]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - net/netlink/af_netlink.c
projects / mirror_ubuntu-hirsute-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
rtnetlink: require unique netns identifier
[mirror_ubuntu-hirsute-kernel.git] / net / netlink / af_netlink.c
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
67 #include <net/net_namespace.h>
68 #include <net/netns/generic.h>
69 #include <net/sock.h>
70 #include <net/scm.h>
71 #include <net/netlink.h>
72
73 #include "af_netlink.h"
74
75 struct listeners {
76 struct rcu_head rcu;
77 unsigned long masks[0];
78 };
79
80 /* state bits */
81 #define NETLINK_S_CONGESTED 0x0
82
83 static inline int netlink_is_kernel(struct sock *sk)
84 {
85 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
86 }
87
88 struct netlink_table *nl_table __read_mostly;
89 EXPORT_SYMBOL_GPL(nl_table);
90
91 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
92
93 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
94
95 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
96 "nlk_cb_mutex-ROUTE",
97 "nlk_cb_mutex-1",
98 "nlk_cb_mutex-USERSOCK",
99 "nlk_cb_mutex-FIREWALL",
100 "nlk_cb_mutex-SOCK_DIAG",
101 "nlk_cb_mutex-NFLOG",
102 "nlk_cb_mutex-XFRM",
103 "nlk_cb_mutex-SELINUX",
104 "nlk_cb_mutex-ISCSI",
105 "nlk_cb_mutex-AUDIT",
106 "nlk_cb_mutex-FIB_LOOKUP",
107 "nlk_cb_mutex-CONNECTOR",
108 "nlk_cb_mutex-NETFILTER",
109 "nlk_cb_mutex-IP6_FW",
110 "nlk_cb_mutex-DNRTMSG",
111 "nlk_cb_mutex-KOBJECT_UEVENT",
112 "nlk_cb_mutex-GENERIC",
113 "nlk_cb_mutex-17",
114 "nlk_cb_mutex-SCSITRANSPORT",
115 "nlk_cb_mutex-ECRYPTFS",
116 "nlk_cb_mutex-RDMA",
117 "nlk_cb_mutex-CRYPTO",
118 "nlk_cb_mutex-SMC",
119 "nlk_cb_mutex-23",
120 "nlk_cb_mutex-24",
121 "nlk_cb_mutex-25",
122 "nlk_cb_mutex-26",
123 "nlk_cb_mutex-27",
124 "nlk_cb_mutex-28",
125 "nlk_cb_mutex-29",
126 "nlk_cb_mutex-30",
127 "nlk_cb_mutex-31",
128 "nlk_cb_mutex-MAX_LINKS"
129 };
130
131 static int netlink_dump(struct sock *sk);
132
133 /* nl_table locking explained:
134 * Lookup and traversal are protected with an RCU read-side lock. Insertion
135 * and removal are protected with per bucket lock while using RCU list
136 * modification primitives and may run in parallel to RCU protected lookups.
137 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
138 * been acquired * either during or after the socket has been removed from
139 * the list and after an RCU grace period.
140 */
141 DEFINE_RWLOCK(nl_table_lock);
142 EXPORT_SYMBOL_GPL(nl_table_lock);
143 static atomic_t nl_table_users = ATOMIC_INIT(0);
144
145 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
146
147 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
148
149
150 static const struct rhashtable_params netlink_rhashtable_params;
151
152 static inline u32 netlink_group_mask(u32 group)
153 {
154 return group ? 1 << (group - 1) : 0;
155 }
156
157 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
158 gfp_t gfp_mask)
159 {
160 unsigned int len = skb_end_offset(skb);
161 struct sk_buff *new;
162
163 new = alloc_skb(len, gfp_mask);
164 if (new == NULL)
165 return NULL;
166
167 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
168 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
169 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
170
171 skb_put_data(new, skb->data, len);
172 return new;
173 }
174
175 static unsigned int netlink_tap_net_id;
176
177 struct netlink_tap_net {
178 struct list_head netlink_tap_all;
179 struct mutex netlink_tap_lock;
180 };
181
182 int netlink_add_tap(struct netlink_tap *nt)
183 {
184 struct net *net = dev_net(nt->dev);
185 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
186
187 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
188 return -EINVAL;
189
190 mutex_lock(&nn->netlink_tap_lock);
191 list_add_rcu(&nt->list, &nn->netlink_tap_all);
192 mutex_unlock(&nn->netlink_tap_lock);
193
194 __module_get(nt->module);
195
196 return 0;
197 }
198 EXPORT_SYMBOL_GPL(netlink_add_tap);
199
200 static int __netlink_remove_tap(struct netlink_tap *nt)
201 {
202 struct net *net = dev_net(nt->dev);
203 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
204 bool found = false;
205 struct netlink_tap *tmp;
206
207 mutex_lock(&nn->netlink_tap_lock);
208
209 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
210 if (nt == tmp) {
211 list_del_rcu(&nt->list);
212 found = true;
213 goto out;
214 }
215 }
216
217 pr_warn("__netlink_remove_tap: %p not found\n", nt);
218 out:
219 mutex_unlock(&nn->netlink_tap_lock);
220
221 if (found)
222 module_put(nt->module);
223
224 return found ? 0 : -ENODEV;
225 }
226
227 int netlink_remove_tap(struct netlink_tap *nt)
228 {
229 int ret;
230
231 ret = __netlink_remove_tap(nt);
232 synchronize_net();
233
234 return ret;
235 }
236 EXPORT_SYMBOL_GPL(netlink_remove_tap);
237
