Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NETLINK Kernel-user communication protocol.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
8 *
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
19 */
20
21#include <linux/module.h>
22
23#include <linux/bpf.h>
24#include <linux/capability.h>
25#include <linux/kernel.h>
26#include <linux/filter.h>
27#include <linux/init.h>
28#include <linux/signal.h>
29#include <linux/sched.h>
30#include <linux/errno.h>
31#include <linux/string.h>
32#include <linux/stat.h>
33#include <linux/socket.h>
34#include <linux/un.h>
35#include <linux/fcntl.h>
36#include <linux/termios.h>
37#include <linux/sockios.h>
38#include <linux/net.h>
39#include <linux/fs.h>
40#include <linux/slab.h>
41#include <linux/uaccess.h>
42#include <linux/skbuff.h>
43#include <linux/netdevice.h>
44#include <linux/rtnetlink.h>
45#include <linux/proc_fs.h>
46#include <linux/seq_file.h>
47#include <linux/notifier.h>
48#include <linux/security.h>
49#include <linux/jhash.h>
50#include <linux/jiffies.h>
51#include <linux/random.h>
52#include <linux/bitops.h>
53#include <linux/mm.h>
54#include <linux/types.h>
55#include <linux/audit.h>
56#include <linux/mutex.h>
57#include <linux/vmalloc.h>
58#include <linux/if_arp.h>
59#include <linux/rhashtable.h>
60#include <asm/cacheflush.h>
61#include <linux/hash.h>
62#include <linux/genetlink.h>
63#include <linux/net_namespace.h>
64#include <linux/nospec.h>
65#include <linux/btf_ids.h>
66
67#include <net/net_namespace.h>
68#include <net/netns/generic.h>
69#include <net/sock.h>
70#include <net/scm.h>
71#include <net/netlink.h>
72#define CREATE_TRACE_POINTS
73#include <trace/events/netlink.h>
74
75#include "af_netlink.h"
76
77struct listeners {
78 struct rcu_head rcu;
79 unsigned long masks[];
80};
81
82/* state bits */
83#define NETLINK_S_CONGESTED 0x0
84
85static inline int netlink_is_kernel(struct sock *sk)
86{
87 return nlk_test_bit(KERNEL_SOCKET, sk);
88}
89
90struct netlink_table *nl_table __read_mostly;
91EXPORT_SYMBOL_GPL(nl_table);
92
93static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94
95static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96
97static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
98 "nlk_cb_mutex-ROUTE",
99 "nlk_cb_mutex-1",
100 "nlk_cb_mutex-USERSOCK",
101 "nlk_cb_mutex-FIREWALL",
102 "nlk_cb_mutex-SOCK_DIAG",
103 "nlk_cb_mutex-NFLOG",
104 "nlk_cb_mutex-XFRM",
105 "nlk_cb_mutex-SELINUX",
106 "nlk_cb_mutex-ISCSI",
107 "nlk_cb_mutex-AUDIT",
108 "nlk_cb_mutex-FIB_LOOKUP",
109 "nlk_cb_mutex-CONNECTOR",
110 "nlk_cb_mutex-NETFILTER",
111 "nlk_cb_mutex-IP6_FW",
112 "nlk_cb_mutex-DNRTMSG",
113 "nlk_cb_mutex-KOBJECT_UEVENT",
114 "nlk_cb_mutex-GENERIC",
115 "nlk_cb_mutex-17",
116 "nlk_cb_mutex-SCSITRANSPORT",
117 "nlk_cb_mutex-ECRYPTFS",
118 "nlk_cb_mutex-RDMA",
119 "nlk_cb_mutex-CRYPTO",
120 "nlk_cb_mutex-SMC",
121 "nlk_cb_mutex-23",
122 "nlk_cb_mutex-24",
123 "nlk_cb_mutex-25",
124 "nlk_cb_mutex-26",
125 "nlk_cb_mutex-27",
126 "nlk_cb_mutex-28",
127 "nlk_cb_mutex-29",
128 "nlk_cb_mutex-30",
129 "nlk_cb_mutex-31",
130 "nlk_cb_mutex-MAX_LINKS"
131};
132
133static int netlink_dump(struct sock *sk);
134
135/* nl_table locking explained:
136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
137 * and removal are protected with per bucket lock while using RCU list
138 * modification primitives and may run in parallel to RCU protected lookups.
139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
140 * been acquired * either during or after the socket has been removed from
141 * the list and after an RCU grace period.
142 */
143DEFINE_RWLOCK(nl_table_lock);
144EXPORT_SYMBOL_GPL(nl_table_lock);
145static atomic_t nl_table_users = ATOMIC_INIT(0);
146
147#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
148
149static BLOCKING_NOTIFIER_HEAD(netlink_chain);
150
151
152static const struct rhashtable_params netlink_rhashtable_params;
153
154void do_trace_netlink_extack(const char *msg)
155{
156 trace_netlink_extack(msg);
157}
158EXPORT_SYMBOL(do_trace_netlink_extack);
159
160static inline u32 netlink_group_mask(u32 group)
161{
162 if (group > 32)
163 return 0;
164 return group ? 1 << (group - 1) : 0;
165}
166
167static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
168 gfp_t gfp_mask)
169{
170 unsigned int len = skb->len;
171 struct sk_buff *new;
172
173 new = alloc_skb(len, gfp_mask);
174 if (new == NULL)
175 return NULL;
176
177 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
178 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
179 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
180
181 skb_put_data(new, skb->data, len);
182 return new;
183}
184
185static unsigned int netlink_tap_net_id;
186
187struct netlink_tap_net {
188 struct list_head netlink_tap_all;
189 struct mutex netlink_tap_lock;
190};
191
192int netlink_add_tap(struct netlink_tap *nt)
193{
194 struct net *net = dev_net(nt->dev);
195 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
196
197 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
198 return -EINVAL;
199
200 mutex_lock(&nn->netlink_tap_lock);
201 list_add_rcu(&nt->list, &nn->netlink_tap_all);
202 mutex_unlock(&nn->netlink_tap_lock);
203
204 __module_get(nt->module);
205
206 return 0;
207}
208EXPORT_SYMBOL_GPL(netlink_add_tap);
209
210static int __netlink_remove_tap(struct netlink_tap *nt)
211{
212 struct net *net = dev_net(nt->dev);
213 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
214 bool found = false;
215 struct netlink_tap *tmp;
216
217 mutex_lock(&nn->netlink_tap_lock);
218
219 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
220 if (nt == tmp) {
221 list_del_rcu(&nt->list);
222 found = true;
223 goto out;
224 }
225 }
226
227 pr_warn("__netlink_remove_tap: %p not found\n", nt);
228out:
229 mutex_unlock(&nn->netlink_tap_lock);
230
231 if (found)
232 module_put(nt->module);
233
234 return found ? 0 : -ENODEV;
235}
236
237int netlink_remove_tap(struct netlink_tap *nt)
238{
239 int ret;
240
241 ret = __netlink_remove_tap(nt);
242 synchronize_net();
243
244 return ret;
245}
246EXPORT_SYMBOL_GPL(netlink_remove_tap);
247
248static __net_init int netlink_tap_init_net(struct net *net)
249{
250 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
251
252 INIT_LIST_HEAD(&nn->netlink_tap_all);
253 mutex_init(&nn->netlink_tap_lock);
254 return 0;
255}
256
257static struct pernet_operations netlink_tap_net_ops = {
258 .init = netlink_tap_init_net,
259 .id = &netlink_tap_net_id,
260 .size = sizeof(struct netlink_tap_net),
261};
262
263static bool netlink_filter_tap(const struct sk_buff *skb)
264{
265 struct sock *sk = skb->sk;
266
267 /* We take the more conservative approach and
268 * whitelist socket protocols that may pass.
269 */
270 switch (sk->sk_protocol) {
271 case NETLINK_ROUTE:
272 case NETLINK_USERSOCK:
273 case NETLINK_SOCK_DIAG:
274 case NETLINK_NFLOG:
275 case NETLINK_XFRM:
276 case NETLINK_FIB_LOOKUP:
277 case NETLINK_NETFILTER:
278 case NETLINK_GENERIC:
279 return true;
280 }
281
282 return false;
283}
284
285static int __netlink_deliver_tap_skb(struct sk_buff *skb,
286 struct net_device *dev)
287{
288 struct sk_buff *nskb;
289 struct sock *sk = skb->sk;
290 int ret = -ENOMEM;
291
292 if (!net_eq(dev_net(dev), sock_net(sk)))
293 return 0;
294
295 dev_hold(dev);
296
297 if (is_vmalloc_addr(skb->head))
298 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
299 else
300 nskb = skb_clone(skb, GFP_ATOMIC);
301 if (nskb) {
302 nskb->dev = dev;
303 nskb->protocol = htons((u16) sk->sk_protocol);
304 nskb->pkt_type = netlink_is_kernel(sk) ?
305 PACKET_KERNEL : PACKET_USER;
306 skb_reset_network_header(nskb);
307 ret = dev_queue_xmit(nskb);
308 if (unlikely(ret > 0))
309 ret = net_xmit_errno(ret);
310 }
311
312 dev_put(dev);
313 return ret;
314}
315
316static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
317{
318 int ret;
319 struct netlink_tap *tmp;
320
321 if (!netlink_filter_tap(skb))
322 return;
323
324 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
325 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
326 if (unlikely(ret))
327 break;
328 }
329}
330
331static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
332{
333 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
334
335 rcu_read_lock();
336
337 if (unlikely(!list_empty(&nn->netlink_tap_all)))
338 __netlink_deliver_tap(skb, nn);
339
340 rcu_read_unlock();
341}
342
343static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
344 struct sk_buff *skb)
345{
346 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
347 netlink_deliver_tap(sock_net(dst), skb);
348}
349
350static void netlink_overrun(struct sock *sk)
351{
352 if (!nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
353 if (!test_and_set_bit(NETLINK_S_CONGESTED,
354 &nlk_sk(sk)->state)) {
355 WRITE_ONCE(sk->sk_err, ENOBUFS);
356 sk_error_report(sk);
357 }
358 }
359 atomic_inc(&sk->sk_drops);
360}
361
362static void netlink_rcv_wake(struct sock *sk)
363{
364 struct netlink_sock *nlk = nlk_sk(sk);
365
366 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
367 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
368 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
369 wake_up_interruptible(&nlk->wait);
370}
371
372static void netlink_skb_destructor(struct sk_buff *skb)
373{
374 if (is_vmalloc_addr(skb->head)) {
375 if (!skb->cloned ||
376 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
377 vfree_atomic(skb->head);
378
379 skb->head = NULL;
380 }
381 if (skb->sk != NULL)
382 sock_rfree(skb);
383}
384
385static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
386{
387 WARN_ON(skb->sk != NULL);
388 skb->sk = sk;
389 skb->destructor = netlink_skb_destructor;
390 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
391 sk_mem_charge(sk, skb->truesize);
392}
393
394static void netlink_sock_destruct(struct sock *sk)
395{
396 struct netlink_sock *nlk = nlk_sk(sk);
397
398 if (nlk->cb_running) {
399 if (nlk->cb.done)
400 nlk->cb.done(&nlk->cb);
401 module_put(nlk->cb.module);
402 kfree_skb(nlk->cb.skb);
403 }
404
405 skb_queue_purge(&sk->sk_receive_queue);
406
407 if (!sock_flag(sk, SOCK_DEAD)) {
408 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
409 return;
410 }
411
412 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
413 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
414 WARN_ON(nlk_sk(sk)->groups);
415}
416
417static void netlink_sock_destruct_work(struct work_struct *work)
418{
419 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
420 work);
421
422 sk_free(&nlk->sk);
423}
424
425/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
426 * SMP. Look, when several writers sleep and reader wakes them up, all but one
427 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
428 * this, _but_ remember, it adds useless work on UP machines.
429 */
430
431void netlink_table_grab(void)
432 __acquires(nl_table_lock)
433{
434 might_sleep();
435
436 write_lock_irq(&nl_table_lock);
437
438 if (atomic_read(&nl_table_users)) {
439 DECLARE_WAITQUEUE(wait, current);
440
441 add_wait_queue_exclusive(&nl_table_wait, &wait);
442 for (;;) {
443 set_current_state(TASK_UNINTERRUPTIBLE);
444 if (atomic_read(&nl_table_users) == 0)
445 break;
446 write_unlock_irq(&nl_table_lock);
447 schedule();
448 write_lock_irq(&nl_table_lock);
449 }
450
451 __set_current_state(TASK_RUNNING);
452 remove_wait_queue(&nl_table_wait, &wait);
453 }
454}
455
456void netlink_table_ungrab(void)
457 __releases(nl_table_lock)
458{
459 write_unlock_irq(&nl_table_lock);
460 wake_up(&nl_table_wait);
461}
462
463static inline void
464netlink_lock_table(void)
465{
466 unsigned long flags;
467
468 /* read_lock() synchronizes us to netlink_table_grab */
469
470 read_lock_irqsave(&nl_table_lock, flags);
471 atomic_inc(&nl_table_users);
472 read_unlock_irqrestore(&nl_table_lock, flags);
473}
474
475static inline void
476netlink_unlock_table(void)
477{
478 if (atomic_dec_and_test(&nl_table_users))
479 wake_up(&nl_table_wait);
480}
481
482struct netlink_compare_arg
483{
484 possible_net_t pnet;
485 u32 portid;
486};
487
488/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
489#define netlink_compare_arg_len \
490 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
491
492static inline int netlink_compare(struct rhashtable_compare_arg *arg,
493 const void *ptr)
494{
495 const struct netlink_compare_arg *x = arg->key;
496 const struct netlink_sock *nlk = ptr;
497
498 return nlk->portid != x->portid ||
499 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
500}
501
502static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
503 struct net *net, u32 portid)
504{
505 memset(arg, 0, sizeof(*arg));
506 write_pnet(&arg->pnet, net);
507 arg->portid = portid;
508}
509
510static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
511 struct net *net)
512{
513 struct netlink_compare_arg arg;
514
515 netlink_compare_arg_init(&arg, net, portid);
516 return rhashtable_lookup_fast(&table->hash, &arg,
517 netlink_rhashtable_params);
518}
519
520static int __netlink_insert(struct netlink_table *table, struct sock *sk)
521{
522 struct netlink_compare_arg arg;
523
524 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
525 return rhashtable_lookup_insert_key(&table->hash, &arg,
526 &nlk_sk(sk)->node,
527 netlink_rhashtable_params);
528}
529
530static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
531{
532 struct netlink_table *table = &nl_table[protocol];
533 struct sock *sk;
534
535 rcu_read_lock();
536 sk = __netlink_lookup(table, portid, net);
537 if (sk)
538 sock_hold(sk);
539 rcu_read_unlock();
540
541 return sk;
542}
543
544static const struct proto_ops netlink_ops;
545
546static void
547netlink_update_listeners(struct sock *sk)
548{
549 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
550 unsigned long mask;
551 unsigned int i;
552 struct listeners *listeners;
553
554 listeners = nl_deref_protected(tbl->listeners);
555 if (!listeners)
556 return;
557
558 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
559 mask = 0;
560 sk_for_each_bound(sk, &tbl->mc_list) {
561 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
562 mask |= nlk_sk(sk)->groups[i];
563 }
564 listeners->masks[i] = mask;
565 }
566 /* this function is only called with the netlink table "grabbed", which
567 * makes sure updates are visible before bind or setsockopt return. */
568}
569
570static int netlink_insert(struct sock *sk, u32 portid)
571{
572 struct netlink_table *table = &nl_table[sk->sk_protocol];
573 int err;
574
575 lock_sock(sk);
576
577 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
578 if (nlk_sk(sk)->bound)
579 goto err;
580
581 /* portid can be read locklessly from netlink_getname(). */
582 WRITE_ONCE(nlk_sk(sk)->portid, portid);
583
584 sock_hold(sk);
585
586 err = __netlink_insert(table, sk);
587 if (err) {
588 /* In case the hashtable backend returns with -EBUSY
589 * from here, it must not escape to the caller.
590 */
591 if (unlikely(err == -EBUSY))
592 err = -EOVERFLOW;
593 if (err == -EEXIST)
594 err = -EADDRINUSE;
595 sock_put(sk);
596 goto err;
597 }
598
599 /* We need to ensure that the socket is hashed and visible. */
600 smp_wmb();
601 /* Paired with lockless reads from netlink_bind(),
602 * netlink_connect() and netlink_sendmsg().
603 */
604 WRITE_ONCE(nlk_sk(sk)->bound, portid);
605
606err:
607 release_sock(sk);
608 return err;
609}
610
611static void netlink_remove(struct sock *sk)
612{
613 struct netlink_table *table;
614
615 table = &nl_table[sk->sk_protocol];
616 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
617 netlink_rhashtable_params)) {
618 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
619 __sock_put(sk);
620 }
621
622 netlink_table_grab();
623 if (nlk_sk(sk)->subscriptions) {
624 __sk_del_bind_node(sk);
625 netlink_update_listeners(sk);
626 }
627 if (sk->sk_protocol == NETLINK_GENERIC)
628 atomic_inc(&genl_sk_destructing_cnt);
629 netlink_table_ungrab();
630}
631
632static struct proto netlink_proto = {
633 .name = "NETLINK",
634 .owner = THIS_MODULE,
635 .obj_size = sizeof(struct netlink_sock),
636};
637
638static int __netlink_create(struct net *net, struct socket *sock,
639 struct mutex *cb_mutex, int protocol,
640 int kern)
641{
642 struct sock *sk;
643 struct netlink_sock *nlk;
644
645 sock->ops = &netlink_ops;
646
647 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
648 if (!sk)
649 return -ENOMEM;
650
651 sock_init_data(sock, sk);
652
653 nlk = nlk_sk(sk);
654 if (cb_mutex) {
655 nlk->cb_mutex = cb_mutex;
656 } else {
657 nlk->cb_mutex = &nlk->cb_def_mutex;
658 mutex_init(nlk->cb_mutex);
659 lockdep_set_class_and_name(nlk->cb_mutex,
660 nlk_cb_mutex_keys + protocol,
661 nlk_cb_mutex_key_strings[protocol]);
662 }
663 init_waitqueue_head(&nlk->wait);
664
665 sk->sk_destruct = netlink_sock_destruct;
666 sk->sk_protocol = protocol;
667 return 0;
668}
669
670static int netlink_create(struct net *net, struct socket *sock, int protocol,
671 int kern)
672{
673 struct module *module = NULL;
674 struct mutex *cb_mutex;
675 struct netlink_sock *nlk;
676 int (*bind)(struct net *net, int group);
677 void (*unbind)(struct net *net, int group);
678 void (*release)(struct sock *sock, unsigned long *groups);
679 int err = 0;
680
681 sock->state = SS_UNCONNECTED;
682
683 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
684 return -ESOCKTNOSUPPORT;
685
686 if (protocol < 0 || protocol >= MAX_LINKS)
687 return -EPROTONOSUPPORT;
688 protocol = array_index_nospec(protocol, MAX_LINKS);
689
690 netlink_lock_table();
691#ifdef CONFIG_MODULES
692 if (!nl_table[protocol].registered) {
693 netlink_unlock_table();
694 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
695 netlink_lock_table();
696 }
697#endif
698 if (nl_table[protocol].registered &&
699 try_module_get(nl_table[protocol].module))
700 module = nl_table[protocol].module;
701 else
702 err = -EPROTONOSUPPORT;
703 cb_mutex = nl_table[protocol].cb_mutex;
704 bind = nl_table[protocol].bind;
705 unbind = nl_table[protocol].unbind;
706 release = nl_table[protocol].release;
707 netlink_unlock_table();
708
709 if (err < 0)
710 goto out;
711
712 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
713 if (err < 0)
714 goto out_module;
715
716 sock_prot_inuse_add(net, &netlink_proto, 1);
717
718 nlk = nlk_sk(sock->sk);
719 nlk->module = module;
720 nlk->netlink_bind = bind;
721 nlk->netlink_unbind = unbind;
722 nlk->netlink_release = release;
723out:
724 return err;
725
726out_module:
727 module_put(module);
728 goto out;
729}
730
731static void deferred_put_nlk_sk(struct rcu_head *head)
732{
733 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
734 struct sock *sk = &nlk->sk;
735
736 kfree(nlk->groups);
737 nlk->groups = NULL;
738
739 if (!refcount_dec_and_test(&sk->sk_refcnt))
740 return;
741
742 if (nlk->cb_running && nlk->cb.done) {
743 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
744 schedule_work(&nlk->work);
745 return;
746 }
747
748 sk_free(sk);
749}
750
751static int netlink_release(struct socket *sock)
752{
753 struct sock *sk = sock->sk;
754 struct netlink_sock *nlk;
755
756 if (!sk)
757 return 0;
758
759 netlink_remove(sk);
760 sock_orphan(sk);
761 nlk = nlk_sk(sk);
762
763 /*
764 * OK. Socket is unlinked, any packets that arrive now
765 * will be purged.
