?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.18.10
[linux/fpc-iii.git] / net / netlink / af_netlink.c
blob56704d95f82d27f5a2bc26714e5001f3868765b2
1 /*
2 * NETLINK Kernel-user communication protocol.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 *
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
23 */
24
25 #include <linux/module.h>
26
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
36 #include <linux/un.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
41 #include <linux/fs.h>
42 #include <linux/slab.h>
43 #include <linux/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
55 #include <linux/mm.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
65 #include <linux/net_namespace.h>
66 #include <linux/nospec.h>
67
68 #include <net/net_namespace.h>
69 #include <net/netns/generic.h>
70 #include <net/sock.h>
71 #include <net/scm.h>
72 #include <net/netlink.h>
73
74 #include "af_netlink.h"
75
76 struct listeners {
77 struct rcu_head rcu;
78 unsigned long masks[0];
79 };
80
81 /* state bits */
82 #define NETLINK_S_CONGESTED 0x0
83
84 static inline int netlink_is_kernel(struct sock *sk)
85 {
86 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
87 }
88
89 struct netlink_table *nl_table __read_mostly;
90 EXPORT_SYMBOL_GPL(nl_table);
91
92 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
93
94 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
95
96 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
97 "nlk_cb_mutex-ROUTE",
98 "nlk_cb_mutex-1",
99 "nlk_cb_mutex-USERSOCK",
100 "nlk_cb_mutex-FIREWALL",
101 "nlk_cb_mutex-SOCK_DIAG",
102 "nlk_cb_mutex-NFLOG",
103 "nlk_cb_mutex-XFRM",
104 "nlk_cb_mutex-SELINUX",
105 "nlk_cb_mutex-ISCSI",
106 "nlk_cb_mutex-AUDIT",
107 "nlk_cb_mutex-FIB_LOOKUP",
108 "nlk_cb_mutex-CONNECTOR",
109 "nlk_cb_mutex-NETFILTER",
110 "nlk_cb_mutex-IP6_FW",
111 "nlk_cb_mutex-DNRTMSG",
112 "nlk_cb_mutex-KOBJECT_UEVENT",
113 "nlk_cb_mutex-GENERIC",
114 "nlk_cb_mutex-17",
115 "nlk_cb_mutex-SCSITRANSPORT",
116 "nlk_cb_mutex-ECRYPTFS",
117 "nlk_cb_mutex-RDMA",
118 "nlk_cb_mutex-CRYPTO",
119 "nlk_cb_mutex-SMC",
120 "nlk_cb_mutex-23",
121 "nlk_cb_mutex-24",
122 "nlk_cb_mutex-25",
123 "nlk_cb_mutex-26",
124 "nlk_cb_mutex-27",
125 "nlk_cb_mutex-28",
126 "nlk_cb_mutex-29",
127 "nlk_cb_mutex-30",
128 "nlk_cb_mutex-31",
129 "nlk_cb_mutex-MAX_LINKS"
130 };
131
132 static int netlink_dump(struct sock *sk);
133
134 /* nl_table locking explained:
135 * Lookup and traversal are protected with an RCU read-side lock. Insertion
136 * and removal are protected with per bucket lock while using RCU list
137 * modification primitives and may run in parallel to RCU protected lookups.
138 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
139 * been acquired * either during or after the socket has been removed from
140 * the list and after an RCU grace period.
141 */
142 DEFINE_RWLOCK(nl_table_lock);
143 EXPORT_SYMBOL_GPL(nl_table_lock);
144 static atomic_t nl_table_users = ATOMIC_INIT(0);
145
146 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
147
148 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
149
150
151 static const struct rhashtable_params netlink_rhashtable_params;
152
153 static inline u32 netlink_group_mask(u32 group)
154 {
155 return group ? 1 << (group - 1) : 0;
156 }
157
158 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
159 gfp_t gfp_mask)
160 {
161 unsigned int len = skb_end_offset(skb);
162 struct sk_buff *new;
163
164 new = alloc_skb(len, gfp_mask);
165 if (new == NULL)
166 return NULL;
167
168 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
169 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
170 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
171
172 skb_put_data(new, skb->data, len);
173 return new;
174 }
175
176 static unsigned int netlink_tap_net_id;
177
178 struct netlink_tap_net {
179 struct list_head netlink_tap_all;
180 struct mutex netlink_tap_lock;
181 };
182
183 int netlink_add_tap(struct netlink_tap *nt)
184 {
185 struct net *net = dev_net(nt->dev);
186 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
187
188 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
189 return -EINVAL;
190
191 mutex_lock(&nn->netlink_tap_lock);
192 list_add_rcu(&nt->list, &nn->netlink_tap_all);
193 mutex_unlock(&nn->netlink_tap_lock);
194
195 __module_get(nt->module);
196
197 return 0;
198 }
199 EXPORT_SYMBOL_GPL(netlink_add_tap);
200
201 static int __netlink_remove_tap(struct netlink_tap *nt)
202 {
203 struct net *net = dev_net(nt->dev);
204 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
205 bool found = false;
206 struct netlink_tap *tmp;
207
208 mutex_lock(&nn->netlink_tap_lock);
209
210 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
211 if (nt == tmp) {
212 list_del_rcu(&nt->list);
213 found = true;
214 goto out;
215 }
216 }
217
218 pr_warn("__netlink_remove_tap: %p not found\n", nt);
219 out:
220 mutex_unlock(&nn->netlink_tap_lock);
221
222 if (found)
223 module_put(nt->module);
224
225 return found ? 0 : -ENODEV;
226 }
227
228 int netlink_remove_tap(struct netlink_tap *nt)
229 {
230 int ret;
231
232 ret = __netlink_remove_tap(nt);
233 synchronize_net();
234
235 return ret;
236 }
237 EXPORT_SYMBOL_GPL(netlink_remove_tap);
238
239 static __net_init int netlink_tap_init_net(struct net *net)
240 {
241 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
242
243 INIT_LIST_HEAD(&nn->netlink_tap_all);
244 mutex_init(&nn->netlink_tap_lock);
245 return 0;
246 }
247
248 static void __net_exit netlink_tap_exit_net(struct net *net)
249 {
250 }
251
252 static struct pernet_operations netlink_tap_net_ops = {
253 .init = netlink_tap_init_net,
254 .exit = netlink_tap_exit_net,
255 .id = &netlink_tap_net_id,
256 .size = sizeof(struct netlink_tap_net),
257 };
258
259 static bool netlink_filter_tap(const struct sk_buff *skb)
260 {
261 struct sock *sk = skb->sk;
262
263 /* We take the more conservative approach and
264 * whitelist socket protocols that may pass.
265 */
266 switch (sk->sk_protocol) {
267 case NETLINK_ROUTE:
268 case NETLINK_USERSOCK:
269 case NETLINK_SOCK_DIAG:
270 case NETLINK_NFLOG:
271 case NETLINK_XFRM:
272 case NETLINK_FIB_LOOKUP:
273 case NETLINK_NETFILTER:
274 case NETLINK_GENERIC:
275 return true;
276 }
277
278 return false;
279 }
280
281 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
282 struct net_device *dev)
283 {
284 struct sk_buff *nskb;
285 struct sock *sk = skb->sk;
286 int ret = -ENOMEM;
287
288 if (!net_eq(dev_net(dev), sock_net(sk)))
289 return 0;
290
291 dev_hold(dev);
292
293 if (is_vmalloc_addr(skb->head))
294 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
295 else
296 nskb = skb_clone(skb, GFP_ATOMIC);
297 if (nskb) {
298 nskb->dev = dev;
299 nskb->protocol = htons((u16) sk->sk_protocol);
300 nskb->pkt_type = netlink_is_kernel(sk) ?
301 PACKET_KERNEL : PACKET_USER;
302 skb_reset_network_header(nskb);
303 ret = dev_queue_xmit(nskb);
304 if (unlikely(ret > 0))
305 ret = net_xmit_errno(ret);
306 }
307
308 dev_put(dev);
309 return ret;
310 }
311
312 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
313 {
314 int ret;
315 struct netlink_tap *tmp;
316
317 if (!netlink_filter_tap(skb))
318 return;
319
320 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
321 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
322 if (unlikely(ret))
323 break;
324 }
325 }
326
327 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
328 {
329 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
330
331 rcu_read_lock();
332
333 if (unlikely(!list_empty(&nn->netlink_tap_all)))
334 __netlink_deliver_tap(skb, nn);
335
336 rcu_read_unlock();
337 }
338
339 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
340 struct sk_buff *skb)
341 {
342 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
343 netlink_deliver_tap(sock_net(dst), skb);
344 }
345
346 static void netlink_overrun(struct sock *sk)
347 {
348 struct netlink_sock *nlk = nlk_sk(sk);
349
350 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
351 if (!test_and_set_bit(NETLINK_S_CONGESTED,
352 &nlk_sk(sk)->state)) {
353 sk->sk_err = ENOBUFS;
354 sk->sk_error_report(sk);
355 }
356 }
357 atomic_inc(&sk->sk_drops);
358 }
359
360 static void netlink_rcv_wake(struct sock *sk)
361 {
362 struct netlink_sock *nlk = nlk_sk(sk);
363
364 if (skb_queue_empty(&sk->sk_receive_queue))
365 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
366 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
367 wake_up_interruptible(&nlk->wait);
368 }
369
370 static void netlink_skb_destructor(struct sk_buff *skb)
371 {
372 if (is_vmalloc_addr(skb->head)) {
373 if (!skb->cloned ||
374 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
375 vfree(skb->head);
376
377 skb->head = NULL;
378 }
379 if (skb->sk != NULL)
380 sock_rfree(skb);
381 }
382
383 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
384 {
385 WARN_ON(skb->sk != NULL);
386 skb->sk = sk;
387 skb->destructor = netlink_skb_destructor;
388 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
389 sk_mem_charge(sk, skb->truesize);
390 }
391
392 static void netlink_sock_destruct(struct sock *sk)
393 {
394 struct netlink_sock *nlk = nlk_sk(sk);
395
396 if (nlk->cb_running) {
397 if (nlk->cb.done)
398 nlk->cb.done(&nlk->cb);
399 module_put(nlk->cb.module);
400 kfree_skb(nlk->cb.skb);
401 }
402
403 skb_queue_purge(&sk->sk_receive_queue);
404
405 if (!sock_flag(sk, SOCK_DEAD)) {
406 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
407 return;
408 }
409
410 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
411 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
412 WARN_ON(nlk_sk(sk)->groups);
413 }
414
415 static void netlink_sock_destruct_work(struct work_struct *work)
416 {
417 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
418 work);
419
420 sk_free(&nlk->sk);
421 }
422
423 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
424 * SMP. Look, when several writers sleep and reader wakes them up, all but one
425 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
426 * this, _but_ remember, it adds useless work on UP machines.