238 static __net_init int netlink_tap_init_net(struct net *net)
239 {
240 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
241
242 INIT_LIST_HEAD(&nn->netlink_tap_all);
243 mutex_init(&nn->netlink_tap_lock);
244 return 0;
245 }
246
247 static void __net_exit netlink_tap_exit_net(struct net *net)
248 {
249 }
250
251 static struct pernet_operations netlink_tap_net_ops = {
252 .init = netlink_tap_init_net,
253 .exit = netlink_tap_exit_net,
254 .id = &netlink_tap_net_id,
255 .size = sizeof(struct netlink_tap_net),
256 };
257
258 static bool netlink_filter_tap(const struct sk_buff *skb)
259 {
260 struct sock *sk = skb->sk;
261
262 /* We take the more conservative approach and
263 * whitelist socket protocols that may pass.
264 */
265 switch (sk->sk_protocol) {
266 case NETLINK_ROUTE:
267 case NETLINK_USERSOCK:
268 case NETLINK_SOCK_DIAG:
269 case NETLINK_NFLOG:
270 case NETLINK_XFRM:
271 case NETLINK_FIB_LOOKUP:
272 case NETLINK_NETFILTER:
273 case NETLINK_GENERIC:
274 return true;
275 }
276
277 return false;
278 }
279
280 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
281 struct net_device *dev)
282 {
283 struct sk_buff *nskb;
284 struct sock *sk = skb->sk;
285 int ret = -ENOMEM;
286
287 if (!net_eq(dev_net(dev), sock_net(sk)))
288 return 0;
289
290 dev_hold(dev);
291
292 if (is_vmalloc_addr(skb->head))
293 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
294 else
295 nskb = skb_clone(skb, GFP_ATOMIC);
296 if (nskb) {
297 nskb->dev = dev;
298 nskb->protocol = htons((u16) sk->sk_protocol);
299 nskb->pkt_type = netlink_is_kernel(sk) ?
300 PACKET_KERNEL : PACKET_USER;
301 skb_reset_network_header(nskb);
302 ret = dev_queue_xmit(nskb);
303 if (unlikely(ret > 0))
304 ret = net_xmit_errno(ret);
305 }
306
307 dev_put(dev);
308 return ret;
309 }
310
311 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
312 {
313 int ret;
314 struct netlink_tap *tmp;
315
316 if (!netlink_filter_tap(skb))
317 return;
318
319 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
320 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
321 if (unlikely(ret))
322 break;
323 }
324 }
325
326 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
327 {
328 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
329
330 rcu_read_lock();
331
332 if (unlikely(!list_empty(&nn->netlink_tap_all)))
333 __netlink_deliver_tap(skb, nn);
334
335 rcu_read_unlock();
336 }
337
338 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
339 struct sk_buff *skb)
340 {
341 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
342 netlink_deliver_tap(sock_net(dst), skb);
343 }
344
345 static void netlink_overrun(struct sock *sk)
346 {
347 struct netlink_sock *nlk = nlk_sk(sk);
348
349 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
350 if (!test_and_set_bit(NETLINK_S_CONGESTED,
351 &nlk_sk(sk)->state)) {
352 sk->sk_err = ENOBUFS;
353 sk->sk_error_report(sk);
354 }
355 }
356 atomic_inc(&sk->sk_drops);
357 }
358
359 static void netlink_rcv_wake(struct sock *sk)
360 {
361 struct netlink_sock *nlk = nlk_sk(sk);
362
363 if (skb_queue_empty(&sk->sk_receive_queue))
364 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
365 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
366 wake_up_interruptible(&nlk->wait);
367 }
368
369 static void netlink_skb_destructor(struct sk_buff *skb)
370 {
371 if (is_vmalloc_addr(skb->head)) {
372 if (!skb->cloned ||
373 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
374 vfree(skb->head);
375
376 skb->head = NULL;
377 }
378 if (skb->sk != NULL)
379 sock_rfree(skb);
380 }
381
382 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
383 {
384 WARN_ON(skb->sk != NULL);
385 skb->sk = sk;
386 skb->destructor = netlink_skb_destructor;
387 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
388 sk_mem_charge(sk, skb->truesize);
389 }
390
391 static void netlink_sock_destruct(struct sock *sk)
392 {
393 struct netlink_sock *nlk = nlk_sk(sk);
394
395 if (nlk->cb_running) {
396 if (nlk->cb.done)
397 nlk->cb.done(&nlk->cb);
398 module_put(nlk->cb.module);
399 kfree_skb(nlk->cb.skb);
400 }
401
402 skb_queue_purge(&sk->sk_receive_queue);
403
404 if (!sock_flag(sk, SOCK_DEAD)) {
405 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
406 return;
407 }
408
409 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
410 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
411 WARN_ON(nlk_sk(sk)->groups);
412 }
413
414 static void netlink_sock_destruct_work(struct work_struct *work)
415 {
416 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
417 work);
418
419 sk_free(&nlk->sk);
420 }
421
422 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
423 * SMP. Look, when several writers sleep and reader wakes them up, all but one
424 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
425 * this, _but_ remember, it adds useless work on UP machines.
