?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / net / netlink / af_netlink.c
blob1bb9f219f07d33a966528b065d80bc6ba50fb6c3
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 /* nl_groups is a u32, so cap the maximum groups we can bind */
1033 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1034 if (!test_bit(group, &groups))
1035 continue;
1036 err = nlk->netlink_bind(net, group + 1);
1037 if (!err)
1038 continue;
1039 netlink_undo_bind(group, groups, sk);
1040 goto unlock;
1041 }
1042 }
1043
1044 /* No need for barriers here as we return to user-space without
1045 * using any of the bound attributes.
1046 */
1047 if (!bound) {
1048 err = nladdr->nl_pid ?
1049 netlink_insert(sk, nladdr->nl_pid) :
1050 netlink_autobind(sock);
1051 if (err) {
1052 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1053 goto unlock;
1054 }
1055 }
1056
1057 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1058 goto unlock;
1059 netlink_unlock_table();
1060
1061 netlink_table_grab();
1062 netlink_update_subscriptions(sk, nlk->subscriptions +
1063 hweight32(groups) -
1064 hweight32(nlk->groups[0]));
1065 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1066 netlink_update_listeners(sk);
1067 netlink_table_ungrab();
1068
1069 return 0;
1070
1071 unlock:
1072 netlink_unlock_table();
1073 return err;
1074 }
1075
1076 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1077 int alen, int flags)
1078 {
1079 int err = 0;
1080 struct sock *sk = sock->sk;
1081 struct netlink_sock *nlk = nlk_sk(sk);
1082 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1083
1084 if (alen < sizeof(addr->sa_family))
1085 return -EINVAL;
1086
1087 if (addr->sa_family == AF_UNSPEC) {
1088 sk->sk_state = NETLINK_UNCONNECTED;
1089 nlk->dst_portid = 0;
1090 nlk->dst_group = 0;
1091 return 0;
1092 }
1093 if (addr->sa_family != AF_NETLINK)
1094 return -EINVAL;
1095
1096 if (alen < sizeof(struct sockaddr_nl))
1097 return -EINVAL;
1098
1099 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1100 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1101 return -EPERM;
1102
1103 /* No need for barriers here as we return to user-space without
1104 * using any of the bound attributes.
1105 */
1106 if (!nlk->bound)
1107 err = netlink_autobind(sock);
1108
1109 if (err == 0) {
1110 sk->sk_state = NETLINK_CONNECTED;
1111 nlk->dst_portid = nladdr->nl_pid;
1112 nlk->dst_group = ffs(nladdr->nl_groups);
1113 }
1114
1115 return err;
1116 }
1117
1118 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1119 int peer)
1120 {
1121 struct sock *sk = sock->sk;
1122 struct netlink_sock *nlk = nlk_sk(sk);
1123 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1124
1125 nladdr->nl_family = AF_NETLINK;
1126 nladdr->nl_pad = 0;
1127
1128 if (peer) {
1129 nladdr->nl_pid = nlk->dst_portid;
1130 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1131 } else {
1132 nladdr->nl_pid = nlk->portid;
1133 netlink_lock_table();
1134 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1135 netlink_unlock_table();
1136 }
1137 return sizeof(*nladdr);
1138 }
1139
1140 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1141 unsigned long arg)
1142 {
1143 /* try to hand this ioctl down to the NIC drivers.
1144 */
1145 return -ENOIOCTLCMD;
1146 }
1147
1148 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1149 {
1150 struct sock *sock;
1151 struct netlink_sock *nlk;
1152
1153 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1154 if (!sock)
1155 return ERR_PTR(-ECONNREFUSED);
1156
1157 /* Don't bother queuing skb if kernel socket has no input function */
1158 nlk = nlk_sk(sock);
1159 if (sock->sk_state == NETLINK_CONNECTED &&
1160 nlk->dst_portid != nlk_sk(ssk)->portid) {
1161 sock_put(sock);
1162 return ERR_PTR(-ECONNREFUSED);
1163 }
1164 return sock;
1165 }
1166
1167 struct sock *netlink_getsockbyfilp(struct file *filp)
1168 {
1169 struct inode *inode = file_inode(filp);
1170 struct sock *sock;
1171
1172 if (!S_ISSOCK(inode->i_mode))
1173 return ERR_PTR(-ENOTSOCK);
1174
1175 sock = SOCKET_I(inode)->sk;
1176 if (sock->sk_family != AF_NETLINK)
1177 return ERR_PTR(-EINVAL);
1178
1179 sock_hold(sock);
1180 return sock;
1181 }
1182
1183 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1184 int broadcast)
1185 {
1186 struct sk_buff *skb;
1187 void *data;
1188
1189 if (size <= NLMSG_GOODSIZE || broadcast)
1190 return alloc_skb(size, GFP_KERNEL);
1191
1192 size = SKB_DATA_ALIGN(size) +
1193 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1194
1195 data = vmalloc(size);
1196 if (data == NULL)
1197 return NULL;
1198
1199 skb = __build_skb(data, size);
1200 if (skb == NULL)
1201 vfree(data);
1202 else
1203 skb->destructor = netlink_skb_destructor;
1204
1205 return skb;
1206 }
1207
1208 /*
1209 * Attach a skb to a netlink socket.