427 */
428
429 void netlink_table_grab(void)
430 __acquires(nl_table_lock)
431 {
432 might_sleep();
433
434 write_lock_irq(&nl_table_lock);
435
436 if (atomic_read(&nl_table_users)) {
437 DECLARE_WAITQUEUE(wait, current);
438
439 add_wait_queue_exclusive(&nl_table_wait, &wait);
440 for (;;) {
441 set_current_state(TASK_UNINTERRUPTIBLE);
442 if (atomic_read(&nl_table_users) == 0)
443 break;
444 write_unlock_irq(&nl_table_lock);
445 schedule();
446 write_lock_irq(&nl_table_lock);
447 }
448
449 __set_current_state(TASK_RUNNING);
450 remove_wait_queue(&nl_table_wait, &wait);
451 }
452 }
453
454 void netlink_table_ungrab(void)
455 __releases(nl_table_lock)
456 {
457 write_unlock_irq(&nl_table_lock);
458 wake_up(&nl_table_wait);
459 }
460
461 static inline void
462 netlink_lock_table(void)
463 {
464 /* read_lock() synchronizes us to netlink_table_grab */
465
466 read_lock(&nl_table_lock);
467 atomic_inc(&nl_table_users);
468 read_unlock(&nl_table_lock);
469 }
470
471 static inline void
472 netlink_unlock_table(void)
473 {
474 if (atomic_dec_and_test(&nl_table_users))
475 wake_up(&nl_table_wait);
476 }
477
478 struct netlink_compare_arg
479 {
480 possible_net_t pnet;
481 u32 portid;
482 };
483
484 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
485 #define netlink_compare_arg_len \
486 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
487
488 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
489 const void *ptr)
490 {
491 const struct netlink_compare_arg *x = arg->key;
492 const struct netlink_sock *nlk = ptr;
493
494 return nlk->portid != x->portid ||
495 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
496 }
497
498 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
499 struct net *net, u32 portid)
500 {
501 memset(arg, 0, sizeof(*arg));
502 write_pnet(&arg->pnet, net);
503 arg->portid = portid;
504 }
505
506 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
507 struct net *net)
508 {
509 struct netlink_compare_arg arg;
510
511 netlink_compare_arg_init(&arg, net, portid);
512 return rhashtable_lookup_fast(&table->hash, &arg,
513 netlink_rhashtable_params);
514 }
515
516 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
517 {
518 struct netlink_compare_arg arg;
519
520 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
521 return rhashtable_lookup_insert_key(&table->hash, &arg,
522 &nlk_sk(sk)->node,
523 netlink_rhashtable_params);
524 }
525
526 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
527 {
528 struct netlink_table *table = &nl_table[protocol];
529 struct sock *sk;
530
531 rcu_read_lock();
532 sk = __netlink_lookup(table, portid, net);
533 if (sk)
534 sock_hold(sk);
535 rcu_read_unlock();
536
537 return sk;
538 }
539
540 static const struct proto_ops netlink_ops;
541
542 static void
543 netlink_update_listeners(struct sock *sk)
544 {
545 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
546 unsigned long mask;
547 unsigned int i;
548 struct listeners *listeners;
549
550 listeners = nl_deref_protected(tbl->listeners);
551 if (!listeners)
552 return;
553
554 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
555 mask = 0;
556 sk_for_each_bound(sk, &tbl->mc_list) {
557 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
558 mask |= nlk_sk(sk)->groups[i];
559 }
560 listeners->masks[i] = mask;
561 }
562 /* this function is only called with the netlink table "grabbed", which
563 * makes sure updates are visible before bind or setsockopt return. */
564 }
565
566 static int netlink_insert(struct sock *sk, u32 portid)
567 {
568 struct netlink_table *table = &nl_table[sk->sk_protocol];
569 int err;
570
571 lock_sock(sk);
572
573 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
574 if (nlk_sk(sk)->bound)
575 goto err;
576
577 err = -ENOMEM;
578 if (BITS_PER_LONG > 32 &&
579 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
580 goto err;
581
582 nlk_sk(sk)->portid = portid;
583 sock_hold(sk);
584
585 err = __netlink_insert(table, sk);
586 if (err) {
587 /* In case the hashtable backend returns with -EBUSY
588 * from here, it must not escape to the caller.
589 */
590 if (unlikely(err == -EBUSY))
591 err = -EOVERFLOW;
592 if (err == -EEXIST)
593 err = -EADDRINUSE;
594 sock_put(sk);
595 goto err;
596 }
597
598 /* We need to ensure that the socket is hashed and visible. */
599 smp_wmb();
600 nlk_sk(sk)->bound = portid;
601
602 err:
603 release_sock(sk);
604 return err;
605 }
606
607 static void netlink_remove(struct sock *sk)
608 {
609 struct netlink_table *table;
610
611 table = &nl_table[sk->sk_protocol];
612 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
613 netlink_rhashtable_params)) {
614 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
615 __sock_put(sk);
616 }
617
618 netlink_table_grab();
619 if (nlk_sk(sk)->subscriptions) {
620 __sk_del_bind_node(sk);
621 netlink_update_listeners(sk);
622 }
623 if (sk->sk_protocol == NETLINK_GENERIC)
624 atomic_inc(&genl_sk_destructing_cnt);
625 netlink_table_ungrab();
626 }
627
628 static struct proto netlink_proto = {
629 .name = "NETLINK",
630 .owner = THIS_MODULE,
631 .obj_size = sizeof(struct netlink_sock),
632 };
633
634 static int __netlink_create(struct net *net, struct socket *sock,
635 struct mutex *cb_mutex, int protocol,
636 int kern)
637 {
638 struct sock *sk;
639 struct netlink_sock *nlk;
640
641 sock->ops = &netlink_ops;
642
643 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
644 if (!sk)
645 return -ENOMEM;
646
647 sock_init_data(sock, sk);
648
649 nlk = nlk_sk(sk);
650 if (cb_mutex) {
651 nlk->cb_mutex = cb_mutex;
652 } else {
653 nlk->cb_mutex = &nlk->cb_def_mutex;
654 mutex_init(nlk->cb_mutex);
655 lockdep_set_class_and_name(nlk->cb_mutex,
656 nlk_cb_mutex_keys + protocol,
657 nlk_cb_mutex_key_strings[protocol]);
658 }
659 init_waitqueue_head(&nlk->wait);
660
661 sk->sk_destruct = netlink_sock_destruct;
662 sk->sk_protocol = protocol;
663 return 0;
664 }
665
666 static int netlink_create(struct net *net, struct socket *sock, int protocol,
667 int kern)
668 {
669 struct module *module = NULL;
670 struct mutex *cb_mutex;
671 struct netlink_sock *nlk;
672 int (*bind)(struct net *net, int group);
673 void (*unbind)(struct net *net, int group);
674 int err = 0;
675
676 sock->state = SS_UNCONNECTED;
677
678 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
679 return -ESOCKTNOSUPPORT;
680
681 if (protocol < 0 || protocol >= MAX_LINKS)
682 return -EPROTONOSUPPORT;
683 protocol = array_index_nospec(protocol, MAX_LINKS);
684
685 netlink_lock_table();
686 #ifdef CONFIG_MODULES
687 if (!nl_table[protocol].registered) {
688 netlink_unlock_table();
689 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
690 netlink_lock_table();
691 }
692 #endif
693 if (nl_table[protocol].registered &&
694 try_module_get(nl_table[protocol].module))
695 module = nl_table[protocol].module;
696 else
697 err = -EPROTONOSUPPORT;
698 cb_mutex = nl_table[protocol].cb_mutex;
699 bind = nl_table[protocol].bind;
700 unbind = nl_table[protocol].unbind;
701 netlink_unlock_table();
702
703 if (err < 0)
704 goto out;
705
706 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
707 if (err < 0)
708 goto out_module;
709
710 local_bh_disable();
711 sock_prot_inuse_add(net, &netlink_proto, 1);
712 local_bh_enable();
713
714 nlk = nlk_sk(sock->sk);
715 nlk->module = module;
716 nlk->netlink_bind = bind;
717 nlk->netlink_unbind = unbind;
718 out:
719 return err;
720
721 out_module:
722 module_put(module);
723 goto out;
724 }
725
726 static void deferred_put_nlk_sk(struct rcu_head *head)
727 {
728 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
729 struct sock *sk = &nlk->sk;
730
731 kfree(nlk->groups);
732 nlk->groups = NULL;
733
734 if (!refcount_dec_and_test(&sk->sk_refcnt))
735 return;
736
737 if (nlk->cb_running && nlk->cb.done) {
738 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
739 schedule_work(&nlk->work);
740 return;
741 }
742
743 sk_free(sk);
744 }
745
746 static int netlink_release(struct socket *sock)
747 {
748 struct sock *sk = sock->sk;
749 struct netlink_sock *nlk;
750
751 if (!sk)
752 return 0;
753
754 netlink_remove(sk);
755 sock_orphan(sk);
756 nlk = nlk_sk(sk);
757
758 /*
759 * OK. Socket is unlinked, any packets that arrive now
760 * will be purged.
