815 pr_warn_once("Zero length message leads to an empty skb\n");
1816 return -ENODATA;
1817 }
1818
1819 err = scm_send(sock, msg, &scm, true);
1820 if (err < 0)
1821 return err;
1822
1823 if (msg->msg_namelen) {
1824 err = -EINVAL;
1825 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1826 goto out;
1827 if (addr->nl_family != AF_NETLINK)
1828 goto out;
1829 dst_portid = addr->nl_pid;
1830 dst_group = ffs(addr->nl_groups);
1831 err = -EPERM;
1832 if ((dst_group || dst_portid) &&
1833 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1834 goto out;
1835 netlink_skb_flags |= NETLINK_SKB_DST;
1836 } else {
1837 /* Paired with WRITE_ONCE() in netlink_connect() */
1838 dst_portid = READ_ONCE(nlk->dst_portid);
1839 dst_group = READ_ONCE(nlk->dst_group);
1840 }
1841
1842 /* Paired with WRITE_ONCE() in netlink_insert() */
1843 if (!READ_ONCE(nlk->bound)) {
1844 err = netlink_autobind(sock);
1845 if (err)
1846 goto out;
1847 } else {
1848 /* Ensure nlk is hashed and visible. */
1849 smp_rmb();
1850 }
1851
1852 err = -EMSGSIZE;
1853 if (len > sk->sk_sndbuf - 32)
1854 goto out;
1855 err = -ENOBUFS;
1856 skb = netlink_alloc_large_skb(len, dst_group);
1857 if (skb == NULL)
1858 goto out;
1859
1860 NETLINK_CB(skb).portid = nlk->portid;
1861 NETLINK_CB(skb).dst_group = dst_group;
1862 NETLINK_CB(skb).creds = scm.creds;
1863 NETLINK_CB(skb).flags = netlink_skb_flags;
1864
1865 err = -EFAULT;
1866 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1867 kfree_skb(skb);
1868 goto out;
1869 }
1870
1871 err = security_netlink_send(sk, skb);
1872 if (err) {
1873 kfree_skb(skb);
1874 goto out;
1875 }
1876
1877 if (dst_group) {
1878 refcount_inc(&skb->users);
1879 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1880 }
1881 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1882
1883out:
1884 scm_destroy(&scm);
1885 return err;
1886}
1887
1888static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1889 int flags)
1890{
1891 struct scm_cookie scm;
1892 struct sock *sk = sock->sk;
1893 struct netlink_sock *nlk = nlk_sk(sk);
1894 size_t copied, max_recvmsg_len;
1895 struct sk_buff *skb, *data_skb;
1896 int err, ret;
1897
1898 if (flags & MSG_OOB)
1899 return -EOPNOTSUPP;
1900
1901 copied = 0;
1902
1903 skb = skb_recv_datagram(sk, flags, &err);
1904 if (skb == NULL)
1905 goto out;
1906
1907 data_skb = skb;
1908
1909#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1910 if (unlikely(skb_shinfo(skb)->frag_list)) {
1911 /*
1912 * If this skb has a frag_list, then here that means that we
1913 * will have to use the frag_list skb's data for compat tasks
1914 * and the regular skb's data for normal (non-compat) tasks.
1915 *
1916 * If we need to send the compat skb, assign it to the
1917 * 'data_skb' variable so that it will be used below for data
1918 * copying. We keep 'skb' for everything else, including
1919 * freeing both later.
1920 */
1921 if (flags & MSG_CMSG_COMPAT)
1922 data_skb = skb_shinfo(skb)->frag_list;
1923 }
1924#endif
1925
1926 /* Record the max length of recvmsg() calls for future allocations */
1927 max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1928 max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1929 SKB_WITH_OVERHEAD(32768));
1930 WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1931
1932 copied = data_skb->len;
1933 if (len < copied) {
1934 msg->msg_flags |= MSG_TRUNC;
1935 copied = len;
1936 }
1937
1938 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1939
1940 if (msg->msg_name) {
1941 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1942 addr->nl_family = AF_NETLINK;
1943 addr->nl_pad = 0;
1944 addr->nl_pid = NETLINK_CB(skb).portid;
1945 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1946 msg->msg_namelen = sizeof(*addr);
1947 }
1948
1949 if (nlk_test_bit(RECV_PKTINFO, sk))
1950 netlink_cmsg_recv_pktinfo(msg, skb);
1951 if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1952 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1953
1954 memset(&scm, 0, sizeof(scm));
1955 scm.creds = *NETLINK_CREDS(skb);
1956 if (flags & MSG_TRUNC)
1957 copied = data_skb->len;
1958
1959 skb_free_datagram(sk, skb);
1960
1961 if (READ_ONCE(nlk->cb_running) &&
1962 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1963 ret = netlink_dump(sk, false);
1964 if (ret) {
1965 WRITE_ONCE(sk->sk_err, -ret);
1966 sk_error_report(sk);
1967 }
1968 }
1969
1970 scm_recv(sock, msg, &scm, flags);
1971out:
1972 netlink_rcv_wake(sk);
1973 return err ? : copied;
1974}
1975
1976static void netlink_data_ready(struct sock *sk)
1977{
1978 BUG();
1979}
1980
1981/*
1982 * We export these functions to other modules. They provide a
1983 * complete set of kernel non-blocking support for message
1984 * queueing.
1985 */
1986
1987struct sock *
1988__netlink_kernel_create(struct net *net, int unit, struct module *module,
1989 struct netlink_kernel_cfg *cfg)
1990{
1991 struct socket *sock;
1992 struct sock *sk;
1993 struct netlink_sock *nlk;
1994 struct listeners *listeners = NULL;
1995 unsigned int groups;
1996
1997 BUG_ON(!nl_table);
1998
1999 if (unit < 0 || unit >= MAX_LINKS)
2000 return NULL;
2001
2002 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2003 return NULL;
2004
2005 if (__netlink_create(net, sock, unit, 1) < 0)
2006 goto out_sock_release_nosk;
2007
2008 sk = sock->sk;
2009
2010 if (!cfg || cfg->groups < 32)
2011 groups = 32;
2012 else
2013 groups = cfg->groups;
2014
2015 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2016 if (!listeners)
2017 goto out_sock_release;
2018
2019 sk->sk_data_ready = netlink_data_ready;
2020 if (cfg && cfg->input)
2021 nlk_sk(sk)->netlink_rcv = cfg->input;
2022
2023 if (netlink_insert(sk, 0))
2024 goto out_sock_release;
2025
2026 nlk = nlk_sk(sk);
2027 set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2028
2029 netlink_table_grab();
2030 if (!nl_table[unit].registered) {
2031 nl_table[unit].groups = groups;
2032 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2033 nl_table[unit].module = module;
2034 if (cfg) {
2035 nl_table[unit].bind = cfg->bind;
2036 nl_table[unit].unbind = cfg->unbind;
2037 nl_table[unit].release = cfg->release;
2038 nl_table[unit].flags = cfg->flags;
2039 }
2040 nl_table[unit].registered = 1;
2041 } else {
2042 kfree(listeners);
2043 nl_table[unit].registered++;
2044 }
2045 netlink_table_ungrab();
2046 return sk;
2047
2048out_sock_release:
2049 kfree(listeners);
2050 netlink_kernel_release(sk);
2051 return NULL;
2052
2053out_sock_release_nosk:
2054 sock_release(sock);
2055 return NULL;
2056}
2057EXPORT_SYMBOL(__netlink_kernel_create);
2058
2059void
2060netlink_kernel_release(struct sock *sk)
2061{
2062 if (sk == NULL || sk->sk_socket == NULL)
2063 return;
2064
2065 sock_release(sk->sk_socket);
2066}
2067EXPORT_SYMBOL(netlink_kernel_release);
2068
2069int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2070{
2071 struct listeners *new, *old;
2072 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2073
2074 if (groups < 32)
2075 groups = 32;
2076
2077 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2078 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2079 if (!new)
2080 return -ENOMEM;
2081 old = nl_deref_protected(tbl->listeners);
2082 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2083 rcu_assign_pointer(tbl->listeners, new);
2084
2085 kfree_rcu(old, rcu);
2086 }
2087 tbl->groups = groups;
2088
2089 return 0;
2090}
2091
2092/**
2093 * netlink_change_ngroups - change number of multicast groups
2094 *
2095 * This changes the number of multicast groups that are available
2096 * on a certain netlink family. Note that it is not possible to
2097 * change the number of groups to below 32. Also note that it does
2098 * not implicitly call netlink_clear_multicast_users() when the
2099 * number of groups is reduced.
2100 *
2101 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2102 * @groups: The new number of groups.
2103 */
2104int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2105{
2106 int err;
2107
2108 netlink_table_grab();
2109 err = __netlink_change_ngroups(sk, groups);
2110 netlink_table_ungrab();
2111
2112 return err;
2113}
2114
2115void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2116{
2117 struct sock *sk;
2118 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2119 struct hlist_node *tmp;
2120
2121 sk_for_each_bound_safe(sk, tmp, &tbl->mc_list)
2122 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2123}
2124
2125struct nlmsghdr *
2126__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2127{
2128 struct nlmsghdr *nlh;
2129 int size = nlmsg_msg_size(len);
2130
2131 nlh = skb_put(skb, NLMSG_ALIGN(size));
2132 nlh->nlmsg_type = type;
2133 nlh->nlmsg_len = size;
2134 nlh->nlmsg_flags = flags;
2135 nlh->nlmsg_pid = portid;
2136 nlh->nlmsg_seq = seq;
2137 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2138 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2139 return nlh;
2140}
2141EXPORT_SYMBOL(__nlmsg_put);
2142
2143static size_t
2144netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2145 const struct netlink_ext_ack *extack)
2146{
2147 size_t tlvlen;
2148
2149 if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2150 return 0;
2151
2152 tlvlen = 0;
2153 if (extack->_msg)
2154 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2155 if (extack->cookie_len)
2156 tlvlen += nla_total_size(extack->cookie_len);
2157
2158 /* Following attributes are only reported as error (not warning) */
2159 if (!err)
2160 return tlvlen;
2161
2162 if (extack->bad_attr)
2163 tlvlen += nla_total_size(sizeof(u32));
2164 if (extack->policy)
2165 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2166 if (extack->miss_type)
2167 tlvlen += nla_total_size(sizeof(u32));
2168 if (extack->miss_nest)
2169 tlvlen += nla_total_size(sizeof(u32));
2170
2171 return tlvlen;
2172}
2173
2174static bool nlmsg_check_in_payload(const struct nlmsghdr *nlh, const void *addr)
2175{
2176 return !WARN_ON(addr < nlmsg_data(nlh) ||
2177 addr - (const void *) nlh >= nlh->nlmsg_len);
2178}
2179
2180static void
2181netlink_ack_tlv_fill(struct sk_buff *skb, const struct nlmsghdr *nlh, int err,
2182 const struct netlink_ext_ack *extack)
2183{
2184 if (extack->_msg)
2185 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2186 if (extack->cookie_len)
2187 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2188 extack->cookie_len, extack->cookie));
2189
2190 if (!err)
2191 return;
2192
2193 if (extack->bad_attr && nlmsg_check_in_payload(nlh, extack->bad_attr))
2194 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2195 (u8 *)extack->bad_attr - (const u8 *)nlh));
2196 if (extack->policy)
2197 netlink_policy_dump_write_attr(skb, extack->policy,
2198 NLMSGERR_ATTR_POLICY);
2199 if (extack->miss_type)
2200 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2201 extack->miss_type));
2202 if (extack->miss_nest && nlmsg_check_in_payload(nlh, extack->miss_nest))
2203 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2204 (u8 *)extack->miss_nest - (const u8 *)nlh));
2205}
2206
2207/*
2208 * It looks a bit ugly.
2209 * It would be better to create kernel thread.
2210 */
2211
2212static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2213 struct netlink_callback *cb,
2214 struct netlink_ext_ack *extack)
2215{
2216 struct nlmsghdr *nlh;
2217 size_t extack_len;
2218
2219 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2220 NLM_F_MULTI | cb->answer_flags);
2221 if (WARN_ON(!nlh))
2222 return -ENOBUFS;
2223
2224 nl_dump_check_consistent(cb, nlh);
2225 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2226
2227 extack_len = netlink_ack_tlv_len(nlk, nlk->dump_done_errno, extack);
2228 if (extack_len) {
2229 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2230 if (skb_tailroom(skb) >= extack_len) {
2231 netlink_ack_tlv_fill(skb, cb->nlh,
2232 nlk->dump_done_errno, extack);
2233 nlmsg_end(skb, nlh);
2234 }
2235 }
2236
2237 return 0;
2238}
2239
2240static int netlink_dump(struct sock *sk, bool lock_taken)
2241{
2242 struct netlink_sock *nlk = nlk_sk(sk);
2243 struct netlink_ext_ack extack = {};
2244 struct netlink_callback *cb;
2245 struct sk_buff *skb = NULL;
2246 size_t max_recvmsg_len;
2247 struct module *module;
2248 int err = -ENOBUFS;
2249 int alloc_min_size;
2250 int alloc_size;
2251
2252 if (!lock_taken)
2253 mutex_lock(&nlk->nl_cb_mutex);
2254 if (!nlk->cb_running) {
2255 err = -EINVAL;
2256 goto errout_skb;
2257 }
2258
2259 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2260 goto errout_skb;
2261
2262 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2263 * required, but it makes sense to _attempt_ a 32KiB allocation
2264 * to reduce number of system calls on dump operations, if user
2265 * ever provided a big enough buffer.
2266 */
2267 cb = &nlk->cb;
2268 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2269
2270 max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2271 if (alloc_min_size < max_recvmsg_len) {
2272 alloc_size = max_recvmsg_len;
2273 skb = alloc_skb(alloc_size,
2274 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2275 __GFP_NOWARN | __GFP_NORETRY);
2276 }
2277 if (!skb) {
2278 alloc_size = alloc_min_size;
2279 skb = alloc_skb(alloc_size, GFP_KERNEL);
2280 }
2281 if (!skb)
2282 goto errout_skb;
2283
2284 /* Trim skb to allocated size. User is expected to provide buffer as
2285 * large as max(min_dump_alloc, 32KiB (max_recvmsg_len capped at
2286 * netlink_recvmsg())). dump will pack as many smaller messages as
2287 * could fit within the allocated skb. skb is typically allocated
2288 * with larger space than required (could be as much as near 2x the
2289 * requested size with align to next power of 2 approach). Allowing
2290 * dump to use the excess space makes it difficult for a user to have a
2291 * reasonable static buffer based on the expected largest dump of a
2292 * single netdev. The outcome is MSG_TRUNC error.
2293 */
2294 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2295
2296 /* Make sure malicious BPF programs can not read unitialized memory
2297 * from skb->head -> skb->data
2298 */
2299 skb_reset_network_header(skb);
2300 skb_reset_mac_header(skb);
2301
2302 netlink_skb_set_owner_r(skb, sk);
2303
2304 if (nlk->dump_done_errno > 0) {
2305 cb->extack = &extack;
2306
2307 nlk->dump_done_errno = cb->dump(skb, cb);
2308
2309 /* EMSGSIZE plus something already in the skb means
2310 * that there's more to dump but current skb has filled up.
2311 * If the callback really wants to return EMSGSIZE to user space
2312 * it needs to do so again, on the next cb->dump() call,
2313 * without putting data in the skb.
2314 */
2315 if (nlk->dump_done_errno == -EMSGSIZE && skb->len)
2316 nlk->dump_done_errno = skb->len;
2317
2318 cb->extack = NULL;
2319 }
2320
2321 if (nlk->dump_done_errno > 0 ||
2322 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2323 mutex_unlock(&nlk->nl_cb_mutex);
2324
2325 if (sk_filter(sk, skb))
2326 kfree_skb(skb);
2327 else
2328 __netlink_sendskb(sk, skb);
2329 return 0;
2330 }
2331
2332 if (netlink_dump_done(nlk, skb, cb, &extack))
2333 goto errout_skb;
2334
2335#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2336 /* frag_list skb's data is used for compat tasks
2337 * and the regular skb's data for normal (non-compat) tasks.