1210 * The caller must hold a reference to the destination socket. On error, the
1211 * reference is dropped. The skb is not send to the destination, just all
1212 * all error checks are performed and memory in the queue is reserved.
1213 * Return values:
1214 * < 0: error. skb freed, reference to sock dropped.
1215 * 0: continue
1216 * 1: repeat lookup - reference dropped while waiting for socket memory.
1217 */
1218 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1219 long *timeo, struct sock *ssk)
1220 {
1221 struct netlink_sock *nlk;
1222
1223 nlk = nlk_sk(sk);
1224
1225 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1226 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1227 DECLARE_WAITQUEUE(wait, current);
1228 if (!*timeo) {
1229 if (!ssk || netlink_is_kernel(ssk))
1230 netlink_overrun(sk);
1231 sock_put(sk);
1232 kfree_skb(skb);
1233 return -EAGAIN;
1234 }
1235
1236 __set_current_state(TASK_INTERRUPTIBLE);
1237 add_wait_queue(&nlk->wait, &wait);
1238
1239 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1240 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1241 !sock_flag(sk, SOCK_DEAD))
1242 *timeo = schedule_timeout(*timeo);
1243
1244 __set_current_state(TASK_RUNNING);
1245 remove_wait_queue(&nlk->wait, &wait);
1246 sock_put(sk);
1247
1248 if (signal_pending(current)) {
1249 kfree_skb(skb);
1250 return sock_intr_errno(*timeo);
1251 }
1252 return 1;
1253 }
1254 netlink_skb_set_owner_r(skb, sk);
1255 return 0;
1256 }
1257
1258 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1259 {
1260 int len = skb->len;
1261
1262 netlink_deliver_tap(sock_net(sk), skb);
1263
1264 skb_queue_tail(&sk->sk_receive_queue, skb);
1265 sk->sk_data_ready(sk);
1266 return len;
1267 }
1268
1269 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1270 {
1271 int len = __netlink_sendskb(sk, skb);
1272
1273 sock_put(sk);
1274 return len;
1275 }
1276
1277 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1278 {
1279 kfree_skb(skb);
1280 sock_put(sk);
1281 }
1282
1283 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1284 {
1285 int delta;
1286
1287 WARN_ON(skb->sk != NULL);
1288 delta = skb->end - skb->tail;
1289 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1290 return skb;
1291
1292 if (skb_shared(skb)) {
1293 struct sk_buff *nskb = skb_clone(skb, allocation);
1294 if (!nskb)
1295 return skb;
1296 consume_skb(skb);
1297 skb = nskb;
1298 }
1299
1300 pskb_expand_head(skb, 0, -delta,
1301 (allocation & ~__GFP_DIRECT_RECLAIM) |
1302 __GFP_NOWARN | __GFP_NORETRY);
1303 return skb;
1304 }
1305
1306 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1307 struct sock *ssk)
1308 {
1309 int ret;
1310 struct netlink_sock *nlk = nlk_sk(sk);
1311
1312 ret = -ECONNREFUSED;
1313 if (nlk->netlink_rcv != NULL) {
1314 ret = skb->len;
1315 netlink_skb_set_owner_r(skb, sk);
1316 NETLINK_CB(skb).sk = ssk;
1317 netlink_deliver_tap_kernel(sk, ssk, skb);
1318 nlk->netlink_rcv(skb);
1319 consume_skb(skb);
1320 } else {
1321 kfree_skb(skb);
1322 }
1323 sock_put(sk);
1324 return ret;
1325 }
1326
1327 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1328 u32 portid, int nonblock)
1329 {
1330 struct sock *sk;
1331 int err;
1332 long timeo;
1333
1334 skb = netlink_trim(skb, gfp_any());
1335
1336 timeo = sock_sndtimeo(ssk, nonblock);
1337 retry:
1338 sk = netlink_getsockbyportid(ssk, portid);
1339 if (IS_ERR(sk)) {
1340 kfree_skb(skb);
1341 return PTR_ERR(sk);
1342 }
1343 if (netlink_is_kernel(sk))
1344 return netlink_unicast_kernel(sk, skb, ssk);
1345
1346 if (sk_filter(sk, skb)) {
1347 err = skb->len;
1348 kfree_skb(skb);
1349 sock_put(sk);
1350 return err;
1351 }
1352
1353 err = netlink_attachskb(sk, skb, &timeo, ssk);
1354 if (err == 1)
1355 goto retry;
1356 if (err)
1357 return err;
1358
1359 return netlink_sendskb(sk, skb);
1360 }
1361 EXPORT_SYMBOL(netlink_unicast);
1362
1363 int netlink_has_listeners(struct sock *sk, unsigned int group)
1364 {
1365 int res = 0;
1366 struct listeners *listeners;
1367
1368 BUG_ON(!netlink_is_kernel(sk));
1369
1370 rcu_read_lock();
1371 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1372
1373 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1374 res = test_bit(group - 1, listeners->masks);