766 */
767 if (nlk->netlink_release)
768 nlk->netlink_release(sk, nlk->groups);
769
770 /* must not acquire netlink_table_lock in any way again before unbind
771 * and notifying genetlink is done as otherwise it might deadlock
772 */
773 if (nlk->netlink_unbind) {
774 int i;
775
776 for (i = 0; i < nlk->ngroups; i++)
777 if (test_bit(i, nlk->groups))
778 nlk->netlink_unbind(sock_net(sk), i + 1);
779 }
780 if (sk->sk_protocol == NETLINK_GENERIC &&
781 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
782 wake_up(&genl_sk_destructing_waitq);
783
784 sock->sk = NULL;
785 wake_up_interruptible_all(&nlk->wait);
786
787 skb_queue_purge(&sk->sk_write_queue);
788
789 if (nlk->portid && nlk->bound) {
790 struct netlink_notify n = {
791 .net = sock_net(sk),
792 .protocol = sk->sk_protocol,
793 .portid = nlk->portid,
794 };
795 blocking_notifier_call_chain(&netlink_chain,
796 NETLINK_URELEASE, &n);
797 }
798
799 module_put(nlk->module);
800
801 if (netlink_is_kernel(sk)) {
802 netlink_table_grab();
803 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
804 if (--nl_table[sk->sk_protocol].registered == 0) {
805 struct listeners *old;
806
807 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
808 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
809 kfree_rcu(old, rcu);
810 nl_table[sk->sk_protocol].module = NULL;
811 nl_table[sk->sk_protocol].bind = NULL;
812 nl_table[sk->sk_protocol].unbind = NULL;
813 nl_table[sk->sk_protocol].flags = 0;
814 nl_table[sk->sk_protocol].registered = 0;
815 }
816 netlink_table_ungrab();
817 }
818
819 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
820
821 /* Because struct net might disappear soon, do not keep a pointer. */
822 if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
823 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
824 /* Because of deferred_put_nlk_sk and use of work queue,
825 * it is possible netns will be freed before this socket.
826 */
827 sock_net_set(sk, &init_net);
828 __netns_tracker_alloc(&init_net, &sk->ns_tracker,
829 false, GFP_KERNEL);
830 }
831 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
832 return 0;
833}
834
835static int netlink_autobind(struct socket *sock)
836{
837 struct sock *sk = sock->sk;
838 struct net *net = sock_net(sk);
839 struct netlink_table *table = &nl_table[sk->sk_protocol];
840 s32 portid = task_tgid_vnr(current);
841 int err;
842 s32 rover = -4096;
843 bool ok;
844
845retry:
846 cond_resched();
847 rcu_read_lock();
848 ok = !__netlink_lookup(table, portid, net);
849 rcu_read_unlock();
850 if (!ok) {
851 /* Bind collision, search negative portid values. */
852 if (rover == -4096)
853 /* rover will be in range [S32_MIN, -4097] */
854 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
855 else if (rover >= -4096)
856 rover = -4097;
857 portid = rover--;
858 goto retry;
859 }
860
861 err = netlink_insert(sk, portid);
862 if (err == -EADDRINUSE)
863 goto retry;
864
865 /* If 2 threads race to autobind, that is fine. */
866 if (err == -EBUSY)
867 err = 0;
868
869 return err;
870}
871
872/**
873 * __netlink_ns_capable - General netlink message capability test
874 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
875 * @user_ns: The user namespace of the capability to use
876 * @cap: The capability to use
877 *
878 * Test to see if the opener of the socket we received the message
879 * from had when the netlink socket was created and the sender of the
880 * message has the capability @cap in the user namespace @user_ns.
881 */
882bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
883 struct user_namespace *user_ns, int cap)
884{
885 return ((nsp->flags & NETLINK_SKB_DST) ||
886 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
887 ns_capable(user_ns, cap);
888}
889EXPORT_SYMBOL(__netlink_ns_capable);
890
891/**
892 * netlink_ns_capable - General netlink message capability test
893 * @skb: socket buffer holding a netlink command from userspace
894 * @user_ns: The user namespace of the capability to use
895 * @cap: The capability to use
896 *
897 * Test to see if the opener of the socket we received the message
898 * from had when the netlink socket was created and the sender of the
899 * message has the capability @cap in the user namespace @user_ns.
900 */
901bool netlink_ns_capable(const struct sk_buff *skb,
902 struct user_namespace *user_ns, int cap)
903{
904 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
905}
906EXPORT_SYMBOL(netlink_ns_capable);
907
908/**
909 * netlink_capable - Netlink global message capability test
910 * @skb: socket buffer holding a netlink command from userspace
911 * @cap: The capability to use
912 *
913 * Test to see if the opener of the socket we received the message
914 * from had when the netlink socket was created and the sender of the
915 * message has the capability @cap in all user namespaces.
916 */
917bool netlink_capable(const struct sk_buff *skb, int cap)
918{
919 return netlink_ns_capable(skb, &init_user_ns, cap);
920}
921EXPORT_SYMBOL(netlink_capable);
922
923/**
924 * netlink_net_capable - Netlink network namespace message capability test
925 * @skb: socket buffer holding a netlink command from userspace
926 * @cap: The capability to use
927 *
928 * Test to see if the opener of the socket we received the message
929 * from had when the netlink socket was created and the sender of the
930 * message has the capability @cap over the network namespace of
931 * the socket we received the message from.
932 */
933bool netlink_net_capable(const struct sk_buff *skb, int cap)
934{
935 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
936}
937EXPORT_SYMBOL(netlink_net_capable);
938
939static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
940{
941 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
942 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
943}
944
945static void
946netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
947{
948 struct netlink_sock *nlk = nlk_sk(sk);
949
950 if (nlk->subscriptions && !subscriptions)
951 __sk_del_bind_node(sk);
952 else if (!nlk->subscriptions && subscriptions)
953 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
954 nlk->subscriptions = subscriptions;
955}
956
957static int netlink_realloc_groups(struct sock *sk)
958{
959 struct netlink_sock *nlk = nlk_sk(sk);
960 unsigned int groups;
961 unsigned long *new_groups;
962 int err = 0;
963
964 netlink_table_grab();
965
966 groups = nl_table[sk->sk_protocol].groups;
967 if (!nl_table[sk->sk_protocol].registered) {
968 err = -ENOENT;
969 goto out_unlock;
970 }
971
972 if (nlk->ngroups >= groups)
973 goto out_unlock;
974
975 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
976 if (new_groups == NULL) {
977 err = -ENOMEM;
978 goto out_unlock;
979 }
980 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
981 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
982
983 nlk->groups = new_groups;
984 nlk->ngroups = groups;
985 out_unlock:
986 netlink_table_ungrab();
987 return err;
988}
989
990static void netlink_undo_bind(int group, long unsigned int groups,
991 struct sock *sk)
992{
993 struct netlink_sock *nlk = nlk_sk(sk);
994 int undo;
995
996 if (!nlk->netlink_unbind)
997 return;
998
999 for (undo = 0; undo < group; undo++)
1000 if (test_bit(undo, &groups))
1001 nlk->netlink_unbind(sock_net(sk), undo + 1);
1002}
1003
1004static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1005 int addr_len)
1006{
1007 struct sock *sk = sock->sk;
1008 struct net *net = sock_net(sk);
1009 struct netlink_sock *nlk = nlk_sk(sk);
1010 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1011 int err = 0;
1012 unsigned long groups;
1013 bool bound;
1014
1015 if (addr_len < sizeof(struct sockaddr_nl))
1016 return -EINVAL;
1017
1018 if (nladdr->nl_family != AF_NETLINK)
1019 return -EINVAL;
1020 groups = nladdr->nl_groups;
1021
1022 /* Only superuser is allowed to listen multicasts */
1023 if (groups) {
1024 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1025 return -EPERM;
1026 err = netlink_realloc_groups(sk);
1027 if (err)
1028 return err;
1029 }
1030
1031 if (nlk->ngroups < BITS_PER_LONG)
1032 groups &= (1UL << nlk->ngroups) - 1;
1033
1034 /* Paired with WRITE_ONCE() in netlink_insert() */
1035 bound = READ_ONCE(nlk->bound);
1036 if (bound) {
1037 /* Ensure nlk->portid is up-to-date. */
1038 smp_rmb();
1039
1040 if (nladdr->nl_pid != nlk->portid)
1041 return -EINVAL;
1042 }
1043
1044 if (nlk->netlink_bind && groups) {
1045 int group;
1046
1047 /* nl_groups is a u32, so cap the maximum groups we can bind */
1048 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1049 if (!test_bit(group, &groups))
1050 continue;
1051 err = nlk->netlink_bind(net, group + 1);
1052 if (!err)
1053 continue;
1054 netlink_undo_bind(group, groups, sk);
1055 return err;
1056 }
1057 }
1058
1059 /* No need for barriers here as we return to user-space without
1060 * using any of the bound attributes.
1061 */
1062 netlink_lock_table();
1063 if (!bound) {
1064 err = nladdr->nl_pid ?
1065 netlink_insert(sk, nladdr->nl_pid) :
1066 netlink_autobind(sock);
1067 if (err) {
1068 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1069 goto unlock;
1070 }
1071 }
1072
1073 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1074 goto unlock;
1075 netlink_unlock_table();
1076
1077 netlink_table_grab();
1078 netlink_update_subscriptions(sk, nlk->subscriptions +
1079 hweight32(groups) -
1080 hweight32(nlk->groups[0]));
1081 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1082 netlink_update_listeners(sk);
1083 netlink_table_ungrab();
1084
1085 return 0;
1086
1087unlock:
1088 netlink_unlock_table();
1089 return err;
1090}
1091
1092static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1093 int alen, int flags)
1094{
1095 int err = 0;
1096 struct sock *sk = sock->sk;
1097 struct netlink_sock *nlk = nlk_sk(sk);
1098 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1099
1100 if (alen < sizeof(addr->sa_family))
1101 return -EINVAL;
1102
1103 if (addr->sa_family == AF_UNSPEC) {
1104 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1105 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1106 /* dst_portid and dst_group can be read locklessly */
1107 WRITE_ONCE(nlk->dst_portid, 0);
1108 WRITE_ONCE(nlk->dst_group, 0);
1109 return 0;
1110 }
1111 if (addr->sa_family != AF_NETLINK)
1112 return -EINVAL;
1113
1114 if (alen < sizeof(struct sockaddr_nl))
1115 return -EINVAL;
1116
1117 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1118 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1119 return -EPERM;
1120
1121 /* No need for barriers here as we return to user-space without
1122 * using any of the bound attributes.
1123 * Paired with WRITE_ONCE() in netlink_insert().
1124 */
1125 if (!READ_ONCE(nlk->bound))
1126 err = netlink_autobind(sock);
1127
1128 if (err == 0) {
1129 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1130 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1131 /* dst_portid and dst_group can be read locklessly */
1132 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1133 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1134 }
1135
1136 return err;
1137}
1138
1139static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1140 int peer)
1141{
1142 struct sock *sk = sock->sk;
1143 struct netlink_sock *nlk = nlk_sk(sk);
1144 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1145
1146 nladdr->nl_family = AF_NETLINK;
1147 nladdr->nl_pad = 0;
1148
1149 if (peer) {
1150 /* Paired with WRITE_ONCE() in netlink_connect() */
1151 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1152 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1153 } else {
1154 /* Paired with WRITE_ONCE() in netlink_insert() */
1155 nladdr->nl_pid = READ_ONCE(nlk->portid);
1156 netlink_lock_table();
1157 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1158 netlink_unlock_table();
1159 }
1160 return sizeof(*nladdr);
1161}
1162
1163static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1164 unsigned long arg)
1165{
1166 /* try to hand this ioctl down to the NIC drivers.
1167 */
1168 return -ENOIOCTLCMD;
1169}
1170
1171static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1172{
1173 struct sock *sock;
1174 struct netlink_sock *nlk;
1175
1176 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1177 if (!sock)
1178 return ERR_PTR(-ECONNREFUSED);
1179
1180 /* Don't bother queuing skb if kernel socket has no input function */
1181 nlk = nlk_sk(sock);
1182 /* dst_portid and sk_state can be changed in netlink_connect() */
1183 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1184 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1185 sock_put(sock);
1186 return ERR_PTR(-ECONNREFUSED);
1187 }
1188 return sock;
1189}
1190
1191struct sock *netlink_getsockbyfilp(struct file *filp)
1192{
1193 struct inode *inode = file_inode(filp);
1194 struct sock *sock;
1195
1196 if (!S_ISSOCK(inode->i_mode))
1197 return ERR_PTR(-ENOTSOCK);
1198
1199 sock = SOCKET_I(inode)->sk;
1200 if (sock->sk_family != AF_NETLINK)
1201 return ERR_PTR(-EINVAL);
1202
1203 sock_hold(sock);
1204 return sock;
1205}
1206
1207struct sk_buff *netlink_alloc_large_skb(unsigned int size, int broadcast)
1208{
1209 struct sk_buff *skb;
1210 void *data;
1211
1212 if (size <= NLMSG_GOODSIZE || broadcast)
1213 return alloc_skb(size, GFP_KERNEL);
1214
1215 size = SKB_DATA_ALIGN(size) +
1216 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1217
1218 data = vmalloc(size);
1219 if (data == NULL)
1220 return NULL;
1221
1222 skb = __build_skb(data, size);
1223 if (skb == NULL)
1224 vfree(data);
1225 else
1226 skb->destructor = netlink_skb_destructor;
1227
1228 return skb;
1229}
1230
1231/*
1232 * Attach a skb to a netlink socket.
1233 * The caller must hold a reference to the destination socket. On error, the
1234 * reference is dropped. The skb is not send to the destination, just all
1235 * all error checks are performed and memory in the queue is reserved.
1236 * Return values:
1237 * < 0: error. skb freed, reference to sock dropped.
1238 * 0: continue
1239 * 1: repeat lookup - reference dropped while waiting for socket memory.
1240 */
1241int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1242 long *timeo, struct sock *ssk)
1243{
1244 struct netlink_sock *nlk;
1245
1246 nlk = nlk_sk(sk);
1247
1248 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1249 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1250 DECLARE_WAITQUEUE(wait, current);
1251 if (!*timeo) {
1252 if (!ssk || netlink_is_kernel(ssk))
1253 netlink_overrun(sk);
1254 sock_put(sk);
1255 kfree_skb(skb);
1256 return -EAGAIN;
1257 }
1258
1259 __set_current_state(TASK_INTERRUPTIBLE);
1260 add_wait_queue(&nlk->wait, &wait);
1261
1262 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1263 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1264 !sock_flag(sk, SOCK_DEAD))
1265 *timeo = schedule_timeout(*timeo);
1266
1267 __set_current_state(TASK_RUNNING);
1268 remove_wait_queue(&nlk->wait, &wait);
1269 sock_put(sk);
1270
1271 if (signal_pending(current)) {
1272 kfree_skb(skb);
1273 return sock_intr_errno(*timeo);
1274 }
1275 return 1;
1276 }
1277 netlink_skb_set_owner_r(skb, sk);
1278 return 0;
1279}
1280
1281static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1282{
1283 int len = skb->len;
1284
1285 netlink_deliver_tap(sock_net(sk), skb);
1286
1287 skb_queue_tail(&sk->sk_receive_queue, skb);
1288 sk->sk_data_ready(sk);
1289 return len;
1290}
1291
1292int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1293{
1294 int len = __netlink_sendskb(sk, skb);
1295
1296 sock_put(sk);
1297 return len;
1298}
1299
1300void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1301{
1302 kfree_skb(skb);
1303 sock_put(sk);
1304}
1305
1306static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1307{
1308 int delta;
1309
1310 WARN_ON(skb->sk != NULL);
1311 delta = skb->end - skb->tail;
1312 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1313 return skb;
1314
1315 if (skb_shared(skb)) {
1316 struct sk_buff *nskb = skb_clone(skb, allocation);
1317 if (!nskb)
1318 return skb;
1319 consume_skb(skb);
1320 skb = nskb;
1321 }
1322
1323 pskb_expand_head(skb, 0, -delta,
1324 (allocation & ~__GFP_DIRECT_RECLAIM) |
1325 __GFP_NOWARN | __GFP_NORETRY);
1326 return skb;
1327}
1328
1329static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1330 struct sock *ssk)
1331{
1332 int ret;
1333 struct netlink_sock *nlk = nlk_sk(sk);
1334
1335 ret = -ECONNREFUSED;
1336 if (nlk->netlink_rcv != NULL) {
1337 ret = skb->len;
1338 netlink_skb_set_owner_r(skb, sk);
1339 NETLINK_CB(skb).sk = ssk;
1340 netlink_deliver_tap_kernel(sk, ssk, skb);
1341 nlk->netlink_rcv(skb);
1342 consume_skb(skb);
1343 } else {
1344 kfree_skb(skb);
1345 }
1346 sock_put(sk);
1347 return ret;
1348}
1349
1350int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1351 u32 portid, int nonblock)
1352{
1353 struct sock *sk;
1354 int err;
1355 long timeo;
1356
1357 skb = netlink_trim(skb, gfp_any());
1358
1359 timeo = sock_sndtimeo(ssk, nonblock);
1360retry:
1361 sk = netlink_getsockbyportid(ssk, portid);
1362 if (IS_ERR(sk)) {
1363 kfree_skb(skb);
1364 return PTR_ERR(sk);
1365 }
1366 if (netlink_is_kernel(sk))
1367 return netlink_unicast_kernel(sk, skb, ssk);
1368
1369 if (sk_filter(sk, skb)) {
1370 err = skb->len;
1371 kfree_skb(skb);
1372 sock_put(sk);
1373 return err;
1374 }
1375
1376 err = netlink_attachskb(sk, skb, &timeo, ssk);
1377 if (err == 1)
1378 goto retry;
1379 if (err)
1380 return err;
1381
1382 return netlink_sendskb(sk, skb);
1383}
1384EXPORT_SYMBOL(netlink_unicast);
1385
1386int netlink_has_listeners(struct sock *sk, unsigned int group)
1387{
1388 int res = 0;
1389 struct listeners *listeners;
1390
1391 BUG_ON(!netlink_is_kernel(sk));
1392
1393 rcu_read_lock();
1394 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1395
1396 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1397 res = test_bit(group - 1, listeners->masks);
1398
1399 rcu_read_unlock();
1400
1401 return res;
1402}
1403EXPORT_SYMBOL_GPL(netlink_has_listeners);
1404
1405bool netlink_strict_get_check(struct sk_buff *skb)
1406{
1407 return nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
1408}
1409EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1410
1411static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1412{
1413 struct netlink_sock *nlk = nlk_sk(sk);
1414
1415 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1416 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1417 netlink_skb_set_owner_r(skb, sk);
1418 __netlink_sendskb(sk, skb);
1419 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1420 }
1421 return -1;
1422}
1423
1424struct netlink_broadcast_data {
1425 struct sock *exclude_sk;
1426 struct net *net;
1427 u32 portid;
1428 u32 group;
1429 int failure;
1430 int delivery_failure;
1431 int congested;
1432 int delivered;
1433 gfp_t allocation;
1434 struct sk_buff *skb, *skb2;
1435 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1436 void *tx_data;
1437};
1438
1439static void do_one_broadcast(struct sock *sk,
1440 struct netlink_broadcast_data *p)
1441{
1442 struct netlink_sock *nlk = nlk_sk(sk);
1443 int val;
1444
1445 if (p->exclude_sk == sk)
1446 return;
1447
1448 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1449 !test_bit(p->group - 1, nlk->groups))
1450 return;
1451
1452 if (!net_eq(sock_net(sk), p->net)) {
1453 if (!nlk_test_bit(LISTEN_ALL_NSID, sk))
1454 return;
1455
1456 if (!peernet_has_id(sock_net(sk), p->net))
1457 return;
1458
1459 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1460 CAP_NET_BROADCAST))
1461 return;
1462 }
1463
1464 if (p->failure) {
1465 netlink_overrun(sk);
1466 return;
1467 }
1468
1469 sock_hold(sk);
1470 if (p->skb2 == NULL) {
1471 if (skb_shared(p->skb)) {
1472 p->skb2 = skb_clone(p->skb, p->allocation);
1473 } else {
1474 p->skb2 = skb_get(p->skb);
1475 /*
1476 * skb ownership may have been set when
1477 * delivered to a previous socket.