2338 * See netlink_recvmsg().
2339 */
2340 if (unlikely(skb_shinfo(skb)->frag_list)) {
2341 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2342 goto errout_skb;
2343 }
2344#endif
2345
2346 if (sk_filter(sk, skb))
2347 kfree_skb(skb);
2348 else
2349 __netlink_sendskb(sk, skb);
2350
2351 if (cb->done)
2352 cb->done(cb);
2353
2354 WRITE_ONCE(nlk->cb_running, false);
2355 module = cb->module;
2356 skb = cb->skb;
2357 mutex_unlock(&nlk->nl_cb_mutex);
2358 module_put(module);
2359 consume_skb(skb);
2360 return 0;
2361
2362errout_skb:
2363 mutex_unlock(&nlk->nl_cb_mutex);
2364 kfree_skb(skb);
2365 return err;
2366}
2367
2368int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2369 const struct nlmsghdr *nlh,
2370 struct netlink_dump_control *control)
2371{
2372 struct netlink_callback *cb;
2373 struct netlink_sock *nlk;
2374 struct sock *sk;
2375 int ret;
2376
2377 refcount_inc(&skb->users);
2378
2379 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2380 if (sk == NULL) {
2381 ret = -ECONNREFUSED;
2382 goto error_free;
2383 }
2384
2385 nlk = nlk_sk(sk);
2386 mutex_lock(&nlk->nl_cb_mutex);
2387 /* A dump is in progress... */
2388 if (nlk->cb_running) {
2389 ret = -EBUSY;
2390 goto error_unlock;
2391 }
2392 /* add reference of module which cb->dump belongs to */
2393 if (!try_module_get(control->module)) {
2394 ret = -EPROTONOSUPPORT;
2395 goto error_unlock;
2396 }
2397
2398 cb = &nlk->cb;
2399 memset(cb, 0, sizeof(*cb));
2400 cb->dump = control->dump;
2401 cb->done = control->done;
2402 cb->nlh = nlh;
2403 cb->data = control->data;
2404 cb->module = control->module;
2405 cb->min_dump_alloc = control->min_dump_alloc;
2406 cb->flags = control->flags;
2407 cb->skb = skb;
2408
2409 cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2410
2411 if (control->start) {
2412 cb->extack = control->extack;
2413 ret = control->start(cb);
2414 cb->extack = NULL;
2415 if (ret)
2416 goto error_put;
2417 }
2418
2419 WRITE_ONCE(nlk->cb_running, true);
2420 nlk->dump_done_errno = INT_MAX;
2421
2422 ret = netlink_dump(sk, true);
2423
2424 sock_put(sk);
2425
2426 if (ret)
2427 return ret;
2428
2429 /* We successfully started a dump, by returning -EINTR we
2430 * signal not to send ACK even if it was requested.
2431 */
2432 return -EINTR;
2433
2434error_put:
2435 module_put(control->module);
2436error_unlock:
2437 sock_put(sk);
2438 mutex_unlock(&nlk->nl_cb_mutex);
2439error_free:
2440 kfree_skb(skb);
2441 return ret;
2442}
2443EXPORT_SYMBOL(__netlink_dump_start);
2444
2445void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2446 const struct netlink_ext_ack *extack)
2447{
2448 struct sk_buff *skb;
2449 struct nlmsghdr *rep;
2450 struct nlmsgerr *errmsg;
2451 size_t payload = sizeof(*errmsg);
2452 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2453 unsigned int flags = 0;
2454 size_t tlvlen;
2455
2456 /* Error messages get the original request appened, unless the user
2457 * requests to cap the error message, and get extra error data if
2458 * requested.
2459 */
2460 if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2461 payload += nlmsg_len(nlh);
2462 else
2463 flags |= NLM_F_CAPPED;
2464
2465 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2466 if (tlvlen)
2467 flags |= NLM_F_ACK_TLVS;
2468
2469 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2470 if (!skb)
2471 goto err_skb;
2472
2473 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2474 NLMSG_ERROR, sizeof(*errmsg), flags);
2475 if (!rep)
2476 goto err_bad_put;
2477 errmsg = nlmsg_data(rep);
2478 errmsg->error = err;
2479 errmsg->msg = *nlh;
2480
2481 if (!(flags & NLM_F_CAPPED)) {
2482 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2483 goto err_bad_put;
2484
2485 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2486 nlmsg_len(nlh));
2487 }
2488
2489 if (tlvlen)
2490 netlink_ack_tlv_fill(skb, nlh, err, extack);
2491
2492 nlmsg_end(skb, rep);
2493
2494 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2495
2496 return;
2497
2498err_bad_put:
2499 nlmsg_free(skb);
2500err_skb:
2501 WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2502 sk_error_report(NETLINK_CB(in_skb).sk);
2503}
2504EXPORT_SYMBOL(netlink_ack);
2505
2506int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2507 struct nlmsghdr *,
2508 struct netlink_ext_ack *))
2509{
2510 struct netlink_ext_ack extack;
2511 struct nlmsghdr *nlh;
2512 int err;
2513
2514 while (skb->len >= nlmsg_total_size(0)) {
2515 int msglen;
2516
2517 memset(&extack, 0, sizeof(extack));
2518 nlh = nlmsg_hdr(skb);
2519 err = 0;
2520
2521 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2522 return 0;
2523
2524 /* Only requests are handled by the kernel */
2525 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2526 goto ack;
2527
2528 /* Skip control messages */
2529 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2530 goto ack;
2531
2532 err = cb(skb, nlh, &extack);
2533 if (err == -EINTR)
2534 goto skip;
2535
2536ack:
2537 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2538 netlink_ack(skb, nlh, err, &extack);
2539
2540skip:
2541 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2542 if (msglen > skb->len)
2543 msglen = skb->len;
2544 skb_pull(skb, msglen);
2545 }
2546
2547 return 0;
2548}
2549EXPORT_SYMBOL(netlink_rcv_skb);
2550
2551/**
2552 * nlmsg_notify - send a notification netlink message
2553 * @sk: netlink socket to use
2554 * @skb: notification message
2555 * @portid: destination netlink portid for reports or 0
2556 * @group: destination multicast group or 0
2557 * @report: 1 to report back, 0 to disable
2558 * @flags: allocation flags
2559 */
2560int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2561 unsigned int group, int report, gfp_t flags)
2562{
2563 int err = 0;
2564
2565 if (group) {
2566 int exclude_portid = 0;
2567
2568 if (report) {
2569 refcount_inc(&skb->users);
2570 exclude_portid = portid;
2571 }
2572
2573 /* errors reported via destination sk->sk_err, but propagate
2574 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2575 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2576 if (err == -ESRCH)
2577 err = 0;
2578 }
2579
2580 if (report) {
2581 int err2;
2582
2583 err2 = nlmsg_unicast(sk, skb, portid);
2584 if (!err)
2585 err = err2;
2586 }
2587
2588 return err;
2589}
2590EXPORT_SYMBOL(nlmsg_notify);
2591
2592#ifdef CONFIG_PROC_FS
2593struct nl_seq_iter {
2594 struct seq_net_private p;
2595 struct rhashtable_iter hti;
2596 int link;
2597};
2598
2599static void netlink_walk_start(struct nl_seq_iter *iter)
2600{
2601 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2602 rhashtable_walk_start(&iter->hti);
2603}
2604
2605static void netlink_walk_stop(struct nl_seq_iter *iter)
2606{
2607 rhashtable_walk_stop(&iter->hti);
2608 rhashtable_walk_exit(&iter->hti);
2609}
2610
2611static void *__netlink_seq_next(struct seq_file *seq)
2612{
2613 struct nl_seq_iter *iter = seq->private;
2614 struct netlink_sock *nlk;
2615
2616 do {
2617 for (;;) {
2618 nlk = rhashtable_walk_next(&iter->hti);
2619
2620 if (IS_ERR(nlk)) {
2621 if (PTR_ERR(nlk) == -EAGAIN)
2622 continue;
2623
2624 return nlk;
2625 }
2626
2627 if (nlk)
2628 break;
2629
2630 netlink_walk_stop(iter);
2631 if (++iter->link >= MAX_LINKS)
2632 return NULL;
2633
2634 netlink_walk_start(iter);
2635 }
2636 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2637
2638 return nlk;
2639}
2640
2641static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2642 __acquires(RCU)
2643{
2644 struct nl_seq_iter *iter = seq->private;
2645 void *obj = SEQ_START_TOKEN;
2646 loff_t pos;
2647
2648 iter->link = 0;
2649
2650 netlink_walk_start(iter);
2651
2652 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2653 obj = __netlink_seq_next(seq);
2654
2655 return obj;
2656}
2657
2658static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2659{
2660 ++*pos;
2661 return __netlink_seq_next(seq);
2662}
2663
2664static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2665{
2666 struct nl_seq_iter *iter = seq->private;
2667
2668 if (iter->link >= MAX_LINKS)
2669 return;
2670
2671 netlink_walk_stop(iter);
2672}
2673
2674
2675static int netlink_native_seq_show(struct seq_file *seq, void *v)
2676{
2677 if (v == SEQ_START_TOKEN) {
2678 seq_puts(seq,
2679 "sk Eth Pid Groups "
2680 "Rmem Wmem Dump Locks Drops Inode\n");
2681 } else {
2682 struct sock *s = v;
2683 struct netlink_sock *nlk = nlk_sk(s);
2684
2685 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2686 s,
2687 s->sk_protocol,
2688 nlk->portid,
2689 nlk->groups ? (u32)nlk->groups[0] : 0,
2690 sk_rmem_alloc_get(s),
2691 sk_wmem_alloc_get(s),
2692 READ_ONCE(nlk->cb_running),
2693 refcount_read(&s->sk_refcnt),
2694 atomic_read(&s->sk_drops),
2695 sock_i_ino(s)
2696 );
2697
2698 }
2699 return 0;
2700}
2701
2702#ifdef CONFIG_BPF_SYSCALL
2703struct bpf_iter__netlink {
2704 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2705 __bpf_md_ptr(struct netlink_sock *, sk);
2706};
2707
2708DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2709
2710static int netlink_prog_seq_show(struct bpf_prog *prog,
2711 struct bpf_iter_meta *meta,
2712 void *v)
2713{
2714 struct bpf_iter__netlink ctx;
2715
2716 meta->seq_num--; /* skip SEQ_START_TOKEN */
2717 ctx.meta = meta;
2718 ctx.sk = nlk_sk((struct sock *)v);
2719 return bpf_iter_run_prog(prog, &ctx);
2720}
2721
2722static int netlink_seq_show(struct seq_file *seq, void *v)
2723{
2724 struct bpf_iter_meta meta;
2725 struct bpf_prog *prog;
2726
2727 meta.seq = seq;
2728 prog = bpf_iter_get_info(&meta, false);
2729 if (!prog)
2730 return netlink_native_seq_show(seq, v);
2731
2732 if (v != SEQ_START_TOKEN)
2733 return netlink_prog_seq_show(prog, &meta, v);
2734
2735 return 0;
2736}
2737
2738static void netlink_seq_stop(struct seq_file *seq, void *v)
2739{
2740 struct bpf_iter_meta meta;
2741 struct bpf_prog *prog;
2742
2743 if (!v) {
2744 meta.seq = seq;
2745 prog = bpf_iter_get_info(&meta, true);
2746 if (prog)
2747 (void)netlink_prog_seq_show(prog, &meta, v);
2748 }
2749
2750 netlink_native_seq_stop(seq, v);
2751}
2752#else
2753static int netlink_seq_show(struct seq_file *seq, void *v)
2754{
2755 return netlink_native_seq_show(seq, v);
2756}
2757
2758static void netlink_seq_stop(struct seq_file *seq, void *v)
2759{
2760 netlink_native_seq_stop(seq, v);
2761}
2762#endif
2763
2764static const struct seq_operations netlink_seq_ops = {
2765 .start = netlink_seq_start,
2766 .next = netlink_seq_next,
2767 .stop = netlink_seq_stop,
2768 .show = netlink_seq_show,
2769};
2770#endif
2771
2772int netlink_register_notifier(struct notifier_block *nb)
2773{
2774 return blocking_notifier_chain_register(&netlink_chain, nb);
2775}
2776EXPORT_SYMBOL(netlink_register_notifier);
2777
2778int netlink_unregister_notifier(struct notifier_block *nb)
2779{
2780 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2781}
2782EXPORT_SYMBOL(netlink_unregister_notifier);
2783
2784static const struct proto_ops netlink_ops = {
2785 .family = PF_NETLINK,
2786 .owner = THIS_MODULE,
2787 .release = netlink_release,
2788 .bind = netlink_bind,
2789 .connect = netlink_connect,
2790 .socketpair = sock_no_socketpair,
2791 .accept = sock_no_accept,
2792 .getname = netlink_getname,
2793 .poll = datagram_poll,
2794 .ioctl = netlink_ioctl,
2795 .listen = sock_no_listen,
2796 .shutdown = sock_no_shutdown,
2797 .setsockopt = netlink_setsockopt,
2798 .getsockopt = netlink_getsockopt,
2799 .sendmsg = netlink_sendmsg,
2800 .recvmsg = netlink_recvmsg,
2801 .mmap = sock_no_mmap,
2802};
2803
2804static const struct net_proto_family netlink_family_ops = {
2805 .family = PF_NETLINK,
2806 .create = netlink_create,
2807 .owner = THIS_MODULE, /* for consistency 8) */
2808};
2809
2810static int __net_init netlink_net_init(struct net *net)
2811{
2812#ifdef CONFIG_PROC_FS
2813 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2814 sizeof(struct nl_seq_iter)))
2815 return -ENOMEM;
2816#endif
2817 return 0;
2818}
2819
2820static void __net_exit netlink_net_exit(struct net *net)
2821{
2822#ifdef CONFIG_PROC_FS
2823 remove_proc_entry("netlink", net->proc_net);
2824#endif
2825}
2826
2827static void __init netlink_add_usersock_entry(void)
2828{
2829 struct listeners *listeners;
2830 int groups = 32;
2831
2832 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2833 if (!listeners)
2834 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2835
2836 netlink_table_grab();
2837
2838 nl_table[NETLINK_USERSOCK].groups = groups;
2839 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2840 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2841 nl_table[NETLINK_USERSOCK].registered = 1;
2842 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2843
2844 netlink_table_ungrab();
2845}
2846
2847static struct pernet_operations __net_initdata netlink_net_ops = {
2848 .init = netlink_net_init,
2849 .exit = netlink_net_exit,
2850};
2851
2852static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2853{
2854 const struct netlink_sock *nlk = data;
2855 struct netlink_compare_arg arg;
2856
2857 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2858 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2859}
2860
2861static const struct rhashtable_params netlink_rhashtable_params = {
2862 .head_offset = offsetof(struct netlink_sock, node),
2863 .key_len = netlink_compare_arg_len,
2864 .obj_hashfn = netlink_hash,
2865 .obj_cmpfn = netlink_compare,
2866 .automatic_shrinking = true,
2867};
2868
2869#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2870BTF_ID_LIST(btf_netlink_sock_id)
2871BTF_ID(struct, netlink_sock)
2872
2873static const struct bpf_iter_seq_info netlink_seq_info = {
2874 .seq_ops = &netlink_seq_ops,
2875 .init_seq_private = bpf_iter_init_seq_net,
2876 .fini_seq_private = bpf_iter_fini_seq_net,
2877 .seq_priv_size = sizeof(struct nl_seq_iter),
2878};
2879
2880static struct bpf_iter_reg netlink_reg_info = {
2881 .target = "netlink",
2882 .ctx_arg_info_size = 1,
2883 .ctx_arg_info = {
2884 { offsetof(struct bpf_iter__netlink, sk),
2885 PTR_TO_BTF_ID_OR_NULL },
2886 },
2887 .seq_info = &netlink_seq_info,
2888};
2889
2890static int __init bpf_iter_register(void)
2891{
2892 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2893 return bpf_iter_reg_target(&netlink_reg_info);
2894}
2895#endif
2896
2897static int __init netlink_proto_init(void)
2898{
2899 int i;
2900 int err = proto_register(&netlink_proto, 0);
2901
2902 if (err != 0)
2903 goto out;
2904
2905#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2906 err = bpf_iter_register();
2907 if (err)
2908 goto out;
2909#endif
2910
2911 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2912
2913 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2914 if (!nl_table)
2915 goto panic;
2916
2917 for (i = 0; i < MAX_LINKS; i++) {
2918 if (rhashtable_init(&nl_table[i].hash,
2919 &netlink_rhashtable_params) < 0)
2920 goto panic;
2921 }
2922
2923 netlink_add_usersock_entry();
2924
2925 sock_register(&netlink_family_ops);
2926 register_pernet_subsys(&netlink_net_ops);
2927 register_pernet_subsys(&netlink_tap_net_ops);
2928 /* The netlink device handler may be needed early. */
2929 rtnetlink_init();
2930out:
2931 return err;
2932panic:
2933 panic("netlink_init: Cannot allocate nl_table\n");
2934}
2935
2936core_initcall(netlink_proto_init);
1/*