426 */
427
428 void netlink_table_grab(void)
429 __acquires(nl_table_lock)
430 {
431 might_sleep();
432
433 write_lock_irq(&nl_table_lock);
434
435 if (atomic_read(&nl_table_users)) {
436 DECLARE_WAITQUEUE(wait, current);
437
438 add_wait_queue_exclusive(&nl_table_wait, &wait);
439 for (;;) {
440 set_current_state(TASK_UNINTERRUPTIBLE);
441 if (atomic_read(&nl_table_users) == 0)
442 break;
443 write_unlock_irq(&nl_table_lock);
444 schedule();
445 write_lock_irq(&nl_table_lock);
446 }
447
448 __set_current_state(TASK_RUNNING);
449 remove_wait_queue(&nl_table_wait, &wait);
450 }
451 }
452
453 void netlink_table_ungrab(void)
454 __releases(nl_table_lock)
455 {
456 write_unlock_irq(&nl_table_lock);
457 wake_up(&nl_table_wait);
458 }
459
460 static inline void
461 netlink_lock_table(void)
462 {
463 /* read_lock() synchronizes us to netlink_table_grab */
464
465 read_lock(&nl_table_lock);
466 atomic_inc(&nl_table_users);
467 read_unlock(&nl_table_lock);
468 }
469
470 static inline void
471 netlink_unlock_table(void)
472 {
473 if (atomic_dec_and_test(&nl_table_users))
474 wake_up(&nl_table_wait);
475 }
476
477 struct netlink_compare_arg
478 {
479 possible_net_t pnet;
480 u32 portid;
481 };
482
483 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
484 #define netlink_compare_arg_len \
485 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
486
487 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
488 const void *ptr)
489 {
490 const struct netlink_compare_arg *x = arg->key;
491 const struct netlink_sock *nlk = ptr;
492
493 return nlk->portid != x->portid ||
494 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
495 }
496
497 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
498 struct net *net, u32 portid)
499 {
500 memset(arg, 0, sizeof(*arg));
501 write_pnet(&arg->pnet, net);
502 arg->portid = portid;
503 }
504
505 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
506 struct net *net)
507 {
508 struct netlink_compare_arg arg;
509
510 netlink_compare_arg_init(&arg, net, portid);
511 return rhashtable_lookup_fast(&table->hash, &arg,
512 netlink_rhashtable_params);
513 }
514
515 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
516 {
517 struct netlink_compare_arg arg;
518
519 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
520 return rhashtable_lookup_insert_key(&table->hash, &arg,
521 &nlk_sk(sk)->node,
522 netlink_rhashtable_params);
523 }
524
525 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
526 {
527 struct netlink_table *table = &nl_table[protocol];
528 struct sock *sk;
529
530 rcu_read_lock();
531 sk = __netlink_lookup(table, portid, net);
532 if (sk)
533 sock_hold(sk);
534 rcu_read_unlock();
535
536 return sk;
537 }
538
539 static const struct proto_ops netlink_ops;
540
541 static void
542 netlink_update_listeners(struct sock *sk)
543 {
544 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
545 unsigned long mask;
546 unsigned int i;
547 struct listeners *listeners;
548
549 listeners = nl_deref_protected(tbl->listeners);
550 if (!listeners)
551 return;
552
553 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
554 mask = 0;
555 sk_for_each_bound(sk, &tbl->mc_list) {
556 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
557 mask |= nlk_sk(sk)->groups[i];
558 }
559 listeners->masks[i] = mask;
560 }
561 /* this function is only called with the netlink table "grabbed", which
562 * makes sure updates are visible before bind or setsockopt return. */
563 }
564
565 static int netlink_insert(struct sock *sk, u32 portid)
566 {
567 struct netlink_table *table = &nl_table[sk->sk_protocol];
568 int err;
569
570 lock_sock(sk);
571
572 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
573 if (nlk_sk(sk)->bound)
574 goto err;
575
576 err = -ENOMEM;
577 if (BITS_PER_LONG > 32 &&
578 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
579 goto err;
580
581 nlk_sk(sk)->portid = portid;
582 sock_hold(sk);
583
584 err = __netlink_insert(table, sk);
585 if (err) {
586 /* In case the hashtable backend returns with -EBUSY
587 * from here, it must not escape to the caller.