1478 */
1479 skb_orphan(p->skb2);
1480 }
1481 }
1482 if (p->skb2 == NULL) {
1483 netlink_overrun(sk);
1484 /* Clone failed. Notify ALL listeners. */
1485 p->failure = 1;
1486 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1487 p->delivery_failure = 1;
1488 goto out;
1489 }
1490
1491 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1492 kfree_skb(p->skb2);
1493 p->skb2 = NULL;
1494 goto out;
1495 }
1496
1497 if (sk_filter(sk, p->skb2)) {
1498 kfree_skb(p->skb2);
1499 p->skb2 = NULL;
1500 goto out;
1501 }
1502 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1503 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1504 NETLINK_CB(p->skb2).nsid_is_set = true;
1505 val = netlink_broadcast_deliver(sk, p->skb2);
1506 if (val < 0) {
1507 netlink_overrun(sk);
1508 if (nlk_test_bit(BROADCAST_SEND_ERROR, sk))
1509 p->delivery_failure = 1;
1510 } else {
1511 p->congested |= val;
1512 p->delivered = 1;
1513 p->skb2 = NULL;
1514 }
1515out:
1516 sock_put(sk);
1517}
1518
1519int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb,
1520 u32 portid,
1521 u32 group, gfp_t allocation,
1522 netlink_filter_fn filter,
1523 void *filter_data)
1524{
1525 struct net *net = sock_net(ssk);
1526 struct netlink_broadcast_data info;
1527 struct sock *sk;
1528
1529 skb = netlink_trim(skb, allocation);
1530
1531 info.exclude_sk = ssk;
1532 info.net = net;
1533 info.portid = portid;
1534 info.group = group;
1535 info.failure = 0;
1536 info.delivery_failure = 0;
1537 info.congested = 0;
1538 info.delivered = 0;
1539 info.allocation = allocation;
1540 info.skb = skb;
1541 info.skb2 = NULL;
1542 info.tx_filter = filter;
1543 info.tx_data = filter_data;
1544
1545 /* While we sleep in clone, do not allow to change socket list */
1546
1547 netlink_lock_table();
1548
1549 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1550 do_one_broadcast(sk, &info);
1551
1552 consume_skb(skb);
1553
1554 netlink_unlock_table();
1555
1556 if (info.delivery_failure) {
1557 kfree_skb(info.skb2);
1558 return -ENOBUFS;
1559 }
1560 consume_skb(info.skb2);
1561
1562 if (info.delivered) {
1563 if (info.congested && gfpflags_allow_blocking(allocation))
1564 yield();
1565 return 0;
1566 }
1567 return -ESRCH;
1568}
1569EXPORT_SYMBOL(netlink_broadcast_filtered);
1570
1571int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1572 u32 group, gfp_t allocation)
1573{
1574 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1575 NULL, NULL);
1576}
1577EXPORT_SYMBOL(netlink_broadcast);
1578
1579struct netlink_set_err_data {
1580 struct sock *exclude_sk;
1581 u32 portid;
1582 u32 group;
1583 int code;
1584};
1585
1586static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1587{
1588 struct netlink_sock *nlk = nlk_sk(sk);
1589 int ret = 0;
1590
1591 if (sk == p->exclude_sk)
1592 goto out;
1593
1594 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1595 goto out;
1596
1597 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1598 !test_bit(p->group - 1, nlk->groups))
1599 goto out;
1600
1601 if (p->code == ENOBUFS && nlk_test_bit(RECV_NO_ENOBUFS, sk)) {
1602 ret = 1;
1603 goto out;
1604 }
1605
1606 WRITE_ONCE(sk->sk_err, p->code);
1607 sk_error_report(sk);
1608out:
1609 return ret;
1610}
1611
1612/**
1613 * netlink_set_err - report error to broadcast listeners
1614 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1615 * @portid: the PORTID of a process that we want to skip (if any)
1616 * @group: the broadcast group that will notice the error
1617 * @code: error code, must be negative (as usual in kernelspace)
1618 *
1619 * This function returns the number of broadcast listeners that have set the
1620 * NETLINK_NO_ENOBUFS socket option.
1621 */
1622int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1623{
1624 struct netlink_set_err_data info;
1625 unsigned long flags;
1626 struct sock *sk;
1627 int ret = 0;
1628
1629 info.exclude_sk = ssk;
1630 info.portid = portid;
1631 info.group = group;
1632 /* sk->sk_err wants a positive error value */
1633 info.code = -code;
1634
1635 read_lock_irqsave(&nl_table_lock, flags);
1636
1637 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1638 ret += do_one_set_err(sk, &info);
1639
1640 read_unlock_irqrestore(&nl_table_lock, flags);
1641 return ret;
1642}
1643EXPORT_SYMBOL(netlink_set_err);
1644
1645/* must be called with netlink table grabbed */
1646static void netlink_update_socket_mc(struct netlink_sock *nlk,
1647 unsigned int group,
1648 int is_new)
1649{
1650 int old, new = !!is_new, subscriptions;
1651
1652 old = test_bit(group - 1, nlk->groups);
1653 subscriptions = nlk->subscriptions - old + new;
1654 __assign_bit(group - 1, nlk->groups, new);
1655 netlink_update_subscriptions(&nlk->sk, subscriptions);
1656 netlink_update_listeners(&nlk->sk);
1657}
1658
1659static int netlink_setsockopt(struct socket *sock, int level, int optname,
1660 sockptr_t optval, unsigned int optlen)
1661{
1662 struct sock *sk = sock->sk;
1663 struct netlink_sock *nlk = nlk_sk(sk);
1664 unsigned int val = 0;
1665 int nr = -1;
1666
1667 if (level != SOL_NETLINK)
1668 return -ENOPROTOOPT;
1669
1670 if (optlen >= sizeof(int) &&
1671 copy_from_sockptr(&val, optval, sizeof(val)))
1672 return -EFAULT;
1673
1674 switch (optname) {
1675 case NETLINK_PKTINFO:
1676 nr = NETLINK_F_RECV_PKTINFO;
1677 break;
1678 case NETLINK_ADD_MEMBERSHIP:
1679 case NETLINK_DROP_MEMBERSHIP: {
1680 int err;
1681
1682 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1683 return -EPERM;
1684 err = netlink_realloc_groups(sk);
1685 if (err)
1686 return err;
1687 if (!val || val - 1 >= nlk->ngroups)
1688 return -EINVAL;
1689 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1690 err = nlk->netlink_bind(sock_net(sk), val);
1691 if (err)
1692 return err;
1693 }
1694 netlink_table_grab();
1695 netlink_update_socket_mc(nlk, val,
1696 optname == NETLINK_ADD_MEMBERSHIP);
1697 netlink_table_ungrab();
1698 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1699 nlk->netlink_unbind(sock_net(sk), val);
1700
1701 break;
1702 }
1703 case NETLINK_BROADCAST_ERROR:
1704 nr = NETLINK_F_BROADCAST_SEND_ERROR;
1705 break;
1706 case NETLINK_NO_ENOBUFS:
1707 assign_bit(NETLINK_F_RECV_NO_ENOBUFS, &nlk->flags, val);
1708 if (val) {
1709 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1710 wake_up_interruptible(&nlk->wait);
1711 }
1712 break;
1713 case NETLINK_LISTEN_ALL_NSID:
1714 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1715 return -EPERM;
1716 nr = NETLINK_F_LISTEN_ALL_NSID;
1717 break;
1718 case NETLINK_CAP_ACK:
1719 nr = NETLINK_F_CAP_ACK;
1720 break;
1721 case NETLINK_EXT_ACK:
1722 nr = NETLINK_F_EXT_ACK;
1723 break;
1724 case NETLINK_GET_STRICT_CHK:
1725 nr = NETLINK_F_STRICT_CHK;
1726 break;
1727 default:
1728 return -ENOPROTOOPT;
1729 }
1730 if (nr >= 0)
1731 assign_bit(nr, &nlk->flags, val);
1732 return 0;
1733}
1734
1735static int netlink_getsockopt(struct socket *sock, int level, int optname,
1736 char __user *optval, int __user *optlen)
1737{
1738 struct sock *sk = sock->sk;
1739 struct netlink_sock *nlk = nlk_sk(sk);
1740 unsigned int flag;
1741 int len, val;
1742
1743 if (level != SOL_NETLINK)
1744 return -ENOPROTOOPT;
1745
1746 if (get_user(len, optlen))
1747 return -EFAULT;
1748 if (len < 0)
1749 return -EINVAL;
1750
1751 switch (optname) {
1752 case NETLINK_PKTINFO:
1753 flag = NETLINK_F_RECV_PKTINFO;
1754 break;
1755 case NETLINK_BROADCAST_ERROR:
1756 flag = NETLINK_F_BROADCAST_SEND_ERROR;
1757 break;
1758 case NETLINK_NO_ENOBUFS:
1759 flag = NETLINK_F_RECV_NO_ENOBUFS;
1760 break;
1761 case NETLINK_LIST_MEMBERSHIPS: {
1762 int pos, idx, shift, err = 0;
1763
1764 netlink_lock_table();
1765 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1766 if (len - pos < sizeof(u32))
1767 break;
1768
1769 idx = pos / sizeof(unsigned long);
1770 shift = (pos % sizeof(unsigned long)) * 8;
1771 if (put_user((u32)(nlk->groups[idx] >> shift),
1772 (u32 __user *)(optval + pos))) {
1773 err = -EFAULT;
1774 break;
1775 }
1776 }
1777 if (put_user(ALIGN(BITS_TO_BYTES(nlk->ngroups), sizeof(u32)), optlen))
1778 err = -EFAULT;
1779 netlink_unlock_table();
1780 return err;
1781 }
1782 case NETLINK_CAP_ACK:
1783 flag = NETLINK_F_CAP_ACK;
1784 break;
1785 case NETLINK_EXT_ACK:
1786 flag = NETLINK_F_EXT_ACK;
1787 break;
1788 case NETLINK_GET_STRICT_CHK:
1789 flag = NETLINK_F_STRICT_CHK;
1790 break;
1791 default:
1792 return -ENOPROTOOPT;
1793 }
1794
1795 if (len < sizeof(int))
1796 return -EINVAL;
1797
1798 len = sizeof(int);
1799 val = test_bit(flag, &nlk->flags);
1800
1801 if (put_user(len, optlen) ||
1802 copy_to_user(optval, &val, len))
1803 return -EFAULT;
1804
1805 return 0;
1806}
1807
1808static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1809{
1810 struct nl_pktinfo info;
1811
1812 info.group = NETLINK_CB(skb).dst_group;
1813 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1814}
1815
1816static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1817 struct sk_buff *skb)
1818{
1819 if (!NETLINK_CB(skb).nsid_is_set)
1820 return;
1821
1822 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1823 &NETLINK_CB(skb).nsid);
1824}
1825
1826static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1827{
1828 struct sock *sk = sock->sk;
1829 struct netlink_sock *nlk = nlk_sk(sk);
1830 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1831 u32 dst_portid;
1832 u32 dst_group;
1833 struct sk_buff *skb;
1834 int err;
1835 struct scm_cookie scm;
1836 u32 netlink_skb_flags = 0;
1837
1838 if (msg->msg_flags & MSG_OOB)
1839 return -EOPNOTSUPP;
1840
1841 if (len == 0) {
1842 pr_warn_once("Zero length message leads to an empty skb\n");
1843 return -ENODATA;
1844 }
1845
1846 err = scm_send(sock, msg, &scm, true);
1847 if (err < 0)
1848 return err;
1849
1850 if (msg->msg_namelen) {
1851 err = -EINVAL;
1852 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1853 goto out;
1854 if (addr->nl_family != AF_NETLINK)
1855 goto out;
1856 dst_portid = addr->nl_pid;
1857 dst_group = ffs(addr->nl_groups);
1858 err = -EPERM;
1859 if ((dst_group || dst_portid) &&
1860 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1861 goto out;
1862 netlink_skb_flags |= NETLINK_SKB_DST;
1863 } else {
1864 /* Paired with WRITE_ONCE() in netlink_connect() */
1865 dst_portid = READ_ONCE(nlk->dst_portid);
1866 dst_group = READ_ONCE(nlk->dst_group);
1867 }
1868
1869 /* Paired with WRITE_ONCE() in netlink_insert() */
1870 if (!READ_ONCE(nlk->bound)) {
1871 err = netlink_autobind(sock);
1872 if (err)
1873 goto out;
1874 } else {
1875 /* Ensure nlk is hashed and visible. */
1876 smp_rmb();
1877 }
1878
1879 err = -EMSGSIZE;
1880 if (len > sk->sk_sndbuf - 32)
1881 goto out;
1882 err = -ENOBUFS;
1883 skb = netlink_alloc_large_skb(len, dst_group);
1884 if (skb == NULL)
1885 goto out;
1886
1887 NETLINK_CB(skb).portid = nlk->portid;
1888 NETLINK_CB(skb).dst_group = dst_group;
1889 NETLINK_CB(skb).creds = scm.creds;
1890 NETLINK_CB(skb).flags = netlink_skb_flags;
1891
1892 err = -EFAULT;
1893 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1894 kfree_skb(skb);
1895 goto out;
1896 }
1897
1898 err = security_netlink_send(sk, skb);
1899 if (err) {
1900 kfree_skb(skb);
1901 goto out;
1902 }
1903
1904 if (dst_group) {
1905 refcount_inc(&skb->users);
1906 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1907 }
1908 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1909
1910out:
1911 scm_destroy(&scm);
1912 return err;
1913}
1914
1915static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1916 int flags)
1917{
1918 struct scm_cookie scm;
1919 struct sock *sk = sock->sk;
1920 struct netlink_sock *nlk = nlk_sk(sk);
1921 size_t copied, max_recvmsg_len;
1922 struct sk_buff *skb, *data_skb;
1923 int err, ret;
1924
1925 if (flags & MSG_OOB)
1926 return -EOPNOTSUPP;
1927
1928 copied = 0;
1929
1930 skb = skb_recv_datagram(sk, flags, &err);
1931 if (skb == NULL)
1932 goto out;
1933
1934 data_skb = skb;
1935
1936#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1937 if (unlikely(skb_shinfo(skb)->frag_list)) {
1938 /*
1939 * If this skb has a frag_list, then here that means that we
1940 * will have to use the frag_list skb's data for compat tasks
1941 * and the regular skb's data for normal (non-compat) tasks.
1942 *
1943 * If we need to send the compat skb, assign it to the
1944 * 'data_skb' variable so that it will be used below for data
1945 * copying. We keep 'skb' for everything else, including
1946 * freeing both later.
1947 */
1948 if (flags & MSG_CMSG_COMPAT)
1949 data_skb = skb_shinfo(skb)->frag_list;
1950 }
1951#endif
1952
1953 /* Record the max length of recvmsg() calls for future allocations */
1954 max_recvmsg_len = max(READ_ONCE(nlk->max_recvmsg_len), len);
1955 max_recvmsg_len = min_t(size_t, max_recvmsg_len,
1956 SKB_WITH_OVERHEAD(32768));
1957 WRITE_ONCE(nlk->max_recvmsg_len, max_recvmsg_len);
1958
1959 copied = data_skb->len;
1960 if (len < copied) {
1961 msg->msg_flags |= MSG_TRUNC;
1962 copied = len;
1963 }
1964
1965 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1966
1967 if (msg->msg_name) {
1968 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1969 addr->nl_family = AF_NETLINK;
1970 addr->nl_pad = 0;
1971 addr->nl_pid = NETLINK_CB(skb).portid;
1972 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1973 msg->msg_namelen = sizeof(*addr);
1974 }
1975
1976 if (nlk_test_bit(RECV_PKTINFO, sk))
1977 netlink_cmsg_recv_pktinfo(msg, skb);
1978 if (nlk_test_bit(LISTEN_ALL_NSID, sk))
1979 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1980
1981 memset(&scm, 0, sizeof(scm));
1982 scm.creds = *NETLINK_CREDS(skb);
1983 if (flags & MSG_TRUNC)
1984 copied = data_skb->len;
1985
1986 skb_free_datagram(sk, skb);
1987
1988 if (READ_ONCE(nlk->cb_running) &&
1989 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1990 ret = netlink_dump(sk);
1991 if (ret) {
1992 WRITE_ONCE(sk->sk_err, -ret);
1993 sk_error_report(sk);
1994 }
1995 }
1996
1997 scm_recv(sock, msg, &scm, flags);
1998out:
1999 netlink_rcv_wake(sk);
2000 return err ? : copied;
2001}
2002
2003static void netlink_data_ready(struct sock *sk)
2004{
2005 BUG();
2006}
2007
2008/*
2009 * We export these functions to other modules. They provide a
2010 * complete set of kernel non-blocking support for message
2011 * queueing.
2012 */
2013
2014struct sock *
2015__netlink_kernel_create(struct net *net, int unit, struct module *module,
2016 struct netlink_kernel_cfg *cfg)
2017{
2018 struct socket *sock;
2019 struct sock *sk;
2020 struct netlink_sock *nlk;
2021 struct listeners *listeners = NULL;
2022 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2023 unsigned int groups;
2024
2025 BUG_ON(!nl_table);
2026
2027 if (unit < 0 || unit >= MAX_LINKS)
2028 return NULL;
2029
2030 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2031 return NULL;
2032
2033 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2034 goto out_sock_release_nosk;
2035
2036 sk = sock->sk;
2037
2038 if (!cfg || cfg->groups < 32)
2039 groups = 32;
2040 else
2041 groups = cfg->groups;
2042
2043 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2044 if (!listeners)
2045 goto out_sock_release;
2046
2047 sk->sk_data_ready = netlink_data_ready;
2048 if (cfg && cfg->input)
2049 nlk_sk(sk)->netlink_rcv = cfg->input;
2050
2051 if (netlink_insert(sk, 0))
2052 goto out_sock_release;
2053
2054 nlk = nlk_sk(sk);
2055 set_bit(NETLINK_F_KERNEL_SOCKET, &nlk->flags);
2056
2057 netlink_table_grab();
2058 if (!nl_table[unit].registered) {
2059 nl_table[unit].groups = groups;
2060 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2061 nl_table[unit].cb_mutex = cb_mutex;
2062 nl_table[unit].module = module;
2063 if (cfg) {
2064 nl_table[unit].bind = cfg->bind;
2065 nl_table[unit].unbind = cfg->unbind;
2066 nl_table[unit].release = cfg->release;
2067 nl_table[unit].flags = cfg->flags;
2068 }
2069 nl_table[unit].registered = 1;
2070 } else {
2071 kfree(listeners);
2072 nl_table[unit].registered++;
2073 }
2074 netlink_table_ungrab();
2075 return sk;
2076
2077out_sock_release:
2078 kfree(listeners);
2079 netlink_kernel_release(sk);
2080 return NULL;
2081
2082out_sock_release_nosk:
2083 sock_release(sock);
2084 return NULL;
2085}
2086EXPORT_SYMBOL(__netlink_kernel_create);
2087
2088void
2089netlink_kernel_release(struct sock *sk)
2090{
2091 if (sk == NULL || sk->sk_socket == NULL)
2092 return;
2093
2094 sock_release(sk->sk_socket);
2095}
2096EXPORT_SYMBOL(netlink_kernel_release);
2097
2098int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2099{
2100 struct listeners *new, *old;
2101 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2102
2103 if (groups < 32)
2104 groups = 32;
2105
2106 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2107 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2108 if (!new)
2109 return -ENOMEM;
2110 old = nl_deref_protected(tbl->listeners);
2111 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2112 rcu_assign_pointer(tbl->listeners, new);
2113
2114 kfree_rcu(old, rcu);
2115 }
2116 tbl->groups = groups;
2117
2118 return 0;
2119}
2120
2121/**
2122 * netlink_change_ngroups - change number of multicast groups
2123 *
2124 * This changes the number of multicast groups that are available
2125 * on a certain netlink family. Note that it is not possible to
2126 * change the number of groups to below 32. Also note that it does
2127 * not implicitly call netlink_clear_multicast_users() when the
2128 * number of groups is reduced.
2129 *
2130 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2131 * @groups: The new number of groups.
2132 */
2133int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2134{
2135 int err;
2136
2137 netlink_table_grab();
2138 err = __netlink_change_ngroups(sk, groups);
2139 netlink_table_ungrab();
2140
2141 return err;
2142}
2143
2144void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2145{
2146 struct sock *sk;
2147 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2148
2149 sk_for_each_bound(sk, &tbl->mc_list)
2150 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2151}
2152
2153struct nlmsghdr *
2154__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2155{
2156 struct nlmsghdr *nlh;
2157 int size = nlmsg_msg_size(len);
2158
2159 nlh = skb_put(skb, NLMSG_ALIGN(size));
2160 nlh->nlmsg_type = type;
2161 nlh->nlmsg_len = size;
2162 nlh->nlmsg_flags = flags;
2163 nlh->nlmsg_pid = portid;
2164 nlh->nlmsg_seq = seq;
2165 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2166 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2167 return nlh;
2168}
2169EXPORT_SYMBOL(__nlmsg_put);
2170
2171/*
2172 * It looks a bit ugly.