2 * NETLINK Kernel-user communication protocol.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
23 */
24
25#include <linux/module.h>
26
27#include <linux/capability.h>
28#include <linux/kernel.h>
29#include <linux/init.h>
30#include <linux/signal.h>
31#include <linux/sched.h>
32#include <linux/errno.h>
33#include <linux/string.h>
34#include <linux/stat.h>
35#include <linux/socket.h>
36#include <linux/un.h>
37#include <linux/fcntl.h>
38#include <linux/termios.h>
39#include <linux/sockios.h>
40#include <linux/net.h>
41#include <linux/fs.h>
42#include <linux/slab.h>
43#include <asm/uaccess.h>
44#include <linux/skbuff.h>
45#include <linux/netdevice.h>
46#include <linux/rtnetlink.h>
47#include <linux/proc_fs.h>
48#include <linux/seq_file.h>
49#include <linux/notifier.h>
50#include <linux/security.h>
51#include <linux/jhash.h>
52#include <linux/jiffies.h>
53#include <linux/random.h>
54#include <linux/bitops.h>
55#include <linux/mm.h>
56#include <linux/types.h>
57#include <linux/audit.h>
58#include <linux/mutex.h>
59#include <linux/vmalloc.h>
60#include <linux/if_arp.h>
61#include <asm/cacheflush.h>
62
63#include <net/net_namespace.h>
64#include <net/sock.h>
65#include <net/scm.h>
66#include <net/netlink.h>
67
68#include "af_netlink.h"
69
70struct listeners {
71 struct rcu_head rcu;
72 unsigned long masks[0];
73};
74
75/* state bits */
76#define NETLINK_CONGESTED 0x0
77
78/* flags */
79#define NETLINK_KERNEL_SOCKET 0x1
80#define NETLINK_RECV_PKTINFO 0x2
81#define NETLINK_BROADCAST_SEND_ERROR 0x4
82#define NETLINK_RECV_NO_ENOBUFS 0x8
83
84static inline int netlink_is_kernel(struct sock *sk)
85{
86 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
87}
88
89struct netlink_table *nl_table;
90EXPORT_SYMBOL_GPL(nl_table);
91
92static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
93
94static int netlink_dump(struct sock *sk);
95static void netlink_skb_destructor(struct sk_buff *skb);
96
97DEFINE_RWLOCK(nl_table_lock);
98EXPORT_SYMBOL_GPL(nl_table_lock);
99static atomic_t nl_table_users = ATOMIC_INIT(0);
100
101#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
102
103static ATOMIC_NOTIFIER_HEAD(netlink_chain);
104
105static DEFINE_SPINLOCK(netlink_tap_lock);
106static struct list_head netlink_tap_all __read_mostly;
107
108static inline u32 netlink_group_mask(u32 group)
109{
110 return group ? 1 << (group - 1) : 0;
111}
112
113static inline struct hlist_head *nl_portid_hashfn(struct nl_portid_hash *hash, u32 portid)
114{
115 return &hash->table[jhash_1word(portid, hash->rnd) & hash->mask];
116}
117
118int netlink_add_tap(struct netlink_tap *nt)
119{
120 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
121 return -EINVAL;
122
123 spin_lock(&netlink_tap_lock);
124 list_add_rcu(&nt->list, &netlink_tap_all);
125 spin_unlock(&netlink_tap_lock);
126
127 if (nt->module)
128 __module_get(nt->module);
129
130 return 0;
131}
132EXPORT_SYMBOL_GPL(netlink_add_tap);
133
134static int __netlink_remove_tap(struct netlink_tap *nt)
135{
136 bool found = false;
137 struct netlink_tap *tmp;
138
139 spin_lock(&netlink_tap_lock);
140
141 list_for_each_entry(tmp, &netlink_tap_all, list) {
142 if (nt == tmp) {
143 list_del_rcu(&nt->list);
144 found = true;
145 goto out;
146 }
147 }
148
149 pr_warn("__netlink_remove_tap: %p not found\n", nt);
150out:
151 spin_unlock(&netlink_tap_lock);
152
153 if (found && nt->module)
154 module_put(nt->module);
155
156 return found ? 0 : -ENODEV;
157}
158
159int netlink_remove_tap(struct netlink_tap *nt)
160{
161 int ret;
162
163 ret = __netlink_remove_tap(nt);
164 synchronize_net();
165
166 return ret;
167}
168EXPORT_SYMBOL_GPL(netlink_remove_tap);
169
170static bool netlink_filter_tap(const struct sk_buff *skb)
171{
172 struct sock *sk = skb->sk;
173 bool pass = false;
174
175 /* We take the more conservative approach and
176 * whitelist socket protocols that may pass.
177 */
178 switch (sk->sk_protocol) {
179 case NETLINK_ROUTE:
180 case NETLINK_USERSOCK:
181 case NETLINK_SOCK_DIAG:
182 case NETLINK_NFLOG:
183 case NETLINK_XFRM:
184 case NETLINK_FIB_LOOKUP:
185 case NETLINK_NETFILTER:
186 case NETLINK_GENERIC:
187 pass = true;
188 break;
189 }
190
191 return pass;
192}
193
194static int __netlink_deliver_tap_skb(struct sk_buff *skb,
195 struct net_device *dev)
196{
197 struct sk_buff *nskb;
198 struct sock *sk = skb->sk;
199 int ret = -ENOMEM;
200
201 dev_hold(dev);
202 nskb = skb_clone(skb, GFP_ATOMIC);
203 if (nskb) {
204 nskb->dev = dev;
205 nskb->protocol = htons((u16) sk->sk_protocol);
206 nskb->pkt_type = netlink_is_kernel(sk) ?
207 PACKET_KERNEL : PACKET_USER;
208
209 ret = dev_queue_xmit(nskb);
210 if (unlikely(ret > 0))
211 ret = net_xmit_errno(ret);
212 }
213
214 dev_put(dev);
215 return ret;
216}
217
218static void __netlink_deliver_tap(struct sk_buff *skb)
219{
220 int ret;
221 struct netlink_tap *tmp;
222
223 if (!netlink_filter_tap(skb))
224 return;
225
226 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
227 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
228 if (unlikely(ret))
229 break;
230 }
231}
232
233static void netlink_deliver_tap(struct sk_buff *skb)
234{
235 rcu_read_lock();
236
237 if (unlikely(!list_empty(&netlink_tap_all)))
238 __netlink_deliver_tap(skb);
239
240 rcu_read_unlock();
241}
242
243static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
244 struct sk_buff *skb)
245{
246 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
247 netlink_deliver_tap(skb);
248}
249
250static void netlink_overrun(struct sock *sk)
251{
252 struct netlink_sock *nlk = nlk_sk(sk);
253
254 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
255 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
256 sk->sk_err = ENOBUFS;
257 sk->sk_error_report(sk);
258 }
259 }
260 atomic_inc(&sk->sk_drops);
261}
262
263static void netlink_rcv_wake(struct sock *sk)
264{
265 struct netlink_sock *nlk = nlk_sk(sk);
266
267 if (skb_queue_empty(&sk->sk_receive_queue))
268 clear_bit(NETLINK_CONGESTED, &nlk->state);
269 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
270 wake_up_interruptible(&nlk->wait);
271}
272
273#ifdef CONFIG_NETLINK_MMAP
274static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
275{
276 return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
277}
278
279static bool netlink_rx_is_mmaped(struct sock *sk)
280{
281 return nlk_sk(sk)->rx_ring.pg_vec != NULL;
282}
283
284static bool netlink_tx_is_mmaped(struct sock *sk)
285{
286 return nlk_sk(sk)->tx_ring.pg_vec != NULL;
287}
288
289static __pure struct page *pgvec_to_page(const void *addr)
290{
291 if (is_vmalloc_addr(addr))
292 return vmalloc_to_page(addr);
293 else
294 return virt_to_page(addr);
295}
296
297static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
298{
299 unsigned int i;
300
301 for (i = 0; i < len; i++) {
302 if (pg_vec[i] != NULL) {
303 if (is_vmalloc_addr(pg_vec[i]))
304 vfree(pg_vec[i]);
305 else
306 free_pages((unsigned long)pg_vec[i], order);
307 }
308 }
309 kfree(pg_vec);
310}
311
312static void *alloc_one_pg_vec_page(unsigned long order)
313{
314 void *buffer;
315 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
316 __GFP_NOWARN | __GFP_NORETRY;
317
318 buffer = (void *)__get_free_pages(gfp_flags, order);
319 if (buffer != NULL)
320 return buffer;
321
322 buffer = vzalloc((1 << order) * PAGE_SIZE);
323 if (buffer != NULL)
324 return buffer;
325
326 gfp_flags &= ~__GFP_NORETRY;
327 return (void *)__get_free_pages(gfp_flags, order);
328}
329
330static void **alloc_pg_vec(struct netlink_sock *nlk,
331 struct nl_mmap_req *req, unsigned int order)
332{
333 unsigned int block_nr = req->nm_block_nr;
334 unsigned int i;
335 void **pg_vec;
336
337 pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
338 if (pg_vec == NULL)
339 return NULL;
340
341 for (i = 0; i < block_nr; i++) {
342 pg_vec[i] = alloc_one_pg_vec_page(order);
343 if (pg_vec[i] == NULL)
344 goto err1;
345 }
346
347 return pg_vec;
348err1:
349 free_pg_vec(pg_vec, order, block_nr);
350 return NULL;
351}
352
353static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
354 bool closing, bool tx_ring)
355{
356 struct netlink_sock *nlk = nlk_sk(sk);
357 struct netlink_ring *ring;
358 struct sk_buff_head *queue;
359 void **pg_vec = NULL;
360 unsigned int order = 0;
361 int err;
362
363 ring = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
364 queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
365
366 if (!closing) {
367 if (atomic_read(&nlk->mapped))
368 return -EBUSY;
369 if (atomic_read(&ring->pending))
370 return -EBUSY;
371 }
372
373 if (req->nm_block_nr) {
374 if (ring->pg_vec != NULL)
375 return -EBUSY;
376
377 if ((int)req->nm_block_size <= 0)
378 return -EINVAL;
379 if (!IS_ALIGNED(req->nm_block_size, PAGE_SIZE))
380 return -EINVAL;
381 if (req->nm_frame_size < NL_MMAP_HDRLEN)
382 return -EINVAL;
383 if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
384 return -EINVAL;
385
386 ring->frames_per_block = req->nm_block_size /
387 req->nm_frame_size;
388 if (ring->frames_per_block == 0)
389 return -EINVAL;
390 if (ring->frames_per_block * req->nm_block_nr !=
391 req->nm_frame_nr)
392 return -EINVAL;
393
394 order = get_order(req->nm_block_size);
395 pg_vec = alloc_pg_vec(nlk, req, order);
396 if (pg_vec == NULL)
397 return -ENOMEM;
398 } else {
399 if (req->nm_frame_nr)
400 return -EINVAL;
401 }
402
403 err = -EBUSY;
404 mutex_lock(&nlk->pg_vec_lock);
405 if (closing || atomic_read(&nlk->mapped) == 0) {
406 err = 0;
407 spin_lock_bh(&queue->lock);
408
409 ring->frame_max = req->nm_frame_nr - 1;
410 ring->head = 0;
411 ring->frame_size = req->nm_frame_size;
412 ring->pg_vec_pages = req->nm_block_size / PAGE_SIZE;
413
414 swap(ring->pg_vec_len, req->nm_block_nr);
415 swap(ring->pg_vec_order, order);
416 swap(ring->pg_vec, pg_vec);
417
418 __skb_queue_purge(queue);
419 spin_unlock_bh(&queue->lock);
420
421 WARN_ON(atomic_read(&nlk->mapped));
422 }
423 mutex_unlock(&nlk->pg_vec_lock);
424
425 if (pg_vec)
426 free_pg_vec(pg_vec, order, req->nm_block_nr);
427 return err;
428}
429
430static void netlink_mm_open(struct vm_area_struct *vma)
431{
432 struct file *file = vma->vm_file;
433 struct socket *sock = file->private_data;
434 struct sock *sk = sock->sk;
435
436 if (sk)
437 atomic_inc(&nlk_sk(sk)->mapped);
438}
439
440static void netlink_mm_close(struct vm_area_struct *vma)
441{
442 struct file *file = vma->vm_file;
443 struct socket *sock = file->private_data;
444 struct sock *sk = sock->sk;
445
446 if (sk)
447 atomic_dec(&nlk_sk(sk)->mapped);
448}
449
450static const struct vm_operations_struct netlink_mmap_ops = {
451 .open = netlink_mm_open,
452 .close = netlink_mm_close,
453};
454
455static int netlink_mmap(struct file *file, struct socket *sock,
456 struct vm_area_struct *vma)
457{
458 struct sock *sk = sock->sk;
459 struct netlink_sock *nlk = nlk_sk(sk);
460 struct netlink_ring *ring;
461 unsigned long start, size, expected;
462 unsigned int i;
463 int err = -EINVAL;
464
465 if (vma->vm_pgoff)
466 return -EINVAL;
467
468 mutex_lock(&nlk->pg_vec_lock);
469
470 expected = 0;
471 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
472 if (ring->pg_vec == NULL)
473 continue;
474 expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
475 }
476
477 if (expected == 0)
478 goto out;
479
480 size = vma->vm_end - vma->vm_start;
481 if (size != expected)
482 goto out;
483
484 start = vma->vm_start;
485 for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
486 if (ring->pg_vec == NULL)
487 continue;
488
489 for (i = 0; i < ring->pg_vec_len; i++) {
490 struct page *page;
491 void *kaddr = ring->pg_vec[i];
492 unsigned int pg_num;
493
494 for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
495 page = pgvec_to_page(kaddr);
496 err = vm_insert_page(vma, start, page);
497 if (err < 0)
498 goto out;
499 start += PAGE_SIZE;
500 kaddr += PAGE_SIZE;
501 }
502 }
503 }
504
505 atomic_inc(&nlk->mapped);
506 vma->vm_ops = &netlink_mmap_ops;
507 err = 0;
508out:
509 mutex_unlock(&nlk->pg_vec_lock);
510 return err;
511}
512
513static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr)
514{
515#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
516 struct page *p_start, *p_end;
517
518 /* First page is flushed through netlink_{get,set}_status */
519 p_start = pgvec_to_page(hdr + PAGE_SIZE);
520 p_end = pgvec_to_page((void *)hdr + NL_MMAP_HDRLEN + hdr->nm_len - 1);
521 while (p_start <= p_end) {
522 flush_dcache_page(p_start);
523 p_start++;
524 }
525#endif
526}
527
528static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
529{
530 smp_rmb();
531 flush_dcache_page(pgvec_to_page(hdr));
532 return hdr->nm_status;
533}
534
535static void netlink_set_status(struct nl_mmap_hdr *hdr,
536 enum nl_mmap_status status)
537{
538 hdr->nm_status = status;
539 flush_dcache_page(pgvec_to_page(hdr));
540 smp_wmb();
541}
542
543static struct nl_mmap_hdr *
544__netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
545{
546 unsigned int pg_vec_pos, frame_off;
547
548 pg_vec_pos = pos / ring->frames_per_block;
549 frame_off = pos % ring->frames_per_block;
550
551 return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
552}
553
554static struct nl_mmap_hdr *
555netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
556 enum nl_mmap_status status)
557{
558 struct nl_mmap_hdr *hdr;
559
560 hdr = __netlink_lookup_frame(ring, pos);
561 if (netlink_get_status(hdr) != status)
562 return NULL;
563
564 return hdr;
565}
566
567static struct nl_mmap_hdr *
568netlink_current_frame(const struct netlink_ring *ring,
569 enum nl_mmap_status status)
570{
571 return netlink_lookup_frame(ring, ring->head, status);
572}
573
574static struct nl_mmap_hdr *
575netlink_previous_frame(const struct netlink_ring *ring,
576 enum nl_mmap_status status)
577{
578 unsigned int prev;
579
580 prev = ring->head ? ring->head - 1 : ring->frame_max;
581 return netlink_lookup_frame(ring, prev, status);
582}
583
584static void netlink_increment_head(struct netlink_ring *ring)
585{
586 ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
587}
588
589static void netlink_forward_ring(struct netlink_ring *ring)
590{
591 unsigned int head = ring->head, pos = head;
592 const struct nl_mmap_hdr *hdr;
593
594 do {
595 hdr = __netlink_lookup_frame(ring, pos);
596 if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
597 break;
598 if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
599 break;
600 netlink_increment_head(ring);
601 } while (ring->head != head);
602}
603
604static bool netlink_dump_space(struct netlink_sock *nlk)
605{
606 struct netlink_ring *ring = &nlk->rx_ring;
607 struct nl_mmap_hdr *hdr;
608 unsigned int n;
609
610 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
611 if (hdr == NULL)
612 return false;
613
614 n = ring->head + ring->frame_max / 2;
615 if (n > ring->frame_max)
616 n -= ring->frame_max;
617
618 hdr = __netlink_lookup_frame(ring, n);
619
620 return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
621}
622
623static unsigned int netlink_poll(struct file *file, struct socket *sock,
624 poll_table *wait)
625{
626 struct sock *sk = sock->sk;
627 struct netlink_sock *nlk = nlk_sk(sk);
628 unsigned int mask;
629 int err;
630
631 if (nlk->rx_ring.pg_vec != NULL) {
632 /* Memory mapped sockets don't call recvmsg(), so flow control
633 * for dumps is performed here. A dump is allowed to continue
634 * if at least half the ring is unused.