1375
1376 rcu_read_unlock();
1377
1378 return res;
1379 }
1380 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1381
1382 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1383 {
1384 struct netlink_sock *nlk = nlk_sk(sk);
1385
1386 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1387 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1388 netlink_skb_set_owner_r(skb, sk);
1389 __netlink_sendskb(sk, skb);
1390 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1391 }
1392 return -1;
1393 }
1394
1395 struct netlink_broadcast_data {
1396 struct sock *exclude_sk;
1397 struct net *net;
1398 u32 portid;
1399 u32 group;
1400 int failure;
1401 int delivery_failure;
1402 int congested;
1403 int delivered;
1404 gfp_t allocation;
1405 struct sk_buff *skb, *skb2;
1406 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1407 void *tx_data;
1408 };
1409
1410 static void do_one_broadcast(struct sock *sk,
1411 struct netlink_broadcast_data *p)
1412 {
1413 struct netlink_sock *nlk = nlk_sk(sk);
1414 int val;
1415
1416 if (p->exclude_sk == sk)
1417 return;
1418
1419 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1420 !test_bit(p->group - 1, nlk->groups))
1421 return;
1422
1423 if (!net_eq(sock_net(sk), p->net)) {
1424 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1425 return;
1426
1427 if (!peernet_has_id(sock_net(sk), p->net))
1428 return;
1429
1430 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1431 CAP_NET_BROADCAST))
1432 return;
1433 }
1434
1435 if (p->failure) {
1436 netlink_overrun(sk);
1437 return;
1438 }
1439
1440 sock_hold(sk);
1441 if (p->skb2 == NULL) {
1442 if (skb_shared(p->skb)) {
1443 p->skb2 = skb_clone(p->skb, p->allocation);
1444 } else {
1445 p->skb2 = skb_get(p->skb);
1446 /*
1447 * skb ownership may have been set when
1448 * delivered to a previous socket.
1449 */
1450 skb_orphan(p->skb2);
1451 }
1452 }
1453 if (p->skb2 == NULL) {
1454 netlink_overrun(sk);
1455 /* Clone failed. Notify ALL listeners. */
1456 p->failure = 1;
1457 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1458 p->delivery_failure = 1;
1459 goto out;
1460 }
1461 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1462 kfree_skb(p->skb2);
1463 p->skb2 = NULL;
1464 goto out;
1465 }
1466 if (sk_filter(sk, p->skb2)) {
1467 kfree_skb(p->skb2);
1468 p->skb2 = NULL;
1469 goto out;
1470 }
1471 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1472 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1473 NETLINK_CB(p->skb2).nsid_is_set = true;
1474 val = netlink_broadcast_deliver(sk, p->skb2);
1475 if (val < 0) {
1476 netlink_overrun(sk);
1477 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1478 p->delivery_failure = 1;
1479 } else {
1480 p->congested |= val;
1481 p->delivered = 1;
1482 p->skb2 = NULL;
1483 }
1484 out:
1485 sock_put(sk);
1486 }
1487
1488 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1489 u32 group, gfp_t allocation,
1490 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1491 void *filter_data)
1492 {
1493 struct net *net = sock_net(ssk);
1494 struct netlink_broadcast_data info;
1495 struct sock *sk;
1496
1497 skb = netlink_trim(skb, allocation);
1498
1499 info.exclude_sk = ssk;
1500 info.net = net;
1501 info.portid = portid;
1502 info.group = group;
1503 info.failure = 0;
1504 info.delivery_failure = 0;
1505 info.congested = 0;
1506 info.delivered = 0;
1507 info.allocation = allocation;
1508 info.skb = skb;
1509 info.skb2 = NULL;
1510 info.tx_filter = filter;
1511 info.tx_data = filter_data;
1512
1513 /* While we sleep in clone, do not allow to change socket list */
1514
1515 netlink_lock_table();
1516
1517 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1518 do_one_broadcast(sk, &info);
1519
1520 consume_skb(skb);
1521
1522 netlink_unlock_table();
1523
1524 if (info.delivery_failure) {
1525 kfree_skb(info.skb2);
1526 return -ENOBUFS;
1527 }
1528 consume_skb(info.skb2);
1529
1530 if (info.delivered) {
1531 if (info.congested && gfpflags_allow_blocking(allocation))
1532 yield();
1533 return 0;
1534 }
1535 return -ESRCH;
1536 }
1537 EXPORT_SYMBOL(netlink_broadcast_filtered);
1538
1539 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1540 u32 group, gfp_t allocation)