2173 * It would be better to create kernel thread.
2174 */
2175
2176static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2177 struct netlink_callback *cb,
2178 struct netlink_ext_ack *extack)
2179{
2180 struct nlmsghdr *nlh;
2181
2182 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2183 NLM_F_MULTI | cb->answer_flags);
2184 if (WARN_ON(!nlh))
2185 return -ENOBUFS;
2186
2187 nl_dump_check_consistent(cb, nlh);
2188 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2189
2190 if (extack->_msg && test_bit(NETLINK_F_EXT_ACK, &nlk->flags)) {
2191 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2192 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2193 nlmsg_end(skb, nlh);
2194 }
2195
2196 return 0;
2197}
2198
2199static int netlink_dump(struct sock *sk)
2200{
2201 struct netlink_sock *nlk = nlk_sk(sk);
2202 struct netlink_ext_ack extack = {};
2203 struct netlink_callback *cb;
2204 struct sk_buff *skb = NULL;
2205 size_t max_recvmsg_len;
2206 struct module *module;
2207 int err = -ENOBUFS;
2208 int alloc_min_size;
2209 int alloc_size;
2210
2211 mutex_lock(nlk->cb_mutex);
2212 if (!nlk->cb_running) {
2213 err = -EINVAL;
2214 goto errout_skb;
2215 }
2216
2217 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2218 goto errout_skb;
2219
2220 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2221 * required, but it makes sense to _attempt_ a 16K bytes allocation
2222 * to reduce number of system calls on dump operations, if user
2223 * ever provided a big enough buffer.
2224 */
2225 cb = &nlk->cb;
2226 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2227
2228 max_recvmsg_len = READ_ONCE(nlk->max_recvmsg_len);
2229 if (alloc_min_size < max_recvmsg_len) {
2230 alloc_size = max_recvmsg_len;
2231 skb = alloc_skb(alloc_size,
2232 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2233 __GFP_NOWARN | __GFP_NORETRY);
2234 }
2235 if (!skb) {
2236 alloc_size = alloc_min_size;
2237 skb = alloc_skb(alloc_size, GFP_KERNEL);
2238 }
2239 if (!skb)
2240 goto errout_skb;
2241
2242 /* Trim skb to allocated size. User is expected to provide buffer as
2243 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2244 * netlink_recvmsg())). dump will pack as many smaller messages as
2245 * could fit within the allocated skb. skb is typically allocated
2246 * with larger space than required (could be as much as near 2x the
2247 * requested size with align to next power of 2 approach). Allowing
2248 * dump to use the excess space makes it difficult for a user to have a
2249 * reasonable static buffer based on the expected largest dump of a
2250 * single netdev. The outcome is MSG_TRUNC error.
2251 */
2252 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2253
2254 /* Make sure malicious BPF programs can not read unitialized memory
2255 * from skb->head -> skb->data
2256 */
2257 skb_reset_network_header(skb);
2258 skb_reset_mac_header(skb);
2259
2260 netlink_skb_set_owner_r(skb, sk);
2261
2262 if (nlk->dump_done_errno > 0) {
2263 cb->extack = &extack;
2264 nlk->dump_done_errno = cb->dump(skb, cb);
2265 cb->extack = NULL;
2266 }
2267
2268 if (nlk->dump_done_errno > 0 ||
2269 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2270 mutex_unlock(nlk->cb_mutex);
2271
2272 if (sk_filter(sk, skb))
2273 kfree_skb(skb);
2274 else
2275 __netlink_sendskb(sk, skb);
2276 return 0;
2277 }
2278
2279 if (netlink_dump_done(nlk, skb, cb, &extack))
2280 goto errout_skb;
2281
2282#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2283 /* frag_list skb's data is used for compat tasks
2284 * and the regular skb's data for normal (non-compat) tasks.
2285 * See netlink_recvmsg().
2286 */
2287 if (unlikely(skb_shinfo(skb)->frag_list)) {
2288 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2289 goto errout_skb;
2290 }
2291#endif
2292
2293 if (sk_filter(sk, skb))
2294 kfree_skb(skb);
2295 else
2296 __netlink_sendskb(sk, skb);
2297
2298 if (cb->done)
2299 cb->done(cb);
2300
2301 WRITE_ONCE(nlk->cb_running, false);
2302 module = cb->module;
2303 skb = cb->skb;
2304 mutex_unlock(nlk->cb_mutex);
2305 module_put(module);
2306 consume_skb(skb);
2307 return 0;
2308
2309errout_skb:
2310 mutex_unlock(nlk->cb_mutex);
2311 kfree_skb(skb);
2312 return err;
2313}
2314
2315int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2316 const struct nlmsghdr *nlh,
2317 struct netlink_dump_control *control)
2318{
2319 struct netlink_callback *cb;
2320 struct netlink_sock *nlk;
2321 struct sock *sk;
2322 int ret;
2323
2324 refcount_inc(&skb->users);
2325
2326 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2327 if (sk == NULL) {
2328 ret = -ECONNREFUSED;
2329 goto error_free;
2330 }
2331
2332 nlk = nlk_sk(sk);
2333 mutex_lock(nlk->cb_mutex);
2334 /* A dump is in progress... */
2335 if (nlk->cb_running) {
2336 ret = -EBUSY;
2337 goto error_unlock;
2338 }
2339 /* add reference of module which cb->dump belongs to */
2340 if (!try_module_get(control->module)) {
2341 ret = -EPROTONOSUPPORT;
2342 goto error_unlock;
2343 }
2344
2345 cb = &nlk->cb;
2346 memset(cb, 0, sizeof(*cb));
2347 cb->dump = control->dump;
2348 cb->done = control->done;
2349 cb->nlh = nlh;
2350 cb->data = control->data;
2351 cb->module = control->module;
2352 cb->min_dump_alloc = control->min_dump_alloc;
2353 cb->skb = skb;
2354
2355 cb->strict_check = nlk_test_bit(STRICT_CHK, NETLINK_CB(skb).sk);
2356
2357 if (control->start) {
2358 cb->extack = control->extack;
2359 ret = control->start(cb);
2360 cb->extack = NULL;
2361 if (ret)
2362 goto error_put;
2363 }
2364
2365 WRITE_ONCE(nlk->cb_running, true);
2366 nlk->dump_done_errno = INT_MAX;
2367
2368 mutex_unlock(nlk->cb_mutex);
2369
2370 ret = netlink_dump(sk);
2371
2372 sock_put(sk);
2373
2374 if (ret)
2375 return ret;
2376
2377 /* We successfully started a dump, by returning -EINTR we
2378 * signal not to send ACK even if it was requested.
2379 */
2380 return -EINTR;
2381
2382error_put:
2383 module_put(control->module);
2384error_unlock:
2385 sock_put(sk);
2386 mutex_unlock(nlk->cb_mutex);
2387error_free:
2388 kfree_skb(skb);
2389 return ret;
2390}
2391EXPORT_SYMBOL(__netlink_dump_start);
2392
2393static size_t
2394netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2395 const struct netlink_ext_ack *extack)
2396{
2397 size_t tlvlen;
2398
2399 if (!extack || !test_bit(NETLINK_F_EXT_ACK, &nlk->flags))
2400 return 0;
2401
2402 tlvlen = 0;
2403 if (extack->_msg)
2404 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2405 if (extack->cookie_len)
2406 tlvlen += nla_total_size(extack->cookie_len);
2407
2408 /* Following attributes are only reported as error (not warning) */
2409 if (!err)
2410 return tlvlen;
2411
2412 if (extack->bad_attr)
2413 tlvlen += nla_total_size(sizeof(u32));
2414 if (extack->policy)
2415 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2416 if (extack->miss_type)
2417 tlvlen += nla_total_size(sizeof(u32));
2418 if (extack->miss_nest)
2419 tlvlen += nla_total_size(sizeof(u32));
2420
2421 return tlvlen;
2422}
2423
2424static void
2425netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2426 struct nlmsghdr *nlh, int err,
2427 const struct netlink_ext_ack *extack)
2428{
2429 if (extack->_msg)
2430 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2431 if (extack->cookie_len)
2432 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2433 extack->cookie_len, extack->cookie));
2434
2435 if (!err)
2436 return;
2437
2438 if (extack->bad_attr &&
2439 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2440 (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2441 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2442 (u8 *)extack->bad_attr - (u8 *)nlh));
2443 if (extack->policy)
2444 netlink_policy_dump_write_attr(skb, extack->policy,
2445 NLMSGERR_ATTR_POLICY);
2446 if (extack->miss_type)
2447 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2448 extack->miss_type));
2449 if (extack->miss_nest &&
2450 !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2451 (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2452 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2453 (u8 *)extack->miss_nest - (u8 *)nlh));
2454}
2455
2456void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2457 const struct netlink_ext_ack *extack)
2458{
2459 struct sk_buff *skb;
2460 struct nlmsghdr *rep;
2461 struct nlmsgerr *errmsg;
2462 size_t payload = sizeof(*errmsg);
2463 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2464 unsigned int flags = 0;
2465 size_t tlvlen;
2466
2467 /* Error messages get the original request appened, unless the user
2468 * requests to cap the error message, and get extra error data if
2469 * requested.
2470 */
2471 if (err && !test_bit(NETLINK_F_CAP_ACK, &nlk->flags))
2472 payload += nlmsg_len(nlh);
2473 else
2474 flags |= NLM_F_CAPPED;
2475
2476 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2477 if (tlvlen)
2478 flags |= NLM_F_ACK_TLVS;
2479
2480 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2481 if (!skb)
2482 goto err_skb;
2483
2484 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2485 NLMSG_ERROR, sizeof(*errmsg), flags);
2486 if (!rep)
2487 goto err_bad_put;
2488 errmsg = nlmsg_data(rep);
2489 errmsg->error = err;
2490 errmsg->msg = *nlh;
2491
2492 if (!(flags & NLM_F_CAPPED)) {
2493 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2494 goto err_bad_put;
2495
2496 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2497 nlmsg_len(nlh));
2498 }
2499
2500 if (tlvlen)
2501 netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2502
2503 nlmsg_end(skb, rep);
2504
2505 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2506
2507 return;
2508
2509err_bad_put:
2510 nlmsg_free(skb);
2511err_skb:
2512 WRITE_ONCE(NETLINK_CB(in_skb).sk->sk_err, ENOBUFS);
2513 sk_error_report(NETLINK_CB(in_skb).sk);
2514}
2515EXPORT_SYMBOL(netlink_ack);
2516
2517int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2518 struct nlmsghdr *,
2519 struct netlink_ext_ack *))
2520{
2521 struct netlink_ext_ack extack;
2522 struct nlmsghdr *nlh;
2523 int err;
2524
2525 while (skb->len >= nlmsg_total_size(0)) {
2526 int msglen;
2527
2528 memset(&extack, 0, sizeof(extack));
2529 nlh = nlmsg_hdr(skb);
2530 err = 0;
2531
2532 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2533 return 0;
2534
2535 /* Only requests are handled by the kernel */
2536 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2537 goto ack;
2538
2539 /* Skip control messages */
2540 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2541 goto ack;
2542
2543 err = cb(skb, nlh, &extack);
2544 if (err == -EINTR)
2545 goto skip;
2546
2547ack:
2548 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2549 netlink_ack(skb, nlh, err, &extack);
2550
2551skip:
2552 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2553 if (msglen > skb->len)
2554 msglen = skb->len;
2555 skb_pull(skb, msglen);
2556 }
2557
2558 return 0;
2559}
2560EXPORT_SYMBOL(netlink_rcv_skb);
2561
2562/**
2563 * nlmsg_notify - send a notification netlink message
2564 * @sk: netlink socket to use
2565 * @skb: notification message
2566 * @portid: destination netlink portid for reports or 0
2567 * @group: destination multicast group or 0
2568 * @report: 1 to report back, 0 to disable
2569 * @flags: allocation flags
2570 */
2571int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2572 unsigned int group, int report, gfp_t flags)
2573{
2574 int err = 0;
2575
2576 if (group) {
2577 int exclude_portid = 0;
2578
2579 if (report) {
2580 refcount_inc(&skb->users);
2581 exclude_portid = portid;
2582 }
2583
2584 /* errors reported via destination sk->sk_err, but propagate
2585 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2586 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2587 if (err == -ESRCH)
2588 err = 0;
2589 }
2590
2591 if (report) {
2592 int err2;
2593
2594 err2 = nlmsg_unicast(sk, skb, portid);
2595 if (!err)
2596 err = err2;
2597 }
2598
2599 return err;
2600}
2601EXPORT_SYMBOL(nlmsg_notify);
2602
2603#ifdef CONFIG_PROC_FS
2604struct nl_seq_iter {
2605 struct seq_net_private p;
2606 struct rhashtable_iter hti;
2607 int link;
2608};
2609
2610static void netlink_walk_start(struct nl_seq_iter *iter)
2611{
2612 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2613 rhashtable_walk_start(&iter->hti);
2614}
2615
2616static void netlink_walk_stop(struct nl_seq_iter *iter)
2617{
2618 rhashtable_walk_stop(&iter->hti);
2619 rhashtable_walk_exit(&iter->hti);
2620}
2621
2622static void *__netlink_seq_next(struct seq_file *seq)
2623{
2624 struct nl_seq_iter *iter = seq->private;
2625 struct netlink_sock *nlk;
2626
2627 do {
2628 for (;;) {
2629 nlk = rhashtable_walk_next(&iter->hti);
2630
2631 if (IS_ERR(nlk)) {
2632 if (PTR_ERR(nlk) == -EAGAIN)
2633 continue;
2634
2635 return nlk;
2636 }
2637
2638 if (nlk)
2639 break;
2640
2641 netlink_walk_stop(iter);
2642 if (++iter->link >= MAX_LINKS)
2643 return NULL;
2644
2645 netlink_walk_start(iter);
2646 }
2647 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2648
2649 return nlk;
2650}
2651
2652static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2653 __acquires(RCU)
2654{
2655 struct nl_seq_iter *iter = seq->private;
2656 void *obj = SEQ_START_TOKEN;
2657 loff_t pos;
2658
2659 iter->link = 0;
2660
2661 netlink_walk_start(iter);
2662
2663 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2664 obj = __netlink_seq_next(seq);
2665
2666 return obj;
2667}
2668
2669static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2670{
2671 ++*pos;
2672 return __netlink_seq_next(seq);
2673}
2674
2675static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2676{
2677 struct nl_seq_iter *iter = seq->private;
2678
2679 if (iter->link >= MAX_LINKS)
2680 return;
2681
2682 netlink_walk_stop(iter);
2683}
2684
2685
2686static int netlink_native_seq_show(struct seq_file *seq, void *v)
2687{
2688 if (v == SEQ_START_TOKEN) {
2689 seq_puts(seq,
2690 "sk Eth Pid Groups "
2691 "Rmem Wmem Dump Locks Drops Inode\n");
2692 } else {
2693 struct sock *s = v;
2694 struct netlink_sock *nlk = nlk_sk(s);
2695
2696 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2697 s,
2698 s->sk_protocol,
2699 nlk->portid,
2700 nlk->groups ? (u32)nlk->groups[0] : 0,
2701 sk_rmem_alloc_get(s),
2702 sk_wmem_alloc_get(s),
2703 READ_ONCE(nlk->cb_running),
2704 refcount_read(&s->sk_refcnt),
2705 atomic_read(&s->sk_drops),
2706 sock_i_ino(s)
2707 );
2708
2709 }
2710 return 0;
2711}
2712
2713#ifdef CONFIG_BPF_SYSCALL
2714struct bpf_iter__netlink {
2715 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2716 __bpf_md_ptr(struct netlink_sock *, sk);
2717};
2718
2719DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2720
2721static int netlink_prog_seq_show(struct bpf_prog *prog,
2722 struct bpf_iter_meta *meta,
2723 void *v)
2724{
2725 struct bpf_iter__netlink ctx;
2726
2727 meta->seq_num--; /* skip SEQ_START_TOKEN */
2728 ctx.meta = meta;
2729 ctx.sk = nlk_sk((struct sock *)v);
2730 return bpf_iter_run_prog(prog, &ctx);
2731}
2732
2733static int netlink_seq_show(struct seq_file *seq, void *v)
2734{
2735 struct bpf_iter_meta meta;
2736 struct bpf_prog *prog;
2737
2738 meta.seq = seq;
2739 prog = bpf_iter_get_info(&meta, false);
2740 if (!prog)
2741 return netlink_native_seq_show(seq, v);
2742
2743 if (v != SEQ_START_TOKEN)
2744 return netlink_prog_seq_show(prog, &meta, v);
2745
2746 return 0;
2747}
2748
2749static void netlink_seq_stop(struct seq_file *seq, void *v)
2750{
2751 struct bpf_iter_meta meta;
2752 struct bpf_prog *prog;
2753
2754 if (!v) {
2755 meta.seq = seq;
2756 prog = bpf_iter_get_info(&meta, true);
2757 if (prog)
2758 (void)netlink_prog_seq_show(prog, &meta, v);
2759 }
2760
2761 netlink_native_seq_stop(seq, v);
2762}
2763#else
2764static int netlink_seq_show(struct seq_file *seq, void *v)
2765{
2766 return netlink_native_seq_show(seq, v);
2767}
2768
2769static void netlink_seq_stop(struct seq_file *seq, void *v)
2770{
2771 netlink_native_seq_stop(seq, v);
2772}
2773#endif
2774
2775static const struct seq_operations netlink_seq_ops = {
2776 .start = netlink_seq_start,
2777 .next = netlink_seq_next,
2778 .stop = netlink_seq_stop,
2779 .show = netlink_seq_show,
2780};
2781#endif
2782
2783int netlink_register_notifier(struct notifier_block *nb)
2784{
2785 return blocking_notifier_chain_register(&netlink_chain, nb);
2786}
2787EXPORT_SYMBOL(netlink_register_notifier);
2788
2789int netlink_unregister_notifier(struct notifier_block *nb)
2790{
2791 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2792}
2793EXPORT_SYMBOL(netlink_unregister_notifier);
2794
2795static const struct proto_ops netlink_ops = {
2796 .family = PF_NETLINK,
2797 .owner = THIS_MODULE,
2798 .release = netlink_release,
2799 .bind = netlink_bind,
2800 .connect = netlink_connect,
2801 .socketpair = sock_no_socketpair,
2802 .accept = sock_no_accept,
2803 .getname = netlink_getname,
2804 .poll = datagram_poll,
2805 .ioctl = netlink_ioctl,
2806 .listen = sock_no_listen,
2807 .shutdown = sock_no_shutdown,
2808 .setsockopt = netlink_setsockopt,
2809 .getsockopt = netlink_getsockopt,
2810 .sendmsg = netlink_sendmsg,
2811 .recvmsg = netlink_recvmsg,
2812 .mmap = sock_no_mmap,
2813};
2814
2815static const struct net_proto_family netlink_family_ops = {
2816 .family = PF_NETLINK,
2817 .create = netlink_create,
2818 .owner = THIS_MODULE, /* for consistency 8) */
2819};
2820
2821static int __net_init netlink_net_init(struct net *net)
2822{
2823#ifdef CONFIG_PROC_FS
2824 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2825 sizeof(struct nl_seq_iter)))
2826 return -ENOMEM;
2827#endif
2828 return 0;
2829}
2830
2831static void __net_exit netlink_net_exit(struct net *net)
2832{
2833#ifdef CONFIG_PROC_FS
2834 remove_proc_entry("netlink", net->proc_net);
2835#endif
2836}
2837
2838static void __init netlink_add_usersock_entry(void)
2839{
2840 struct listeners *listeners;
2841 int groups = 32;
2842
2843 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2844 if (!listeners)
2845 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2846
2847 netlink_table_grab();
2848
2849 nl_table[NETLINK_USERSOCK].groups = groups;
2850 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2851 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2852 nl_table[NETLINK_USERSOCK].registered = 1;
2853 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2854
2855 netlink_table_ungrab();
2856}
2857
2858static struct pernet_operations __net_initdata netlink_net_ops = {
2859 .init = netlink_net_init,
2860 .exit = netlink_net_exit,
2861};
2862
2863static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2864{
2865 const struct netlink_sock *nlk = data;
2866 struct netlink_compare_arg arg;
2867
2868 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2869 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2870}
2871
2872static const struct rhashtable_params netlink_rhashtable_params = {
2873 .head_offset = offsetof(struct netlink_sock, node),