635 */
636 while (nlk->cb_running && netlink_dump_space(nlk)) {
637 err = netlink_dump(sk);
638 if (err < 0) {
639 sk->sk_err = err;
640 sk->sk_error_report(sk);
641 break;
642 }
643 }
644 netlink_rcv_wake(sk);
645 }
646
647 mask = datagram_poll(file, sock, wait);
648
649 spin_lock_bh(&sk->sk_receive_queue.lock);
650 if (nlk->rx_ring.pg_vec) {
651 netlink_forward_ring(&nlk->rx_ring);
652 if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
653 mask |= POLLIN | POLLRDNORM;
654 }
655 spin_unlock_bh(&sk->sk_receive_queue.lock);
656
657 spin_lock_bh(&sk->sk_write_queue.lock);
658 if (nlk->tx_ring.pg_vec) {
659 if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
660 mask |= POLLOUT | POLLWRNORM;
661 }
662 spin_unlock_bh(&sk->sk_write_queue.lock);
663
664 return mask;
665}
666
667static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
668{
669 return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
670}
671
672static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
673 struct netlink_ring *ring,
674 struct nl_mmap_hdr *hdr)
675{
676 unsigned int size;
677 void *data;
678
679 size = ring->frame_size - NL_MMAP_HDRLEN;
680 data = (void *)hdr + NL_MMAP_HDRLEN;
681
682 skb->head = data;
683 skb->data = data;
684 skb_reset_tail_pointer(skb);
685 skb->end = skb->tail + size;
686 skb->len = 0;
687
688 skb->destructor = netlink_skb_destructor;
689 NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
690 NETLINK_CB(skb).sk = sk;
691}
692
693static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
694 u32 dst_portid, u32 dst_group,
695 struct sock_iocb *siocb)
696{
697 struct netlink_sock *nlk = nlk_sk(sk);
698 struct netlink_ring *ring;
699 struct nl_mmap_hdr *hdr;
700 struct sk_buff *skb;
701 unsigned int maxlen;
702 bool excl = true;
703 int err = 0, len = 0;
704
705 /* Netlink messages are validated by the receiver before processing.
706 * In order to avoid userspace changing the contents of the message
707 * after validation, the socket and the ring may only be used by a
708 * single process, otherwise we fall back to copying.
709 */
710 if (atomic_long_read(&sk->sk_socket->file->f_count) > 2 ||
711 atomic_read(&nlk->mapped) > 1)
712 excl = false;
713
714 mutex_lock(&nlk->pg_vec_lock);
715
716 ring = &nlk->tx_ring;
717 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
718
719 do {
720 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
721 if (hdr == NULL) {
722 if (!(msg->msg_flags & MSG_DONTWAIT) &&
723 atomic_read(&nlk->tx_ring.pending))
724 schedule();
725 continue;
726 }
727 if (hdr->nm_len > maxlen) {
728 err = -EINVAL;
729 goto out;
730 }
731
732 netlink_frame_flush_dcache(hdr);
733
734 if (likely(dst_portid == 0 && dst_group == 0 && excl)) {
735 skb = alloc_skb_head(GFP_KERNEL);
736 if (skb == NULL) {
737 err = -ENOBUFS;
738 goto out;
739 }
740 sock_hold(sk);
741 netlink_ring_setup_skb(skb, sk, ring, hdr);
742 NETLINK_CB(skb).flags |= NETLINK_SKB_TX;
743 __skb_put(skb, hdr->nm_len);
744 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
745 atomic_inc(&ring->pending);
746 } else {
747 skb = alloc_skb(hdr->nm_len, GFP_KERNEL);
748 if (skb == NULL) {
749 err = -ENOBUFS;
750 goto out;
751 }
752 __skb_put(skb, hdr->nm_len);
753 memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, hdr->nm_len);
754 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
755 }
756
757 netlink_increment_head(ring);
758
759 NETLINK_CB(skb).portid = nlk->portid;
760 NETLINK_CB(skb).dst_group = dst_group;
761 NETLINK_CB(skb).creds = siocb->scm->creds;
762
763 err = security_netlink_send(sk, skb);
764 if (err) {
765 kfree_skb(skb);
766 goto out;
767 }
768
769 if (unlikely(dst_group)) {
770 atomic_inc(&skb->users);
771 netlink_broadcast(sk, skb, dst_portid, dst_group,
772 GFP_KERNEL);
773 }
774 err = netlink_unicast(sk, skb, dst_portid,
775 msg->msg_flags & MSG_DONTWAIT);
776 if (err < 0)
777 goto out;
778 len += err;
779
780 } while (hdr != NULL ||
781 (!(msg->msg_flags & MSG_DONTWAIT) &&
782 atomic_read(&nlk->tx_ring.pending)));
783
784 if (len > 0)
785 err = len;
786out:
787 mutex_unlock(&nlk->pg_vec_lock);
788 return err;
789}
790
791static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
792{
793 struct nl_mmap_hdr *hdr;
794
795 hdr = netlink_mmap_hdr(skb);
796 hdr->nm_len = skb->len;
797 hdr->nm_group = NETLINK_CB(skb).dst_group;
798 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
799 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
800 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
801 netlink_frame_flush_dcache(hdr);
802 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
803
804 NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
805 kfree_skb(skb);
806}
807
808static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
809{
810 struct netlink_sock *nlk = nlk_sk(sk);
811 struct netlink_ring *ring = &nlk->rx_ring;
812 struct nl_mmap_hdr *hdr;
813
814 spin_lock_bh(&sk->sk_receive_queue.lock);
815 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
816 if (hdr == NULL) {
817 spin_unlock_bh(&sk->sk_receive_queue.lock);
818 kfree_skb(skb);
819 netlink_overrun(sk);
820 return;
821 }
822 netlink_increment_head(ring);
823 __skb_queue_tail(&sk->sk_receive_queue, skb);
824 spin_unlock_bh(&sk->sk_receive_queue.lock);
825
826 hdr->nm_len = skb->len;
827 hdr->nm_group = NETLINK_CB(skb).dst_group;
828 hdr->nm_pid = NETLINK_CB(skb).creds.pid;
829 hdr->nm_uid = from_kuid(sk_user_ns(sk), NETLINK_CB(skb).creds.uid);
830 hdr->nm_gid = from_kgid(sk_user_ns(sk), NETLINK_CB(skb).creds.gid);
831 netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
832}
833
834#else /* CONFIG_NETLINK_MMAP */
835#define netlink_skb_is_mmaped(skb) false
836#define netlink_rx_is_mmaped(sk) false
837#define netlink_tx_is_mmaped(sk) false
838#define netlink_mmap sock_no_mmap
839#define netlink_poll datagram_poll
840#define netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group, siocb) 0
841#endif /* CONFIG_NETLINK_MMAP */
842
843static void netlink_skb_destructor(struct sk_buff *skb)
844{
845#ifdef CONFIG_NETLINK_MMAP
846 struct nl_mmap_hdr *hdr;
847 struct netlink_ring *ring;
848 struct sock *sk;
849
850 /* If a packet from the kernel to userspace was freed because of an
851 * error without being delivered to userspace, the kernel must reset
852 * the status. In the direction userspace to kernel, the status is
853 * always reset here after the packet was processed and freed.
854 */
855 if (netlink_skb_is_mmaped(skb)) {
856 hdr = netlink_mmap_hdr(skb);
857 sk = NETLINK_CB(skb).sk;
858
859 if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
860 netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
861 ring = &nlk_sk(sk)->tx_ring;
862 } else {
863 if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
864 hdr->nm_len = 0;
865 netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
866 }
867 ring = &nlk_sk(sk)->rx_ring;
868 }
869
870 WARN_ON(atomic_read(&ring->pending) == 0);
871 atomic_dec(&ring->pending);
872 sock_put(sk);
873
874 skb->head = NULL;
875 }
876#endif
877 if (is_vmalloc_addr(skb->head)) {
878 if (!skb->cloned ||
879 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
880 vfree(skb->head);
881
882 skb->head = NULL;
883 }
884 if (skb->sk != NULL)
885 sock_rfree(skb);
886}
887
888static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
889{
890 WARN_ON(skb->sk != NULL);
891 skb->sk = sk;
892 skb->destructor = netlink_skb_destructor;
893 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
894 sk_mem_charge(sk, skb->truesize);
895}
896
897static void netlink_sock_destruct(struct sock *sk)
898{
899 struct netlink_sock *nlk = nlk_sk(sk);
900
901 if (nlk->cb_running) {
902 if (nlk->cb.done)
903 nlk->cb.done(&nlk->cb);
904
905 module_put(nlk->cb.module);
906 kfree_skb(nlk->cb.skb);
907 }
908
909 skb_queue_purge(&sk->sk_receive_queue);
910#ifdef CONFIG_NETLINK_MMAP
911 if (1) {
912 struct nl_mmap_req req;
913
914 memset(&req, 0, sizeof(req));
915 if (nlk->rx_ring.pg_vec)
916 netlink_set_ring(sk, &req, true, false);
917 memset(&req, 0, sizeof(req));
918 if (nlk->tx_ring.pg_vec)
919 netlink_set_ring(sk, &req, true, true);
920 }
921#endif /* CONFIG_NETLINK_MMAP */
922
923 if (!sock_flag(sk, SOCK_DEAD)) {
924 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
925 return;
926 }
927
928 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
929 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
930 WARN_ON(nlk_sk(sk)->groups);
931}
932
933/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
934 * SMP. Look, when several writers sleep and reader wakes them up, all but one
935 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
936 * this, _but_ remember, it adds useless work on UP machines.