1541 {
1542 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1543 NULL, NULL);
1544 }
1545 EXPORT_SYMBOL(netlink_broadcast);
1546
1547 struct netlink_set_err_data {
1548 struct sock *exclude_sk;
1549 u32 portid;
1550 u32 group;
1551 int code;
1552 };
1553
1554 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1555 {
1556 struct netlink_sock *nlk = nlk_sk(sk);
1557 int ret = 0;
1558
1559 if (sk == p->exclude_sk)
1560 goto out;
1561
1562 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1563 goto out;
1564
1565 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1566 !test_bit(p->group - 1, nlk->groups))
1567 goto out;
1568
1569 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1570 ret = 1;
1571 goto out;
1572 }
1573
1574 sk->sk_err = p->code;
1575 sk->sk_error_report(sk);
1576 out:
1577 return ret;
1578 }
1579
1580 /**
1581 * netlink_set_err - report error to broadcast listeners
1582 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1583 * @portid: the PORTID of a process that we want to skip (if any)
1584 * @group: the broadcast group that will notice the error
1585 * @code: error code, must be negative (as usual in kernelspace)
1586 *
1587 * This function returns the number of broadcast listeners that have set the
1588 * NETLINK_NO_ENOBUFS socket option.
1589 */
1590 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1591 {
1592 struct netlink_set_err_data info;
1593 struct sock *sk;
1594 int ret = 0;
1595
1596 info.exclude_sk = ssk;
1597 info.portid = portid;
1598 info.group = group;
1599 /* sk->sk_err wants a positive error value */
1600 info.code = -code;
1601
1602 read_lock(&nl_table_lock);
1603
1604 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1605 ret += do_one_set_err(sk, &info);
1606
1607 read_unlock(&nl_table_lock);
1608 return ret;
1609 }
1610 EXPORT_SYMBOL(netlink_set_err);
1611
1612 /* must be called with netlink table grabbed */
1613 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1614 unsigned int group,
1615 int is_new)
1616 {
1617 int old, new = !!is_new, subscriptions;
1618
1619 old = test_bit(group - 1, nlk->groups);
1620 subscriptions = nlk->subscriptions - old + new;
1621 if (new)
1622 __set_bit(group - 1, nlk->groups);
1623 else
1624 __clear_bit(group - 1, nlk->groups);
1625 netlink_update_subscriptions(&nlk->sk, subscriptions);
1626 netlink_update_listeners(&nlk->sk);
1627 }
1628
1629 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1630 char __user *optval, unsigned int optlen)
1631 {
1632 struct sock *sk = sock->sk;
1633 struct netlink_sock *nlk = nlk_sk(sk);
1634 unsigned int val = 0;
1635 int err;
1636
1637 if (level != SOL_NETLINK)
1638 return -ENOPROTOOPT;
1639
1640 if (optlen >= sizeof(int) &&
1641 get_user(val, (unsigned int __user *)optval))
1642 return -EFAULT;
1643
1644 switch (optname) {
1645 case NETLINK_PKTINFO:
1646 if (val)
1647 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1648 else
1649 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1650 err = 0;
1651 break;
1652 case NETLINK_ADD_MEMBERSHIP:
1653 case NETLINK_DROP_MEMBERSHIP: {
1654 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1655 return -EPERM;
1656 err = netlink_realloc_groups(sk);
1657 if (err)
1658 return err;
1659 if (!val || val - 1 >= nlk->ngroups)
1660 return -EINVAL;
1661 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1662 err = nlk->netlink_bind(sock_net(sk), val);
1663 if (err)
1664 return err;
1665 }
1666 netlink_table_grab();
1667 netlink_update_socket_mc(nlk, val,
1668 optname == NETLINK_ADD_MEMBERSHIP);
1669 netlink_table_ungrab();
1670 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1671 nlk->netlink_unbind(sock_net(sk), val);
1672
1673 err = 0;
1674 break;
1675 }
1676 case NETLINK_BROADCAST_ERROR:
1677 if (val)
1678 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1679 else
1680 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1681 err = 0;
1682 break;
1683 case NETLINK_NO_ENOBUFS:
1684 if (val) {
1685 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1686 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1687 wake_up_interruptible(&nlk->wait);
1688 } else {
1689 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1690 }
1691 err = 0;