2874 .key_len = netlink_compare_arg_len,
2875 .obj_hashfn = netlink_hash,
2876 .obj_cmpfn = netlink_compare,
2877 .automatic_shrinking = true,
2878};
2879
2880#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2881BTF_ID_LIST(btf_netlink_sock_id)
2882BTF_ID(struct, netlink_sock)
2883
2884static const struct bpf_iter_seq_info netlink_seq_info = {
2885 .seq_ops = &netlink_seq_ops,
2886 .init_seq_private = bpf_iter_init_seq_net,
2887 .fini_seq_private = bpf_iter_fini_seq_net,
2888 .seq_priv_size = sizeof(struct nl_seq_iter),
2889};
2890
2891static struct bpf_iter_reg netlink_reg_info = {
2892 .target = "netlink",
2893 .ctx_arg_info_size = 1,
2894 .ctx_arg_info = {
2895 { offsetof(struct bpf_iter__netlink, sk),
2896 PTR_TO_BTF_ID_OR_NULL },
2897 },
2898 .seq_info = &netlink_seq_info,
2899};
2900
2901static int __init bpf_iter_register(void)
2902{
2903 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2904 return bpf_iter_reg_target(&netlink_reg_info);
2905}
2906#endif
2907
2908static int __init netlink_proto_init(void)
2909{
2910 int i;
2911 int err = proto_register(&netlink_proto, 0);
2912
2913 if (err != 0)
2914 goto out;
2915
2916#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2917 err = bpf_iter_register();
2918 if (err)
2919 goto out;
2920#endif
2921
2922 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2923
2924 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2925 if (!nl_table)
2926 goto panic;
2927
2928 for (i = 0; i < MAX_LINKS; i++) {
2929 if (rhashtable_init(&nl_table[i].hash,
2930 &netlink_rhashtable_params) < 0) {
2931 while (--i > 0)
2932 rhashtable_destroy(&nl_table[i].hash);
2933 kfree(nl_table);
2934 goto panic;
2935 }
2936 }
2937
2938 netlink_add_usersock_entry();
2939
2940 sock_register(&netlink_family_ops);
2941 register_pernet_subsys(&netlink_net_ops);
2942 register_pernet_subsys(&netlink_tap_net_ops);
2943 /* The netlink device handler may be needed early. */
2944 rtnetlink_init();
2945out:
2946 return err;
2947panic:
2948 panic("netlink_init: Cannot allocate nl_table\n");
2949}
2950
2951core_initcall(netlink_proto_init);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * NETLINK Kernel-user communication protocol.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
8 *
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
19 */
20
21#include <linux/module.h>
22
23#include <linux/bpf.h>
24#include <linux/capability.h>
25#include <linux/kernel.h>
26#include <linux/filter.h>
27#include <linux/init.h>
28#include <linux/signal.h>
29#include <linux/sched.h>
30#include <linux/errno.h>
31#include <linux/string.h>
32#include <linux/stat.h>
33#include <linux/socket.h>
34#include <linux/un.h>
35#include <linux/fcntl.h>
36#include <linux/termios.h>
37#include <linux/sockios.h>
38#include <linux/net.h>
39#include <linux/fs.h>
40#include <linux/slab.h>
41#include <linux/uaccess.h>
42#include <linux/skbuff.h>
43#include <linux/netdevice.h>
44#include <linux/rtnetlink.h>
45#include <linux/proc_fs.h>
46#include <linux/seq_file.h>
47#include <linux/notifier.h>
48#include <linux/security.h>
49#include <linux/jhash.h>
50#include <linux/jiffies.h>
51#include <linux/random.h>
52#include <linux/bitops.h>
53#include <linux/mm.h>
54#include <linux/types.h>
55#include <linux/audit.h>
56#include <linux/mutex.h>
57#include <linux/vmalloc.h>
58#include <linux/if_arp.h>
59#include <linux/rhashtable.h>
60#include <asm/cacheflush.h>
61#include <linux/hash.h>
62#include <linux/genetlink.h>
63#include <linux/net_namespace.h>
64#include <linux/nospec.h>
65#include <linux/btf_ids.h>
66
67#include <net/net_namespace.h>
68#include <net/netns/generic.h>
69#include <net/sock.h>
70#include <net/scm.h>
71#include <net/netlink.h>
72#define CREATE_TRACE_POINTS
73#include <trace/events/netlink.h>
74
75#include "af_netlink.h"
76
77struct listeners {
78 struct rcu_head rcu;
79 unsigned long masks[];
80};
81
82/* state bits */
83#define NETLINK_S_CONGESTED 0x0
84
85static inline int netlink_is_kernel(struct sock *sk)
86{
87 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
88}
89
90struct netlink_table *nl_table __read_mostly;
91EXPORT_SYMBOL_GPL(nl_table);
92
93static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
94
95static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
96
97static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
98 "nlk_cb_mutex-ROUTE",
99 "nlk_cb_mutex-1",
100 "nlk_cb_mutex-USERSOCK",
101 "nlk_cb_mutex-FIREWALL",
102 "nlk_cb_mutex-SOCK_DIAG",
103 "nlk_cb_mutex-NFLOG",
104 "nlk_cb_mutex-XFRM",
105 "nlk_cb_mutex-SELINUX",
106 "nlk_cb_mutex-ISCSI",
107 "nlk_cb_mutex-AUDIT",
108 "nlk_cb_mutex-FIB_LOOKUP",
109 "nlk_cb_mutex-CONNECTOR",
110 "nlk_cb_mutex-NETFILTER",
111 "nlk_cb_mutex-IP6_FW",
112 "nlk_cb_mutex-DNRTMSG",
113 "nlk_cb_mutex-KOBJECT_UEVENT",
114 "nlk_cb_mutex-GENERIC",
115 "nlk_cb_mutex-17",
116 "nlk_cb_mutex-SCSITRANSPORT",
117 "nlk_cb_mutex-ECRYPTFS",
118 "nlk_cb_mutex-RDMA",
119 "nlk_cb_mutex-CRYPTO",
120 "nlk_cb_mutex-SMC",
121 "nlk_cb_mutex-23",
122 "nlk_cb_mutex-24",
123 "nlk_cb_mutex-25",
124 "nlk_cb_mutex-26",
125 "nlk_cb_mutex-27",
126 "nlk_cb_mutex-28",
127 "nlk_cb_mutex-29",
128 "nlk_cb_mutex-30",
129 "nlk_cb_mutex-31",
130 "nlk_cb_mutex-MAX_LINKS"
131};
132
133static int netlink_dump(struct sock *sk);
134
135/* nl_table locking explained:
136 * Lookup and traversal are protected with an RCU read-side lock. Insertion
137 * and removal are protected with per bucket lock while using RCU list
138 * modification primitives and may run in parallel to RCU protected lookups.
139 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
140 * been acquired * either during or after the socket has been removed from
141 * the list and after an RCU grace period.
142 */
143DEFINE_RWLOCK(nl_table_lock);
144EXPORT_SYMBOL_GPL(nl_table_lock);
145static atomic_t nl_table_users = ATOMIC_INIT(0);
146
147#define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
148
149static BLOCKING_NOTIFIER_HEAD(netlink_chain);
150
151
152static const struct rhashtable_params netlink_rhashtable_params;
153
154void do_trace_netlink_extack(const char *msg)
155{
156 trace_netlink_extack(msg);
157}
158EXPORT_SYMBOL(do_trace_netlink_extack);
159
160static inline u32 netlink_group_mask(u32 group)
161{
162 if (group > 32)
163 return 0;
164 return group ? 1 << (group - 1) : 0;
165}
166
167static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
168 gfp_t gfp_mask)
169{
170 unsigned int len = skb_end_offset(skb);
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 struct netlink_sock *nlk = nlk_sk(sk);
353
354 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
355 if (!test_and_set_bit(NETLINK_S_CONGESTED,
356 &nlk_sk(sk)->state)) {
357 sk->sk_err = ENOBUFS;
358 sk_error_report(sk);
359 }
360 }
361 atomic_inc(&sk->sk_drops);
362}
363
364static void netlink_rcv_wake(struct sock *sk)
365{
366 struct netlink_sock *nlk = nlk_sk(sk);
367
368 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
369 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
370 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
371 wake_up_interruptible(&nlk->wait);
372}
373
374static void netlink_skb_destructor(struct sk_buff *skb)
375{
376 if (is_vmalloc_addr(skb->head)) {
377 if (!skb->cloned ||
378 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
379 vfree(skb->head);
380
381 skb->head = NULL;
382 }
383 if (skb->sk != NULL)
384 sock_rfree(skb);
385}
386
387static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
388{
389 WARN_ON(skb->sk != NULL);
390 skb->sk = sk;
391 skb->destructor = netlink_skb_destructor;
392 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
393 sk_mem_charge(sk, skb->truesize);
394}
395
396static void netlink_sock_destruct(struct sock *sk)
397{
398 struct netlink_sock *nlk = nlk_sk(sk);
399
400 if (nlk->cb_running) {
401 if (nlk->cb.done)
402 nlk->cb.done(&nlk->cb);
403 module_put(nlk->cb.module);
404 kfree_skb(nlk->cb.skb);
405 }
406
407 skb_queue_purge(&sk->sk_receive_queue);
408
409 if (!sock_flag(sk, SOCK_DEAD)) {
410 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
411 return;
412 }
413
414 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
415 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
416 WARN_ON(nlk_sk(sk)->groups);
417}
418
419static void netlink_sock_destruct_work(struct work_struct *work)
420{
421 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
422 work);
423
424 sk_free(&nlk->sk);
425}
426
427/* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
428 * SMP. Look, when several writers sleep and reader wakes them up, all but one
429 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
430 * this, _but_ remember, it adds useless work on UP machines.
431 */
432
433void netlink_table_grab(void)
434 __acquires(nl_table_lock)
435{
436 might_sleep();
437
438 write_lock_irq(&nl_table_lock);
439
440 if (atomic_read(&nl_table_users)) {
441 DECLARE_WAITQUEUE(wait, current);
442
443 add_wait_queue_exclusive(&nl_table_wait, &wait);
444 for (;;) {
445 set_current_state(TASK_UNINTERRUPTIBLE);
446 if (atomic_read(&nl_table_users) == 0)
447 break;
448 write_unlock_irq(&nl_table_lock);
449 schedule();
450 write_lock_irq(&nl_table_lock);
451 }
452
453 __set_current_state(TASK_RUNNING);
454 remove_wait_queue(&nl_table_wait, &wait);
455 }
456}
457
458void netlink_table_ungrab(void)
459 __releases(nl_table_lock)
460{
461 write_unlock_irq(&nl_table_lock);
462 wake_up(&nl_table_wait);
463}
464
465static inline void
466netlink_lock_table(void)
467{
468 unsigned long flags;
469
470 /* read_lock() synchronizes us to netlink_table_grab */
471
472 read_lock_irqsave(&nl_table_lock, flags);
473 atomic_inc(&nl_table_users);
474 read_unlock_irqrestore(&nl_table_lock, flags);
475}
476
477static inline void
478netlink_unlock_table(void)
479{
480 if (atomic_dec_and_test(&nl_table_users))
481 wake_up(&nl_table_wait);
482}
483
484struct netlink_compare_arg
485{
486 possible_net_t pnet;
487 u32 portid;
488};
489
490/* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
491#define netlink_compare_arg_len \
492 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
493
494static inline int netlink_compare(struct rhashtable_compare_arg *arg,
495 const void *ptr)
496{
497 const struct netlink_compare_arg *x = arg->key;
498 const struct netlink_sock *nlk = ptr;
499
500 return nlk->portid != x->portid ||
501 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
502}
503
504static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
505 struct net *net, u32 portid)
506{
507 memset(arg, 0, sizeof(*arg));
508 write_pnet(&arg->pnet, net);
509 arg->portid = portid;
510}
511
512static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
513 struct net *net)
514{
515 struct netlink_compare_arg arg;
516
517 netlink_compare_arg_init(&arg, net, portid);
518 return rhashtable_lookup_fast(&table->hash, &arg,
519 netlink_rhashtable_params);
520}
521
522static int __netlink_insert(struct netlink_table *table, struct sock *sk)
523{
524 struct netlink_compare_arg arg;
525
526 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
527 return rhashtable_lookup_insert_key(&table->hash, &arg,
528 &nlk_sk(sk)->node,
529 netlink_rhashtable_params);
530}
531
532static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
533{
534 struct netlink_table *table = &nl_table[protocol];
535 struct sock *sk;
536
537 rcu_read_lock();
538 sk = __netlink_lookup(table, portid, net);
539 if (sk)
540 sock_hold(sk);
541 rcu_read_unlock();
542
543 return sk;
544}
545
546static const struct proto_ops netlink_ops;
547
548static void
549netlink_update_listeners(struct sock *sk)
550{
551 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
552 unsigned long mask;
553 unsigned int i;
554 struct listeners *listeners;
555
556 listeners = nl_deref_protected(tbl->listeners);
557 if (!listeners)
558 return;
559
560 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
561 mask = 0;
562 sk_for_each_bound(sk, &tbl->mc_list) {
563 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
564 mask |= nlk_sk(sk)->groups[i];
565 }
566 listeners->masks[i] = mask;
567 }
568 /* this function is only called with the netlink table "grabbed", which
569 * makes sure updates are visible before bind or setsockopt return. */
570}
571
572static int netlink_insert(struct sock *sk, u32 portid)
573{
574 struct netlink_table *table = &nl_table[sk->sk_protocol];
575 int err;
576
577 lock_sock(sk);
578
579 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
580 if (nlk_sk(sk)->bound)
581 goto err;
582
583 /* portid can be read locklessly from netlink_getname(). */
584 WRITE_ONCE(nlk_sk(sk)->portid, portid);
585
586 sock_hold(sk);
587
588 err = __netlink_insert(table, sk);
589 if (err) {
590 /* In case the hashtable backend returns with -EBUSY
591 * from here, it must not escape to the caller.
592 */
593 if (unlikely(err == -EBUSY))
594 err = -EOVERFLOW;
595 if (err == -EEXIST)
596 err = -EADDRINUSE;
597 sock_put(sk);
598 goto err;
599 }
600
601 /* We need to ensure that the socket is hashed and visible. */
602 smp_wmb();
603 /* Paired with lockless reads from netlink_bind(),
604 * netlink_connect() and netlink_sendmsg().
605 */
606 WRITE_ONCE(nlk_sk(sk)->bound, portid);
607
608err:
609 release_sock(sk);
610 return err;
611}
612
613static void netlink_remove(struct sock *sk)
614{
615 struct netlink_table *table;
616
617 table = &nl_table[sk->sk_protocol];
618 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
619 netlink_rhashtable_params)) {
620 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
621 __sock_put(sk);
622 }
623
624 netlink_table_grab();
625 if (nlk_sk(sk)->subscriptions) {
626 __sk_del_bind_node(sk);
627 netlink_update_listeners(sk);
628 }
629 if (sk->sk_protocol == NETLINK_GENERIC)
630 atomic_inc(&genl_sk_destructing_cnt);
631 netlink_table_ungrab();
632}
633
634static struct proto netlink_proto = {
635 .name = "NETLINK",
636 .owner = THIS_MODULE,
637 .obj_size = sizeof(struct netlink_sock),
638};
639
640static int __netlink_create(struct net *net, struct socket *sock,
641 struct mutex *cb_mutex, int protocol,
642 int kern)
643{
644 struct sock *sk;
645 struct netlink_sock *nlk;
646
647 sock->ops = &netlink_ops;
648
649 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
650 if (!sk)
651 return -ENOMEM;
652
653 sock_init_data(sock, sk);
654
655 nlk = nlk_sk(sk);
656 if (cb_mutex) {
657 nlk->cb_mutex = cb_mutex;
658 } else {
659 nlk->cb_mutex = &nlk->cb_def_mutex;
660 mutex_init(nlk->cb_mutex);
661 lockdep_set_class_and_name(nlk->cb_mutex,
662 nlk_cb_mutex_keys + protocol,
663 nlk_cb_mutex_key_strings[protocol]);
664 }
665 init_waitqueue_head(&nlk->wait);
666
667 sk->sk_destruct = netlink_sock_destruct;
668 sk->sk_protocol = protocol;
669 return 0;
670}
671
672static int netlink_create(struct net *net, struct socket *sock, int protocol,
673 int kern)
674{
675 struct module *module = NULL;
676 struct mutex *cb_mutex;
677 struct netlink_sock *nlk;
678 int (*bind)(struct net *net, int group);
679 void (*unbind)(struct net *net, int group);
680 int err = 0;
681
682 sock->state = SS_UNCONNECTED;
683
684 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
685 return -ESOCKTNOSUPPORT;
686
687 if (protocol < 0 || protocol >= MAX_LINKS)
688 return -EPROTONOSUPPORT;
689 protocol = array_index_nospec(protocol, MAX_LINKS);
690
691 netlink_lock_table();
692#ifdef CONFIG_MODULES
693 if (!nl_table[protocol].registered) {
694 netlink_unlock_table();
695 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
696 netlink_lock_table();
697 }
698#endif
699 if (nl_table[protocol].registered &&
700 try_module_get(nl_table[protocol].module))
701 module = nl_table[protocol].module;
702 else
703 err = -EPROTONOSUPPORT;
704 cb_mutex = nl_table[protocol].cb_mutex;
705 bind = nl_table[protocol].bind;
706 unbind = nl_table[protocol].unbind;
707 netlink_unlock_table();
708
709 if (err < 0)
710 goto out;
711
712 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
713 if (err < 0)
714 goto out_module;
715
716 sock_prot_inuse_add(net, &netlink_proto, 1);
717
718 nlk = nlk_sk(sock->sk);
719 nlk->module = module;
720 nlk->netlink_bind = bind;
721 nlk->netlink_unbind = unbind;
722out:
723 return err;
724
725out_module:
726 module_put(module);
727 goto out;
728}
729
730static void deferred_put_nlk_sk(struct rcu_head *head)
731{
732 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
733 struct sock *sk = &nlk->sk;
734
735 kfree(nlk->groups);
736 nlk->groups = NULL;
737
738 if (!refcount_dec_and_test(&sk->sk_refcnt))
739 return;
740
741 if (nlk->cb_running && nlk->cb.done) {
742 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
743 schedule_work(&nlk->work);
744 return;
745 }
746
747 sk_free(sk);
748}
749
750static int netlink_release(struct socket *sock)
751{
752 struct sock *sk = sock->sk;
753 struct netlink_sock *nlk;
754
755 if (!sk)
756 return 0;
757
758 netlink_remove(sk);
759 sock_orphan(sk);
760 nlk = nlk_sk(sk);
761
762 /*
763 * OK. Socket is unlinked, any packets that arrive now
764 * will be purged.
765 */
766
767 /* must not acquire netlink_table_lock in any way again before unbind
768 * and notifying genetlink is done as otherwise it might deadlock
769 */
770 if (nlk->netlink_unbind) {
771 int i;
772
773 for (i = 0; i < nlk->ngroups; i++)
774 if (test_bit(i, nlk->groups))
775 nlk->netlink_unbind(sock_net(sk), i + 1);
776 }
777 if (sk->sk_protocol == NETLINK_GENERIC &&
778 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
779 wake_up(&genl_sk_destructing_waitq);
780
781 sock->sk = NULL;
782 wake_up_interruptible_all(&nlk->wait);
783
784 skb_queue_purge(&sk->sk_write_queue);
785
786 if (nlk->portid && nlk->bound) {
787 struct netlink_notify n = {
788 .net = sock_net(sk),
789 .protocol = sk->sk_protocol,
790 .portid = nlk->portid,
791 };
792 blocking_notifier_call_chain(&netlink_chain,
793 NETLINK_URELEASE, &n);
794 }
795
796 module_put(nlk->module);
797
798 if (netlink_is_kernel(sk)) {
799 netlink_table_grab();
800 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
801 if (--nl_table[sk->sk_protocol].registered == 0) {
802 struct listeners *old;
803
804 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
805 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
806 kfree_rcu(old, rcu);
807 nl_table[sk->sk_protocol].module = NULL;
808 nl_table[sk->sk_protocol].bind = NULL;
809 nl_table[sk->sk_protocol].unbind = NULL;
810 nl_table[sk->sk_protocol].flags = 0;
811 nl_table[sk->sk_protocol].registered = 0;
812 }
813 netlink_table_ungrab();
814 }
815
816 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
817
818 /* Because struct net might disappear soon, do not keep a pointer. */
819 if (!sk->sk_net_refcnt && sock_net(sk) != &init_net) {
820 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false);
821 /* Because of deferred_put_nlk_sk and use of work queue,
822 * it is possible netns will be freed before this socket.