937 */
938
939void netlink_table_grab(void)
940 __acquires(nl_table_lock)
941{
942 might_sleep();
943
944 write_lock_irq(&nl_table_lock);
945
946 if (atomic_read(&nl_table_users)) {
947 DECLARE_WAITQUEUE(wait, current);
948
949 add_wait_queue_exclusive(&nl_table_wait, &wait);
950 for (;;) {
951 set_current_state(TASK_UNINTERRUPTIBLE);
952 if (atomic_read(&nl_table_users) == 0)
953 break;
954 write_unlock_irq(&nl_table_lock);
955 schedule();
956 write_lock_irq(&nl_table_lock);
957 }
958
959 __set_current_state(TASK_RUNNING);
960 remove_wait_queue(&nl_table_wait, &wait);
961 }
962}
963
964void netlink_table_ungrab(void)
965 __releases(nl_table_lock)
966{
967 write_unlock_irq(&nl_table_lock);
968 wake_up(&nl_table_wait);
969}
970
971static inline void
972netlink_lock_table(void)
973{
974 /* read_lock() synchronizes us to netlink_table_grab */
975
976 read_lock(&nl_table_lock);
977 atomic_inc(&nl_table_users);
978 read_unlock(&nl_table_lock);
979}
980
981static inline void
982netlink_unlock_table(void)
983{
984 if (atomic_dec_and_test(&nl_table_users))
985 wake_up(&nl_table_wait);
986}
987
988static bool netlink_compare(struct net *net, struct sock *sk)
989{
990 return net_eq(sock_net(sk), net);
991}
992
993static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
994{
995 struct netlink_table *table = &nl_table[protocol];
996 struct nl_portid_hash *hash = &table->hash;
997 struct hlist_head *head;
998 struct sock *sk;
999
1000 read_lock(&nl_table_lock);
1001 head = nl_portid_hashfn(hash, portid);
1002 sk_for_each(sk, head) {
1003 if (table->compare(net, sk) &&
1004 (nlk_sk(sk)->portid == portid)) {
1005 sock_hold(sk);
1006 goto found;
1007 }
1008 }
1009 sk = NULL;
1010found:
1011 read_unlock(&nl_table_lock);
1012 return sk;
1013}
1014
1015static struct hlist_head *nl_portid_hash_zalloc(size_t size)
1016{
1017 if (size <= PAGE_SIZE)
1018 return kzalloc(size, GFP_ATOMIC);
1019 else
1020 return (struct hlist_head *)
1021 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
1022 get_order(size));
1023}
1024
1025static void nl_portid_hash_free(struct hlist_head *table, size_t size)
1026{
1027 if (size <= PAGE_SIZE)
1028 kfree(table);
1029 else
1030 free_pages((unsigned long)table, get_order(size));
1031}
1032
1033static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
1034{
1035 unsigned int omask, mask, shift;
1036 size_t osize, size;
1037 struct hlist_head *otable, *table;
1038 int i;
1039
1040 omask = mask = hash->mask;
1041 osize = size = (mask + 1) * sizeof(*table);
1042 shift = hash->shift;
1043
1044 if (grow) {
1045 if (++shift > hash->max_shift)
1046 return 0;
1047 mask = mask * 2 + 1;
1048 size *= 2;
1049 }
1050
1051 table = nl_portid_hash_zalloc(size);
1052 if (!table)
1053 return 0;
1054
1055 otable = hash->table;
1056 hash->table = table;
1057 hash->mask = mask;
1058 hash->shift = shift;
1059 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
1060
1061 for (i = 0; i <= omask; i++) {
1062 struct sock *sk;
1063 struct hlist_node *tmp;
1064
1065 sk_for_each_safe(sk, tmp, &otable[i])
1066 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
1067 }
1068
1069 nl_portid_hash_free(otable, osize);
1070 hash->rehash_time = jiffies + 10 * 60 * HZ;
1071 return 1;
1072}
1073
1074static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
1075{
1076 int avg = hash->entries >> hash->shift;
1077
1078 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
1079 return 1;
1080
1081 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
1082 nl_portid_hash_rehash(hash, 0);
1083 return 1;
1084 }
1085
1086 return 0;
1087}
1088
1089static const struct proto_ops netlink_ops;
1090
1091static void
1092netlink_update_listeners(struct sock *sk)
1093{
1094 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1095 unsigned long mask;
1096 unsigned int i;
1097 struct listeners *listeners;
1098
1099 listeners = nl_deref_protected(tbl->listeners);
1100 if (!listeners)
1101 return;
1102
1103 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
1104 mask = 0;
1105 sk_for_each_bound(sk, &tbl->mc_list) {
1106 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
1107 mask |= nlk_sk(sk)->groups[i];
1108 }
1109 listeners->masks[i] = mask;
1110 }
1111 /* this function is only called with the netlink table "grabbed", which
1112 * makes sure updates are visible before bind or setsockopt return. */
1113}
1114
1115static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
1116{
1117 struct netlink_table *table = &nl_table[sk->sk_protocol];
1118 struct nl_portid_hash *hash = &table->hash;
1119 struct hlist_head *head;
1120 int err = -EADDRINUSE;
1121 struct sock *osk;
1122 int len;
1123
1124 netlink_table_grab();
1125 head = nl_portid_hashfn(hash, portid);
1126 len = 0;
1127 sk_for_each(osk, head) {
1128 if (table->compare(net, osk) &&
1129 (nlk_sk(osk)->portid == portid))
1130 break;
1131 len++;
1132 }
1133 if (osk)
1134 goto err;
1135
1136 err = -EBUSY;
1137 if (nlk_sk(sk)->portid)
1138 goto err;
1139
1140 err = -ENOMEM;
1141 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
1142 goto err;
1143
1144 if (len && nl_portid_hash_dilute(hash, len))
1145 head = nl_portid_hashfn(hash, portid);
1146 hash->entries++;
1147 nlk_sk(sk)->portid = portid;
1148 sk_add_node(sk, head);
1149 err = 0;
1150
1151err:
1152 netlink_table_ungrab();
1153 return err;
1154}
1155
1156static void netlink_remove(struct sock *sk)
1157{
1158 netlink_table_grab();
1159 if (sk_del_node_init(sk))
1160 nl_table[sk->sk_protocol].hash.entries--;
1161 if (nlk_sk(sk)->subscriptions)
1162 __sk_del_bind_node(sk);
1163 netlink_table_ungrab();
1164}
1165
1166static struct proto netlink_proto = {
1167 .name = "NETLINK",
1168 .owner = THIS_MODULE,
1169 .obj_size = sizeof(struct netlink_sock),
1170};
1171
1172static int __netlink_create(struct net *net, struct socket *sock,
1173 struct mutex *cb_mutex, int protocol)
1174{
1175 struct sock *sk;
1176 struct netlink_sock *nlk;
1177
1178 sock->ops = &netlink_ops;
1179
1180 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
1181 if (!sk)
1182 return -ENOMEM;
1183
1184 sock_init_data(sock, sk);
1185
1186 nlk = nlk_sk(sk);
1187 if (cb_mutex) {
1188 nlk->cb_mutex = cb_mutex;
1189 } else {
1190 nlk->cb_mutex = &nlk->cb_def_mutex;
1191 mutex_init(nlk->cb_mutex);
1192 }
1193 init_waitqueue_head(&nlk->wait);
1194#ifdef CONFIG_NETLINK_MMAP
1195 mutex_init(&nlk->pg_vec_lock);
1196#endif
1197
1198 sk->sk_destruct = netlink_sock_destruct;
1199 sk->sk_protocol = protocol;
1200 return 0;
1201}
1202
1203static int netlink_create(struct net *net, struct socket *sock, int protocol,
1204 int kern)
1205{
1206 struct module *module = NULL;
1207 struct mutex *cb_mutex;
1208 struct netlink_sock *nlk;
1209 void (*bind)(int group);
1210 int err = 0;
1211
1212 sock->state = SS_UNCONNECTED;
1213
1214 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
1215 return -ESOCKTNOSUPPORT;
1216
1217 if (protocol < 0 || protocol >= MAX_LINKS)
1218 return -EPROTONOSUPPORT;
1219
1220 netlink_lock_table();
1221#ifdef CONFIG_MODULES
1222 if (!nl_table[protocol].registered) {
1223 netlink_unlock_table();
1224 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
1225 netlink_lock_table();
1226 }
1227#endif
1228 if (nl_table[protocol].registered &&
1229 try_module_get(nl_table[protocol].module))
1230 module = nl_table[protocol].module;
1231 else
1232 err = -EPROTONOSUPPORT;
1233 cb_mutex = nl_table[protocol].cb_mutex;
1234 bind = nl_table[protocol].bind;
1235 netlink_unlock_table();
1236
1237 if (err < 0)
1238 goto out;
1239
1240 err = __netlink_create(net, sock, cb_mutex, protocol);
1241 if (err < 0)
1242 goto out_module;
1243
1244 local_bh_disable();
1245 sock_prot_inuse_add(net, &netlink_proto, 1);
1246 local_bh_enable();
1247
1248 nlk = nlk_sk(sock->sk);
1249 nlk->module = module;
1250 nlk->netlink_bind = bind;
1251out:
1252 return err;
1253
1254out_module:
1255 module_put(module);
1256 goto out;
1257}
1258
1259static int netlink_release(struct socket *sock)
1260{
1261 struct sock *sk = sock->sk;
1262 struct netlink_sock *nlk;
1263
1264 if (!sk)
1265 return 0;
1266
1267 netlink_remove(sk);
1268 sock_orphan(sk);
1269 nlk = nlk_sk(sk);
1270
1271 /*
1272 * OK. Socket is unlinked, any packets that arrive now
1273 * will be purged.
1274 */
1275
1276 sock->sk = NULL;
1277 wake_up_interruptible_all(&nlk->wait);
1278
1279 skb_queue_purge(&sk->sk_write_queue);
1280
1281 if (nlk->portid) {
1282 struct netlink_notify n = {
1283 .net = sock_net(sk),
1284 .protocol = sk->sk_protocol,
1285 .portid = nlk->portid,
1286 };
1287 atomic_notifier_call_chain(&netlink_chain,
1288 NETLINK_URELEASE, &n);
1289 }
1290
1291 module_put(nlk->module);
1292
1293 netlink_table_grab();
1294 if (netlink_is_kernel(sk)) {
1295 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
1296 if (--nl_table[sk->sk_protocol].registered == 0) {
1297 struct listeners *old;
1298
1299 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
1300 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
1301 kfree_rcu(old, rcu);
1302 nl_table[sk->sk_protocol].module = NULL;
1303 nl_table[sk->sk_protocol].bind = NULL;
1304 nl_table[sk->sk_protocol].flags = 0;
1305 nl_table[sk->sk_protocol].registered = 0;
1306 }
1307 } else if (nlk->subscriptions) {
1308 netlink_update_listeners(sk);
1309 }
1310 netlink_table_ungrab();
1311
1312 kfree(nlk->groups);
1313 nlk->groups = NULL;
1314
1315 local_bh_disable();
1316 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
1317 local_bh_enable();
1318 sock_put(sk);
1319 return 0;
1320}
1321
1322static int netlink_autobind(struct socket *sock)
1323{
1324 struct sock *sk = sock->sk;
1325 struct net *net = sock_net(sk);
1326 struct netlink_table *table = &nl_table[sk->sk_protocol];
1327 struct nl_portid_hash *hash = &table->hash;
1328 struct hlist_head *head;
1329 struct sock *osk;
1330 s32 portid = task_tgid_vnr(current);
1331 int err;
1332 static s32 rover = -4097;
1333
1334retry:
1335 cond_resched();
1336 netlink_table_grab();
1337 head = nl_portid_hashfn(hash, portid);
1338 sk_for_each(osk, head) {
1339 if (!table->compare(net, osk))
1340 continue;
1341 if (nlk_sk(osk)->portid == portid) {
1342 /* Bind collision, search negative portid values. */
1343 portid = rover--;
1344 if (rover > -4097)
1345 rover = -4097;
1346 netlink_table_ungrab();
1347 goto retry;
1348 }
1349 }
1350 netlink_table_ungrab();
1351
1352 err = netlink_insert(sk, net, portid);
1353 if (err == -EADDRINUSE)
1354 goto retry;
1355
1356 /* If 2 threads race to autobind, that is fine. */
1357 if (err == -EBUSY)
1358 err = 0;
1359
1360 return err;
1361}
1362
1363/**
1364 * __netlink_ns_capable - General netlink message capability test
1365 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
1366 * @user_ns: The user namespace of the capability to use
1367 * @cap: The capability to use
1368 *
1369 * Test to see if the opener of the socket we received the message
1370 * from had when the netlink socket was created and the sender of the
1371 * message has has the capability @cap in the user namespace @user_ns.
1372 */
1373bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
1374 struct user_namespace *user_ns, int cap)
1375{
1376 return ((nsp->flags & NETLINK_SKB_DST) ||
1377 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
1378 ns_capable(user_ns, cap);
1379}
1380EXPORT_SYMBOL(__netlink_ns_capable);
1381
1382/**
1383 * netlink_ns_capable - General netlink message capability test
1384 * @skb: socket buffer holding a netlink command from userspace
1385 * @user_ns: The user namespace of the capability to use
1386 * @cap: The capability to use
1387 *
1388 * Test to see if the opener of the socket we received the message
1389 * from had when the netlink socket was created and the sender of the
1390 * message has has the capability @cap in the user namespace @user_ns.
1391 */
1392bool netlink_ns_capable(const struct sk_buff *skb,
1393 struct user_namespace *user_ns, int cap)
1394{
1395 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
1396}
1397EXPORT_SYMBOL(netlink_ns_capable);
1398
1399/**
1400 * netlink_capable - Netlink global message capability test
1401 * @skb: socket buffer holding a netlink command from userspace
1402 * @cap: The capability to use
1403 *
1404 * Test to see if the opener of the socket we received the message
1405 * from had when the netlink socket was created and the sender of the
1406 * message has has the capability @cap in all user namespaces.
1407 */
1408bool netlink_capable(const struct sk_buff *skb, int cap)
1409{
1410 return netlink_ns_capable(skb, &init_user_ns, cap);
1411}
1412EXPORT_SYMBOL(netlink_capable);
1413
1414/**
1415 * netlink_net_capable - Netlink network namespace message capability test
1416 * @skb: socket buffer holding a netlink command from userspace
1417 * @cap: The capability to use
1418 *
1419 * Test to see if the opener of the socket we received the message
1420 * from had when the netlink socket was created and the sender of the
1421 * message has has the capability @cap over the network namespace of
1422 * the socket we received the message from.
1423 */
1424bool netlink_net_capable(const struct sk_buff *skb, int cap)
1425{
1426 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
1427}
1428EXPORT_SYMBOL(netlink_net_capable);
1429
1430static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
1431{
1432 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
1433 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
1434}
1435
1436static void
1437netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
1438{
1439 struct netlink_sock *nlk = nlk_sk(sk);
1440
1441 if (nlk->subscriptions && !subscriptions)
1442 __sk_del_bind_node(sk);
1443 else if (!nlk->subscriptions && subscriptions)
1444 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
1445 nlk->subscriptions = subscriptions;
1446}
1447
1448static int netlink_realloc_groups(struct sock *sk)
1449{
1450 struct netlink_sock *nlk = nlk_sk(sk);
1451 unsigned int groups;
1452 unsigned long *new_groups;
1453 int err = 0;
1454
1455 netlink_table_grab();
1456
1457 groups = nl_table[sk->sk_protocol].groups;
1458 if (!nl_table[sk->sk_protocol].registered) {
1459 err = -ENOENT;
1460 goto out_unlock;
1461 }
1462
1463 if (nlk->ngroups >= groups)
1464 goto out_unlock;
1465
1466 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
1467 if (new_groups == NULL) {
1468 err = -ENOMEM;
1469 goto out_unlock;
1470 }
1471 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
1472 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
1473
1474 nlk->groups = new_groups;
1475 nlk->ngroups = groups;
1476 out_unlock:
1477 netlink_table_ungrab();
1478 return err;
1479}
1480
1481static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1482 int addr_len)
1483{
1484 struct sock *sk = sock->sk;
1485 struct net *net = sock_net(sk);
1486 struct netlink_sock *nlk = nlk_sk(sk);
1487 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1488 int err;
1489
1490 if (addr_len < sizeof(struct sockaddr_nl))
1491 return -EINVAL;
1492
1493 if (nladdr->nl_family != AF_NETLINK)
1494 return -EINVAL;
1495
1496 /* Only superuser is allowed to listen multicasts */
1497 if (nladdr->nl_groups) {
1498 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1499 return -EPERM;
1500 err = netlink_realloc_groups(sk);
1501 if (err)
1502 return err;
1503 }
1504
1505 if (nlk->portid) {
1506 if (nladdr->nl_pid != nlk->portid)
1507 return -EINVAL;
1508 } else {
1509 err = nladdr->nl_pid ?