1692 break;
1693 case NETLINK_LISTEN_ALL_NSID:
1694 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1695 return -EPERM;
1696
1697 if (val)
1698 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1699 else
1700 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1701 err = 0;
1702 break;
1703 case NETLINK_CAP_ACK:
1704 if (val)
1705 nlk->flags |= NETLINK_F_CAP_ACK;
1706 else
1707 nlk->flags &= ~NETLINK_F_CAP_ACK;
1708 err = 0;
1709 break;
1710 case NETLINK_EXT_ACK:
1711 if (val)
1712 nlk->flags |= NETLINK_F_EXT_ACK;
1713 else
1714 nlk->flags &= ~NETLINK_F_EXT_ACK;
1715 err = 0;
1716 break;
1717 default:
1718 err = -ENOPROTOOPT;
1719 }
1720 return err;
1721 }
1722
1723 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1724 char __user *optval, int __user *optlen)
1725 {
1726 struct sock *sk = sock->sk;
1727 struct netlink_sock *nlk = nlk_sk(sk);
1728 int len, val, err;
1729
1730 if (level != SOL_NETLINK)
1731 return -ENOPROTOOPT;
1732
1733 if (get_user(len, optlen))
1734 return -EFAULT;
1735 if (len < 0)
1736 return -EINVAL;
1737
1738 switch (optname) {
1739 case NETLINK_PKTINFO:
1740 if (len < sizeof(int))
1741 return -EINVAL;
1742 len = sizeof(int);
1743 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1744 if (put_user(len, optlen) ||
1745 put_user(val, optval))
1746 return -EFAULT;
1747 err = 0;
1748 break;
1749 case NETLINK_BROADCAST_ERROR:
1750 if (len < sizeof(int))
1751 return -EINVAL;
1752 len = sizeof(int);
1753 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1754 if (put_user(len, optlen) ||
1755 put_user(val, optval))
1756 return -EFAULT;
1757 err = 0;
1758 break;
1759 case NETLINK_NO_ENOBUFS:
1760 if (len < sizeof(int))
1761 return -EINVAL;
1762 len = sizeof(int);
1763 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1764 if (put_user(len, optlen) ||
1765 put_user(val, optval))
1766 return -EFAULT;
1767 err = 0;
1768 break;
1769 case NETLINK_LIST_MEMBERSHIPS: {
1770 int pos, idx, shift;
1771
1772 err = 0;
1773 netlink_lock_table();
1774 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1775 if (len - pos < sizeof(u32))
1776 break;
1777
1778 idx = pos / sizeof(unsigned long);
1779 shift = (pos % sizeof(unsigned long)) * 8;
1780 if (put_user((u32)(nlk->groups[idx] >> shift),
1781 (u32 __user *)(optval + pos))) {
1782 err = -EFAULT;
1783 break;
1784 }
1785 }
1786 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1787 err = -EFAULT;
1788 netlink_unlock_table();
1789 break;
1790 }
1791 case NETLINK_CAP_ACK:
1792 if (len < sizeof(int))
1793 return -EINVAL;
1794 len = sizeof(int);
1795 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1796 if (put_user(len, optlen) ||
1797 put_user(val, optval))
1798 return -EFAULT;
1799 err = 0;
1800 break;
1801 case NETLINK_EXT_ACK:
1802 if (len < sizeof(int))
1803 return -EINVAL;
1804 len = sizeof(int);
1805 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1806 if (put_user(len, optlen) || put_user(val, optval))
1807 return -EFAULT;
1808 err = 0;
1809 break;
1810 default:
1811 err = -ENOPROTOOPT;
1812 }
1813 return err;
1814 }
1815
1816 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1817 {
1818 struct nl_pktinfo info;
1819
1820 info.group = NETLINK_CB(skb).dst_group;
1821 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1822 }
1823
1824 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1825 struct sk_buff *skb)
1826 {
1827 if (!NETLINK_CB(skb).nsid_is_set)
1828 return;
1829
1830 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1831 &NETLINK_CB(skb).nsid);
1832 }
1833
1834 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1835 {
1836 struct sock *sk = sock->sk;
1837 struct netlink_sock *nlk = nlk_sk(sk);
1838 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1839 u32 dst_portid;
1840 u32 dst_group;
1841 struct sk_buff *skb;
1842 int err;
1843 struct scm_cookie scm;
1844 u32 netlink_skb_flags = 0;
1845
1846 if (msg->msg_flags&MSG_OOB)
1847 return -EOPNOTSUPP;
1848
1849 err = scm_send(sock, msg, &scm, true);
1850 if (err < 0)
1851 return err;
1852
1853 if (msg->msg_namelen) {
1854 err = -EINVAL;
1855 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1856 goto out;
1857 if (addr->nl_family != AF_NETLINK)
1858 goto out;