823 */
824 sock_net_set(sk, &init_net);
825 __netns_tracker_alloc(&init_net, &sk->ns_tracker,
826 false, GFP_KERNEL);
827 }
828 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
829 return 0;
830}
831
832static int netlink_autobind(struct socket *sock)
833{
834 struct sock *sk = sock->sk;
835 struct net *net = sock_net(sk);
836 struct netlink_table *table = &nl_table[sk->sk_protocol];
837 s32 portid = task_tgid_vnr(current);
838 int err;
839 s32 rover = -4096;
840 bool ok;
841
842retry:
843 cond_resched();
844 rcu_read_lock();
845 ok = !__netlink_lookup(table, portid, net);
846 rcu_read_unlock();
847 if (!ok) {
848 /* Bind collision, search negative portid values. */
849 if (rover == -4096)
850 /* rover will be in range [S32_MIN, -4097] */
851 rover = S32_MIN + get_random_u32_below(-4096 - S32_MIN);
852 else if (rover >= -4096)
853 rover = -4097;
854 portid = rover--;
855 goto retry;
856 }
857
858 err = netlink_insert(sk, portid);
859 if (err == -EADDRINUSE)
860 goto retry;
861
862 /* If 2 threads race to autobind, that is fine. */
863 if (err == -EBUSY)
864 err = 0;
865
866 return err;
867}
868
869/**
870 * __netlink_ns_capable - General netlink message capability test
871 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
872 * @user_ns: The user namespace of the capability to use
873 * @cap: The capability to use
874 *
875 * Test to see if the opener of the socket we received the message
876 * from had when the netlink socket was created and the sender of the
877 * message has the capability @cap in the user namespace @user_ns.
878 */
879bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
880 struct user_namespace *user_ns, int cap)
881{
882 return ((nsp->flags & NETLINK_SKB_DST) ||
883 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
884 ns_capable(user_ns, cap);
885}
886EXPORT_SYMBOL(__netlink_ns_capable);
887
888/**
889 * netlink_ns_capable - General netlink message capability test
890 * @skb: socket buffer holding a netlink command from userspace
891 * @user_ns: The user namespace of the capability to use
892 * @cap: The capability to use
893 *
894 * Test to see if the opener of the socket we received the message
895 * from had when the netlink socket was created and the sender of the
896 * message has the capability @cap in the user namespace @user_ns.
897 */
898bool netlink_ns_capable(const struct sk_buff *skb,
899 struct user_namespace *user_ns, int cap)
900{
901 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
902}
903EXPORT_SYMBOL(netlink_ns_capable);
904
905/**
906 * netlink_capable - Netlink global message capability test
907 * @skb: socket buffer holding a netlink command from userspace
908 * @cap: The capability to use
909 *
910 * Test to see if the opener of the socket we received the message
911 * from had when the netlink socket was created and the sender of the
912 * message has the capability @cap in all user namespaces.
913 */
914bool netlink_capable(const struct sk_buff *skb, int cap)
915{
916 return netlink_ns_capable(skb, &init_user_ns, cap);
917}
918EXPORT_SYMBOL(netlink_capable);
919
920/**
921 * netlink_net_capable - Netlink network namespace message capability test
922 * @skb: socket buffer holding a netlink command from userspace
923 * @cap: The capability to use
924 *
925 * Test to see if the opener of the socket we received the message
926 * from had when the netlink socket was created and the sender of the
927 * message has the capability @cap over the network namespace of
928 * the socket we received the message from.
929 */
930bool netlink_net_capable(const struct sk_buff *skb, int cap)
931{
932 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
933}
934EXPORT_SYMBOL(netlink_net_capable);
935
936static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
937{
938 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
939 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
940}
941
942static void
943netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
944{
945 struct netlink_sock *nlk = nlk_sk(sk);
946
947 if (nlk->subscriptions && !subscriptions)
948 __sk_del_bind_node(sk);
949 else if (!nlk->subscriptions && subscriptions)
950 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
951 nlk->subscriptions = subscriptions;
952}
953
954static int netlink_realloc_groups(struct sock *sk)
955{
956 struct netlink_sock *nlk = nlk_sk(sk);
957 unsigned int groups;
958 unsigned long *new_groups;
959 int err = 0;
960
961 netlink_table_grab();
962
963 groups = nl_table[sk->sk_protocol].groups;
964 if (!nl_table[sk->sk_protocol].registered) {
965 err = -ENOENT;
966 goto out_unlock;
967 }
968
969 if (nlk->ngroups >= groups)
970 goto out_unlock;
971
972 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
973 if (new_groups == NULL) {
974 err = -ENOMEM;
975 goto out_unlock;
976 }
977 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
978 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
979
980 nlk->groups = new_groups;
981 nlk->ngroups = groups;
982 out_unlock:
983 netlink_table_ungrab();
984 return err;
985}
986
987static void netlink_undo_bind(int group, long unsigned int groups,
988 struct sock *sk)
989{
990 struct netlink_sock *nlk = nlk_sk(sk);
991 int undo;
992
993 if (!nlk->netlink_unbind)
994 return;
995
996 for (undo = 0; undo < group; undo++)
997 if (test_bit(undo, &groups))
998 nlk->netlink_unbind(sock_net(sk), undo + 1);
999}
1000
1001static int netlink_bind(struct socket *sock, struct sockaddr *addr,
1002 int addr_len)
1003{
1004 struct sock *sk = sock->sk;
1005 struct net *net = sock_net(sk);
1006 struct netlink_sock *nlk = nlk_sk(sk);
1007 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1008 int err = 0;
1009 unsigned long groups;
1010 bool bound;
1011
1012 if (addr_len < sizeof(struct sockaddr_nl))
1013 return -EINVAL;
1014
1015 if (nladdr->nl_family != AF_NETLINK)
1016 return -EINVAL;
1017 groups = nladdr->nl_groups;
1018
1019 /* Only superuser is allowed to listen multicasts */
1020 if (groups) {
1021 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1022 return -EPERM;
1023 err = netlink_realloc_groups(sk);
1024 if (err)
1025 return err;
1026 }
1027
1028 if (nlk->ngroups < BITS_PER_LONG)
1029 groups &= (1UL << nlk->ngroups) - 1;
1030
1031 /* Paired with WRITE_ONCE() in netlink_insert() */
1032 bound = READ_ONCE(nlk->bound);
1033 if (bound) {
1034 /* Ensure nlk->portid is up-to-date. */
1035 smp_rmb();
1036
1037 if (nladdr->nl_pid != nlk->portid)
1038 return -EINVAL;
1039 }
1040
1041 if (nlk->netlink_bind && groups) {
1042 int group;
1043
1044 /* nl_groups is a u32, so cap the maximum groups we can bind */
1045 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1046 if (!test_bit(group, &groups))
1047 continue;
1048 err = nlk->netlink_bind(net, group + 1);
1049 if (!err)
1050 continue;
1051 netlink_undo_bind(group, groups, sk);
1052 return err;
1053 }
1054 }
1055
1056 /* No need for barriers here as we return to user-space without
1057 * using any of the bound attributes.
1058 */
1059 netlink_lock_table();
1060 if (!bound) {
1061 err = nladdr->nl_pid ?
1062 netlink_insert(sk, nladdr->nl_pid) :
1063 netlink_autobind(sock);
1064 if (err) {
1065 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1066 goto unlock;
1067 }
1068 }
1069
1070 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1071 goto unlock;
1072 netlink_unlock_table();
1073
1074 netlink_table_grab();
1075 netlink_update_subscriptions(sk, nlk->subscriptions +
1076 hweight32(groups) -
1077 hweight32(nlk->groups[0]));
1078 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1079 netlink_update_listeners(sk);
1080 netlink_table_ungrab();
1081
1082 return 0;
1083
1084unlock:
1085 netlink_unlock_table();
1086 return err;
1087}
1088
1089static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1090 int alen, int flags)
1091{
1092 int err = 0;
1093 struct sock *sk = sock->sk;
1094 struct netlink_sock *nlk = nlk_sk(sk);
1095 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1096
1097 if (alen < sizeof(addr->sa_family))
1098 return -EINVAL;
1099
1100 if (addr->sa_family == AF_UNSPEC) {
1101 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1102 WRITE_ONCE(sk->sk_state, NETLINK_UNCONNECTED);
1103 /* dst_portid and dst_group can be read locklessly */
1104 WRITE_ONCE(nlk->dst_portid, 0);
1105 WRITE_ONCE(nlk->dst_group, 0);
1106 return 0;
1107 }
1108 if (addr->sa_family != AF_NETLINK)
1109 return -EINVAL;
1110
1111 if (alen < sizeof(struct sockaddr_nl))
1112 return -EINVAL;
1113
1114 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1115 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1116 return -EPERM;
1117
1118 /* No need for barriers here as we return to user-space without
1119 * using any of the bound attributes.
1120 * Paired with WRITE_ONCE() in netlink_insert().
1121 */
1122 if (!READ_ONCE(nlk->bound))
1123 err = netlink_autobind(sock);
1124
1125 if (err == 0) {
1126 /* paired with READ_ONCE() in netlink_getsockbyportid() */
1127 WRITE_ONCE(sk->sk_state, NETLINK_CONNECTED);
1128 /* dst_portid and dst_group can be read locklessly */
1129 WRITE_ONCE(nlk->dst_portid, nladdr->nl_pid);
1130 WRITE_ONCE(nlk->dst_group, ffs(nladdr->nl_groups));
1131 }
1132
1133 return err;
1134}
1135
1136static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1137 int peer)
1138{
1139 struct sock *sk = sock->sk;
1140 struct netlink_sock *nlk = nlk_sk(sk);
1141 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1142
1143 nladdr->nl_family = AF_NETLINK;
1144 nladdr->nl_pad = 0;
1145
1146 if (peer) {
1147 /* Paired with WRITE_ONCE() in netlink_connect() */
1148 nladdr->nl_pid = READ_ONCE(nlk->dst_portid);
1149 nladdr->nl_groups = netlink_group_mask(READ_ONCE(nlk->dst_group));
1150 } else {
1151 /* Paired with WRITE_ONCE() in netlink_insert() */
1152 nladdr->nl_pid = READ_ONCE(nlk->portid);
1153 netlink_lock_table();
1154 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1155 netlink_unlock_table();
1156 }
1157 return sizeof(*nladdr);
1158}
1159
1160static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1161 unsigned long arg)
1162{
1163 /* try to hand this ioctl down to the NIC drivers.
1164 */
1165 return -ENOIOCTLCMD;
1166}
1167
1168static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1169{
1170 struct sock *sock;
1171 struct netlink_sock *nlk;
1172
1173 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1174 if (!sock)
1175 return ERR_PTR(-ECONNREFUSED);
1176
1177 /* Don't bother queuing skb if kernel socket has no input function */
1178 nlk = nlk_sk(sock);
1179 /* dst_portid and sk_state can be changed in netlink_connect() */
1180 if (READ_ONCE(sock->sk_state) == NETLINK_CONNECTED &&
1181 READ_ONCE(nlk->dst_portid) != nlk_sk(ssk)->portid) {
1182 sock_put(sock);
1183 return ERR_PTR(-ECONNREFUSED);
1184 }
1185 return sock;
1186}
1187
1188struct sock *netlink_getsockbyfilp(struct file *filp)
1189{
1190 struct inode *inode = file_inode(filp);
1191 struct sock *sock;
1192
1193 if (!S_ISSOCK(inode->i_mode))
1194 return ERR_PTR(-ENOTSOCK);
1195
1196 sock = SOCKET_I(inode)->sk;
1197 if (sock->sk_family != AF_NETLINK)
1198 return ERR_PTR(-EINVAL);
1199
1200 sock_hold(sock);
1201 return sock;
1202}
1203
1204static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1205 int broadcast)
1206{
1207 struct sk_buff *skb;
1208 void *data;
1209
1210 if (size <= NLMSG_GOODSIZE || broadcast)
1211 return alloc_skb(size, GFP_KERNEL);
1212
1213 size = SKB_DATA_ALIGN(size) +
1214 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1215
1216 data = vmalloc(size);
1217 if (data == NULL)
1218 return NULL;
1219
1220 skb = __build_skb(data, size);
1221 if (skb == NULL)
1222 vfree(data);
1223 else
1224 skb->destructor = netlink_skb_destructor;
1225
1226 return skb;
1227}
1228
1229/*
1230 * Attach a skb to a netlink socket.
1231 * The caller must hold a reference to the destination socket. On error, the
1232 * reference is dropped. The skb is not send to the destination, just all
1233 * all error checks are performed and memory in the queue is reserved.
1234 * Return values:
1235 * < 0: error. skb freed, reference to sock dropped.
1236 * 0: continue
1237 * 1: repeat lookup - reference dropped while waiting for socket memory.
1238 */
1239int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1240 long *timeo, struct sock *ssk)
1241{
1242 struct netlink_sock *nlk;
1243
1244 nlk = nlk_sk(sk);
1245
1246 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1247 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1248 DECLARE_WAITQUEUE(wait, current);
1249 if (!*timeo) {
1250 if (!ssk || netlink_is_kernel(ssk))
1251 netlink_overrun(sk);
1252 sock_put(sk);
1253 kfree_skb(skb);
1254 return -EAGAIN;
1255 }
1256
1257 __set_current_state(TASK_INTERRUPTIBLE);
1258 add_wait_queue(&nlk->wait, &wait);
1259
1260 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1261 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1262 !sock_flag(sk, SOCK_DEAD))
1263 *timeo = schedule_timeout(*timeo);
1264
1265 __set_current_state(TASK_RUNNING);
1266 remove_wait_queue(&nlk->wait, &wait);
1267 sock_put(sk);
1268
1269 if (signal_pending(current)) {
1270 kfree_skb(skb);
1271 return sock_intr_errno(*timeo);
1272 }
1273 return 1;
1274 }
1275 netlink_skb_set_owner_r(skb, sk);
1276 return 0;
1277}
1278
1279static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1280{
1281 int len = skb->len;
1282
1283 netlink_deliver_tap(sock_net(sk), skb);
1284
1285 skb_queue_tail(&sk->sk_receive_queue, skb);
1286 sk->sk_data_ready(sk);
1287 return len;
1288}
1289
1290int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1291{
1292 int len = __netlink_sendskb(sk, skb);
1293
1294 sock_put(sk);
1295 return len;
1296}
1297
1298void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1299{
1300 kfree_skb(skb);
1301 sock_put(sk);
1302}
1303
1304static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1305{
1306 int delta;
1307
1308 WARN_ON(skb->sk != NULL);
1309 delta = skb->end - skb->tail;
1310 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1311 return skb;
1312
1313 if (skb_shared(skb)) {
1314 struct sk_buff *nskb = skb_clone(skb, allocation);
1315 if (!nskb)
1316 return skb;
1317 consume_skb(skb);
1318 skb = nskb;
1319 }
1320
1321 pskb_expand_head(skb, 0, -delta,
1322 (allocation & ~__GFP_DIRECT_RECLAIM) |
1323 __GFP_NOWARN | __GFP_NORETRY);
1324 return skb;
1325}
1326
1327static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1328 struct sock *ssk)
1329{
1330 int ret;
1331 struct netlink_sock *nlk = nlk_sk(sk);
1332
1333 ret = -ECONNREFUSED;
1334 if (nlk->netlink_rcv != NULL) {
1335 ret = skb->len;
1336 netlink_skb_set_owner_r(skb, sk);
1337 NETLINK_CB(skb).sk = ssk;
1338 netlink_deliver_tap_kernel(sk, ssk, skb);
1339 nlk->netlink_rcv(skb);
1340 consume_skb(skb);
1341 } else {
1342 kfree_skb(skb);
1343 }
1344 sock_put(sk);
1345 return ret;
1346}
1347
1348int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1349 u32 portid, int nonblock)
1350{
1351 struct sock *sk;
1352 int err;
1353 long timeo;
1354
1355 skb = netlink_trim(skb, gfp_any());
1356
1357 timeo = sock_sndtimeo(ssk, nonblock);
1358retry:
1359 sk = netlink_getsockbyportid(ssk, portid);
1360 if (IS_ERR(sk)) {
1361 kfree_skb(skb);
1362 return PTR_ERR(sk);
1363 }
1364 if (netlink_is_kernel(sk))
1365 return netlink_unicast_kernel(sk, skb, ssk);
1366
1367 if (sk_filter(sk, skb)) {
1368 err = skb->len;
1369 kfree_skb(skb);
1370 sock_put(sk);
1371 return err;
1372 }
1373
1374 err = netlink_attachskb(sk, skb, &timeo, ssk);
1375 if (err == 1)
1376 goto retry;
1377 if (err)
1378 return err;
1379
1380 return netlink_sendskb(sk, skb);
1381}
1382EXPORT_SYMBOL(netlink_unicast);
1383
1384int netlink_has_listeners(struct sock *sk, unsigned int group)
1385{
1386 int res = 0;
1387 struct listeners *listeners;
1388
1389 BUG_ON(!netlink_is_kernel(sk));
1390
1391 rcu_read_lock();
1392 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1393
1394 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1395 res = test_bit(group - 1, listeners->masks);
1396
1397 rcu_read_unlock();
1398
1399 return res;
1400}
1401EXPORT_SYMBOL_GPL(netlink_has_listeners);
1402
1403bool netlink_strict_get_check(struct sk_buff *skb)
1404{
1405 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1406
1407 return nlk->flags & NETLINK_F_STRICT_CHK;
1408}
1409EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1410
1411static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1412{
1413 struct netlink_sock *nlk = nlk_sk(sk);
1414
1415 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1416 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1417 netlink_skb_set_owner_r(skb, sk);
1418 __netlink_sendskb(sk, skb);
1419 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1420 }
1421 return -1;
1422}
1423
1424struct netlink_broadcast_data {
1425 struct sock *exclude_sk;
1426 struct net *net;
1427 u32 portid;
1428 u32 group;
1429 int failure;
1430 int delivery_failure;
1431 int congested;
1432 int delivered;
1433 gfp_t allocation;
1434 struct sk_buff *skb, *skb2;
1435};
1436
1437static void do_one_broadcast(struct sock *sk,
1438 struct netlink_broadcast_data *p)
1439{
1440 struct netlink_sock *nlk = nlk_sk(sk);
1441 int val;
1442
1443 if (p->exclude_sk == sk)
1444 return;
1445
1446 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1447 !test_bit(p->group - 1, nlk->groups))
1448 return;
1449
1450 if (!net_eq(sock_net(sk), p->net)) {
1451 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1452 return;
1453
1454 if (!peernet_has_id(sock_net(sk), p->net))
1455 return;
1456
1457 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1458 CAP_NET_BROADCAST))
1459 return;
1460 }
1461
1462 if (p->failure) {
1463 netlink_overrun(sk);
1464 return;
1465 }
1466
1467 sock_hold(sk);
1468 if (p->skb2 == NULL) {
1469 if (skb_shared(p->skb)) {
1470 p->skb2 = skb_clone(p->skb, p->allocation);
1471 } else {
1472 p->skb2 = skb_get(p->skb);
1473 /*
1474 * skb ownership may have been set when
1475 * delivered to a previous socket.
1476 */
1477 skb_orphan(p->skb2);
1478 }
1479 }
1480 if (p->skb2 == NULL) {
1481 netlink_overrun(sk);
1482 /* Clone failed. Notify ALL listeners. */
1483 p->failure = 1;
1484 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1485 p->delivery_failure = 1;
1486 goto out;
1487 }
1488 if (sk_filter(sk, p->skb2)) {
1489 kfree_skb(p->skb2);
1490 p->skb2 = NULL;
1491 goto out;
1492 }
1493 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1494 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1495 NETLINK_CB(p->skb2).nsid_is_set = true;
1496 val = netlink_broadcast_deliver(sk, p->skb2);
1497 if (val < 0) {
1498 netlink_overrun(sk);
1499 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1500 p->delivery_failure = 1;
1501 } else {
1502 p->congested |= val;
1503 p->delivered = 1;
1504 p->skb2 = NULL;
1505 }
1506out:
1507 sock_put(sk);
1508}
1509
1510int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1511 u32 group, gfp_t allocation)
1512{
1513 struct net *net = sock_net(ssk);
1514 struct netlink_broadcast_data info;
1515 struct sock *sk;
1516
1517 skb = netlink_trim(skb, allocation);
1518
1519 info.exclude_sk = ssk;
1520 info.net = net;
1521 info.portid = portid;
1522 info.group = group;
1523 info.failure = 0;
1524 info.delivery_failure = 0;
1525 info.congested = 0;
1526 info.delivered = 0;
1527 info.allocation = allocation;
1528 info.skb = skb;
1529 info.skb2 = NULL;
1530
1531 /* While we sleep in clone, do not allow to change socket list */
1532
1533 netlink_lock_table();
1534
1535 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1536 do_one_broadcast(sk, &info);
1537
1538 consume_skb(skb);
1539
1540 netlink_unlock_table();
1541
1542 if (info.delivery_failure) {
1543 kfree_skb(info.skb2);
1544 return -ENOBUFS;
1545 }
1546 consume_skb(info.skb2);
1547
1548 if (info.delivered) {
1549 if (info.congested && gfpflags_allow_blocking(allocation))
1550 yield();
1551 return 0;
1552 }
1553 return -ESRCH;
1554}
1555EXPORT_SYMBOL(netlink_broadcast);
1556
1557struct netlink_set_err_data {
1558 struct sock *exclude_sk;
1559 u32 portid;
1560 u32 group;
1561 int code;
1562};
1563
1564static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1565{
1566 struct netlink_sock *nlk = nlk_sk(sk);
1567 int ret = 0;
1568
1569 if (sk == p->exclude_sk)
1570 goto out;
1571
1572 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1573 goto out;
1574
1575 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1576 !test_bit(p->group - 1, nlk->groups))
1577 goto out;
1578
1579 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1580 ret = 1;
1581 goto out;
1582 }
1583
1584 sk->sk_err = p->code;
1585 sk_error_report(sk);
1586out:
1587 return ret;
1588}
1589
1590/**
1591 * netlink_set_err - report error to broadcast listeners
1592 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1593 * @portid: the PORTID of a process that we want to skip (if any)
1594 * @group: the broadcast group that will notice the error
1595 * @code: error code, must be negative (as usual in kernelspace)
1596 *
1597 * This function returns the number of broadcast listeners that have set the
1598 * NETLINK_NO_ENOBUFS socket option.