1510 netlink_insert(sk, net, nladdr->nl_pid) :
1511 netlink_autobind(sock);
1512 if (err)
1513 return err;
1514 }
1515
1516 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1517 return 0;
1518
1519 netlink_table_grab();
1520 netlink_update_subscriptions(sk, nlk->subscriptions +
1521 hweight32(nladdr->nl_groups) -
1522 hweight32(nlk->groups[0]));
1523 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
1524 netlink_update_listeners(sk);
1525 netlink_table_ungrab();
1526
1527 if (nlk->netlink_bind && nlk->groups[0]) {
1528 int i;
1529
1530 for (i = 0; i < nlk->ngroups; i++) {
1531 if (test_bit(i, nlk->groups))
1532 nlk->netlink_bind(i);
1533 }
1534 }
1535
1536 return 0;
1537}
1538
1539static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1540 int alen, int flags)
1541{
1542 int err = 0;
1543 struct sock *sk = sock->sk;
1544 struct netlink_sock *nlk = nlk_sk(sk);
1545 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1546
1547 if (alen < sizeof(addr->sa_family))
1548 return -EINVAL;
1549
1550 if (addr->sa_family == AF_UNSPEC) {
1551 sk->sk_state = NETLINK_UNCONNECTED;
1552 nlk->dst_portid = 0;
1553 nlk->dst_group = 0;
1554 return 0;
1555 }
1556 if (addr->sa_family != AF_NETLINK)
1557 return -EINVAL;
1558
1559 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1560 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1561 return -EPERM;
1562
1563 if (!nlk->portid)
1564 err = netlink_autobind(sock);
1565
1566 if (err == 0) {
1567 sk->sk_state = NETLINK_CONNECTED;
1568 nlk->dst_portid = nladdr->nl_pid;
1569 nlk->dst_group = ffs(nladdr->nl_groups);
1570 }
1571
1572 return err;
1573}
1574
1575static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1576 int *addr_len, int peer)
1577{
1578 struct sock *sk = sock->sk;
1579 struct netlink_sock *nlk = nlk_sk(sk);
1580 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1581
1582 nladdr->nl_family = AF_NETLINK;
1583 nladdr->nl_pad = 0;
1584 *addr_len = sizeof(*nladdr);
1585
1586 if (peer) {
1587 nladdr->nl_pid = nlk->dst_portid;
1588 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1589 } else {
1590 nladdr->nl_pid = nlk->portid;
1591 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1592 }
1593 return 0;
1594}
1595
1596static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1597{
1598 struct sock *sock;
1599 struct netlink_sock *nlk;
1600
1601 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1602 if (!sock)
1603 return ERR_PTR(-ECONNREFUSED);
1604
1605 /* Don't bother queuing skb if kernel socket has no input function */
1606 nlk = nlk_sk(sock);
1607 if (sock->sk_state == NETLINK_CONNECTED &&
1608 nlk->dst_portid != nlk_sk(ssk)->portid) {
1609 sock_put(sock);
1610 return ERR_PTR(-ECONNREFUSED);
1611 }
1612 return sock;
1613}
1614
1615struct sock *netlink_getsockbyfilp(struct file *filp)
1616{
1617 struct inode *inode = file_inode(filp);
1618 struct sock *sock;
1619
1620 if (!S_ISSOCK(inode->i_mode))
1621 return ERR_PTR(-ENOTSOCK);
1622
1623 sock = SOCKET_I(inode)->sk;
1624 if (sock->sk_family != AF_NETLINK)
1625 return ERR_PTR(-EINVAL);
1626
1627 sock_hold(sock);
1628 return sock;
1629}
1630
1631static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1632 int broadcast)
1633{
1634 struct sk_buff *skb;
1635 void *data;
1636
1637 if (size <= NLMSG_GOODSIZE || broadcast)
1638 return alloc_skb(size, GFP_KERNEL);
1639
1640 size = SKB_DATA_ALIGN(size) +
1641 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1642
1643 data = vmalloc(size);
1644 if (data == NULL)
1645 return NULL;
1646
1647 skb = build_skb(data, size);
1648 if (skb == NULL)
1649 vfree(data);
1650 else {
1651 skb->head_frag = 0;
1652 skb->destructor = netlink_skb_destructor;
1653 }
1654
1655 return skb;
1656}
1657
1658/*
1659 * Attach a skb to a netlink socket.
1660 * The caller must hold a reference to the destination socket. On error, the
1661 * reference is dropped. The skb is not send to the destination, just all
1662 * all error checks are performed and memory in the queue is reserved.
1663 * Return values:
1664 * < 0: error. skb freed, reference to sock dropped.
1665 * 0: continue
1666 * 1: repeat lookup - reference dropped while waiting for socket memory.
1667 */
1668int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1669 long *timeo, struct sock *ssk)
1670{
1671 struct netlink_sock *nlk;
1672
1673 nlk = nlk_sk(sk);
1674
1675 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1676 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1677 !netlink_skb_is_mmaped(skb)) {
1678 DECLARE_WAITQUEUE(wait, current);
1679 if (!*timeo) {
1680 if (!ssk || netlink_is_kernel(ssk))
1681 netlink_overrun(sk);
1682 sock_put(sk);
1683 kfree_skb(skb);
1684 return -EAGAIN;
1685 }
1686
1687 __set_current_state(TASK_INTERRUPTIBLE);
1688 add_wait_queue(&nlk->wait, &wait);
1689
1690 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1691 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
1692 !sock_flag(sk, SOCK_DEAD))
1693 *timeo = schedule_timeout(*timeo);
1694
1695 __set_current_state(TASK_RUNNING);
1696 remove_wait_queue(&nlk->wait, &wait);
1697 sock_put(sk);
1698
1699 if (signal_pending(current)) {
1700 kfree_skb(skb);
1701 return sock_intr_errno(*timeo);
1702 }
1703 return 1;
1704 }
1705 netlink_skb_set_owner_r(skb, sk);
1706 return 0;
1707}
1708
1709static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1710{
1711 int len = skb->len;
1712
1713 netlink_deliver_tap(skb);
1714
1715#ifdef CONFIG_NETLINK_MMAP
1716 if (netlink_skb_is_mmaped(skb))
1717 netlink_queue_mmaped_skb(sk, skb);
1718 else if (netlink_rx_is_mmaped(sk))
1719 netlink_ring_set_copied(sk, skb);
1720 else
1721#endif /* CONFIG_NETLINK_MMAP */
1722 skb_queue_tail(&sk->sk_receive_queue, skb);
1723 sk->sk_data_ready(sk);
1724 return len;
1725}
1726
1727int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1728{
1729 int len = __netlink_sendskb(sk, skb);
1730
1731 sock_put(sk);
1732 return len;
1733}
1734
1735void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1736{
1737 kfree_skb(skb);
1738 sock_put(sk);
1739}
1740
1741static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1742{
1743 int delta;
1744
1745 WARN_ON(skb->sk != NULL);
1746 if (netlink_skb_is_mmaped(skb))
1747 return skb;
1748
1749 delta = skb->end - skb->tail;
1750 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1751 return skb;
1752
1753 if (skb_shared(skb)) {
1754 struct sk_buff *nskb = skb_clone(skb, allocation);
1755 if (!nskb)
1756 return skb;
1757 consume_skb(skb);
1758 skb = nskb;
1759 }
1760
1761 if (!pskb_expand_head(skb, 0, -delta, allocation))
1762 skb->truesize -= delta;
1763
1764 return skb;
1765}
1766
1767static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1768 struct sock *ssk)
1769{
1770 int ret;
1771 struct netlink_sock *nlk = nlk_sk(sk);
1772
1773 ret = -ECONNREFUSED;
1774 if (nlk->netlink_rcv != NULL) {
1775 ret = skb->len;
1776 netlink_skb_set_owner_r(skb, sk);
1777 NETLINK_CB(skb).sk = ssk;
1778 netlink_deliver_tap_kernel(sk, ssk, skb);
1779 nlk->netlink_rcv(skb);
1780 consume_skb(skb);
1781 } else {
1782 kfree_skb(skb);
1783 }
1784 sock_put(sk);
1785 return ret;
1786}
1787
1788int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1789 u32 portid, int nonblock)
1790{
1791 struct sock *sk;
1792 int err;
1793 long timeo;
1794
1795 skb = netlink_trim(skb, gfp_any());
1796
1797 timeo = sock_sndtimeo(ssk, nonblock);
1798retry:
1799 sk = netlink_getsockbyportid(ssk, portid);
1800 if (IS_ERR(sk)) {
1801 kfree_skb(skb);
1802 return PTR_ERR(sk);
1803 }
1804 if (netlink_is_kernel(sk))
1805 return netlink_unicast_kernel(sk, skb, ssk);
1806
1807 if (sk_filter(sk, skb)) {
1808 err = skb->len;
1809 kfree_skb(skb);
1810 sock_put(sk);
1811 return err;
1812 }
1813
1814 err = netlink_attachskb(sk, skb, &timeo, ssk);
1815 if (err == 1)
1816 goto retry;
1817 if (err)
1818 return err;
1819
1820 return netlink_sendskb(sk, skb);
1821}
1822EXPORT_SYMBOL(netlink_unicast);
1823
1824struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
1825 u32 dst_portid, gfp_t gfp_mask)
1826{
1827#ifdef CONFIG_NETLINK_MMAP
1828 struct sock *sk = NULL;
1829 struct sk_buff *skb;
1830 struct netlink_ring *ring;
1831 struct nl_mmap_hdr *hdr;
1832 unsigned int maxlen;
1833
1834 sk = netlink_getsockbyportid(ssk, dst_portid);
1835 if (IS_ERR(sk))
1836 goto out;
1837
1838 ring = &nlk_sk(sk)->rx_ring;
1839 /* fast-path without atomic ops for common case: non-mmaped receiver */
1840 if (ring->pg_vec == NULL)
1841 goto out_put;
1842
1843 if (ring->frame_size - NL_MMAP_HDRLEN < size)
1844 goto out_put;
1845
1846 skb = alloc_skb_head(gfp_mask);
1847 if (skb == NULL)
1848 goto err1;
1849
1850 spin_lock_bh(&sk->sk_receive_queue.lock);
1851 /* check again under lock */
1852 if (ring->pg_vec == NULL)
1853 goto out_free;
1854
1855 /* check again under lock */
1856 maxlen = ring->frame_size - NL_MMAP_HDRLEN;
1857 if (maxlen < size)
1858 goto out_free;
1859
1860 netlink_forward_ring(ring);
1861 hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
1862 if (hdr == NULL)
1863 goto err2;
1864 netlink_ring_setup_skb(skb, sk, ring, hdr);
1865 netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
1866 atomic_inc(&ring->pending);
1867 netlink_increment_head(ring);
1868
1869 spin_unlock_bh(&sk->sk_receive_queue.lock);
1870 return skb;
1871
1872err2:
1873 kfree_skb(skb);
1874 spin_unlock_bh(&sk->sk_receive_queue.lock);
1875 netlink_overrun(sk);
1876err1:
1877 sock_put(sk);
1878 return NULL;
1879
1880out_free:
1881 kfree_skb(skb);
1882 spin_unlock_bh(&sk->sk_receive_queue.lock);
1883out_put:
1884 sock_put(sk);
1885out:
1886#endif
1887 return alloc_skb(size, gfp_mask);
1888}
1889EXPORT_SYMBOL_GPL(netlink_alloc_skb);
1890
1891int netlink_has_listeners(struct sock *sk, unsigned int group)
1892{
1893 int res = 0;
1894 struct listeners *listeners;
1895
1896 BUG_ON(!netlink_is_kernel(sk));
1897
1898 rcu_read_lock();
1899 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1900
1901 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1902 res = test_bit(group - 1, listeners->masks);
1903
1904 rcu_read_unlock();
1905
1906 return res;
1907}
1908EXPORT_SYMBOL_GPL(netlink_has_listeners);
1909
1910static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1911{
1912 struct netlink_sock *nlk = nlk_sk(sk);
1913
1914 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1915 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
1916 netlink_skb_set_owner_r(skb, sk);
1917 __netlink_sendskb(sk, skb);
1918 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1919 }
1920 return -1;
1921}
1922
1923struct netlink_broadcast_data {
1924 struct sock *exclude_sk;
1925 struct net *net;
1926 u32 portid;
1927 u32 group;
1928 int failure;
1929 int delivery_failure;
1930 int congested;
1931 int delivered;
1932 gfp_t allocation;
1933 struct sk_buff *skb, *skb2;
1934 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1935 void *tx_data;
1936};
1937
1938static int do_one_broadcast(struct sock *sk,
1939 struct netlink_broadcast_data *p)
1940{
1941 struct netlink_sock *nlk = nlk_sk(sk);
1942 int val;
1943
1944 if (p->exclude_sk == sk)
1945 goto out;
1946
1947 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1948 !test_bit(p->group - 1, nlk->groups))
1949 goto out;
1950
1951 if (!net_eq(sock_net(sk), p->net))
1952 goto out;
1953
1954 if (p->failure) {
1955 netlink_overrun(sk);
1956 goto out;
1957 }
1958
1959 sock_hold(sk);
1960 if (p->skb2 == NULL) {
1961 if (skb_shared(p->skb)) {
1962 p->skb2 = skb_clone(p->skb, p->allocation);
1963 } else {
1964 p->skb2 = skb_get(p->skb);
1965 /*
1966 * skb ownership may have been set when
1967 * delivered to a previous socket.
1968 */
1969 skb_orphan(p->skb2);
1970 }
1971 }
1972 if (p->skb2 == NULL) {
1973 netlink_overrun(sk);
1974 /* Clone failed. Notify ALL listeners. */
1975 p->failure = 1;
1976 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1977 p->delivery_failure = 1;
1978 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1979 kfree_skb(p->skb2);
1980 p->skb2 = NULL;
1981 } else if (sk_filter(sk, p->skb2)) {
1982 kfree_skb(p->skb2);
1983 p->skb2 = NULL;
1984 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1985 netlink_overrun(sk);
1986 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1987 p->delivery_failure = 1;
1988 } else {
1989 p->congested |= val;
1990 p->delivered = 1;
1991 p->skb2 = NULL;
1992 }
1993 sock_put(sk);
1994
1995out:
1996 return 0;
1997}
1998
1999int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
2000 u32 group, gfp_t allocation,
2001 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
2002 void *filter_data)
2003{
2004 struct net *net = sock_net(ssk);
2005 struct netlink_broadcast_data info;
2006 struct sock *sk;
2007
2008 skb = netlink_trim(skb, allocation);
2009
2010 info.exclude_sk = ssk;
2011 info.net = net;
2012 info.portid = portid;
2013 info.group = group;
2014 info.failure = 0;
2015 info.delivery_failure = 0;
2016 info.congested = 0;
2017 info.delivered = 0;
2018 info.allocation = allocation;
2019 info.skb = skb;
2020 info.skb2 = NULL;
2021 info.tx_filter = filter;
2022 info.tx_data = filter_data;
2023
2024 /* While we sleep in clone, do not allow to change socket list */
2025
2026 netlink_lock_table();
2027
2028 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2029 do_one_broadcast(sk, &info);
2030
2031 consume_skb(skb);
2032
2033 netlink_unlock_table();
2034
2035 if (info.delivery_failure) {
2036 kfree_skb(info.skb2);
2037 return -ENOBUFS;
2038 }
2039 consume_skb(info.skb2);
2040
2041 if (info.delivered) {
2042 if (info.congested && (allocation & __GFP_WAIT))
2043 yield();
2044 return 0;
2045 }
2046 return -ESRCH;
2047}
2048EXPORT_SYMBOL(netlink_broadcast_filtered);
2049
2050int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
2051 u32 group, gfp_t allocation)
2052{
2053 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
2054 NULL, NULL);
2055}
2056EXPORT_SYMBOL(netlink_broadcast);
2057
2058struct netlink_set_err_data {
2059 struct sock *exclude_sk;
2060 u32 portid;
2061 u32 group;
2062 int code;
2063};
2064
2065static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
2066{
2067 struct netlink_sock *nlk = nlk_sk(sk);
2068 int ret = 0;
2069
2070 if (sk == p->exclude_sk)
2071 goto out;
2072
2073 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
2074 goto out;
2075
2076 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
2077 !test_bit(p->group - 1, nlk->groups))
2078 goto out;
2079
2080 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
2081 ret = 1;
2082 goto out;
2083 }
2084
2085 sk->sk_err = p->code;
2086 sk->sk_error_report(sk);
2087out:
2088 return ret;
2089}
2090
2091/**
2092 * netlink_set_err - report error to broadcast listeners
2093 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
2094 * @portid: the PORTID of a process that we want to skip (if any)
2095 * @group: the broadcast group that will notice the error
2096 * @code: error code, must be negative (as usual in kernelspace)
2097 *
2098 * This function returns the number of broadcast listeners that have set the
2099 * NETLINK_RECV_NO_ENOBUFS socket option.