1859 dst_portid = addr->nl_pid;
1860 dst_group = ffs(addr->nl_groups);
1861 err = -EPERM;
1862 if ((dst_group || dst_portid) &&
1863 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1864 goto out;
1865 netlink_skb_flags |= NETLINK_SKB_DST;
1866 } else {
1867 dst_portid = nlk->dst_portid;
1868 dst_group = nlk->dst_group;
1869 }
1870
1871 if (!nlk->bound) {
1872 err = netlink_autobind(sock);
1873 if (err)
1874 goto out;
1875 } else {
1876 /* Ensure nlk is hashed and visible. */
1877 smp_rmb();
1878 }
1879
1880 err = -EMSGSIZE;
1881 if (len > sk->sk_sndbuf - 32)
1882 goto out;
1883 err = -ENOBUFS;
1884 skb = netlink_alloc_large_skb(len, dst_group);
1885 if (skb == NULL)
1886 goto out;
1887
1888 NETLINK_CB(skb).portid = nlk->portid;
1889 NETLINK_CB(skb).dst_group = dst_group;
1890 NETLINK_CB(skb).creds = scm.creds;
1891 NETLINK_CB(skb).flags = netlink_skb_flags;
1892
1893 err = -EFAULT;
1894 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1895 kfree_skb(skb);
1896 goto out;
1897 }
1898
1899 err = security_netlink_send(sk, skb);
1900 if (err) {
1901 kfree_skb(skb);
1902 goto out;
1903 }
1904
1905 if (dst_group) {
1906 refcount_inc(&skb->users);
1907 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1908 }
1909 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1910
1911 out:
1912 scm_destroy(&scm);
1913 return err;
1914 }
1915
1916 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1917 int flags)
1918 {
1919 struct scm_cookie scm;
1920 struct sock *sk = sock->sk;
1921 struct netlink_sock *nlk = nlk_sk(sk);
1922 int noblock = flags&MSG_DONTWAIT;
1923 size_t copied;
1924 struct sk_buff *skb, *data_skb;
1925 int err, ret;
1926
1927 if (flags&MSG_OOB)
1928 return -EOPNOTSUPP;
1929
1930 copied = 0;
1931
1932 skb = skb_recv_datagram(sk, flags, noblock, &err);
1933 if (skb == NULL)
1934 goto out;
1935
1936 data_skb = skb;
1937
1938 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1939 if (unlikely(skb_shinfo(skb)->frag_list)) {
1940 /*
1941 * If this skb has a frag_list, then here that means that we
1942 * will have to use the frag_list skb's data for compat tasks
1943 * and the regular skb's data for normal (non-compat) tasks.
1944 *
1945 * If we need to send the compat skb, assign it to the
1946 * 'data_skb' variable so that it will be used below for data
1947 * copying. We keep 'skb' for everything else, including
1948 * freeing both later.
1949 */
1950 if (flags & MSG_CMSG_COMPAT)
1951 data_skb = skb_shinfo(skb)->frag_list;
1952 }
1953 #endif
1954
1955 /* Record the max length of recvmsg() calls for future allocations */
1956 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1957 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1958 SKB_WITH_OVERHEAD(32768));
1959
1960 copied = data_skb->len;
1961 if (len < copied) {
1962 msg->msg_flags |= MSG_TRUNC;
1963 copied = len;
1964 }
1965
1966 skb_reset_transport_header(data_skb);
1967 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1968
1969 if (msg->msg_name) {
1970 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1971 addr->nl_family = AF_NETLINK;
1972 addr->nl_pad = 0;
1973 addr->nl_pid = NETLINK_CB(skb).portid;
1974 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1975 msg->msg_namelen = sizeof(*addr);
1976 }
1977
1978 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1979 netlink_cmsg_recv_pktinfo(msg, skb);
1980 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1981 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1982
1983 memset(&scm, 0, sizeof(scm));
1984 scm.creds = *NETLINK_CREDS(skb);
1985 if (flags & MSG_TRUNC)
1986 copied = data_skb->len;
1987
1988 skb_free_datagram(sk, skb);
1989
1990 if (nlk->cb_running &&
1991 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1992 ret = netlink_dump(sk);
1993 if (ret) {
1994 sk->sk_err = -ret;
1995 sk->sk_error_report(sk);
1996 }
1997 }
1998
1999 scm_recv(sock, msg, &scm, flags);
2000 out:
2001 netlink_rcv_wake(sk);
2002 return err ? : copied;
2003 }
2004
2005 static void netlink_data_ready(struct sock *sk)
2006 {
2007 BUG();
2008 }
2009
2010 /*
2011 * We export these functions to other modules. They provide a
2012 * complete set of kernel non-blocking support for message
2013 * queueing.