1599 */
1600int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1601{
1602 struct netlink_set_err_data info;
1603 struct sock *sk;
1604 int ret = 0;
1605
1606 info.exclude_sk = ssk;
1607 info.portid = portid;
1608 info.group = group;
1609 /* sk->sk_err wants a positive error value */
1610 info.code = -code;
1611
1612 read_lock(&nl_table_lock);
1613
1614 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1615 ret += do_one_set_err(sk, &info);
1616
1617 read_unlock(&nl_table_lock);
1618 return ret;
1619}
1620EXPORT_SYMBOL(netlink_set_err);
1621
1622/* must be called with netlink table grabbed */
1623static void netlink_update_socket_mc(struct netlink_sock *nlk,
1624 unsigned int group,
1625 int is_new)
1626{
1627 int old, new = !!is_new, subscriptions;
1628
1629 old = test_bit(group - 1, nlk->groups);
1630 subscriptions = nlk->subscriptions - old + new;
1631 if (new)
1632 __set_bit(group - 1, nlk->groups);
1633 else
1634 __clear_bit(group - 1, nlk->groups);
1635 netlink_update_subscriptions(&nlk->sk, subscriptions);
1636 netlink_update_listeners(&nlk->sk);
1637}
1638
1639static int netlink_setsockopt(struct socket *sock, int level, int optname,
1640 sockptr_t optval, unsigned int optlen)
1641{
1642 struct sock *sk = sock->sk;
1643 struct netlink_sock *nlk = nlk_sk(sk);
1644 unsigned int val = 0;
1645 int err;
1646
1647 if (level != SOL_NETLINK)
1648 return -ENOPROTOOPT;
1649
1650 if (optlen >= sizeof(int) &&
1651 copy_from_sockptr(&val, optval, sizeof(val)))
1652 return -EFAULT;
1653
1654 switch (optname) {
1655 case NETLINK_PKTINFO:
1656 if (val)
1657 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1658 else
1659 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1660 err = 0;
1661 break;
1662 case NETLINK_ADD_MEMBERSHIP:
1663 case NETLINK_DROP_MEMBERSHIP: {
1664 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1665 return -EPERM;
1666 err = netlink_realloc_groups(sk);
1667 if (err)
1668 return err;
1669 if (!val || val - 1 >= nlk->ngroups)
1670 return -EINVAL;
1671 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1672 err = nlk->netlink_bind(sock_net(sk), val);
1673 if (err)
1674 return err;
1675 }
1676 netlink_table_grab();
1677 netlink_update_socket_mc(nlk, val,
1678 optname == NETLINK_ADD_MEMBERSHIP);
1679 netlink_table_ungrab();
1680 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1681 nlk->netlink_unbind(sock_net(sk), val);
1682
1683 err = 0;
1684 break;
1685 }
1686 case NETLINK_BROADCAST_ERROR:
1687 if (val)
1688 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1689 else
1690 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1691 err = 0;
1692 break;
1693 case NETLINK_NO_ENOBUFS:
1694 if (val) {
1695 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1696 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1697 wake_up_interruptible(&nlk->wait);
1698 } else {
1699 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1700 }
1701 err = 0;
1702 break;
1703 case NETLINK_LISTEN_ALL_NSID:
1704 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1705 return -EPERM;
1706
1707 if (val)
1708 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1709 else
1710 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1711 err = 0;
1712 break;
1713 case NETLINK_CAP_ACK:
1714 if (val)
1715 nlk->flags |= NETLINK_F_CAP_ACK;
1716 else
1717 nlk->flags &= ~NETLINK_F_CAP_ACK;
1718 err = 0;
1719 break;
1720 case NETLINK_EXT_ACK:
1721 if (val)
1722 nlk->flags |= NETLINK_F_EXT_ACK;
1723 else
1724 nlk->flags &= ~NETLINK_F_EXT_ACK;
1725 err = 0;
1726 break;
1727 case NETLINK_GET_STRICT_CHK:
1728 if (val)
1729 nlk->flags |= NETLINK_F_STRICT_CHK;
1730 else
1731 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1732 err = 0;
1733 break;
1734 default:
1735 err = -ENOPROTOOPT;
1736 }
1737 return err;
1738}
1739
1740static int netlink_getsockopt(struct socket *sock, int level, int optname,
1741 char __user *optval, int __user *optlen)
1742{
1743 struct sock *sk = sock->sk;
1744 struct netlink_sock *nlk = nlk_sk(sk);
1745 int len, val, err;
1746
1747 if (level != SOL_NETLINK)
1748 return -ENOPROTOOPT;
1749
1750 if (get_user(len, optlen))
1751 return -EFAULT;
1752 if (len < 0)
1753 return -EINVAL;
1754
1755 switch (optname) {
1756 case NETLINK_PKTINFO:
1757 if (len < sizeof(int))
1758 return -EINVAL;
1759 len = sizeof(int);
1760 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1761 if (put_user(len, optlen) ||
1762 put_user(val, optval))
1763 return -EFAULT;
1764 err = 0;
1765 break;
1766 case NETLINK_BROADCAST_ERROR:
1767 if (len < sizeof(int))
1768 return -EINVAL;
1769 len = sizeof(int);
1770 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1771 if (put_user(len, optlen) ||
1772 put_user(val, optval))
1773 return -EFAULT;
1774 err = 0;
1775 break;
1776 case NETLINK_NO_ENOBUFS:
1777 if (len < sizeof(int))
1778 return -EINVAL;
1779 len = sizeof(int);
1780 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1781 if (put_user(len, optlen) ||
1782 put_user(val, optval))
1783 return -EFAULT;
1784 err = 0;
1785 break;
1786 case NETLINK_LIST_MEMBERSHIPS: {
1787 int pos, idx, shift;
1788
1789 err = 0;
1790 netlink_lock_table();
1791 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1792 if (len - pos < sizeof(u32))
1793 break;
1794
1795 idx = pos / sizeof(unsigned long);
1796 shift = (pos % sizeof(unsigned long)) * 8;
1797 if (put_user((u32)(nlk->groups[idx] >> shift),
1798 (u32 __user *)(optval + pos))) {
1799 err = -EFAULT;
1800 break;
1801 }
1802 }
1803 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1804 err = -EFAULT;
1805 netlink_unlock_table();
1806 break;
1807 }
1808 case NETLINK_CAP_ACK:
1809 if (len < sizeof(int))
1810 return -EINVAL;
1811 len = sizeof(int);
1812 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1813 if (put_user(len, optlen) ||
1814 put_user(val, optval))
1815 return -EFAULT;
1816 err = 0;
1817 break;
1818 case NETLINK_EXT_ACK:
1819 if (len < sizeof(int))
1820 return -EINVAL;
1821 len = sizeof(int);
1822 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1823 if (put_user(len, optlen) || put_user(val, optval))
1824 return -EFAULT;
1825 err = 0;
1826 break;
1827 case NETLINK_GET_STRICT_CHK:
1828 if (len < sizeof(int))
1829 return -EINVAL;
1830 len = sizeof(int);
1831 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1832 if (put_user(len, optlen) || put_user(val, optval))
1833 return -EFAULT;
1834 err = 0;
1835 break;
1836 default:
1837 err = -ENOPROTOOPT;
1838 }
1839 return err;
1840}
1841
1842static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1843{
1844 struct nl_pktinfo info;
1845
1846 info.group = NETLINK_CB(skb).dst_group;
1847 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1848}
1849
1850static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1851 struct sk_buff *skb)
1852{
1853 if (!NETLINK_CB(skb).nsid_is_set)
1854 return;
1855
1856 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1857 &NETLINK_CB(skb).nsid);
1858}
1859
1860static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1861{
1862 struct sock *sk = sock->sk;
1863 struct netlink_sock *nlk = nlk_sk(sk);
1864 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1865 u32 dst_portid;
1866 u32 dst_group;
1867 struct sk_buff *skb;
1868 int err;
1869 struct scm_cookie scm;
1870 u32 netlink_skb_flags = 0;
1871
1872 if (msg->msg_flags & MSG_OOB)
1873 return -EOPNOTSUPP;
1874
1875 if (len == 0) {
1876 pr_warn_once("Zero length message leads to an empty skb\n");
1877 return -ENODATA;
1878 }
1879
1880 err = scm_send(sock, msg, &scm, true);
1881 if (err < 0)
1882 return err;
1883
1884 if (msg->msg_namelen) {
1885 err = -EINVAL;
1886 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1887 goto out;
1888 if (addr->nl_family != AF_NETLINK)
1889 goto out;
1890 dst_portid = addr->nl_pid;
1891 dst_group = ffs(addr->nl_groups);
1892 err = -EPERM;
1893 if ((dst_group || dst_portid) &&
1894 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1895 goto out;
1896 netlink_skb_flags |= NETLINK_SKB_DST;
1897 } else {
1898 /* Paired with WRITE_ONCE() in netlink_connect() */
1899 dst_portid = READ_ONCE(nlk->dst_portid);
1900 dst_group = READ_ONCE(nlk->dst_group);
1901 }
1902
1903 /* Paired with WRITE_ONCE() in netlink_insert() */
1904 if (!READ_ONCE(nlk->bound)) {
1905 err = netlink_autobind(sock);
1906 if (err)
1907 goto out;
1908 } else {
1909 /* Ensure nlk is hashed and visible. */
1910 smp_rmb();
1911 }
1912
1913 err = -EMSGSIZE;
1914 if (len > sk->sk_sndbuf - 32)
1915 goto out;
1916 err = -ENOBUFS;
1917 skb = netlink_alloc_large_skb(len, dst_group);
1918 if (skb == NULL)
1919 goto out;
1920
1921 NETLINK_CB(skb).portid = nlk->portid;
1922 NETLINK_CB(skb).dst_group = dst_group;
1923 NETLINK_CB(skb).creds = scm.creds;
1924 NETLINK_CB(skb).flags = netlink_skb_flags;
1925
1926 err = -EFAULT;
1927 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1928 kfree_skb(skb);
1929 goto out;
1930 }
1931
1932 err = security_netlink_send(sk, skb);
1933 if (err) {
1934 kfree_skb(skb);
1935 goto out;
1936 }
1937
1938 if (dst_group) {
1939 refcount_inc(&skb->users);
1940 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1941 }
1942 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1943
1944out:
1945 scm_destroy(&scm);
1946 return err;
1947}
1948
1949static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1950 int flags)
1951{
1952 struct scm_cookie scm;
1953 struct sock *sk = sock->sk;
1954 struct netlink_sock *nlk = nlk_sk(sk);
1955 size_t copied;
1956 struct sk_buff *skb, *data_skb;
1957 int err, ret;
1958
1959 if (flags & MSG_OOB)
1960 return -EOPNOTSUPP;
1961
1962 copied = 0;
1963
1964 skb = skb_recv_datagram(sk, flags, &err);
1965 if (skb == NULL)
1966 goto out;
1967
1968 data_skb = skb;
1969
1970#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1971 if (unlikely(skb_shinfo(skb)->frag_list)) {
1972 /*
1973 * If this skb has a frag_list, then here that means that we
1974 * will have to use the frag_list skb's data for compat tasks
1975 * and the regular skb's data for normal (non-compat) tasks.
1976 *
1977 * If we need to send the compat skb, assign it to the
1978 * 'data_skb' variable so that it will be used below for data
1979 * copying. We keep 'skb' for everything else, including
1980 * freeing both later.
1981 */
1982 if (flags & MSG_CMSG_COMPAT)
1983 data_skb = skb_shinfo(skb)->frag_list;
1984 }
1985#endif
1986
1987 /* Record the max length of recvmsg() calls for future allocations */
1988 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1989 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1990 SKB_WITH_OVERHEAD(32768));
1991
1992 copied = data_skb->len;
1993 if (len < copied) {
1994 msg->msg_flags |= MSG_TRUNC;
1995 copied = len;
1996 }
1997
1998 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1999
2000 if (msg->msg_name) {
2001 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
2002 addr->nl_family = AF_NETLINK;
2003 addr->nl_pad = 0;
2004 addr->nl_pid = NETLINK_CB(skb).portid;
2005 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
2006 msg->msg_namelen = sizeof(*addr);
2007 }
2008
2009 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
2010 netlink_cmsg_recv_pktinfo(msg, skb);
2011 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
2012 netlink_cmsg_listen_all_nsid(sk, msg, skb);
2013
2014 memset(&scm, 0, sizeof(scm));
2015 scm.creds = *NETLINK_CREDS(skb);
2016 if (flags & MSG_TRUNC)
2017 copied = data_skb->len;
2018
2019 skb_free_datagram(sk, skb);
2020
2021 if (nlk->cb_running &&
2022 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2023 ret = netlink_dump(sk);
2024 if (ret) {
2025 sk->sk_err = -ret;
2026 sk_error_report(sk);
2027 }
2028 }
2029
2030 scm_recv(sock, msg, &scm, flags);
2031out:
2032 netlink_rcv_wake(sk);
2033 return err ? : copied;
2034}
2035
2036static void netlink_data_ready(struct sock *sk)
2037{
2038 BUG();
2039}
2040
2041/*
2042 * We export these functions to other modules. They provide a
2043 * complete set of kernel non-blocking support for message
2044 * queueing.
2045 */
2046
2047struct sock *
2048__netlink_kernel_create(struct net *net, int unit, struct module *module,
2049 struct netlink_kernel_cfg *cfg)
2050{
2051 struct socket *sock;
2052 struct sock *sk;
2053 struct netlink_sock *nlk;
2054 struct listeners *listeners = NULL;
2055 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2056 unsigned int groups;
2057
2058 BUG_ON(!nl_table);
2059
2060 if (unit < 0 || unit >= MAX_LINKS)
2061 return NULL;
2062
2063 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2064 return NULL;
2065
2066 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2067 goto out_sock_release_nosk;
2068
2069 sk = sock->sk;
2070
2071 if (!cfg || cfg->groups < 32)
2072 groups = 32;
2073 else
2074 groups = cfg->groups;
2075
2076 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2077 if (!listeners)
2078 goto out_sock_release;
2079
2080 sk->sk_data_ready = netlink_data_ready;
2081 if (cfg && cfg->input)
2082 nlk_sk(sk)->netlink_rcv = cfg->input;
2083
2084 if (netlink_insert(sk, 0))
2085 goto out_sock_release;
2086
2087 nlk = nlk_sk(sk);
2088 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2089
2090 netlink_table_grab();
2091 if (!nl_table[unit].registered) {
2092 nl_table[unit].groups = groups;
2093 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2094 nl_table[unit].cb_mutex = cb_mutex;
2095 nl_table[unit].module = module;
2096 if (cfg) {
2097 nl_table[unit].bind = cfg->bind;
2098 nl_table[unit].unbind = cfg->unbind;
2099 nl_table[unit].flags = cfg->flags;
2100 if (cfg->compare)
2101 nl_table[unit].compare = cfg->compare;
2102 }
2103 nl_table[unit].registered = 1;
2104 } else {
2105 kfree(listeners);
2106 nl_table[unit].registered++;
2107 }
2108 netlink_table_ungrab();
2109 return sk;
2110
2111out_sock_release:
2112 kfree(listeners);
2113 netlink_kernel_release(sk);
2114 return NULL;
2115
2116out_sock_release_nosk:
2117 sock_release(sock);
2118 return NULL;
2119}
2120EXPORT_SYMBOL(__netlink_kernel_create);
2121
2122void
2123netlink_kernel_release(struct sock *sk)
2124{
2125 if (sk == NULL || sk->sk_socket == NULL)
2126 return;
2127
2128 sock_release(sk->sk_socket);
2129}
2130EXPORT_SYMBOL(netlink_kernel_release);
2131
2132int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2133{
2134 struct listeners *new, *old;
2135 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2136
2137 if (groups < 32)
2138 groups = 32;
2139
2140 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2141 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2142 if (!new)
2143 return -ENOMEM;
2144 old = nl_deref_protected(tbl->listeners);
2145 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2146 rcu_assign_pointer(tbl->listeners, new);
2147
2148 kfree_rcu(old, rcu);
2149 }
2150 tbl->groups = groups;
2151
2152 return 0;
2153}
2154
2155/**
2156 * netlink_change_ngroups - change number of multicast groups
2157 *
2158 * This changes the number of multicast groups that are available
2159 * on a certain netlink family. Note that it is not possible to
2160 * change the number of groups to below 32. Also note that it does
2161 * not implicitly call netlink_clear_multicast_users() when the
2162 * number of groups is reduced.
2163 *
2164 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2165 * @groups: The new number of groups.
2166 */
2167int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2168{
2169 int err;
2170
2171 netlink_table_grab();
2172 err = __netlink_change_ngroups(sk, groups);
2173 netlink_table_ungrab();
2174
2175 return err;
2176}
2177
2178void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2179{
2180 struct sock *sk;
2181 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2182
2183 sk_for_each_bound(sk, &tbl->mc_list)
2184 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2185}
2186
2187struct nlmsghdr *
2188__nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2189{
2190 struct nlmsghdr *nlh;
2191 int size = nlmsg_msg_size(len);
2192
2193 nlh = skb_put(skb, NLMSG_ALIGN(size));
2194 nlh->nlmsg_type = type;
2195 nlh->nlmsg_len = size;
2196 nlh->nlmsg_flags = flags;
2197 nlh->nlmsg_pid = portid;
2198 nlh->nlmsg_seq = seq;
2199 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2200 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2201 return nlh;
2202}
2203EXPORT_SYMBOL(__nlmsg_put);
2204
2205/*
2206 * It looks a bit ugly.
2207 * It would be better to create kernel thread.
2208 */
2209
2210static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2211 struct netlink_callback *cb,
2212 struct netlink_ext_ack *extack)
2213{
2214 struct nlmsghdr *nlh;
2215
2216 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2217 NLM_F_MULTI | cb->answer_flags);
2218 if (WARN_ON(!nlh))
2219 return -ENOBUFS;
2220
2221 nl_dump_check_consistent(cb, nlh);
2222 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2223
2224 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2225 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2226 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2227 nlmsg_end(skb, nlh);
2228 }
2229
2230 return 0;
2231}
2232
2233static int netlink_dump(struct sock *sk)
2234{
2235 struct netlink_sock *nlk = nlk_sk(sk);
2236 struct netlink_ext_ack extack = {};
2237 struct netlink_callback *cb;
2238 struct sk_buff *skb = NULL;
2239 struct module *module;
2240 int err = -ENOBUFS;
2241 int alloc_min_size;
2242 int alloc_size;
2243
2244 mutex_lock(nlk->cb_mutex);
2245 if (!nlk->cb_running) {
2246 err = -EINVAL;
2247 goto errout_skb;
2248 }
2249
2250 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2251 goto errout_skb;
2252
2253 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2254 * required, but it makes sense to _attempt_ a 16K bytes allocation
2255 * to reduce number of system calls on dump operations, if user
2256 * ever provided a big enough buffer.
2257 */
2258 cb = &nlk->cb;
2259 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2260
2261 if (alloc_min_size < nlk->max_recvmsg_len) {
2262 alloc_size = nlk->max_recvmsg_len;
2263 skb = alloc_skb(alloc_size,
2264 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2265 __GFP_NOWARN | __GFP_NORETRY);
2266 }
2267 if (!skb) {
2268 alloc_size = alloc_min_size;
2269 skb = alloc_skb(alloc_size, GFP_KERNEL);
2270 }
2271 if (!skb)
2272 goto errout_skb;
2273
2274 /* Trim skb to allocated size. User is expected to provide buffer as
2275 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2276 * netlink_recvmsg())). dump will pack as many smaller messages as
2277 * could fit within the allocated skb. skb is typically allocated
2278 * with larger space than required (could be as much as near 2x the
2279 * requested size with align to next power of 2 approach). Allowing
2280 * dump to use the excess space makes it difficult for a user to have a
2281 * reasonable static buffer based on the expected largest dump of a
2282 * single netdev. The outcome is MSG_TRUNC error.