761 */
762
763 /* must not acquire netlink_table_lock in any way again before unbind
764 * and notifying genetlink is done as otherwise it might deadlock
765 */
766 if (nlk->netlink_unbind) {
767 int i;
768
769 for (i = 0; i < nlk->ngroups; i++)
770 if (test_bit(i, nlk->groups))
771 nlk->netlink_unbind(sock_net(sk), i + 1);
772 }
773 if (sk->sk_protocol == NETLINK_GENERIC &&
774 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
775 wake_up(&genl_sk_destructing_waitq);
776
777 sock->sk = NULL;
778 wake_up_interruptible_all(&nlk->wait);
779
780 skb_queue_purge(&sk->sk_write_queue);
781
782 if (nlk->portid && nlk->bound) {
783 struct netlink_notify n = {
784 .net = sock_net(sk),
785 .protocol = sk->sk_protocol,
786 .portid = nlk->portid,
787 };
788 blocking_notifier_call_chain(&netlink_chain,
789 NETLINK_URELEASE, &n);
790 }
791
792 module_put(nlk->module);
793
794 if (netlink_is_kernel(sk)) {
795 netlink_table_grab();
796 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
797 if (--nl_table[sk->sk_protocol].registered == 0) {
798 struct listeners *old;
799
800 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
801 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
802 kfree_rcu(old, rcu);
803 nl_table[sk->sk_protocol].module = NULL;
804 nl_table[sk->sk_protocol].bind = NULL;
805 nl_table[sk->sk_protocol].unbind = NULL;
806 nl_table[sk->sk_protocol].flags = 0;
807 nl_table[sk->sk_protocol].registered = 0;
808 }
809 netlink_table_ungrab();
810 }
811
812 local_bh_disable();
813 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
814 local_bh_enable();
815 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
816 return 0;
817 }
818
819 static int netlink_autobind(struct socket *sock)
820 {
821 struct sock *sk = sock->sk;
822 struct net *net = sock_net(sk);
823 struct netlink_table *table = &nl_table[sk->sk_protocol];
824 s32 portid = task_tgid_vnr(current);
825 int err;
826 s32 rover = -4096;
827 bool ok;
828
829 retry:
830 cond_resched();
831 rcu_read_lock();
832 ok = !__netlink_lookup(table, portid, net);
833 rcu_read_unlock();
834 if (!ok) {
835 /* Bind collision, search negative portid values. */
836 if (rover == -4096)
837 /* rover will be in range [S32_MIN, -4097] */
838 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
839 else if (rover >= -4096)
840 rover = -4097;
841 portid = rover--;
842 goto retry;
843 }
844
845 err = netlink_insert(sk, portid);
846 if (err == -EADDRINUSE)
847 goto retry;
848
849 /* If 2 threads race to autobind, that is fine. */
850 if (err == -EBUSY)
851 err = 0;
852
853 return err;
854 }
855
856 /**
857 * __netlink_ns_capable - General netlink message capability test
858 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
859 * @user_ns: The user namespace of the capability to use
860 * @cap: The capability to use
861 *
862 * Test to see if the opener of the socket we received the message
863 * from had when the netlink socket was created and the sender of the
864 * message has has the capability @cap in the user namespace @user_ns.
865 */
866 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
867 struct user_namespace *user_ns, int cap)
868 {
869 return ((nsp->flags & NETLINK_SKB_DST) ||
870 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
871 ns_capable(user_ns, cap);
872 }
873 EXPORT_SYMBOL(__netlink_ns_capable);
874
875 /**
876 * netlink_ns_capable - General netlink message capability test
877 * @skb: socket buffer holding a netlink command from userspace
878 * @user_ns: The user namespace of the capability to use
879 * @cap: The capability to use
880 *
881 * Test to see if the opener of the socket we received the message
882 * from had when the netlink socket was created and the sender of the
883 * message has has the capability @cap in the user namespace @user_ns.
884 */
885 bool netlink_ns_capable(const struct sk_buff *skb,
886 struct user_namespace *user_ns, int cap)
887 {
888 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
889 }
890 EXPORT_SYMBOL(netlink_ns_capable);
891
892 /**
893 * netlink_capable - Netlink global message capability test
894 * @skb: socket buffer holding a netlink command from userspace
895 * @cap: The capability to use
896 *
897 * Test to see if the opener of the socket we received the message
898 * from had when the netlink socket was created and the sender of the
899 * message has has the capability @cap in all user namespaces.
900 */
901 bool netlink_capable(const struct sk_buff *skb, int cap)
902 {
903 return netlink_ns_capable(skb, &init_user_ns, cap);
904 }
905 EXPORT_SYMBOL(netlink_capable);
906
907 /**
908 * netlink_net_capable - Netlink network namespace message capability test
909 * @skb: socket buffer holding a netlink command from userspace
910 * @cap: The capability to use
911 *
912 * Test to see if the opener of the socket we received the message
913 * from had when the netlink socket was created and the sender of the
914 * message has has the capability @cap over the network namespace of
915 * the socket we received the message from.
916 */
917 bool netlink_net_capable(const struct sk_buff *skb, int cap)
918 {
919 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
920 }
921 EXPORT_SYMBOL(netlink_net_capable);
922
923 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
924 {
925 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
926 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
927 }
928
929 static void
930 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
931 {
932 struct netlink_sock *nlk = nlk_sk(sk);
933
934 if (nlk->subscriptions && !subscriptions)
935 __sk_del_bind_node(sk);
936 else if (!nlk->subscriptions && subscriptions)
937 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
938 nlk->subscriptions = subscriptions;
939 }
940
941 static int netlink_realloc_groups(struct sock *sk)
942 {
943 struct netlink_sock *nlk = nlk_sk(sk);
944 unsigned int groups;
945 unsigned long *new_groups;
946 int err = 0;
947
948 netlink_table_grab();
949
950 groups = nl_table[sk->sk_protocol].groups;
951 if (!nl_table[sk->sk_protocol].registered) {
952 err = -ENOENT;
953 goto out_unlock;
954 }
955
956 if (nlk->ngroups >= groups)
957 goto out_unlock;
958
959 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
960 if (new_groups == NULL) {
961 err = -ENOMEM;
962 goto out_unlock;
963 }
964 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
965 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
966
967 nlk->groups = new_groups;
968 nlk->ngroups = groups;
969 out_unlock:
970 netlink_table_ungrab();
971 return err;
972 }
973
974 static void netlink_undo_bind(int group, long unsigned int groups,
975 struct sock *sk)
976 {
977 struct netlink_sock *nlk = nlk_sk(sk);
978 int undo;
979
980 if (!nlk->netlink_unbind)
981 return;
982
983 for (undo = 0; undo < group; undo++)
984 if (test_bit(undo, &groups))
985 nlk->netlink_unbind(sock_net(sk), undo + 1);
986 }
987
988 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
989 int addr_len)
990 {
991 struct sock *sk = sock->sk;
992 struct net *net = sock_net(sk);
993 struct netlink_sock *nlk = nlk_sk(sk);
994 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
995 int err = 0;
996 long unsigned int groups = nladdr->nl_groups;
997 bool bound;
998
999 if (addr_len < sizeof(struct sockaddr_nl))
1000 return -EINVAL;
1001
1002 if (nladdr->nl_family != AF_NETLINK)
1003 return -EINVAL;
1004
1005 /* Only superuser is allowed to listen multicasts */
1006 if (groups) {
1007 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1008 return -EPERM;
1009 err = netlink_realloc_groups(sk);
1010 if (err)
1011 return err;
1012 }
1013
1014 if (nlk->ngroups == 0)
1015 groups = 0;
1016 else if (nlk->ngroups < 8*sizeof(groups))
1017 groups &= (1UL << nlk->ngroups) - 1;
1018
1019 bound = nlk->bound;
1020 if (bound) {
1021 /* Ensure nlk->portid is up-to-date. */
1022 smp_rmb();
1023
1024 if (nladdr->nl_pid != nlk->portid)
1025 return -EINVAL;
1026 }
1027
1028 netlink_lock_table();
1029 if (nlk->netlink_bind && groups) {
1030 int group;
1031
1032 for (group = 0; group < nlk->ngroups; group++) {
1033 if (!test_bit(group, &groups))
1034 continue;
1035 err = nlk->netlink_bind(net, group + 1);
1036 if (!err)
1037 continue;
1038 netlink_undo_bind(group, groups, sk);
1039 goto unlock;
1040 }
1041 }
1042
1043 /* No need for barriers here as we return to user-space without
1044 * using any of the bound attributes.
1045 */
1046 if (!bound) {
1047 err = nladdr->nl_pid ?