2014 */
2015
2016 struct sock *
2017 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2018 struct netlink_kernel_cfg *cfg)
2019 {
2020 struct socket *sock;
2021 struct sock *sk;
2022 struct netlink_sock *nlk;
2023 struct listeners *listeners = NULL;
2024 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2025 unsigned int groups;
2026
2027 BUG_ON(!nl_table);
2028
2029 if (unit < 0 || unit >= MAX_LINKS)
2030 return NULL;
2031
2032 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2033 return NULL;
2034
2035 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2036 goto out_sock_release_nosk;
2037
2038 sk = sock->sk;
2039
2040 if (!cfg || cfg->groups < 32)
2041 groups = 32;
2042 else
2043 groups = cfg->groups;
2044
2045 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2046 if (!listeners)
2047 goto out_sock_release;
2048
2049 sk->sk_data_ready = netlink_data_ready;
2050 if (cfg && cfg->input)
2051 nlk_sk(sk)->netlink_rcv = cfg->input;
2052
2053 if (netlink_insert(sk, 0))
2054 goto out_sock_release;
2055
2056 nlk = nlk_sk(sk);
2057 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2058
2059 netlink_table_grab();
2060 if (!nl_table[unit].registered) {
2061 nl_table[unit].groups = groups;
2062 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2063 nl_table[unit].cb_mutex = cb_mutex;
2064 nl_table[unit].module = module;
2065 if (cfg) {
2066 nl_table[unit].bind = cfg->bind;
2067 nl_table[unit].unbind = cfg->unbind;
2068 nl_table[unit].flags = cfg->flags;
2069 if (cfg->compare)
2070 nl_table[unit].compare = cfg->compare;
2071 }
2072 nl_table[unit].registered = 1;
2073 } else {
2074 kfree(listeners);
2075 nl_table[unit].registered++;
2076 }
2077 netlink_table_ungrab();
2078 return sk;
2079
2080 out_sock_release:
2081 kfree(listeners);
2082 netlink_kernel_release(sk);
2083 return NULL;
2084
2085 out_sock_release_nosk:
2086 sock_release(sock);
2087 return NULL;
2088 }
2089 EXPORT_SYMBOL(__netlink_kernel_create);
2090
2091 void
2092 netlink_kernel_release(struct sock *sk)
2093 {
2094 if (sk == NULL || sk->sk_socket == NULL)
2095 return;
2096
2097 sock_release(sk->sk_socket);
2098 }
2099 EXPORT_SYMBOL(netlink_kernel_release);
2100
2101 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2102 {
2103 struct listeners *new, *old;
2104 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2105
2106 if (groups < 32)
2107 groups = 32;
2108
2109 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2110 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2111 if (!new)
2112 return -ENOMEM;
2113 old = nl_deref_protected(tbl->listeners);
2114 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2115 rcu_assign_pointer(tbl->listeners, new);
2116
2117 kfree_rcu(old, rcu);
2118 }
2119 tbl->groups = groups;
2120
2121 return 0;
2122 }
2123
2124 /**
2125 * netlink_change_ngroups - change number of multicast groups
2126 *
2127 * This changes the number of multicast groups that are available
2128 * on a certain netlink family. Note that it is not possible to
2129 * change the number of groups to below 32. Also note that it does
2130 * not implicitly call netlink_clear_multicast_users() when the
2131 * number of groups is reduced.
2132 *
2133 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2134 * @groups: The new number of groups.
2135 */
2136 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2137 {
2138 int err;
2139
2140 netlink_table_grab();
2141 err = __netlink_change_ngroups(sk, groups);
2142 netlink_table_ungrab();
2143
2144 return err;
2145 }
2146
2147 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2148 {
2149 struct sock *sk;
2150 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2151
2152 sk_for_each_bound(sk, &tbl->mc_list)
2153 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2154 }
2155
2156 struct nlmsghdr *
2157 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2158 {
2159 struct nlmsghdr *nlh;
2160 int size = nlmsg_msg_size(len);
2161
2162 nlh = skb_put(skb, NLMSG_ALIGN(size));
2163 nlh->nlmsg_type = type;
2164 nlh->nlmsg_len = size;
2165 nlh->nlmsg_flags = flags;
2166 nlh->nlmsg_pid = portid;
2167 nlh->nlmsg_seq = seq;
2168 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2169 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2170 return nlh;
2171 }
2172 EXPORT_SYMBOL(__nlmsg_put);
2173
2174 /*
2175 * It looks a bit ugly.