588 */
589 if (unlikely(err == -EBUSY))
590 err = -EOVERFLOW;
591 if (err == -EEXIST)
592 err = -EADDRINUSE;
593 sock_put(sk);
594 goto err;
595 }
596
597 /* We need to ensure that the socket is hashed and visible. */
598 smp_wmb();
599 nlk_sk(sk)->bound = portid;
600
601 err:
602 release_sock(sk);
603 return err;
604 }
605
606 static void netlink_remove(struct sock *sk)
607 {
608 struct netlink_table *table;
609
610 table = &nl_table[sk->sk_protocol];
611 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
612 netlink_rhashtable_params)) {
613 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
614 __sock_put(sk);
615 }
616
617 netlink_table_grab();
618 if (nlk_sk(sk)->subscriptions) {
619 __sk_del_bind_node(sk);
620 netlink_update_listeners(sk);
621 }
622 if (sk->sk_protocol == NETLINK_GENERIC)
623 atomic_inc(&genl_sk_destructing_cnt);
624 netlink_table_ungrab();
625 }
626
627 static struct proto netlink_proto = {
628 .name = "NETLINK",
629 .owner = THIS_MODULE,
630 .obj_size = sizeof(struct netlink_sock),
631 };
632
633 static int __netlink_create(struct net *net, struct socket *sock,
634 struct mutex *cb_mutex, int protocol,
635 int kern)
636 {
637 struct sock *sk;
638 struct netlink_sock *nlk;
639
640 sock->ops = &netlink_ops;
641
642 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
643 if (!sk)
644 return -ENOMEM;
645
646 sock_init_data(sock, sk);
647
648 nlk = nlk_sk(sk);
649 if (cb_mutex) {
650 nlk->cb_mutex = cb_mutex;
651 } else {
652 nlk->cb_mutex = &nlk->cb_def_mutex;
653 mutex_init(nlk->cb_mutex);
654 lockdep_set_class_and_name(nlk->cb_mutex,
655 nlk_cb_mutex_keys + protocol,
656 nlk_cb_mutex_key_strings[protocol]);
657 }
658 init_waitqueue_head(&nlk->wait);
659
660 sk->sk_destruct = netlink_sock_destruct;
661 sk->sk_protocol = protocol;
662 return 0;
663 }
664
665 static int netlink_create(struct net *net, struct socket *sock, int protocol,
666 int kern)
667 {
668 struct module *module = NULL;
669 struct mutex *cb_mutex;
670 struct netlink_sock *nlk;
671 int (*bind)(struct net *net, int group);
672 void (*unbind)(struct net *net, int group);
673 int err = 0;
674
675 sock->state = SS_UNCONNECTED;
676
677 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
678 return -ESOCKTNOSUPPORT;
679
680 if (protocol < 0 || protocol >= MAX_LINKS)
681 return -EPROTONOSUPPORT;
682
683 netlink_lock_table();
684 #ifdef CONFIG_MODULES
685 if (!nl_table[protocol].registered) {
686 netlink_unlock_table();
687 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
688 netlink_lock_table();
689 }
690 #endif
691 if (nl_table[protocol].registered &&
692 try_module_get(nl_table[protocol].module))
693 module = nl_table[protocol].module;
694 else
695 err = -EPROTONOSUPPORT;
696 cb_mutex = nl_table[protocol].cb_mutex;
697 bind = nl_table[protocol].bind;
698 unbind = nl_table[protocol].unbind;
699 netlink_unlock_table();
700
701 if (err < 0)
702 goto out;
703
704 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
705 if (err < 0)
706 goto out_module;
707
708 local_bh_disable();
709 sock_prot_inuse_add(net, &netlink_proto, 1);
710 local_bh_enable();
711
712 nlk = nlk_sk(sock->sk);
713 nlk->module = module;
714 nlk->netlink_bind = bind;
715 nlk->netlink_unbind = unbind;
716 out:
717 return err;
718
719 out_module:
720 module_put(module);
721 goto out;
722 }
723
724 static void deferred_put_nlk_sk(struct rcu_head *head)
725 {
726 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
727 struct sock *sk = &nlk->sk;
728
729 kfree(nlk->groups);
730 nlk->groups = NULL;
731
732 if (!refcount_dec_and_test(&sk->sk_refcnt))
733 return;
734
735 if (nlk->cb_running && nlk->cb.done) {
736 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
737 schedule_work(&nlk->work);
738 return;
739 }
740
741 sk_free(sk);
742 }
743
744 static int netlink_release(struct socket *sock)
745 {
746 struct sock *sk = sock->sk;
747 struct netlink_sock *nlk;
748
749 if (!sk)
750 return 0;
751
752 netlink_remove(sk);
753 sock_orphan(sk);
754 nlk = nlk_sk(sk);
755
756 /*
757 * OK. Socket is unlinked, any packets that arrive now
758 * will be purged.