1048 netlink_insert(sk, nladdr->nl_pid) :
1049 netlink_autobind(sock);
1050 if (err) {
1051 netlink_undo_bind(nlk->ngroups, groups, sk);
1052 goto unlock;
1053 }
1054 }
1055
1056 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1057 goto unlock;
1058 netlink_unlock_table();
1059
1060 netlink_table_grab();
1061 netlink_update_subscriptions(sk, nlk->subscriptions +
1062 hweight32(groups) -
1063 hweight32(nlk->groups[0]));
1064 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1065 netlink_update_listeners(sk);
1066 netlink_table_ungrab();
1067
1068 return 0;
1069
1070 unlock:
1071 netlink_unlock_table();
1072 return err;
1073 }
1074
1075 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1076 int alen, int flags)
1077 {
1078 int err = 0;
1079 struct sock *sk = sock->sk;
1080 struct netlink_sock *nlk = nlk_sk(sk);
1081 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1082
1083 if (alen < sizeof(addr->sa_family))
1084 return -EINVAL;
1085
1086 if (addr->sa_family == AF_UNSPEC) {
1087 sk->sk_state = NETLINK_UNCONNECTED;
1088 nlk->dst_portid = 0;
1089 nlk->dst_group = 0;
1090 return 0;
1091 }
1092 if (addr->sa_family != AF_NETLINK)
1093 return -EINVAL;
1094
1095 if (alen < sizeof(struct sockaddr_nl))
1096 return -EINVAL;
1097
1098 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1099 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1100 return -EPERM;
1101
1102 /* No need for barriers here as we return to user-space without
1103 * using any of the bound attributes.
1104 */
1105 if (!nlk->bound)
1106 err = netlink_autobind(sock);
1107
1108 if (err == 0) {
1109 sk->sk_state = NETLINK_CONNECTED;
1110 nlk->dst_portid = nladdr->nl_pid;
1111 nlk->dst_group = ffs(nladdr->nl_groups);
1112 }
1113
1114 return err;
1115 }
1116
1117 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1118 int peer)
1119 {
1120 struct sock *sk = sock->sk;
1121 struct netlink_sock *nlk = nlk_sk(sk);
1122 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1123
1124 nladdr->nl_family = AF_NETLINK;
1125 nladdr->nl_pad = 0;
1126
1127 if (peer) {
1128 nladdr->nl_pid = nlk->dst_portid;
1129 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1130 } else {
1131 nladdr->nl_pid = nlk->portid;
1132 netlink_lock_table();
1133 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1134 netlink_unlock_table();
1135 }
1136 return sizeof(*nladdr);
1137 }
1138
1139 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1140 unsigned long arg)
1141 {
1142 /* try to hand this ioctl down to the NIC drivers.
1143 */
1144 return -ENOIOCTLCMD;
1145 }
1146
1147 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1148 {
1149 struct sock *sock;
1150 struct netlink_sock *nlk;
1151
1152 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1153 if (!sock)
1154 return ERR_PTR(-ECONNREFUSED);
1155
1156 /* Don't bother queuing skb if kernel socket has no input function */
1157 nlk = nlk_sk(sock);
1158 if (sock->sk_state == NETLINK_CONNECTED &&
1159 nlk->dst_portid != nlk_sk(ssk)->portid) {
1160 sock_put(sock);
1161 return ERR_PTR(-ECONNREFUSED);
1162 }
1163 return sock;
1164 }
1165
1166 struct sock *netlink_getsockbyfilp(struct file *filp)
1167 {
1168 struct inode *inode = file_inode(filp);
1169 struct sock *sock;
1170
1171 if (!S_ISSOCK(inode->i_mode))
1172 return ERR_PTR(-ENOTSOCK);
1173
1174 sock = SOCKET_I(inode)->sk;
1175 if (sock->sk_family != AF_NETLINK)
1176 return ERR_PTR(-EINVAL);
1177
1178 sock_hold(sock);
1179 return sock;
1180 }
1181
1182 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1183 int broadcast)
1184 {
1185 struct sk_buff *skb;
1186 void *data;
1187
1188 if (size <= NLMSG_GOODSIZE || broadcast)
1189 return alloc_skb(size, GFP_KERNEL);
1190
1191 size = SKB_DATA_ALIGN(size) +
1192 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1193
1194 data = vmalloc(size);
1195 if (data == NULL)
1196 return NULL;
1197
1198 skb = __build_skb(data, size);
1199 if (skb == NULL)
1200 vfree(data);
1201 else
1202 skb->destructor = netlink_skb_destructor;
1203
1204 return skb;
1205 }
1206
1207 /*
1208 * Attach a skb to a netlink socket.
1209 * The caller must hold a reference to the destination socket. On error, the
1210 * reference is dropped. The skb is not send to the destination, just all
1211 * all error checks are performed and memory in the queue is reserved.
1212 * Return values:
1213 * < 0: error. skb freed, reference to sock dropped.
1214 * 0: continue
1215 * 1: repeat lookup - reference dropped while waiting for socket memory.
1216 */
1217 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1218 long *timeo, struct sock *ssk)
1219 {
1220 struct netlink_sock *nlk;
1221
1222 nlk = nlk_sk(sk);
1223
1224 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1225 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1226 DECLARE_WAITQUEUE(wait, current);
1227 if (!*timeo) {
1228 if (!ssk || netlink_is_kernel(ssk))
1229 netlink_overrun(sk);
1230 sock_put(sk);
1231 kfree_skb(skb);
1232 return -EAGAIN;
1233 }
1234
1235 __set_current_state(TASK_INTERRUPTIBLE);
1236 add_wait_queue(&nlk->wait, &wait);
1237
1238 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1239 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1240 !sock_flag(sk, SOCK_DEAD))
1241 *timeo = schedule_timeout(*timeo);
1242
1243 __set_current_state(TASK_RUNNING);
1244 remove_wait_queue(&nlk->wait, &wait);
1245 sock_put(sk);
1246
1247 if (signal_pending(current)) {
1248 kfree_skb(skb);
1249 return sock_intr_errno(*timeo);
1250 }
1251 return 1;
1252 }
1253 netlink_skb_set_owner_r(skb, sk);
1254 return 0;
1255 }
1256
1257 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1258 {
1259 int len = skb->len;
1260
1261 netlink_deliver_tap(sock_net(sk), skb);
1262
1263 skb_queue_tail(&sk->sk_receive_queue, skb);
1264 sk->sk_data_ready(sk);
1265 return len;
1266 }
1267
1268 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1269 {
1270 int len = __netlink_sendskb(sk, skb);
1271
1272 sock_put(sk);
1273 return len;
1274 }
1275
1276 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1277 {
1278 kfree_skb(skb);
1279 sock_put(sk);
1280 }
1281
1282 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1283 {
1284 int delta;
1285
1286 WARN_ON(skb->sk != NULL);
1287 delta = skb->end - skb->tail;
1288 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1289 return skb;
1290
1291 if (skb_shared(skb)) {
1292 struct sk_buff *nskb = skb_clone(skb, allocation);
1293 if (!nskb)
1294 return skb;
1295 consume_skb(skb);
1296 skb = nskb;
1297 }
1298
1299 pskb_expand_head(skb, 0, -delta,
1300 (allocation & ~__GFP_DIRECT_RECLAIM) |
1301 __GFP_NOWARN | __GFP_NORETRY);
1302 return skb;
1303 }
1304
1305 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1306 struct sock *ssk)
1307 {
1308 int ret;
1309 struct netlink_sock *nlk = nlk_sk(sk);
1310
1311 ret = -ECONNREFUSED;
1312 if (nlk->netlink_rcv != NULL) {
1313 ret = skb->len;
1314 netlink_skb_set_owner_r(skb, sk);
1315 NETLINK_CB(skb).sk = ssk;
1316 netlink_deliver_tap_kernel(sk, ssk, skb);
1317 nlk->netlink_rcv(skb);
1318 consume_skb(skb);
1319 } else {
1320 kfree_skb(skb);
1321 }
1322 sock_put(sk);
1323 return ret;
1324 }
1325
1326 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1327 u32 portid, int nonblock)
1328 {
1329 struct sock *sk;
1330 int err;
1331 long timeo;
1332
1333 skb = netlink_trim(skb, gfp_any());
1334
1335 timeo = sock_sndtimeo(ssk, nonblock);
1336 retry:
1337 sk = netlink_getsockbyportid(ssk, portid);
1338 if (IS_ERR(sk)) {
1339 kfree_skb(skb);
1340 return PTR_ERR(sk);
1341 }
1342 if (netlink_is_kernel(sk))
1343 return netlink_unicast_kernel(sk, skb, ssk);
1344
1345 if (sk_filter(sk, skb)) {
1346 err = skb->len;
1347 kfree_skb(skb);
1348 sock_put(sk);
1349 return err;
1350 }
1351
1352 err = netlink_attachskb(sk, skb, &timeo, ssk);
1353 if (err == 1)
1354 goto retry;
1355 if (err)
1356 return err;
1357
1358 return netlink_sendskb(sk, skb);
1359 }
1360 EXPORT_SYMBOL(netlink_unicast);
1361
1362 int netlink_has_listeners(struct sock *sk, unsigned int group)
1363 {
1364 int res = 0;
1365 struct listeners *listeners;
1366
1367 BUG_ON(!netlink_is_kernel(sk));
1368
1369 rcu_read_lock();
1370 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1371
1372 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1373 res = test_bit(group - 1, listeners->masks);
1374
1375 rcu_read_unlock();
1376
1377 return res;
1378 }
1379 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1380
1381 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1382 {
1383 struct netlink_sock *nlk = nlk_sk(sk);
1384
1385 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1386 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1387 netlink_skb_set_owner_r(skb, sk);
1388 __netlink_sendskb(sk, skb);
1389 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1390 }
1391 return -1;
1392 }
1393
1394 struct netlink_broadcast_data {
1395 struct sock *exclude_sk;
1396 struct net *net;
1397 u32 portid;
1398 u32 group;
1399 int failure;
1400 int delivery_failure;
1401 int congested;
1402 int delivered;
1403 gfp_t allocation;
1404 struct sk_buff *skb, *skb2;
1405 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1406 void *tx_data;
1407 };
1408
1409 static void do_one_broadcast(struct sock *sk,
1410 struct netlink_broadcast_data *p)
1411 {
1412 struct netlink_sock *nlk = nlk_sk(sk);
1413 int val;
1414
1415 if (p->exclude_sk == sk)
1416 return;
1417
1418 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1419 !test_bit(p->group - 1, nlk->groups))
1420 return;
1421
1422 if (!net_eq(sock_net(sk), p->net)) {
1423 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1424 return;
1425
1426 if (!peernet_has_id(sock_net(sk), p->net))
1427 return;
1428
1429 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1430 CAP_NET_BROADCAST))
1431 return;
1432 }
1433
1434 if (p->failure) {
1435 netlink_overrun(sk);
1436 return;
1437 }
1438
1439 sock_hold(sk);
1440 if (p->skb2 == NULL) {
1441 if (skb_shared(p->skb)) {
1442 p->skb2 = skb_clone(p->skb, p->allocation);
1443 } else {
1444 p->skb2 = skb_get(p->skb);
1445 /*
1446 * skb ownership may have been set when
1447 * delivered to a previous socket.