2283 */
2284 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2285
2286 /* Make sure malicious BPF programs can not read unitialized memory
2287 * from skb->head -> skb->data
2288 */
2289 skb_reset_network_header(skb);
2290 skb_reset_mac_header(skb);
2291
2292 netlink_skb_set_owner_r(skb, sk);
2293
2294 if (nlk->dump_done_errno > 0) {
2295 cb->extack = &extack;
2296 nlk->dump_done_errno = cb->dump(skb, cb);
2297 cb->extack = NULL;
2298 }
2299
2300 if (nlk->dump_done_errno > 0 ||
2301 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2302 mutex_unlock(nlk->cb_mutex);
2303
2304 if (sk_filter(sk, skb))
2305 kfree_skb(skb);
2306 else
2307 __netlink_sendskb(sk, skb);
2308 return 0;
2309 }
2310
2311 if (netlink_dump_done(nlk, skb, cb, &extack))
2312 goto errout_skb;
2313
2314#ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2315 /* frag_list skb's data is used for compat tasks
2316 * and the regular skb's data for normal (non-compat) tasks.
2317 * See netlink_recvmsg().
2318 */
2319 if (unlikely(skb_shinfo(skb)->frag_list)) {
2320 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2321 goto errout_skb;
2322 }
2323#endif
2324
2325 if (sk_filter(sk, skb))
2326 kfree_skb(skb);
2327 else
2328 __netlink_sendskb(sk, skb);
2329
2330 if (cb->done)
2331 cb->done(cb);
2332
2333 nlk->cb_running = false;
2334 module = cb->module;
2335 skb = cb->skb;
2336 mutex_unlock(nlk->cb_mutex);
2337 module_put(module);
2338 consume_skb(skb);
2339 return 0;
2340
2341errout_skb:
2342 mutex_unlock(nlk->cb_mutex);
2343 kfree_skb(skb);
2344 return err;
2345}
2346
2347int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2348 const struct nlmsghdr *nlh,
2349 struct netlink_dump_control *control)
2350{
2351 struct netlink_sock *nlk, *nlk2;
2352 struct netlink_callback *cb;
2353 struct sock *sk;
2354 int ret;
2355
2356 refcount_inc(&skb->users);
2357
2358 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2359 if (sk == NULL) {
2360 ret = -ECONNREFUSED;
2361 goto error_free;
2362 }
2363
2364 nlk = nlk_sk(sk);
2365 mutex_lock(nlk->cb_mutex);
2366 /* A dump is in progress... */
2367 if (nlk->cb_running) {
2368 ret = -EBUSY;
2369 goto error_unlock;
2370 }
2371 /* add reference of module which cb->dump belongs to */
2372 if (!try_module_get(control->module)) {
2373 ret = -EPROTONOSUPPORT;
2374 goto error_unlock;
2375 }
2376
2377 cb = &nlk->cb;
2378 memset(cb, 0, sizeof(*cb));
2379 cb->dump = control->dump;
2380 cb->done = control->done;
2381 cb->nlh = nlh;
2382 cb->data = control->data;
2383 cb->module = control->module;
2384 cb->min_dump_alloc = control->min_dump_alloc;
2385 cb->skb = skb;
2386
2387 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2388 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2389
2390 if (control->start) {
2391 ret = control->start(cb);
2392 if (ret)
2393 goto error_put;
2394 }
2395
2396 nlk->cb_running = true;
2397 nlk->dump_done_errno = INT_MAX;
2398
2399 mutex_unlock(nlk->cb_mutex);
2400
2401 ret = netlink_dump(sk);
2402
2403 sock_put(sk);
2404
2405 if (ret)
2406 return ret;
2407
2408 /* We successfully started a dump, by returning -EINTR we
2409 * signal not to send ACK even if it was requested.
2410 */
2411 return -EINTR;
2412
2413error_put:
2414 module_put(control->module);
2415error_unlock:
2416 sock_put(sk);
2417 mutex_unlock(nlk->cb_mutex);
2418error_free:
2419 kfree_skb(skb);
2420 return ret;
2421}
2422EXPORT_SYMBOL(__netlink_dump_start);
2423
2424static size_t
2425netlink_ack_tlv_len(struct netlink_sock *nlk, int err,
2426 const struct netlink_ext_ack *extack)
2427{
2428 size_t tlvlen;
2429
2430 if (!extack || !(nlk->flags & NETLINK_F_EXT_ACK))
2431 return 0;
2432
2433 tlvlen = 0;
2434 if (extack->_msg)
2435 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2436 if (extack->cookie_len)
2437 tlvlen += nla_total_size(extack->cookie_len);
2438
2439 /* Following attributes are only reported as error (not warning) */
2440 if (!err)
2441 return tlvlen;
2442
2443 if (extack->bad_attr)
2444 tlvlen += nla_total_size(sizeof(u32));
2445 if (extack->policy)
2446 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2447 if (extack->miss_type)
2448 tlvlen += nla_total_size(sizeof(u32));
2449 if (extack->miss_nest)
2450 tlvlen += nla_total_size(sizeof(u32));
2451
2452 return tlvlen;
2453}
2454
2455static void
2456netlink_ack_tlv_fill(struct sk_buff *in_skb, struct sk_buff *skb,
2457 struct nlmsghdr *nlh, int err,
2458 const struct netlink_ext_ack *extack)
2459{
2460 if (extack->_msg)
2461 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg));
2462 if (extack->cookie_len)
2463 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2464 extack->cookie_len, extack->cookie));
2465
2466 if (!err)
2467 return;
2468
2469 if (extack->bad_attr &&
2470 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2471 (u8 *)extack->bad_attr >= in_skb->data + in_skb->len))
2472 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2473 (u8 *)extack->bad_attr - (u8 *)nlh));
2474 if (extack->policy)
2475 netlink_policy_dump_write_attr(skb, extack->policy,
2476 NLMSGERR_ATTR_POLICY);
2477 if (extack->miss_type)
2478 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_TYPE,
2479 extack->miss_type));
2480 if (extack->miss_nest &&
2481 !WARN_ON((u8 *)extack->miss_nest < in_skb->data ||
2482 (u8 *)extack->miss_nest > in_skb->data + in_skb->len))
2483 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_MISS_NEST,
2484 (u8 *)extack->miss_nest - (u8 *)nlh));
2485}
2486
2487void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2488 const struct netlink_ext_ack *extack)
2489{
2490 struct sk_buff *skb;
2491 struct nlmsghdr *rep;
2492 struct nlmsgerr *errmsg;
2493 size_t payload = sizeof(*errmsg);
2494 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2495 unsigned int flags = 0;
2496 size_t tlvlen;
2497
2498 /* Error messages get the original request appened, unless the user
2499 * requests to cap the error message, and get extra error data if
2500 * requested.
2501 */
2502 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2503 payload += nlmsg_len(nlh);
2504 else
2505 flags |= NLM_F_CAPPED;
2506
2507 tlvlen = netlink_ack_tlv_len(nlk, err, extack);
2508 if (tlvlen)
2509 flags |= NLM_F_ACK_TLVS;
2510
2511 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2512 if (!skb)
2513 goto err_skb;
2514
2515 rep = nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2516 NLMSG_ERROR, sizeof(*errmsg), flags);
2517 if (!rep)
2518 goto err_bad_put;
2519 errmsg = nlmsg_data(rep);
2520 errmsg->error = err;
2521 errmsg->msg = *nlh;
2522
2523 if (!(flags & NLM_F_CAPPED)) {
2524 if (!nlmsg_append(skb, nlmsg_len(nlh)))
2525 goto err_bad_put;
2526
2527 memcpy(nlmsg_data(&errmsg->msg), nlmsg_data(nlh),
2528 nlmsg_len(nlh));
2529 }
2530
2531 if (tlvlen)
2532 netlink_ack_tlv_fill(in_skb, skb, nlh, err, extack);
2533
2534 nlmsg_end(skb, rep);
2535
2536 nlmsg_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid);
2537
2538 return;
2539
2540err_bad_put:
2541 nlmsg_free(skb);
2542err_skb:
2543 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2544 sk_error_report(NETLINK_CB(in_skb).sk);
2545}
2546EXPORT_SYMBOL(netlink_ack);
2547
2548int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2549 struct nlmsghdr *,
2550 struct netlink_ext_ack *))
2551{
2552 struct netlink_ext_ack extack;
2553 struct nlmsghdr *nlh;
2554 int err;
2555
2556 while (skb->len >= nlmsg_total_size(0)) {
2557 int msglen;
2558
2559 memset(&extack, 0, sizeof(extack));
2560 nlh = nlmsg_hdr(skb);
2561 err = 0;
2562
2563 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2564 return 0;
2565
2566 /* Only requests are handled by the kernel */
2567 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2568 goto ack;
2569
2570 /* Skip control messages */
2571 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2572 goto ack;
2573
2574 err = cb(skb, nlh, &extack);
2575 if (err == -EINTR)
2576 goto skip;
2577
2578ack:
2579 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2580 netlink_ack(skb, nlh, err, &extack);
2581
2582skip:
2583 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2584 if (msglen > skb->len)
2585 msglen = skb->len;
2586 skb_pull(skb, msglen);
2587 }
2588
2589 return 0;
2590}
2591EXPORT_SYMBOL(netlink_rcv_skb);
2592
2593/**
2594 * nlmsg_notify - send a notification netlink message
2595 * @sk: netlink socket to use
2596 * @skb: notification message
2597 * @portid: destination netlink portid for reports or 0
2598 * @group: destination multicast group or 0
2599 * @report: 1 to report back, 0 to disable
2600 * @flags: allocation flags
2601 */
2602int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2603 unsigned int group, int report, gfp_t flags)
2604{
2605 int err = 0;
2606
2607 if (group) {
2608 int exclude_portid = 0;
2609
2610 if (report) {
2611 refcount_inc(&skb->users);
2612 exclude_portid = portid;
2613 }
2614
2615 /* errors reported via destination sk->sk_err, but propagate
2616 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2617 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2618 if (err == -ESRCH)
2619 err = 0;
2620 }
2621
2622 if (report) {
2623 int err2;
2624
2625 err2 = nlmsg_unicast(sk, skb, portid);
2626 if (!err)
2627 err = err2;
2628 }
2629
2630 return err;
2631}
2632EXPORT_SYMBOL(nlmsg_notify);
2633
2634#ifdef CONFIG_PROC_FS
2635struct nl_seq_iter {
2636 struct seq_net_private p;
2637 struct rhashtable_iter hti;
2638 int link;
2639};
2640
2641static void netlink_walk_start(struct nl_seq_iter *iter)
2642{
2643 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2644 rhashtable_walk_start(&iter->hti);
2645}
2646
2647static void netlink_walk_stop(struct nl_seq_iter *iter)
2648{
2649 rhashtable_walk_stop(&iter->hti);
2650 rhashtable_walk_exit(&iter->hti);
2651}
2652
2653static void *__netlink_seq_next(struct seq_file *seq)
2654{
2655 struct nl_seq_iter *iter = seq->private;
2656 struct netlink_sock *nlk;
2657
2658 do {
2659 for (;;) {
2660 nlk = rhashtable_walk_next(&iter->hti);
2661
2662 if (IS_ERR(nlk)) {
2663 if (PTR_ERR(nlk) == -EAGAIN)
2664 continue;
2665
2666 return nlk;
2667 }
2668
2669 if (nlk)
2670 break;
2671
2672 netlink_walk_stop(iter);
2673 if (++iter->link >= MAX_LINKS)
2674 return NULL;
2675
2676 netlink_walk_start(iter);
2677 }
2678 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2679
2680 return nlk;
2681}
2682
2683static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2684 __acquires(RCU)
2685{
2686 struct nl_seq_iter *iter = seq->private;
2687 void *obj = SEQ_START_TOKEN;
2688 loff_t pos;
2689
2690 iter->link = 0;
2691
2692 netlink_walk_start(iter);
2693
2694 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2695 obj = __netlink_seq_next(seq);
2696
2697 return obj;
2698}
2699
2700static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2701{
2702 ++*pos;
2703 return __netlink_seq_next(seq);
2704}
2705
2706static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2707{
2708 struct nl_seq_iter *iter = seq->private;
2709
2710 if (iter->link >= MAX_LINKS)
2711 return;
2712
2713 netlink_walk_stop(iter);
2714}
2715
2716
2717static int netlink_native_seq_show(struct seq_file *seq, void *v)
2718{
2719 if (v == SEQ_START_TOKEN) {
2720 seq_puts(seq,
2721 "sk Eth Pid Groups "
2722 "Rmem Wmem Dump Locks Drops Inode\n");
2723 } else {
2724 struct sock *s = v;
2725 struct netlink_sock *nlk = nlk_sk(s);
2726
2727 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2728 s,
2729 s->sk_protocol,
2730 nlk->portid,
2731 nlk->groups ? (u32)nlk->groups[0] : 0,
2732 sk_rmem_alloc_get(s),
2733 sk_wmem_alloc_get(s),
2734 nlk->cb_running,
2735 refcount_read(&s->sk_refcnt),
2736 atomic_read(&s->sk_drops),
2737 sock_i_ino(s)
2738 );
2739
2740 }
2741 return 0;
2742}
2743
2744#ifdef CONFIG_BPF_SYSCALL
2745struct bpf_iter__netlink {
2746 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2747 __bpf_md_ptr(struct netlink_sock *, sk);
2748};
2749
2750DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2751
2752static int netlink_prog_seq_show(struct bpf_prog *prog,
2753 struct bpf_iter_meta *meta,
2754 void *v)
2755{
2756 struct bpf_iter__netlink ctx;
2757
2758 meta->seq_num--; /* skip SEQ_START_TOKEN */
2759 ctx.meta = meta;
2760 ctx.sk = nlk_sk((struct sock *)v);
2761 return bpf_iter_run_prog(prog, &ctx);
2762}
2763
2764static int netlink_seq_show(struct seq_file *seq, void *v)
2765{
2766 struct bpf_iter_meta meta;
2767 struct bpf_prog *prog;
2768
2769 meta.seq = seq;
2770 prog = bpf_iter_get_info(&meta, false);
2771 if (!prog)
2772 return netlink_native_seq_show(seq, v);
2773
2774 if (v != SEQ_START_TOKEN)
2775 return netlink_prog_seq_show(prog, &meta, v);
2776
2777 return 0;
2778}
2779
2780static void netlink_seq_stop(struct seq_file *seq, void *v)
2781{
2782 struct bpf_iter_meta meta;
2783 struct bpf_prog *prog;
2784
2785 if (!v) {
2786 meta.seq = seq;
2787 prog = bpf_iter_get_info(&meta, true);
2788 if (prog)
2789 (void)netlink_prog_seq_show(prog, &meta, v);
2790 }
2791
2792 netlink_native_seq_stop(seq, v);
2793}
2794#else
2795static int netlink_seq_show(struct seq_file *seq, void *v)
2796{
2797 return netlink_native_seq_show(seq, v);
2798}
2799
2800static void netlink_seq_stop(struct seq_file *seq, void *v)
2801{
2802 netlink_native_seq_stop(seq, v);
2803}
2804#endif
2805
2806static const struct seq_operations netlink_seq_ops = {
2807 .start = netlink_seq_start,
2808 .next = netlink_seq_next,
2809 .stop = netlink_seq_stop,
2810 .show = netlink_seq_show,
2811};
2812#endif
2813
2814int netlink_register_notifier(struct notifier_block *nb)
2815{
2816 return blocking_notifier_chain_register(&netlink_chain, nb);
2817}
2818EXPORT_SYMBOL(netlink_register_notifier);
2819
2820int netlink_unregister_notifier(struct notifier_block *nb)
2821{
2822 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2823}
2824EXPORT_SYMBOL(netlink_unregister_notifier);
2825
2826static const struct proto_ops netlink_ops = {
2827 .family = PF_NETLINK,
2828 .owner = THIS_MODULE,
2829 .release = netlink_release,
2830 .bind = netlink_bind,
2831 .connect = netlink_connect,
2832 .socketpair = sock_no_socketpair,
2833 .accept = sock_no_accept,
2834 .getname = netlink_getname,
2835 .poll = datagram_poll,
2836 .ioctl = netlink_ioctl,
2837 .listen = sock_no_listen,
2838 .shutdown = sock_no_shutdown,
2839 .setsockopt = netlink_setsockopt,
2840 .getsockopt = netlink_getsockopt,
2841 .sendmsg = netlink_sendmsg,
2842 .recvmsg = netlink_recvmsg,
2843 .mmap = sock_no_mmap,
2844 .sendpage = sock_no_sendpage,
2845};
2846
2847static const struct net_proto_family netlink_family_ops = {
2848 .family = PF_NETLINK,
2849 .create = netlink_create,
2850 .owner = THIS_MODULE, /* for consistency 8) */
2851};
2852
2853static int __net_init netlink_net_init(struct net *net)
2854{
2855#ifdef CONFIG_PROC_FS
2856 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2857 sizeof(struct nl_seq_iter)))
2858 return -ENOMEM;
2859#endif
2860 return 0;
2861}
2862
2863static void __net_exit netlink_net_exit(struct net *net)
2864{
2865#ifdef CONFIG_PROC_FS
2866 remove_proc_entry("netlink", net->proc_net);
2867#endif
2868}
2869
2870static void __init netlink_add_usersock_entry(void)
2871{
2872 struct listeners *listeners;
2873 int groups = 32;
2874
2875 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2876 if (!listeners)
2877 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2878
2879 netlink_table_grab();
2880
2881 nl_table[NETLINK_USERSOCK].groups = groups;
2882 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2883 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2884 nl_table[NETLINK_USERSOCK].registered = 1;
2885 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2886
2887 netlink_table_ungrab();
2888}
2889
2890static struct pernet_operations __net_initdata netlink_net_ops = {
2891 .init = netlink_net_init,
2892 .exit = netlink_net_exit,
2893};
2894
2895static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2896{
2897 const struct netlink_sock *nlk = data;
2898 struct netlink_compare_arg arg;
2899
2900 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2901 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2902}
2903
2904static const struct rhashtable_params netlink_rhashtable_params = {
2905 .head_offset = offsetof(struct netlink_sock, node),
2906 .key_len = netlink_compare_arg_len,
2907 .obj_hashfn = netlink_hash,
2908 .obj_cmpfn = netlink_compare,
2909 .automatic_shrinking = true,
2910};
2911
2912#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2913BTF_ID_LIST(btf_netlink_sock_id)
2914BTF_ID(struct, netlink_sock)
2915
2916static const struct bpf_iter_seq_info netlink_seq_info = {
2917 .seq_ops = &netlink_seq_ops,
2918 .init_seq_private = bpf_iter_init_seq_net,
2919 .fini_seq_private = bpf_iter_fini_seq_net,
2920 .seq_priv_size = sizeof(struct nl_seq_iter),
2921};
2922
2923static struct bpf_iter_reg netlink_reg_info = {
2924 .target = "netlink",
2925 .ctx_arg_info_size = 1,
2926 .ctx_arg_info = {
2927 { offsetof(struct bpf_iter__netlink, sk),
2928 PTR_TO_BTF_ID_OR_NULL },
2929 },
2930 .seq_info = &netlink_seq_info,
2931};
2932
2933static int __init bpf_iter_register(void)
2934{
2935 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2936 return bpf_iter_reg_target(&netlink_reg_info);
2937}
2938#endif
2939
2940static int __init netlink_proto_init(void)
2941{
2942 int i;
2943 int err = proto_register(&netlink_proto, 0);
2944
2945 if (err != 0)
2946 goto out;
2947
2948#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2949 err = bpf_iter_register();
2950 if (err)
2951 goto out;
2952#endif
2953
2954 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2955
2956 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2957 if (!nl_table)
2958 goto panic;
2959
2960 for (i = 0; i < MAX_LINKS; i++) {
2961 if (rhashtable_init(&nl_table[i].hash,
2962 &netlink_rhashtable_params) < 0) {
2963 while (--i > 0)
2964 rhashtable_destroy(&nl_table[i].hash);
2965 kfree(nl_table);
2966 goto panic;
2967 }
2968 }
2969
2970 netlink_add_usersock_entry();
2971
2972 sock_register(&netlink_family_ops);
2973 register_pernet_subsys(&netlink_net_ops);
2974 register_pernet_subsys(&netlink_tap_net_ops);
2975 /* The netlink device handler may be needed early. */
2976 rtnetlink_init();
2977out:
2978 return err;
2979panic:
2980 panic("netlink_init: Cannot allocate nl_table\n");
2981}
2982
2983core_initcall(netlink_proto_init);