759 */
760
761 /* must not acquire netlink_table_lock in any way again before unbind
762 * and notifying genetlink is done as otherwise it might deadlock
763 */
764 if (nlk->netlink_unbind) {
765 int i;
766
767 for (i = 0; i < nlk->ngroups; i++)
768 if (test_bit(i, nlk->groups))
769 nlk->netlink_unbind(sock_net(sk), i + 1);
770 }
771 if (sk->sk_protocol == NETLINK_GENERIC &&
772 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
773 wake_up(&genl_sk_destructing_waitq);
774
775 sock->sk = NULL;
776 wake_up_interruptible_all(&nlk->wait);
777
778 skb_queue_purge(&sk->sk_write_queue);
779
780 if (nlk->portid && nlk->bound) {
781 struct netlink_notify n = {
782 .net = sock_net(sk),
783 .protocol = sk->sk_protocol,
784 .portid = nlk->portid,
785 };
786 blocking_notifier_call_chain(&netlink_chain,
787 NETLINK_URELEASE, &n);
788 }
789
790 module_put(nlk->module);
791
792 if (netlink_is_kernel(sk)) {
793 netlink_table_grab();
794 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
795 if (--nl_table[sk->sk_protocol].registered == 0) {
796 struct listeners *old;
797
798 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
799 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
800 kfree_rcu(old, rcu);
801 nl_table[sk->sk_protocol].module = NULL;
802 nl_table[sk->sk_protocol].bind = NULL;
803 nl_table[sk->sk_protocol].unbind = NULL;
804 nl_table[sk->sk_protocol].flags = 0;
805 nl_table[sk->sk_protocol].registered = 0;
806 }
807 netlink_table_ungrab();
808 }
809
810 local_bh_disable();
811 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
812 local_bh_enable();
813 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
814 return 0;
815 }
816
817 static int netlink_autobind(struct socket *sock)
818 {
819 struct sock *sk = sock->sk;
820 struct net *net = sock_net(sk);
821 struct netlink_table *table = &nl_table[sk->sk_protocol];
822 s32 portid = task_tgid_vnr(current);
823 int err;
824 s32 rover = -4096;
825 bool ok;
826
827 retry:
828 cond_resched();
829 rcu_read_lock();
830 ok = !__netlink_lookup(table, portid, net);
831 rcu_read_unlock();
832 if (!ok) {
833 /* Bind collision, search negative portid values. */
834 if (rover == -4096)
835 /* rover will be in range [S32_MIN, -4097] */
836 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
837 else if (rover >= -4096)
838 rover = -4097;
839 portid = rover--;
840 goto retry;
841 }
842
843 err = netlink_insert(sk, portid);
844 if (err == -EADDRINUSE)
845 goto retry;
846
847 /* If 2 threads race to autobind, that is fine. */
848 if (err == -EBUSY)
849 err = 0;
850
851 return err;
852 }
853
854 /**
855 * __netlink_ns_capable - General netlink message capability test
856 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
857 * @user_ns: The user namespace of the capability to use
858 * @cap: The capability to use
859 *
860 * Test to see if the opener of the socket we received the message
861 * from had when the netlink socket was created and the sender of the
862 * message has has the capability @cap in the user namespace @user_ns.
863 */
864 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
865 struct user_namespace *user_ns, int cap)
866 {
867 return ((nsp->flags & NETLINK_SKB_DST) ||
868 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
869 ns_capable(user_ns, cap);
870 }
871 EXPORT_SYMBOL(__netlink_ns_capable);
872
873 /**
874 * netlink_ns_capable - General netlink message capability test
875 * @skb: socket buffer holding a netlink command from userspace
876 * @user_ns: The user namespace of the capability to use
877 * @cap: The capability to use
878 *
879 * Test to see if the opener of the socket we received the message
880 * from had when the netlink socket was created and the sender of the
881 * message has has the capability @cap in the user namespace @user_ns.