2176 * It would be better to create kernel thread.
2177 */
2178
2179 static int netlink_dump(struct sock *sk)
2180 {
2181 struct netlink_sock *nlk = nlk_sk(sk);
2182 struct netlink_callback *cb;
2183 struct sk_buff *skb = NULL;
2184 struct nlmsghdr *nlh;
2185 struct module *module;
2186 int err = -ENOBUFS;
2187 int alloc_min_size;
2188 int alloc_size;
2189
2190 mutex_lock(nlk->cb_mutex);
2191 if (!nlk->cb_running) {
2192 err = -EINVAL;
2193 goto errout_skb;
2194 }
2195
2196 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2197 goto errout_skb;
2198
2199 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2200 * required, but it makes sense to _attempt_ a 16K bytes allocation
2201 * to reduce number of system calls on dump operations, if user
2202 * ever provided a big enough buffer.
2203 */
2204 cb = &nlk->cb;
2205 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2206
2207 if (alloc_min_size < nlk->max_recvmsg_len) {
2208 alloc_size = nlk->max_recvmsg_len;
2209 skb = alloc_skb(alloc_size,
2210 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2211 __GFP_NOWARN | __GFP_NORETRY);
2212 }
2213 if (!skb) {
2214 alloc_size = alloc_min_size;
2215 skb = alloc_skb(alloc_size, GFP_KERNEL);
2216 }
2217 if (!skb)
2218 goto errout_skb;
2219
2220 /* Trim skb to allocated size. User is expected to provide buffer as
2221 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2222 * netlink_recvmsg())). dump will pack as many smaller messages as
2223 * could fit within the allocated skb. skb is typically allocated
2224 * with larger space than required (could be as much as near 2x the
2225 * requested size with align to next power of 2 approach). Allowing
2226 * dump to use the excess space makes it difficult for a user to have a
2227 * reasonable static buffer based on the expected largest dump of a
2228 * single netdev. The outcome is MSG_TRUNC error.
2229 */
2230 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2231 netlink_skb_set_owner_r(skb, sk);
2232
2233 if (nlk->dump_done_errno > 0)
2234 nlk->dump_done_errno = cb->dump(skb, cb);
2235
2236 if (nlk->dump_done_errno > 0 ||
2237 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2238 mutex_unlock(nlk->cb_mutex);
2239
2240 if (sk_filter(sk, skb))
2241 kfree_skb(skb);
2242 else
2243 __netlink_sendskb(sk, skb);
2244 return 0;
2245 }
2246
2247 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2248 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2249 if (WARN_ON(!nlh))
2250 goto errout_skb;
2251
2252 nl_dump_check_consistent(cb, nlh);
2253
2254 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2255 sizeof(nlk->dump_done_errno));
2256
2257 if (sk_filter(sk, skb))
2258 kfree_skb(skb);
2259 else
2260 __netlink_sendskb(sk, skb);
2261
2262 if (cb->done)
2263 cb->done(cb);
2264
2265 nlk->cb_running = false;
2266 module = cb->module;
2267 skb = cb->skb;
2268 mutex_unlock(nlk->cb_mutex);
2269 module_put(module);
2270 consume_skb(skb);
2271 return 0;
2272
2273 errout_skb:
2274 mutex_unlock(nlk->cb_mutex);
2275 kfree_skb(skb);
2276 return err;
2277 }
2278
2279 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2280 const struct nlmsghdr *nlh,
2281 struct netlink_dump_control *control)
2282 {
2283 struct netlink_callback *cb;
2284 struct sock *sk;
2285 struct netlink_sock *nlk;
2286 int ret;
2287
2288 refcount_inc(&skb->users);
2289
2290 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2291 if (sk == NULL) {
2292 ret = -ECONNREFUSED;
2293 goto error_free;
2294 }
2295
2296 nlk = nlk_sk(sk);
2297 mutex_lock(nlk->cb_mutex);
2298 /* A dump is in progress... */
2299 if (nlk->cb_running) {
2300 ret = -EBUSY;
2301 goto error_unlock;
2302 }
2303 /* add reference of module which cb->dump belongs to */
2304 if (!try_module_get(control->module)) {
2305 ret = -EPROTONOSUPPORT;
2306 goto error_unlock;
2307 }
2308
2309 cb = &nlk->cb;
2310 memset(cb, 0, sizeof(*cb));
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 (control->start) {
2320 ret = control->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 (u8 *)nlh));
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