2100 */
2101int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
2102{
2103 struct netlink_set_err_data info;
2104 struct sock *sk;
2105 int ret = 0;
2106
2107 info.exclude_sk = ssk;
2108 info.portid = portid;
2109 info.group = group;
2110 /* sk->sk_err wants a positive error value */
2111 info.code = -code;
2112
2113 read_lock(&nl_table_lock);
2114
2115 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
2116 ret += do_one_set_err(sk, &info);
2117
2118 read_unlock(&nl_table_lock);
2119 return ret;
2120}
2121EXPORT_SYMBOL(netlink_set_err);
2122
2123/* must be called with netlink table grabbed */
2124static void netlink_update_socket_mc(struct netlink_sock *nlk,
2125 unsigned int group,
2126 int is_new)
2127{
2128 int old, new = !!is_new, subscriptions;
2129
2130 old = test_bit(group - 1, nlk->groups);
2131 subscriptions = nlk->subscriptions - old + new;
2132 if (new)
2133 __set_bit(group - 1, nlk->groups);
2134 else
2135 __clear_bit(group - 1, nlk->groups);
2136 netlink_update_subscriptions(&nlk->sk, subscriptions);
2137 netlink_update_listeners(&nlk->sk);
2138}
2139
2140static int netlink_setsockopt(struct socket *sock, int level, int optname,
2141 char __user *optval, unsigned int optlen)
2142{
2143 struct sock *sk = sock->sk;
2144 struct netlink_sock *nlk = nlk_sk(sk);
2145 unsigned int val = 0;
2146 int err;
2147
2148 if (level != SOL_NETLINK)
2149 return -ENOPROTOOPT;
2150
2151 if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
2152 optlen >= sizeof(int) &&
2153 get_user(val, (unsigned int __user *)optval))
2154 return -EFAULT;
2155
2156 switch (optname) {
2157 case NETLINK_PKTINFO:
2158 if (val)
2159 nlk->flags |= NETLINK_RECV_PKTINFO;
2160 else
2161 nlk->flags &= ~NETLINK_RECV_PKTINFO;
2162 err = 0;
2163 break;
2164 case NETLINK_ADD_MEMBERSHIP:
2165 case NETLINK_DROP_MEMBERSHIP: {
2166 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
2167 return -EPERM;
2168 err = netlink_realloc_groups(sk);
2169 if (err)
2170 return err;
2171 if (!val || val - 1 >= nlk->ngroups)
2172 return -EINVAL;
2173 netlink_table_grab();
2174 netlink_update_socket_mc(nlk, val,
2175 optname == NETLINK_ADD_MEMBERSHIP);
2176 netlink_table_ungrab();
2177
2178 if (nlk->netlink_bind)
2179 nlk->netlink_bind(val);
2180
2181 err = 0;
2182 break;
2183 }
2184 case NETLINK_BROADCAST_ERROR:
2185 if (val)
2186 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
2187 else
2188 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
2189 err = 0;
2190 break;
2191 case NETLINK_NO_ENOBUFS:
2192 if (val) {
2193 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
2194 clear_bit(NETLINK_CONGESTED, &nlk->state);
2195 wake_up_interruptible(&nlk->wait);
2196 } else {
2197 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
2198 }
2199 err = 0;
2200 break;
2201#ifdef CONFIG_NETLINK_MMAP
2202 case NETLINK_RX_RING:
2203 case NETLINK_TX_RING: {
2204 struct nl_mmap_req req;
2205
2206 /* Rings might consume more memory than queue limits, require
2207 * CAP_NET_ADMIN.
2208 */
2209 if (!capable(CAP_NET_ADMIN))
2210 return -EPERM;
2211 if (optlen < sizeof(req))
2212 return -EINVAL;
2213 if (copy_from_user(&req, optval, sizeof(req)))
2214 return -EFAULT;
2215 err = netlink_set_ring(sk, &req, false,
2216 optname == NETLINK_TX_RING);
2217 break;
2218 }
2219#endif /* CONFIG_NETLINK_MMAP */
2220 default:
2221 err = -ENOPROTOOPT;
2222 }
2223 return err;
2224}
2225
2226static int netlink_getsockopt(struct socket *sock, int level, int optname,
2227 char __user *optval, int __user *optlen)
2228{
2229 struct sock *sk = sock->sk;
2230 struct netlink_sock *nlk = nlk_sk(sk);
2231 int len, val, err;
2232
2233 if (level != SOL_NETLINK)
2234 return -ENOPROTOOPT;
2235
2236 if (get_user(len, optlen))
2237 return -EFAULT;
2238 if (len < 0)
2239 return -EINVAL;
2240
2241 switch (optname) {
2242 case NETLINK_PKTINFO:
2243 if (len < sizeof(int))
2244 return -EINVAL;
2245 len = sizeof(int);
2246 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
2247 if (put_user(len, optlen) ||
2248 put_user(val, optval))
2249 return -EFAULT;
2250 err = 0;
2251 break;
2252 case NETLINK_BROADCAST_ERROR:
2253 if (len < sizeof(int))
2254 return -EINVAL;
2255 len = sizeof(int);
2256 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
2257 if (put_user(len, optlen) ||
2258 put_user(val, optval))
2259 return -EFAULT;
2260 err = 0;
2261 break;
2262 case NETLINK_NO_ENOBUFS:
2263 if (len < sizeof(int))
2264 return -EINVAL;
2265 len = sizeof(int);
2266 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
2267 if (put_user(len, optlen) ||
2268 put_user(val, optval))
2269 return -EFAULT;
2270 err = 0;
2271 break;
2272 default:
2273 err = -ENOPROTOOPT;
2274 }
2275 return err;
2276}
2277
2278static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
2279{
2280 struct nl_pktinfo info;
2281
2282 info.group = NETLINK_CB(skb).dst_group;
2283 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
2284}
2285
2286static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
2287 struct msghdr *msg, size_t len)
2288{
2289 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2290 struct sock *sk = sock->sk;
2291 struct netlink_sock *nlk = nlk_sk(sk);
2292 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2293 u32 dst_portid;
2294 u32 dst_group;
2295 struct sk_buff *skb;
2296 int err;
2297 struct scm_cookie scm;
2298 u32 netlink_skb_flags = 0;
2299
2300 if (msg->msg_flags&MSG_OOB)
2301 return -EOPNOTSUPP;
2302
2303 if (NULL == siocb->scm)
2304 siocb->scm = &scm;
2305
2306 err = scm_send(sock, msg, siocb->scm, true);
2307 if (err < 0)
2308 return err;
2309
2310 if (msg->msg_namelen) {
2311 err = -EINVAL;
2312 if (addr->nl_family != AF_NETLINK)
2313 goto out;
2314 dst_portid = addr->nl_pid;
2315 dst_group = ffs(addr->nl_groups);
2316 err = -EPERM;
2317 if ((dst_group || dst_portid) &&
2318 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
2319 goto out;
2320 netlink_skb_flags |= NETLINK_SKB_DST;
2321 } else {
2322 dst_portid = nlk->dst_portid;
2323 dst_group = nlk->dst_group;
2324 }
2325
2326 if (!nlk->portid) {
2327 err = netlink_autobind(sock);
2328 if (err)
2329 goto out;
2330 }
2331
2332 if (netlink_tx_is_mmaped(sk) &&
2333 msg->msg_iov->iov_base == NULL) {
2334 err = netlink_mmap_sendmsg(sk, msg, dst_portid, dst_group,
2335 siocb);
2336 goto out;
2337 }
2338
2339 err = -EMSGSIZE;
2340 if (len > sk->sk_sndbuf - 32)
2341 goto out;
2342 err = -ENOBUFS;
2343 skb = netlink_alloc_large_skb(len, dst_group);
2344 if (skb == NULL)
2345 goto out;
2346
2347 NETLINK_CB(skb).portid = nlk->portid;
2348 NETLINK_CB(skb).dst_group = dst_group;
2349 NETLINK_CB(skb).creds = siocb->scm->creds;
2350 NETLINK_CB(skb).flags = netlink_skb_flags;
2351
2352 err = -EFAULT;
2353 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
2354 kfree_skb(skb);
2355 goto out;
2356 }
2357
2358 err = security_netlink_send(sk, skb);
2359 if (err) {
2360 kfree_skb(skb);
2361 goto out;
2362 }
2363
2364 if (dst_group) {
2365 atomic_inc(&skb->users);
2366 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
2367 }
2368 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
2369
2370out:
2371 scm_destroy(siocb->scm);
2372 return err;
2373}
2374
2375static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
2376 struct msghdr *msg, size_t len,
2377 int flags)
2378{
2379 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
2380 struct scm_cookie scm;
2381 struct sock *sk = sock->sk;
2382 struct netlink_sock *nlk = nlk_sk(sk);
2383 int noblock = flags&MSG_DONTWAIT;
2384 size_t copied;
2385 struct sk_buff *skb, *data_skb;
2386 int err, ret;
2387
2388 if (flags&MSG_OOB)
2389 return -EOPNOTSUPP;
2390
2391 copied = 0;
2392
2393 skb = skb_recv_datagram(sk, flags, noblock, &err);
2394 if (skb == NULL)
2395 goto out;
2396
2397 data_skb = skb;
2398
2399#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2400 if (unlikely(skb_shinfo(skb)->frag_list)) {
2401 /*
2402 * If this skb has a frag_list, then here that means that we
2403 * will have to use the frag_list skb's data for compat tasks
2404 * and the regular skb's data for normal (non-compat) tasks.
2405 *
2406 * If we need to send the compat skb, assign it to the
2407 * 'data_skb' variable so that it will be used below for data
2408 * copying. We keep 'skb' for everything else, including
2409 * freeing both later.
2410 */
2411 if (flags & MSG_CMSG_COMPAT)
2412 data_skb = skb_shinfo(skb)->frag_list;
2413 }
2414#endif
2415
2416 /* Record the max length of recvmsg() calls for future allocations */
2417 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
2418 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
2419 16384);
2420
2421 copied = data_skb->len;
2422 if (len < copied) {
2423 msg->msg_flags |= MSG_TRUNC;
2424 copied = len;
2425 }
2426
2427 skb_reset_transport_header(data_skb);
2428 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
2429
2430 if (msg->msg_name) {
2431 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2432 addr->nl_family = AF_NETLINK;
2433 addr->nl_pad = 0;
2434 addr->nl_pid = NETLINK_CB(skb).portid;
2435 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2436 msg->msg_namelen = sizeof(*addr);
2437 }
2438
2439 if (nlk->flags & NETLINK_RECV_PKTINFO)
2440 netlink_cmsg_recv_pktinfo(msg, skb);
2441
2442 if (NULL == siocb->scm) {
2443 memset(&scm, 0, sizeof(scm));
2444 siocb->scm = &scm;
2445 }
2446 siocb->scm->creds = *NETLINK_CREDS(skb);
2447 if (flags & MSG_TRUNC)
2448 copied = data_skb->len;
2449
2450 skb_free_datagram(sk, skb);
2451
2452 if (nlk->cb_running &&
2453 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2454 ret = netlink_dump(sk);
2455 if (ret) {
2456 sk->sk_err = ret;
2457 sk->sk_error_report(sk);
2458 }
2459 }
2460
2461 scm_recv(sock, msg, siocb->scm, flags);
2462out:
2463 netlink_rcv_wake(sk);
2464 return err ? : copied;
2465}
2466
2467static void netlink_data_ready(struct sock *sk)
2468{
2469 BUG();
2470}
2471
2472/*
2473 * We export these functions to other modules. They provide a
2474 * complete set of kernel non-blocking support for message
2475 * queueing.
2476 */
2477
2478struct sock *
2479__netlink_kernel_create(struct net *net, int unit, struct module *module,
2480 struct netlink_kernel_cfg *cfg)
2481{
2482 struct socket *sock;
2483 struct sock *sk;
2484 struct netlink_sock *nlk;
2485 struct listeners *listeners = NULL;
2486 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2487 unsigned int groups;
2488
2489 BUG_ON(!nl_table);
2490
2491 if (unit < 0 || unit >= MAX_LINKS)
2492 return NULL;
2493
2494 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2495 return NULL;
2496
2497 /*
2498 * We have to just have a reference on the net from sk, but don't
2499 * get_net it. Besides, we cannot get and then put the net here.
2500 * So we create one inside init_net and the move it to net.
2501 */
2502
2503 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
2504 goto out_sock_release_nosk;
2505
2506 sk = sock->sk;
2507 sk_change_net(sk, net);
2508
2509 if (!cfg || cfg->groups < 32)
2510 groups = 32;
2511 else
2512 groups = cfg->groups;
2513
2514 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2515 if (!listeners)
2516 goto out_sock_release;
2517
2518 sk->sk_data_ready = netlink_data_ready;
2519 if (cfg && cfg->input)
2520 nlk_sk(sk)->netlink_rcv = cfg->input;
2521
2522 if (netlink_insert(sk, net, 0))
2523 goto out_sock_release;
2524
2525 nlk = nlk_sk(sk);
2526 nlk->flags |= NETLINK_KERNEL_SOCKET;
2527
2528 netlink_table_grab();
2529 if (!nl_table[unit].registered) {
2530 nl_table[unit].groups = groups;
2531 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2532 nl_table[unit].cb_mutex = cb_mutex;
2533 nl_table[unit].module = module;
2534 if (cfg) {
2535 nl_table[unit].bind = cfg->bind;
2536 nl_table[unit].flags = cfg->flags;
2537 if (cfg->compare)
2538 nl_table[unit].compare = cfg->compare;
2539 }
2540 nl_table[unit].registered = 1;
2541 } else {
2542 kfree(listeners);
2543 nl_table[unit].registered++;
2544 }
2545 netlink_table_ungrab();
2546 return sk;
2547
2548out_sock_release:
2549 kfree(listeners);
2550 netlink_kernel_release(sk);
2551 return NULL;
2552
2553out_sock_release_nosk:
2554 sock_release(sock);
2555 return NULL;
2556}
2557EXPORT_SYMBOL(__netlink_kernel_create);
2558
2559void
2560netlink_kernel_release(struct sock *sk)
2561{
2562 sk_release_kernel(sk);
2563}
2564EXPORT_SYMBOL(netlink_kernel_release);
2565
2566int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2567{
2568 struct listeners *new, *old;
2569 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2570
2571 if (groups < 32)
2572 groups = 32;
2573
2574 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2575 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2576 if (!new)
2577 return -ENOMEM;
2578 old = nl_deref_protected(tbl->listeners);
2579 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2580 rcu_assign_pointer(tbl->listeners, new);
2581
2582 kfree_rcu(old, rcu);
2583 }
2584 tbl->groups = groups;
2585
2586 return 0;
2587}
2588
2589/**
2590 * netlink_change_ngroups - change number of multicast groups
2591 *
2592 * This changes the number of multicast groups that are available
2593 * on a certain netlink family. Note that it is not possible to
2594 * change the number of groups to below 32. Also note that it does
2595 * not implicitly call netlink_clear_multicast_users() when the
2596 * number of groups is reduced.
2597 *
2598 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2599 * @groups: The new number of groups.
2600 */
2601int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2602{
2603 int err;
2604
2605 netlink_table_grab();
2606 err = __netlink_change_ngroups(sk, groups);
2607 netlink_table_ungrab();
2608
2609 return err;
2610}
2611
2612void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2613{
2614 struct sock *sk;
2615 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2616
2617 sk_for_each_bound(sk, &tbl->mc_list)
2618 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2619}
2620
2621struct nlmsghdr *
2622__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2623{
2624 struct nlmsghdr *nlh;
2625 int size = nlmsg_msg_size(len);
2626
2627 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2628 nlh->nlmsg_type = type;
2629 nlh->nlmsg_len = size;
2630 nlh->nlmsg_flags = flags;
2631 nlh->nlmsg_pid = portid;
2632 nlh->nlmsg_seq = seq;
2633 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2634 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2635 return nlh;
2636}
2637EXPORT_SYMBOL(__nlmsg_put);
2638
2639/*
2640 * It looks a bit ugly.
2641 * It would be better to create kernel thread.
2642 */
2643
2644static int netlink_dump(struct sock *sk)
2645{
2646 struct netlink_sock *nlk = nlk_sk(sk);
2647 struct netlink_callback *cb;
2648 struct sk_buff *skb = NULL;
2649 struct nlmsghdr *nlh;
2650 int len, err = -ENOBUFS;
2651 int alloc_size;
2652
2653 mutex_lock(nlk->cb_mutex);
2654 if (!nlk->cb_running) {
2655 err = -EINVAL;
2656 goto errout_skb;
2657 }
2658
2659 cb = &nlk->cb;
2660 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2661
2662 if (!netlink_rx_is_mmaped(sk) &&
2663 atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2664 goto errout_skb;
2665
2666 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2667 * required, but it makes sense to _attempt_ a 16K bytes allocation
2668 * to reduce number of system calls on dump operations, if user
2669 * ever provided a big enough buffer.