882 */
883 bool netlink_ns_capable(const struct sk_buff *skb,
884 struct user_namespace *user_ns, int cap)
885 {
886 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
887 }
888 EXPORT_SYMBOL(netlink_ns_capable);
889
890 /**
891 * netlink_capable - Netlink global message capability test
892 * @skb: socket buffer holding a netlink command from userspace
893 * @cap: The capability to use
894 *
895 * Test to see if the opener of the socket we received the message
896 * from had when the netlink socket was created and the sender of the
897 * message has has the capability @cap in all user namespaces.
898 */
899 bool netlink_capable(const struct sk_buff *skb, int cap)
900 {
901 return netlink_ns_capable(skb, &init_user_ns, cap);
902 }
903 EXPORT_SYMBOL(netlink_capable);
904
905 /**
906 * netlink_net_capable - Netlink network namespace message capability test
907 * @skb: socket buffer holding a netlink command from userspace
908 * @cap: The capability to use
909 *
910 * Test to see if the opener of the socket we received the message
911 * from had when the netlink socket was created and the sender of the
912 * message has has the capability @cap over the network namespace of
913 * the socket we received the message from.
914 */
915 bool netlink_net_capable(const struct sk_buff *skb, int cap)
916 {
917 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
918 }
919 EXPORT_SYMBOL(netlink_net_capable);
920
921 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
922 {
923 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
924 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
925 }
926
927 static void
928 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
929 {
930 struct netlink_sock *nlk = nlk_sk(sk);
931
932 if (nlk->subscriptions && !subscriptions)
933 __sk_del_bind_node(sk);
934 else if (!nlk->subscriptions && subscriptions)
935 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
936 nlk->subscriptions = subscriptions;
937 }
938
939 static int netlink_realloc_groups(struct sock *sk)
940 {
941 struct netlink_sock *nlk = nlk_sk(sk);
942 unsigned int groups;
943 unsigned long *new_groups;
944 int err = 0;
945
946 netlink_table_grab();
947
948 groups = nl_table[sk->sk_protocol].groups;
949 if (!nl_table[sk->sk_protocol].registered) {
950 err = -ENOENT;
951 goto out_unlock;
952 }
953
954 if (nlk->ngroups >= groups)
955 goto out_unlock;
956
957 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
958 if (new_groups == NULL) {
959 err = -ENOMEM;
960 goto out_unlock;
961 }
962 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
963 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
964
965 nlk->groups = new_groups;
966 nlk->ngroups = groups;
967 out_unlock:
968 netlink_table_ungrab();
969 return err;
970 }
971
972 static void netlink_undo_bind(int group, long unsigned int groups,
973 struct sock *sk)
974 {
975 struct netlink_sock *nlk = nlk_sk(sk);
976 int undo;
977
978 if (!nlk->netlink_unbind)
979 return;
980
981 for (undo = 0; undo < group; undo++)
982 if (test_bit(undo, &groups))
983 nlk->netlink_unbind(sock_net(sk), undo + 1);
984 }
985
986 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
987 int addr_len)
988 {
989 struct sock *sk = sock->sk;
990 struct net *net = sock_net(sk);
991 struct netlink_sock *nlk = nlk_sk(sk);
992 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
993 int err = 0;
994 long unsigned int groups = nladdr->nl_groups;
995 bool bound;
996
997 if (addr_len < sizeof(struct sockaddr_nl))
998 return -EINVAL;
999
1000 if (nladdr->nl_family != AF_NETLINK)
1001 return -EINVAL;
1002
1003 /* Only superuser is allowed to listen multicasts */
1004 if (groups) {
1005 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1006 return -EPERM;
1007 err = netlink_realloc_groups(sk);
1008 if (err)
1009 return err;
1010 }
1011
1012 bound = nlk->bound;
1013 if (bound) {
1014 /* Ensure nlk->portid is up-to-date. */
1015 smp_rmb();
1016
1017 if (nladdr->nl_pid != nlk->portid)
1018 return -EINVAL;
1019 }
1020
1021 netlink_lock_table();
1022 if (nlk->netlink_bind && groups) {
1023 int group;
1024
1025 for (group = 0; group < nlk->ngroups; group++) {
1026 if (!test_bit(group, &groups))
1027 continue;
1028 err = nlk->netlink_bind(net, group + 1);
1029 if (!err)
1030 continue;
1031 netlink_undo_bind(group, groups, sk);
1032 goto unlock;
1033 }
1034 }
1035
1036 /* No need for barriers here as we return to user-space without
1037 * using any of the bound attributes.