1448 */
1449 skb_orphan(p->skb2);
1450 }
1451 }
1452 if (p->skb2 == NULL) {
1453 netlink_overrun(sk);
1454 /* Clone failed. Notify ALL listeners. */
1455 p->failure = 1;
1456 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1457 p->delivery_failure = 1;
1458 goto out;
1459 }
1460 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1461 kfree_skb(p->skb2);
1462 p->skb2 = NULL;
1463 goto out;
1464 }
1465 if (sk_filter(sk, p->skb2)) {
1466 kfree_skb(p->skb2);
1467 p->skb2 = NULL;
1468 goto out;
1469 }
1470 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1471 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1472 NETLINK_CB(p->skb2).nsid_is_set = true;
1473 val = netlink_broadcast_deliver(sk, p->skb2);
1474 if (val < 0) {
1475 netlink_overrun(sk);
1476 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1477 p->delivery_failure = 1;
1478 } else {
1479 p->congested |= val;
1480 p->delivered = 1;
1481 p->skb2 = NULL;
1482 }
1483 out:
1484 sock_put(sk);
1485 }
1486
1487 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1488 u32 group, gfp_t allocation,
1489 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1490 void *filter_data)
1491 {
1492 struct net *net = sock_net(ssk);
1493 struct netlink_broadcast_data info;
1494 struct sock *sk;
1495
1496 skb = netlink_trim(skb, allocation);
1497
1498 info.exclude_sk = ssk;
1499 info.net = net;
1500 info.portid = portid;
1501 info.group = group;
1502 info.failure = 0;
1503 info.delivery_failure = 0;
1504 info.congested = 0;
1505 info.delivered = 0;
1506 info.allocation = allocation;
1507 info.skb = skb;
1508 info.skb2 = NULL;
1509 info.tx_filter = filter;
1510 info.tx_data = filter_data;
1511
1512 /* While we sleep in clone, do not allow to change socket list */
1513
1514 netlink_lock_table();
1515
1516 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1517 do_one_broadcast(sk, &info);
1518
1519 consume_skb(skb);
1520
1521 netlink_unlock_table();
1522
1523 if (info.delivery_failure) {
1524 kfree_skb(info.skb2);
1525 return -ENOBUFS;
1526 }
1527 consume_skb(info.skb2);
1528
1529 if (info.delivered) {
1530 if (info.congested && gfpflags_allow_blocking(allocation))
1531 yield();
1532 return 0;
1533 }
1534 return -ESRCH;
1535 }
1536 EXPORT_SYMBOL(netlink_broadcast_filtered);
1537
1538 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1539 u32 group, gfp_t allocation)
1540 {
1541 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1542 NULL, NULL);
1543 }
1544 EXPORT_SYMBOL(netlink_broadcast);
1545
1546 struct netlink_set_err_data {
1547 struct sock *exclude_sk;
1548 u32 portid;
1549 u32 group;
1550 int code;
1551 };
1552
1553 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1554 {
1555 struct netlink_sock *nlk = nlk_sk(sk);
1556 int ret = 0;
1557
1558 if (sk == p->exclude_sk)
1559 goto out;
1560
1561 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1562 goto out;
1563
1564 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1565 !test_bit(p->group - 1, nlk->groups))
1566 goto out;
1567
1568 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1569 ret = 1;
1570 goto out;
1571 }
1572
1573 sk->sk_err = p->code;
1574 sk->sk_error_report(sk);
1575 out:
1576 return ret;
1577 }
1578
1579 /**
1580 * netlink_set_err - report error to broadcast listeners
1581 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1582 * @portid: the PORTID of a process that we want to skip (if any)
1583 * @group: the broadcast group that will notice the error
1584 * @code: error code, must be negative (as usual in kernelspace)
1585 *
1586 * This function returns the number of broadcast listeners that have set the
1587 * NETLINK_NO_ENOBUFS socket option.
1588 */
1589 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1590 {
1591 struct netlink_set_err_data info;
1592 struct sock *sk;
1593 int ret = 0;
1594
1595 info.exclude_sk = ssk;
1596 info.portid = portid;
1597 info.group = group;
1598 /* sk->sk_err wants a positive error value */
1599 info.code = -code;
1600
1601 read_lock(&nl_table_lock);
1602
1603 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1604 ret += do_one_set_err(sk, &info);
1605
1606 read_unlock(&nl_table_lock);
1607 return ret;
1608 }
1609 EXPORT_SYMBOL(netlink_set_err);
1610
1611 /* must be called with netlink table grabbed */
1612 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1613 unsigned int group,
1614 int is_new)
1615 {
1616 int old, new = !!is_new, subscriptions;
1617
1618 old = test_bit(group - 1, nlk->groups);
1619 subscriptions = nlk->subscriptions - old + new;
1620 if (new)
1621 __set_bit(group - 1, nlk->groups);
1622 else
1623 __clear_bit(group - 1, nlk->groups);
1624 netlink_update_subscriptions(&nlk->sk, subscriptions);
1625 netlink_update_listeners(&nlk->sk);
1626 }
1627
1628 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1629 char __user *optval, unsigned int optlen)
1630 {
1631 struct sock *sk = sock->sk;
1632 struct netlink_sock *nlk = nlk_sk(sk);
1633 unsigned int val = 0;
1634 int err;
1635
1636 if (level != SOL_NETLINK)
1637 return -ENOPROTOOPT;
1638
1639 if (optlen >= sizeof(int) &&
1640 get_user(val, (unsigned int __user *)optval))
1641 return -EFAULT;
1642
1643 switch (optname) {
1644 case NETLINK_PKTINFO:
1645 if (val)
1646 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1647 else
1648 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1649 err = 0;
1650 break;
1651 case NETLINK_ADD_MEMBERSHIP:
1652 case NETLINK_DROP_MEMBERSHIP: {
1653 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1654 return -EPERM;
1655 err = netlink_realloc_groups(sk);
1656 if (err)
1657 return err;
1658 if (!val || val - 1 >= nlk->ngroups)
1659 return -EINVAL;
1660 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1661 err = nlk->netlink_bind(sock_net(sk), val);
1662 if (err)
1663 return err;
1664 }
1665 netlink_table_grab();
1666 netlink_update_socket_mc(nlk, val,
1667 optname == NETLINK_ADD_MEMBERSHIP);
1668 netlink_table_ungrab();
1669 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1670 nlk->netlink_unbind(sock_net(sk), val);
1671
1672 err = 0;
1673 break;
1674 }
1675 case NETLINK_BROADCAST_ERROR:
1676 if (val)
1677 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1678 else
1679 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1680 err = 0;
1681 break;
1682 case NETLINK_NO_ENOBUFS:
1683 if (val) {
1684 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1685 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1686 wake_up_interruptible(&nlk->wait);
1687 } else {
1688 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1689 }
1690 err = 0;
1691 break;
1692 case NETLINK_LISTEN_ALL_NSID:
1693 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1694 return -EPERM;
1695
1696 if (val)
1697 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1698 else
1699 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1700 err = 0;
1701 break;
1702 case NETLINK_CAP_ACK:
1703 if (val)
1704 nlk->flags |= NETLINK_F_CAP_ACK;
1705 else
1706 nlk->flags &= ~NETLINK_F_CAP_ACK;
1707 err = 0;
1708 break;
1709 case NETLINK_EXT_ACK:
1710 if (val)
1711 nlk->flags |= NETLINK_F_EXT_ACK;
1712 else
1713 nlk->flags &= ~NETLINK_F_EXT_ACK;
1714 err = 0;
1715 break;
1716 default:
1717 err = -ENOPROTOOPT;
1718 }
1719 return err;
1720 }
1721
1722 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1723 char __user *optval, int __user *optlen)
1724 {
1725 struct sock *sk = sock->sk;
1726 struct netlink_sock *nlk = nlk_sk(sk);
1727 int len, val, err;
1728
1729 if (level != SOL_NETLINK)
1730 return -ENOPROTOOPT;
1731
1732 if (get_user(len, optlen))
1733 return -EFAULT;
1734 if (len < 0)
1735 return -EINVAL;
1736
1737 switch (optname) {
1738 case NETLINK_PKTINFO:
1739 if (len < sizeof(int))
1740 return -EINVAL;
1741 len = sizeof(int);
1742 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1743 if (put_user(len, optlen) ||
1744 put_user(val, optval))
1745 return -EFAULT;
1746 err = 0;
1747 break;
1748 case NETLINK_BROADCAST_ERROR:
1749 if (len < sizeof(int))
1750 return -EINVAL;
1751 len = sizeof(int);
1752 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1753 if (put_user(len, optlen) ||
1754 put_user(val, optval))
1755 return -EFAULT;
1756 err = 0;
1757 break;
1758 case NETLINK_NO_ENOBUFS:
1759 if (len < sizeof(int))
1760 return -EINVAL;
1761 len = sizeof(int);
1762 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1763 if (put_user(len, optlen) ||
1764 put_user(val, optval))
1765 return -EFAULT;
1766 err = 0;
1767 break;
1768 case NETLINK_LIST_MEMBERSHIPS: {
1769 int pos, idx, shift;
1770
1771 err = 0;
1772 netlink_lock_table();