2670 */
2671 if (alloc_size < nlk->max_recvmsg_len) {
2672 skb = netlink_alloc_skb(sk,
2673 nlk->max_recvmsg_len,
2674 nlk->portid,
2675 GFP_KERNEL |
2676 __GFP_NOWARN |
2677 __GFP_NORETRY);
2678 /* available room should be exact amount to avoid MSG_TRUNC */
2679 if (skb)
2680 skb_reserve(skb, skb_tailroom(skb) -
2681 nlk->max_recvmsg_len);
2682 }
2683 if (!skb)
2684 skb = netlink_alloc_skb(sk, alloc_size, nlk->portid,
2685 GFP_KERNEL);
2686 if (!skb)
2687 goto errout_skb;
2688 netlink_skb_set_owner_r(skb, sk);
2689
2690 len = cb->dump(skb, cb);
2691
2692 if (len > 0) {
2693 mutex_unlock(nlk->cb_mutex);
2694
2695 if (sk_filter(sk, skb))
2696 kfree_skb(skb);
2697 else
2698 __netlink_sendskb(sk, skb);
2699 return 0;
2700 }
2701
2702 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2703 if (!nlh)
2704 goto errout_skb;
2705
2706 nl_dump_check_consistent(cb, nlh);
2707
2708 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2709
2710 if (sk_filter(sk, skb))
2711 kfree_skb(skb);
2712 else
2713 __netlink_sendskb(sk, skb);
2714
2715 if (cb->done)
2716 cb->done(cb);
2717
2718 nlk->cb_running = false;
2719 mutex_unlock(nlk->cb_mutex);
2720 module_put(cb->module);
2721 consume_skb(cb->skb);
2722 return 0;
2723
2724errout_skb:
2725 mutex_unlock(nlk->cb_mutex);
2726 kfree_skb(skb);
2727 return err;
2728}
2729
2730int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2731 const struct nlmsghdr *nlh,
2732 struct netlink_dump_control *control)
2733{
2734 struct netlink_callback *cb;
2735 struct sock *sk;
2736 struct netlink_sock *nlk;
2737 int ret;
2738
2739 /* Memory mapped dump requests need to be copied to avoid looping
2740 * on the pending state in netlink_mmap_sendmsg() while the CB hold
2741 * a reference to the skb.
2742 */
2743 if (netlink_skb_is_mmaped(skb)) {
2744 skb = skb_copy(skb, GFP_KERNEL);
2745 if (skb == NULL)
2746 return -ENOBUFS;
2747 } else
2748 atomic_inc(&skb->users);
2749
2750 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2751 if (sk == NULL) {
2752 ret = -ECONNREFUSED;
2753 goto error_free;
2754 }
2755
2756 nlk = nlk_sk(sk);
2757 mutex_lock(nlk->cb_mutex);
2758 /* A dump is in progress... */
2759 if (nlk->cb_running) {
2760 ret = -EBUSY;
2761 goto error_unlock;
2762 }
2763 /* add reference of module which cb->dump belongs to */
2764 if (!try_module_get(control->module)) {
2765 ret = -EPROTONOSUPPORT;
2766 goto error_unlock;
2767 }
2768
2769 cb = &nlk->cb;
2770 memset(cb, 0, sizeof(*cb));
2771 cb->dump = control->dump;
2772 cb->done = control->done;
2773 cb->nlh = nlh;
2774 cb->data = control->data;
2775 cb->module = control->module;
2776 cb->min_dump_alloc = control->min_dump_alloc;
2777 cb->skb = skb;
2778
2779 nlk->cb_running = true;
2780
2781 mutex_unlock(nlk->cb_mutex);
2782
2783 ret = netlink_dump(sk);
2784 sock_put(sk);
2785
2786 if (ret)
2787 return ret;
2788
2789 /* We successfully started a dump, by returning -EINTR we
2790 * signal not to send ACK even if it was requested.
2791 */
2792 return -EINTR;
2793
2794error_unlock:
2795 sock_put(sk);
2796 mutex_unlock(nlk->cb_mutex);
2797error_free:
2798 kfree_skb(skb);
2799 return ret;
2800}
2801EXPORT_SYMBOL(__netlink_dump_start);
2802
2803void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2804{
2805 struct sk_buff *skb;
2806 struct nlmsghdr *rep;
2807 struct nlmsgerr *errmsg;
2808 size_t payload = sizeof(*errmsg);
2809
2810 /* error messages get the original request appened */
2811 if (err)
2812 payload += nlmsg_len(nlh);
2813
2814 skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
2815 NETLINK_CB(in_skb).portid, GFP_KERNEL);
2816 if (!skb) {
2817 struct sock *sk;
2818
2819 sk = netlink_lookup(sock_net(in_skb->sk),
2820 in_skb->sk->sk_protocol,
2821 NETLINK_CB(in_skb).portid);
2822 if (sk) {
2823 sk->sk_err = ENOBUFS;
2824 sk->sk_error_report(sk);
2825 sock_put(sk);
2826 }
2827 return;
2828 }
2829
2830 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2831 NLMSG_ERROR, payload, 0);
2832 errmsg = nlmsg_data(rep);
2833 errmsg->error = err;
2834 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
2835 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2836}
2837EXPORT_SYMBOL(netlink_ack);
2838
2839int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2840 struct nlmsghdr *))
2841{
2842 struct nlmsghdr *nlh;
2843 int err;
2844
2845 while (skb->len >= nlmsg_total_size(0)) {
2846 int msglen;
2847
2848 nlh = nlmsg_hdr(skb);
2849 err = 0;
2850
2851 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2852 return 0;
2853
2854 /* Only requests are handled by the kernel */
2855 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2856 goto ack;
2857
2858 /* Skip control messages */
2859 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2860 goto ack;
2861
2862 err = cb(skb, nlh);
2863 if (err == -EINTR)
2864 goto skip;
2865
2866ack:
2867 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2868 netlink_ack(skb, nlh, err);
2869
2870skip:
2871 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2872 if (msglen > skb->len)
2873 msglen = skb->len;
2874 skb_pull(skb, msglen);
2875 }
2876
2877 return 0;
2878}
2879EXPORT_SYMBOL(netlink_rcv_skb);
2880
2881/**
2882 * nlmsg_notify - send a notification netlink message
2883 * @sk: netlink socket to use
2884 * @skb: notification message
2885 * @portid: destination netlink portid for reports or 0
2886 * @group: destination multicast group or 0
2887 * @report: 1 to report back, 0 to disable
2888 * @flags: allocation flags
2889 */
2890int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2891 unsigned int group, int report, gfp_t flags)
2892{
2893 int err = 0;
2894
2895 if (group) {
2896 int exclude_portid = 0;
2897
2898 if (report) {
2899 atomic_inc(&skb->users);
2900 exclude_portid = portid;
2901 }
2902
2903 /* errors reported via destination sk->sk_err, but propagate
2904 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2905 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2906 }
2907
2908 if (report) {
2909 int err2;
2910
2911 err2 = nlmsg_unicast(sk, skb, portid);
2912 if (!err || err == -ESRCH)
2913 err = err2;
2914 }
2915
2916 return err;
2917}
2918EXPORT_SYMBOL(nlmsg_notify);
2919
2920#ifdef CONFIG_PROC_FS
2921struct nl_seq_iter {
2922 struct seq_net_private p;
2923 int link;
2924 int hash_idx;
2925};
2926
2927static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
2928{
2929 struct nl_seq_iter *iter = seq->private;
2930 int i, j;
2931 struct sock *s;
2932 loff_t off = 0;
2933
2934 for (i = 0; i < MAX_LINKS; i++) {
2935 struct nl_portid_hash *hash = &nl_table[i].hash;
2936
2937 for (j = 0; j <= hash->mask; j++) {
2938 sk_for_each(s, &hash->table[j]) {
2939 if (sock_net(s) != seq_file_net(seq))
2940 continue;
2941 if (off == pos) {
2942 iter->link = i;
2943 iter->hash_idx = j;
2944 return s;
2945 }
2946 ++off;
2947 }
2948 }
2949 }
2950 return NULL;
2951}
2952
2953static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
2954 __acquires(nl_table_lock)
2955{
2956 read_lock(&nl_table_lock);
2957 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2958}
2959
2960static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2961{
2962 struct sock *s;
2963 struct nl_seq_iter *iter;
2964 struct net *net;
2965 int i, j;
2966
2967 ++*pos;
2968
2969 if (v == SEQ_START_TOKEN)
2970 return netlink_seq_socket_idx(seq, 0);
2971
2972 net = seq_file_net(seq);
2973 iter = seq->private;
2974 s = v;
2975 do {
2976 s = sk_next(s);
2977 } while (s && !nl_table[s->sk_protocol].compare(net, s));
2978 if (s)
2979 return s;
2980
2981 i = iter->link;
2982 j = iter->hash_idx + 1;
2983
2984 do {
2985 struct nl_portid_hash *hash = &nl_table[i].hash;
2986
2987 for (; j <= hash->mask; j++) {
2988 s = sk_head(&hash->table[j]);
2989
2990 while (s && !nl_table[s->sk_protocol].compare(net, s))
2991 s = sk_next(s);
2992 if (s) {
2993 iter->link = i;
2994 iter->hash_idx = j;
2995 return s;
2996 }
2997 }
2998
2999 j = 0;
3000 } while (++i < MAX_LINKS);
3001
3002 return NULL;
3003}
3004
3005static void netlink_seq_stop(struct seq_file *seq, void *v)
3006 __releases(nl_table_lock)
3007{
3008 read_unlock(&nl_table_lock);
3009}
3010
3011
3012static int netlink_seq_show(struct seq_file *seq, void *v)
3013{
3014 if (v == SEQ_START_TOKEN) {
3015 seq_puts(seq,
3016 "sk Eth Pid Groups "
3017 "Rmem Wmem Dump Locks Drops Inode\n");
3018 } else {
3019 struct sock *s = v;
3020 struct netlink_sock *nlk = nlk_sk(s);
3021
3022 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
3023 s,
3024 s->sk_protocol,
3025 nlk->portid,
3026 nlk->groups ? (u32)nlk->groups[0] : 0,
3027 sk_rmem_alloc_get(s),
3028 sk_wmem_alloc_get(s),
3029 nlk->cb_running,
3030 atomic_read(&s->sk_refcnt),
3031 atomic_read(&s->sk_drops),
3032 sock_i_ino(s)
3033 );
3034
3035 }
3036 return 0;
3037}
3038
3039static const struct seq_operations netlink_seq_ops = {
3040 .start = netlink_seq_start,
3041 .next = netlink_seq_next,
3042 .stop = netlink_seq_stop,
3043 .show = netlink_seq_show,
3044};
3045
3046
3047static int netlink_seq_open(struct inode *inode, struct file *file)
3048{
3049 return seq_open_net(inode, file, &netlink_seq_ops,
3050 sizeof(struct nl_seq_iter));
3051}
3052
3053static const struct file_operations netlink_seq_fops = {
3054 .owner = THIS_MODULE,
3055 .open = netlink_seq_open,
3056 .read = seq_read,
3057 .llseek = seq_lseek,
3058 .release = seq_release_net,
3059};
3060
3061#endif
3062
3063int netlink_register_notifier(struct notifier_block *nb)
3064{
3065 return atomic_notifier_chain_register(&netlink_chain, nb);
3066}
3067EXPORT_SYMBOL(netlink_register_notifier);
3068
3069int netlink_unregister_notifier(struct notifier_block *nb)
3070{
3071 return atomic_notifier_chain_unregister(&netlink_chain, nb);
3072}
3073EXPORT_SYMBOL(netlink_unregister_notifier);
3074
3075static const struct proto_ops netlink_ops = {
3076 .family = PF_NETLINK,
3077 .owner = THIS_MODULE,
3078 .release = netlink_release,
3079 .bind = netlink_bind,
3080 .connect = netlink_connect,
3081 .socketpair = sock_no_socketpair,
3082 .accept = sock_no_accept,
3083 .getname = netlink_getname,
3084 .poll = netlink_poll,
3085 .ioctl = sock_no_ioctl,
3086 .listen = sock_no_listen,
3087 .shutdown = sock_no_shutdown,
3088 .setsockopt = netlink_setsockopt,
3089 .getsockopt = netlink_getsockopt,
3090 .sendmsg = netlink_sendmsg,
3091 .recvmsg = netlink_recvmsg,
3092 .mmap = netlink_mmap,
3093 .sendpage = sock_no_sendpage,
3094};
3095
3096static const struct net_proto_family netlink_family_ops = {
3097 .family = PF_NETLINK,
3098 .create = netlink_create,
3099 .owner = THIS_MODULE, /* for consistency 8) */
3100};
3101
3102static int __net_init netlink_net_init(struct net *net)
3103{
3104#ifdef CONFIG_PROC_FS
3105 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
3106 return -ENOMEM;
3107#endif
3108 return 0;
3109}
3110
3111static void __net_exit netlink_net_exit(struct net *net)
3112{
3113#ifdef CONFIG_PROC_FS
3114 remove_proc_entry("netlink", net->proc_net);
3115#endif
3116}
3117
3118static void __init netlink_add_usersock_entry(void)
3119{
3120 struct listeners *listeners;
3121 int groups = 32;
3122
3123 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
3124 if (!listeners)
3125 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
3126
3127 netlink_table_grab();
3128
3129 nl_table[NETLINK_USERSOCK].groups = groups;
3130 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
3131 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
3132 nl_table[NETLINK_USERSOCK].registered = 1;
3133 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
3134
3135 netlink_table_ungrab();
3136}
3137
3138static struct pernet_operations __net_initdata netlink_net_ops = {
3139 .init = netlink_net_init,
3140 .exit = netlink_net_exit,
3141};
3142
3143static int __init netlink_proto_init(void)
3144{
3145 int i;
3146 unsigned long limit;
3147 unsigned int order;
3148 int err = proto_register(&netlink_proto, 0);
3149
3150 if (err != 0)
3151 goto out;
3152
3153 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
3154
3155 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
3156 if (!nl_table)
3157 goto panic;
3158
3159 if (totalram_pages >= (128 * 1024))
3160 limit = totalram_pages >> (21 - PAGE_SHIFT);
3161 else
3162 limit = totalram_pages >> (23 - PAGE_SHIFT);
3163
3164 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
3165 limit = (1UL << order) / sizeof(struct hlist_head);
3166 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
3167
3168 for (i = 0; i < MAX_LINKS; i++) {
3169 struct nl_portid_hash *hash = &nl_table[i].hash;
3170
3171 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
3172 if (!hash->table) {
3173 while (i-- > 0)
3174 nl_portid_hash_free(nl_table[i].hash.table,
3175 1 * sizeof(*hash->table));
3176 kfree(nl_table);
3177 goto panic;
3178 }
3179 hash->max_shift = order;
3180 hash->shift = 0;
3181 hash->mask = 0;
3182 hash->rehash_time = jiffies;
3183
3184 nl_table[i].compare = netlink_compare;
3185 }
3186
3187 INIT_LIST_HEAD(&netlink_tap_all);
3188
3189 netlink_add_usersock_entry();
3190
3191 sock_register(&netlink_family_ops);
3192 register_pernet_subsys(&netlink_net_ops);
3193 /* The netlink device handler may be needed early. */
3194 rtnetlink_init();
3195out:
3196 return err;
3197panic:
3198 panic("netlink_init: Cannot allocate nl_table\n");
3199}
3200
3201core_initcall(netlink_proto_init);