1038 */
1039 if (!bound) {
1040 err = nladdr->nl_pid ?
1041 netlink_insert(sk, nladdr->nl_pid) :
1042 netlink_autobind(sock);
1043 if (err) {
1044 netlink_undo_bind(nlk->ngroups, groups, sk);
1045 goto unlock;
1046 }
1047 }
1048
1049 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1050 goto unlock;
1051 netlink_unlock_table();
1052
1053 netlink_table_grab();
1054 netlink_update_subscriptions(sk, nlk->subscriptions +
1055 hweight32(groups) -
1056 hweight32(nlk->groups[0]));
1057 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1058 netlink_update_listeners(sk);
1059 netlink_table_ungrab();
1060
1061 return 0;
1062
1063 unlock:
1064 netlink_unlock_table();
1065 return err;
1066 }
1067
1068 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1069 int alen, int flags)
1070 {
1071 int err = 0;
1072 struct sock *sk = sock->sk;
1073 struct netlink_sock *nlk = nlk_sk(sk);
1074 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1075
1076 if (alen < sizeof(addr->sa_family))
1077 return -EINVAL;
1078
1079 if (addr->sa_family == AF_UNSPEC) {
1080 sk->sk_state = NETLINK_UNCONNECTED;
1081 nlk->dst_portid = 0;
1082 nlk->dst_group = 0;
1083 return 0;
1084 }
1085 if (addr->sa_family != AF_NETLINK)
1086 return -EINVAL;
1087
1088 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1089 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1090 return -EPERM;
1091
1092 /* No need for barriers here as we return to user-space without
1093 * using any of the bound attributes.
1094 */
1095 if (!nlk->bound)
1096 err = netlink_autobind(sock);
1097
1098 if (err == 0) {
1099 sk->sk_state = NETLINK_CONNECTED;
1100 nlk->dst_portid = nladdr->nl_pid;
1101 nlk->dst_group = ffs(nladdr->nl_groups);
1102 }
1103
1104 return err;
1105 }
1106
1107 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1108 int *addr_len, int peer)
1109 {
1110 struct sock *sk = sock->sk;
1111 struct netlink_sock *nlk = nlk_sk(sk);
1112 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1113
1114 nladdr->nl_family = AF_NETLINK;
1115 nladdr->nl_pad = 0;
1116 *addr_len = sizeof(*nladdr);
1117
1118 if (peer) {
1119 nladdr->nl_pid = nlk->dst_portid;
1120 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1121 } else {
1122 nladdr->nl_pid = nlk->portid;
1123 netlink_lock_table();
1124 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1125 netlink_unlock_table();
1126 }
1127 return 0;
1128 }
1129
1130 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1131 unsigned long arg)
1132 {
1133 /* try to hand this ioctl down to the NIC drivers.
1134 */
1135 return -ENOIOCTLCMD;
1136 }
1137
1138 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1139 {
1140 struct sock *sock;
1141 struct netlink_sock *nlk;
1142
1143 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1144 if (!sock)
1145 return ERR_PTR(-ECONNREFUSED);
1146
1147 /* Don't bother queuing skb if kernel socket has no input function */
1148 nlk = nlk_sk(sock);
1149 if (sock->sk_state == NETLINK_CONNECTED &&
1150 nlk->dst_portid != nlk_sk(ssk)->portid) {
1151 sock_put(sock);
1152 return ERR_PTR(-ECONNREFUSED);
1153 }
1154 return sock;
1155 }
1156
1157 struct sock *netlink_getsockbyfilp(struct file *filp)
1158 {
1159 struct inode *inode = file_inode(filp);
1160 struct sock *sock;
1161
1162 if (!S_ISSOCK(inode->i_mode))
1163 return ERR_PTR(-ENOTSOCK);
1164
1165 sock = SOCKET_I(inode)->sk;
1166 if (sock->sk_family != AF_NETLINK)
1167 return ERR_PTR(-EINVAL);
1168
1169 sock_hold(sock);
1170 return sock;
1171 }
1172
1173 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1174 int broadcast)
1175 {
1176 struct sk_buff *skb;
1177 void *data;
1178
1179 if (size <= NLMSG_GOODSIZE || broadcast)
1180 return alloc_skb(size, GFP_KERNEL);
1181
1182 size = SKB_DATA_ALIGN(size) +
1183 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1184
1185 data = vmalloc(size);
1186 if (data == NULL)
1187 return NULL;
1188
1189 skb = __build_skb(data, size);
1190 if (skb == NULL)
1191 vfree(data);
1192 else
1193 skb->destructor = netlink_skb_destructor;
1194
1195 return skb;
1196 }
1197
1198 /*
1199 * Attach a skb to a netlink socket.