1773 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1774 if (len - pos < sizeof(u32))
1775 break;
1776
1777 idx = pos / sizeof(unsigned long);
1778 shift = (pos % sizeof(unsigned long)) * 8;
1779 if (put_user((u32)(nlk->groups[idx] >> shift),
1780 (u32 __user *)(optval + pos))) {
1781 err = -EFAULT;
1782 break;
1783 }
1784 }
1785 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1786 err = -EFAULT;
1787 netlink_unlock_table();
1788 break;
1789 }
1790 case NETLINK_CAP_ACK:
1791 if (len < sizeof(int))
1792 return -EINVAL;
1793 len = sizeof(int);
1794 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1795 if (put_user(len, optlen) ||
1796 put_user(val, optval))
1797 return -EFAULT;
1798 err = 0;
1799 break;
1800 case NETLINK_EXT_ACK:
1801 if (len < sizeof(int))
1802 return -EINVAL;
1803 len = sizeof(int);
1804 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1805 if (put_user(len, optlen) || put_user(val, optval))
1806 return -EFAULT;
1807 err = 0;
1808 break;
1809 default:
1810 err = -ENOPROTOOPT;
1811 }
1812 return err;
1813 }
1814
1815 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1816 {
1817 struct nl_pktinfo info;
1818
1819 info.group = NETLINK_CB(skb).dst_group;
1820 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1821 }
1822
1823 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1824 struct sk_buff *skb)
1825 {
1826 if (!NETLINK_CB(skb).nsid_is_set)
1827 return;
1828
1829 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1830 &NETLINK_CB(skb).nsid);
1831 }
1832
1833 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1834 {
1835 struct sock *sk = sock->sk;
1836 struct netlink_sock *nlk = nlk_sk(sk);
1837 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1838 u32 dst_portid;
1839 u32 dst_group;
1840 struct sk_buff *skb;
1841 int err;
1842 struct scm_cookie scm;
1843 u32 netlink_skb_flags = 0;
1844
1845 if (msg->msg_flags&MSG_OOB)
1846 return -EOPNOTSUPP;
1847
1848 err = scm_send(sock, msg, &scm, true);
1849 if (err < 0)
1850 return err;
1851
1852 if (msg->msg_namelen) {
1853 err = -EINVAL;
1854 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1855 goto out;
1856 if (addr->nl_family != AF_NETLINK)
1857 goto out;
1858 dst_portid = addr->nl_pid;
1859 dst_group = ffs(addr->nl_groups);
1860 err = -EPERM;
1861 if ((dst_group || dst_portid) &&
1862 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1863 goto out;
1864 netlink_skb_flags |= NETLINK_SKB_DST;
1865 } else {
1866 dst_portid = nlk->dst_portid;
1867 dst_group = nlk->dst_group;
1868 }
1869
1870 if (!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
1910 out:
1911 scm_destroy(&scm);
1912 return err;
1913 }
1914
1915 static 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 int noblock = flags&MSG_DONTWAIT;
1922 size_t copied;
1923 struct sk_buff *skb, *data_skb;
1924 int err, ret;
1925
1926 if (flags&MSG_OOB)
1927 return -EOPNOTSUPP;
1928
1929 copied = 0;
1930
1931 skb = skb_recv_datagram(sk, flags, noblock, &err);
1932 if (skb == NULL)
1933 goto out;
1934
1935 data_skb = skb;
1936
1937 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1938 if (unlikely(skb_shinfo(skb)->frag_list)) {
1939 /*
1940 * If this skb has a frag_list, then here that means that we
1941 * will have to use the frag_list skb's data for compat tasks
1942 * and the regular skb's data for normal (non-compat) tasks.
1943 *
1944 * If we need to send the compat skb, assign it to the
1945 * 'data_skb' variable so that it will be used below for data
1946 * copying. We keep 'skb' for everything else, including
1947 * freeing both later.
1948 */
1949 if (flags & MSG_CMSG_COMPAT)
1950 data_skb = skb_shinfo(skb)->frag_list;
1951 }
1952 #endif
1953
1954 /* Record the max length of recvmsg() calls for future allocations */
1955 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1956 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1957 SKB_WITH_OVERHEAD(32768));
1958
1959 copied = data_skb->len;
1960 if (len < copied) {
1961 msg->msg_flags |= MSG_TRUNC;
1962 copied = len;
1963 }
1964
1965 skb_reset_transport_header(data_skb);
1966 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1967
1968 if (msg->msg_name) {
1969 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1970 addr->nl_family = AF_NETLINK;
1971 addr->nl_pad = 0;
1972 addr->nl_pid = NETLINK_CB(skb).portid;
1973 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1974 msg->msg_namelen = sizeof(*addr);
1975 }
1976
1977 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1978 netlink_cmsg_recv_pktinfo(msg, skb);
1979 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1980 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1981
1982 memset(&scm, 0, sizeof(scm));
1983 scm.creds = *NETLINK_CREDS(skb);
1984 if (flags & MSG_TRUNC)
1985 copied = data_skb->len;
1986
1987 skb_free_datagram(sk, skb);
1988
1989 if (nlk->cb_running &&
1990 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1991 ret = netlink_dump(sk);
1992 if (ret) {
1993 sk->sk_err = -ret;
1994 sk->sk_error_report(sk);
1995 }
1996 }
1997
1998 scm_recv(sock, msg, &scm, flags);
1999 out:
2000 netlink_rcv_wake(sk);
2001 return err ? : copied;
2002 }
2003
2004 static void netlink_data_ready(struct sock *sk)
2005 {
2006 BUG();
2007 }
2008
2009 /*
2010 * We export these functions to other modules. They provide a
2011 * complete set of kernel non-blocking support for message
2012 * queueing.
2013 */
2014
2015 struct sock *
2016 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2017 struct netlink_kernel_cfg *cfg)
2018 {
2019 struct socket *sock;
2020 struct sock *sk;
2021 struct netlink_sock *nlk;
2022 struct listeners *listeners = NULL;
2023 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2024 unsigned int groups;
2025
2026 BUG_ON(!nl_table);
2027
2028 if (unit < 0 || unit >= MAX_LINKS)
2029 return NULL;
2030
2031 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2032 return NULL;
2033
2034 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2035 goto out_sock_release_nosk;
2036
2037 sk = sock->sk;
2038
2039 if (!cfg || cfg->groups < 32)
2040 groups = 32;
2041 else
2042 groups = cfg->groups;
2043
2044 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2045 if (!listeners)
2046 goto out_sock_release;
2047
2048 sk->sk_data_ready = netlink_data_ready;
2049 if (cfg && cfg->input)
2050 nlk_sk(sk)->netlink_rcv = cfg->input;
2051
2052 if (netlink_insert(sk, 0))
2053 goto out_sock_release;
2054
2055 nlk = nlk_sk(sk);
2056 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2057
2058 netlink_table_grab();
2059 if (!nl_table[unit].registered) {
2060 nl_table[unit].groups = groups;
2061 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2062 nl_table[unit].cb_mutex = cb_mutex;
2063 nl_table[unit].module = module;
2064 if (cfg) {
2065 nl_table[unit].bind = cfg->bind;
2066 nl_table[unit].unbind = cfg->unbind;
2067 nl_table[unit].flags = cfg->flags;
2068 if (cfg->compare)
2069 nl_table[unit].compare = cfg->compare;
2070 }
2071 nl_table[unit].registered = 1;
2072 } else {
2073 kfree(listeners);
2074 nl_table[unit].registered++;
2075 }
2076 netlink_table_ungrab();
2077 return sk;
2078
2079 out_sock_release:
2080 kfree(listeners);
2081 netlink_kernel_release(sk);
2082 return NULL;
2083
2084 out_sock_release_nosk:
2085 sock_release(sock);
2086 return NULL;
2087 }
2088 EXPORT_SYMBOL(__netlink_kernel_create);
2089
2090 void
2091 netlink_kernel_release(struct sock *sk)
2092 {
2093 if (sk == NULL || sk->sk_socket == NULL)
2094 return;
2095
2096 sock_release(sk->sk_socket);
2097 }
2098 EXPORT_SYMBOL(netlink_kernel_release);
2099
2100 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2101 {
2102 struct listeners *new, *old;
2103 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2104
2105 if (groups < 32)
2106 groups = 32;
2107
2108 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2109 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2110 if (!new)
2111 return -ENOMEM;
2112 old = nl_deref_protected(tbl->listeners);
2113 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2114 rcu_assign_pointer(tbl->listeners, new);
2115
2116 kfree_rcu(old, rcu);
2117 }
2118 tbl->groups = groups;
2119
2120 return 0;
2121 }
2122
2123 /**
2124 * netlink_change_ngroups - change number of multicast groups
2125 *
2126 * This changes the number of multicast groups that are available
2127 * on a certain netlink family. Note that it is not possible to
2128 * change the number of groups to below 32. Also note that it does
2129 * not implicitly call netlink_clear_multicast_users() when the
2130 * number of groups is reduced.
2131 *
2132 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2133 * @groups: The new number of groups.
2134 */
2135 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2136 {
2137 int err;
2138
2139 netlink_table_grab();
2140 err = __netlink_change_ngroups(sk, groups);
2141 netlink_table_ungrab();
2142
2143 return err;
2144 }
2145
2146 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2147 {
2148 struct sock *sk;
2149 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2150
2151 sk_for_each_bound(sk, &tbl->mc_list)
2152 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2153 }
2154
2155 struct nlmsghdr *
2156 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2157 {
2158 struct nlmsghdr *nlh;
2159 int size = nlmsg_msg_size(len);
2160
2161 nlh = skb_put(skb, NLMSG_ALIGN(size));
2162 nlh->nlmsg_type = type;
2163 nlh->nlmsg_len = size;
2164 nlh->nlmsg_flags = flags;
2165 nlh->nlmsg_pid = portid;
2166 nlh->nlmsg_seq = seq;
2167 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2168 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2169 return nlh;
2170 }
2171 EXPORT_SYMBOL(__nlmsg_put);
2172
2173 /*
2174 * It looks a bit ugly.
2175 * It would be better to create kernel thread.
2176 */
2177
2178 static int netlink_dump(struct sock *sk)
2179 {
2180 struct netlink_sock *nlk = nlk_sk(sk);
2181 struct netlink_callback *cb;
2182 struct sk_buff *skb = NULL;
2183 struct nlmsghdr *nlh;
2184 struct module *module;
2185 int err = -ENOBUFS;
2186 int alloc_min_size;
2187 int alloc_size;
2188
2189 mutex_lock(nlk->cb_mutex);
2190 if (!nlk->cb_running) {
2191 err = -EINVAL;
2192 goto errout_skb;
2193 }
2194
2195 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2196 goto errout_skb;
2197
2198 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2199 * required, but it makes sense to _attempt_ a 16K bytes allocation
2200 * to reduce number of system calls on dump operations, if user
2201 * ever provided a big enough buffer.
2202 */
2203 cb = &nlk->cb;
2204 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2205
2206 if (alloc_min_size < nlk->max_recvmsg_len) {
2207 alloc_size = nlk->max_recvmsg_len;
2208 skb = alloc_skb(alloc_size,
2209 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2210 __GFP_NOWARN | __GFP_NORETRY);
2211 }
2212 if (!skb) {
2213 alloc_size = alloc_min_size;
2214 skb = alloc_skb(alloc_size, GFP_KERNEL);
2215 }
2216 if (!skb)
2217 goto errout_skb;
2218
2219 /* Trim skb to allocated size. User is expected to provide buffer as
2220 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2221 * netlink_recvmsg())). dump will pack as many smaller messages as
2222 * could fit within the allocated skb. skb is typically allocated
2223 * with larger space than required (could be as much as near 2x the
2224 * requested size with align to next power of 2 approach). Allowing
2225 * dump to use the excess space makes it difficult for a user to have a
2226 * reasonable static buffer based on the expected largest dump of a
2227 * single netdev. The outcome is MSG_TRUNC error.
2228 */
2229 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2230 netlink_skb_set_owner_r(skb, sk);
2231
2232 if (nlk->dump_done_errno > 0)
2233 nlk->dump_done_errno = cb->dump(skb, cb);
2234
2235 if (nlk->dump_done_errno > 0 ||
2236 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2237 mutex_unlock(nlk->cb_mutex);
2238
2239 if (sk_filter(sk, skb))
2240 kfree_skb(skb);
2241 else
2242 __netlink_sendskb(sk, skb);
2243 return 0;
2244 }
2245
2246 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2247 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2248 if (WARN_ON(!nlh))
2249 goto errout_skb;
2250
2251 nl_dump_check_consistent(cb, nlh);
2252
2253 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2254 sizeof(nlk->dump_done_errno));
2255
2256 if (sk_filter(sk, skb))
2257 kfree_skb(skb);
2258 else
2259 __netlink_sendskb(sk, skb);
2260
2261 if (cb->done)
2262 cb->done(cb);
2263
2264 nlk->cb_running = false;
2265 module = cb->module;
2266 skb = cb->skb;
2267 mutex_unlock(nlk->cb_mutex);
2268 module_put(module);
2269 consume_skb(skb);
2270 return 0;
2271
2272 errout_skb:
2273 mutex_unlock(nlk->cb_mutex);
2274 kfree_skb(skb);
2275 return err;
2276 }
2277
2278 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2279 const struct nlmsghdr *nlh,
2280 struct netlink_dump_control *control)
2281 {
2282 struct netlink_callback *cb;
2283 struct sock *sk;
2284 struct netlink_sock *nlk;
2285 int ret;
2286
2287 refcount_inc(&skb->users);
2288
2289 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2290 if (sk == NULL) {
2291 ret = -ECONNREFUSED;
2292 goto error_free;
2293 }
2294
2295 nlk = nlk_sk(sk);
2296 mutex_lock(nlk->cb_mutex);
2297 /* A dump is in progress... */
2298 if (nlk->cb_running) {
2299 ret = -EBUSY;
2300 goto error_unlock;
2301 }
2302 /* add reference of module which cb->dump belongs to */
2303 if (!try_module_get(control->module)) {
2304 ret = -EPROTONOSUPPORT;
2305 goto error_unlock;
2306 }
2307
2308 cb = &nlk->cb;
2309 memset(cb, 0, sizeof(*cb));
2310 cb->start = control->start;
2311 cb->dump = control->dump;
2312 cb->done = control->done;
2313 cb->nlh = nlh;
2314 cb->data = control->data;
2315 cb->module = control->module;
2316 cb->min_dump_alloc = control->min_dump_alloc;
2317 cb->skb = skb;
2318
2319 if (cb->start) {
2320 ret = cb->start(cb);
2321 if (ret)
2322 goto error_put;
2323 }
2324
2325 nlk->cb_running = true;
2326 nlk->dump_done_errno = INT_MAX;
2327
2328 mutex_unlock(nlk->cb_mutex);
2329
2330 ret = netlink_dump(sk);
2331
2332 sock_put(sk);
2333
2334 if (ret)
2335 return ret;
2336
2337 /* We successfully started a dump, by returning -EINTR we
2338 * signal not to send ACK even if it was requested.
2339 */
2340 return -EINTR;
2341
2342 error_put:
2343 module_put(control->module);
2344 error_unlock:
2345 sock_put(sk);
2346 mutex_unlock(nlk->cb_mutex);
2347 error_free:
2348 kfree_skb(skb);
2349 return ret;
2350 }
2351 EXPORT_SYMBOL(__netlink_dump_start);
2352
2353 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2354 const struct netlink_ext_ack *extack)
2355 {
2356 struct sk_buff *skb;
2357 struct nlmsghdr *rep;
2358 struct nlmsgerr *errmsg;
2359 size_t payload = sizeof(*errmsg);
2360 size_t tlvlen = 0;
2361 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2362 unsigned int flags = 0;
2363 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2364
2365 /* Error messages get the original request appened, unless the user
2366 * requests to cap the error message, and get extra error data if
2367 * requested.
2368 */
2369 if (nlk_has_extack && extack && extack->_msg)
2370 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2371
2372 if (err) {
2373 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2374 payload += nlmsg_len(nlh);
2375 else
2376 flags |= NLM_F_CAPPED;
2377 if (nlk_has_extack && extack && extack->bad_attr)
2378 tlvlen += nla_total_size(sizeof(u32));
2379 } else {
2380 flags |= NLM_F_CAPPED;
2381
2382 if (nlk_has_extack && extack && extack->cookie_len)
2383 tlvlen += nla_total_size(extack->cookie_len);
2384 }
2385
2386 if (tlvlen)
2387 flags |= NLM_F_ACK_TLVS;
2388
2389 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2390 if (!skb) {
2391 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2392 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2393 return;
2394 }
2395
2396 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2397 NLMSG_ERROR, payload, flags);
2398 errmsg = nlmsg_data(rep);
2399 errmsg->error = err;
2400 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2401
2402 if (nlk_has_extack && extack) {
2403 if (extack->_msg) {
2404 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2405 extack->_msg));
2406 }
2407 if (err) {
2408 if (extack->bad_attr &&
2409 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2410 (u8 *)extack->bad_attr >= in_skb->data +
2411 in_skb->len))
2412 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2413 (u8 *)extack->bad_attr -
2414 in_skb->data));
2415 } else {
2416 if (extack->cookie_len)
2417 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2418 extack->cookie_len,
2419 extack->cookie));
2420 }
2421 }
2422
2423 nlmsg_end(skb, rep);
2424
2425 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2426 }
2427 EXPORT_SYMBOL(netlink_ack);
2428
2429 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2430 struct nlmsghdr *,
2431 struct netlink_ext_ack *))
2432 {
2433 struct netlink_ext_ack extack;
2434 struct nlmsghdr *nlh;
2435 int err;
2436
2437 while (skb->len >= nlmsg_total_size(0)) {
2438 int msglen;
2439
2440 memset(&extack, 0, sizeof(extack));
2441 nlh = nlmsg_hdr(skb);
2442 err = 0;
2443
2444 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2445 return 0;
2446
2447 /* Only requests are handled by the kernel */
2448 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2449 goto ack;
2450
2451 /* Skip control messages */
2452 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2453 goto ack;
2454
2455 err = cb(skb, nlh, &extack);
2456 if (err == -EINTR)
2457 goto skip;
2458
2459 ack:
2460 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2461 netlink_ack(skb, nlh, err, &extack);
2462
2463 skip:
2464 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2465 if (msglen > skb->len)
2466 msglen = skb->len;
2467 skb_pull(skb, msglen);
2468 }
2469
2470 return 0;
2471 }
2472 EXPORT_SYMBOL(netlink_rcv_skb);
2473
2474 /**
2475 * nlmsg_notify - send a notification netlink message
2476 * @sk: netlink socket to use
2477 * @skb: notification message
2478 * @portid: destination netlink portid for reports or 0
2479 * @group: destination multicast group or 0
2480 * @report: 1 to report back, 0 to disable
2481 * @flags: allocation flags
2482 */
2483 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2484 unsigned int group, int report, gfp_t flags)
2485 {
2486 int err = 0;
2487
2488 if (group) {
2489 int exclude_portid = 0;
2490
2491 if (report) {
2492 refcount_inc(&skb->users);
2493 exclude_portid = portid;
2494 }
2495
2496 /* errors reported via destination sk->sk_err, but propagate
2497 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2498 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2499 }
2500
2501 if (report) {
2502 int err2;
2503
2504 err2 = nlmsg_unicast(sk, skb, portid);
2505 if (!err || err == -ESRCH)
2506 err = err2;
2507 }
2508
2509 return err;
2510 }
2511 EXPORT_SYMBOL(nlmsg_notify);
2512
2513 #ifdef CONFIG_PROC_FS
2514 struct nl_seq_iter {
2515 struct seq_net_private p;
2516 struct rhashtable_iter hti;
2517 int link;
2518 };
2519
2520 static int netlink_walk_start(struct nl_seq_iter *iter)
2521 {
2522 int err;
2523
2524 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2525 GFP_KERNEL);
2526 if (err) {
2527 iter->link = MAX_LINKS;
2528 return err;
2529 }
2530
2531 rhashtable_walk_start(&iter->hti);
2532
2533 return 0;
2534 }
2535
2536 static void netlink_walk_stop(struct nl_seq_iter *iter)
2537 {
2538 rhashtable_walk_stop(&iter->hti);
2539 rhashtable_walk_exit(&iter->hti);
2540 }
2541
2542 static void *__netlink_seq_next(struct seq_file *seq)
2543 {
2544 struct nl_seq_iter *iter = seq->private;
2545 struct netlink_sock *nlk;
2546
2547 do {
2548 for (;;) {
2549 int err;
2550
2551 nlk = rhashtable_walk_next(&iter->hti);
2552
2553 if (IS_ERR(nlk)) {
2554 if (PTR_ERR(nlk) == -EAGAIN)
2555 continue;
2556
2557 return nlk;
2558 }
2559
2560 if (nlk)
2561 break;
2562
2563 netlink_walk_stop(iter);
2564 if (++iter->link >= MAX_LINKS)
2565 return NULL;
2566
2567 err = netlink_walk_start(iter);
2568 if (err)
2569 return ERR_PTR(err);
2570 }
2571 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2572
2573 return nlk;
2574 }
2575
2576 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2577 {
2578 struct nl_seq_iter *iter = seq->private;
2579 void *obj = SEQ_START_TOKEN;
2580 loff_t pos;
2581 int err;
2582
2583 iter->link = 0;
2584
2585 err = netlink_walk_start(iter);
2586 if (err)
2587 return ERR_PTR(err);
2588
2589 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2590 obj = __netlink_seq_next(seq);
2591
2592 return obj;
2593 }
2594
2595 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2596 {
2597 ++*pos;
2598 return __netlink_seq_next(seq);
2599 }
2600
2601 static void netlink_seq_stop(struct seq_file *seq, void *v)
2602 {
2603 struct nl_seq_iter *iter = seq->private;
2604
2605 if (iter->link >= MAX_LINKS)
2606 return;
2607
2608 netlink_walk_stop(iter);
2609 }
2610
2611
2612 static int netlink_seq_show(struct seq_file *seq, void *v)
2613 {
2614 if (v == SEQ_START_TOKEN) {
2615 seq_puts(seq,
2616 "sk Eth Pid Groups "
2617 "Rmem Wmem Dump Locks Drops Inode\n");
2618 } else {
2619 struct sock *s = v;
2620 struct netlink_sock *nlk = nlk_sk(s);
2621
2622 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8d %-8lu\n",
2623 s,
2624 s->sk_protocol,
2625 nlk->portid,
2626 nlk->groups ? (u32)nlk->groups[0] : 0,
2627 sk_rmem_alloc_get(s),
2628 sk_wmem_alloc_get(s),
2629 nlk->cb_running,
2630 refcount_read(&s->sk_refcnt),
2631 atomic_read(&s->sk_drops),
2632 sock_i_ino(s)
2633 );
2634
2635 }
2636 return 0;
2637 }
2638
2639 static const struct seq_operations netlink_seq_ops = {
2640 .start = netlink_seq_start,
2641 .next = netlink_seq_next,
2642 .stop = netlink_seq_stop,
2643 .show = netlink_seq_show,
2644 };
2645 #endif
2646
2647 int netlink_register_notifier(struct notifier_block *nb)
2648 {
2649 return blocking_notifier_chain_register(&netlink_chain, nb);
2650 }
2651 EXPORT_SYMBOL(netlink_register_notifier);
2652
2653 int netlink_unregister_notifier(struct notifier_block *nb)
2654 {
2655 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2656 }
2657 EXPORT_SYMBOL(netlink_unregister_notifier);
2658
2659 static const struct proto_ops netlink_ops = {
2660 .family = PF_NETLINK,
2661 .owner = THIS_MODULE,
2662 .release = netlink_release,
2663 .bind = netlink_bind,
2664 .connect = netlink_connect,
2665 .socketpair = sock_no_socketpair,
2666 .accept = sock_no_accept,
2667 .getname = netlink_getname,
2668 .poll = datagram_poll,
2669 .ioctl = netlink_ioctl,
2670 .listen = sock_no_listen,
2671 .shutdown = sock_no_shutdown,
2672 .setsockopt = netlink_setsockopt,
2673 .getsockopt = netlink_getsockopt,
2674 .sendmsg = netlink_sendmsg,
2675 .recvmsg = netlink_recvmsg,
2676 .mmap = sock_no_mmap,
2677 .sendpage = sock_no_sendpage,
2678 };
2679
2680 static const struct net_proto_family netlink_family_ops = {
2681 .family = PF_NETLINK,
2682 .create = netlink_create,
2683 .owner = THIS_MODULE, /* for consistency 8) */
2684 };
2685
2686 static int __net_init netlink_net_init(struct net *net)
2687 {
2688 #ifdef CONFIG_PROC_FS
2689 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2690 sizeof(struct nl_seq_iter)))
2691 return -ENOMEM;
2692 #endif
2693 return 0;
2694 }
2695
2696 static void __net_exit netlink_net_exit(struct net *net)
2697 {
2698 #ifdef CONFIG_PROC_FS
2699 remove_proc_entry("netlink", net->proc_net);
2700 #endif
2701 }
2702
2703 static void __init netlink_add_usersock_entry(void)
2704 {
2705 struct listeners *listeners;
2706 int groups = 32;
2707
2708 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2709 if (!listeners)
2710 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2711
2712 netlink_table_grab();
2713
2714 nl_table[NETLINK_USERSOCK].groups = groups;
2715 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2716 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2717 nl_table[NETLINK_USERSOCK].registered = 1;
2718 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2719
2720 netlink_table_ungrab();
2721 }
2722
2723 static struct pernet_operations __net_initdata netlink_net_ops = {
2724 .init = netlink_net_init,
2725 .exit = netlink_net_exit,
2726 };
2727
2728 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2729 {
2730 const struct netlink_sock *nlk = data;
2731 struct netlink_compare_arg arg;
2732
2733 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2734 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2735 }
2736
2737 static const struct rhashtable_params netlink_rhashtable_params = {
2738 .head_offset = offsetof(struct netlink_sock, node),
2739 .key_len = netlink_compare_arg_len,
2740 .obj_hashfn = netlink_hash,
2741 .obj_cmpfn = netlink_compare,
2742 .automatic_shrinking = true,
2743 };
2744
2745 static int __init netlink_proto_init(void)
2746 {
2747 int i;
2748 int err = proto_register(&netlink_proto, 0);
2749
2750 if (err != 0)
2751 goto out;
2752
2753 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2754
2755 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2756 if (!nl_table)
2757 goto panic;
2758
2759 for (i = 0; i < MAX_LINKS; i++) {
2760 if (rhashtable_init(&nl_table[i].hash,
2761 &netlink_rhashtable_params) < 0) {
2762 while (--i > 0)
2763 rhashtable_destroy(&nl_table[i].hash);
2764 kfree(nl_table);
2765 goto panic;
2766 }
2767 }
2768
2769 netlink_add_usersock_entry();
2770
2771 sock_register(&netlink_family_ops);
2772 register_pernet_subsys(&netlink_net_ops);
2773 register_pernet_subsys(&netlink_tap_net_ops);
2774 /* The netlink device handler may be needed early. */
2775 rtnetlink_init();
2776 out:
2777 return err;
2778 panic:
2779 panic("netlink_init: Cannot allocate nl_table\n");
2780 }
2781
2782 core_initcall(netlink_proto_init);
Linux with added FPC-III support