1200 * The caller must hold a reference to the destination socket. On error, the
1201 * reference is dropped. The skb is not send to the destination, just all
1202 * all error checks are performed and memory in the queue is reserved.
1203 * Return values:
1204 * < 0: error. skb freed, reference to sock dropped.
1205 * 0: continue
1206 * 1: repeat lookup - reference dropped while waiting for socket memory.
1207 */
1208 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1209 long *timeo, struct sock *ssk)
1210 {
1211 struct netlink_sock *nlk;
1212
1213 nlk = nlk_sk(sk);
1214
1215 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1216 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1217 DECLARE_WAITQUEUE(wait, current);
1218 if (!*timeo) {
1219 if (!ssk || netlink_is_kernel(ssk))
1220 netlink_overrun(sk);
1221 sock_put(sk);
1222 kfree_skb(skb);
1223 return -EAGAIN;
1224 }
1225
1226 __set_current_state(TASK_INTERRUPTIBLE);
1227 add_wait_queue(&nlk->wait, &wait);
1228
1229 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1230 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1231 !sock_flag(sk, SOCK_DEAD))
1232 *timeo = schedule_timeout(*timeo);
1233
1234 __set_current_state(TASK_RUNNING);
1235 remove_wait_queue(&nlk->wait, &wait);
1236 sock_put(sk);
1237
1238 if (signal_pending(current)) {
1239 kfree_skb(skb);
1240 return sock_intr_errno(*timeo);
1241 }
1242 return 1;
1243 }
1244 netlink_skb_set_owner_r(skb, sk);
1245 return 0;
1246 }
1247
1248 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1249 {
1250 int len = skb->len;
1251
1252 netlink_deliver_tap(sock_net(sk), skb);
1253
1254 skb_queue_tail(&sk->sk_receive_queue, skb);
1255 sk->sk_data_ready(sk);
1256 return len;
1257 }
1258
1259 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1260 {
1261 int len = __netlink_sendskb(sk, skb);
1262
1263 sock_put(sk);
1264 return len;
1265 }
1266
1267 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1268 {
1269 kfree_skb(skb);
1270 sock_put(sk);
1271 }
1272
1273 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1274 {
1275 int delta;
1276
1277 WARN_ON(skb->sk != NULL);
1278 delta = skb->end - skb->tail;
1279 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1280 return skb;
1281
1282 if (skb_shared(skb)) {
1283 struct sk_buff *nskb = skb_clone(skb, allocation);
1284 if (!nskb)
1285 return skb;
1286 consume_skb(skb);
1287 skb = nskb;
1288 }
1289
1290 pskb_expand_head(skb, 0, -delta,
1291 (allocation & ~__GFP_DIRECT_RECLAIM) |
1292 __GFP_NOWARN | __GFP_NORETRY);
1293 return skb;
1294 }
1295
1296 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1297 struct sock *ssk)
1298 {
1299 int ret;
1300 struct netlink_sock *nlk = nlk_sk(sk);
1301
1302 ret = -ECONNREFUSED;
1303 if (nlk->netlink_rcv != NULL) {
1304 ret = skb->len;
1305 netlink_skb_set_owner_r(skb, sk);
1306 NETLINK_CB(skb).sk = ssk;
1307 netlink_deliver_tap_kernel(sk, ssk, skb);
1308 nlk->netlink_rcv(skb);
1309 consume_skb(skb);
1310 } else {
1311 kfree_skb(skb);
1312 }
1313 sock_put(sk);
1314 return ret;
1315 }
1316
1317 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1318 u32 portid, int nonblock)
1319 {
1320 struct sock *sk;
1321 int err;
1322 long timeo;
1323
1324 skb = netlink_trim(skb, gfp_any());
1325
1326 timeo = sock_sndtimeo(ssk, nonblock);
1327 retry:
1328 sk = netlink_getsockbyportid(ssk, portid);
1329 if (IS_ERR(sk)) {
1330 kfree_skb(skb);
1331 return PTR_ERR(sk);
1332 }
1333 if (netlink_is_kernel(sk))
1334 return netlink_unicast_kernel(sk, skb, ssk);
1335
1336 if (sk_filter(sk, skb)) {
1337 err = skb->len;
1338 kfree_skb(skb);
1339 sock_put(sk);
1340 return err;
1341 }
1342
1343 err = netlink_attachskb(sk, skb, &timeo, ssk);
1344 if (err == 1)
1345 goto retry;
1346 if (err)
1347 return err;
1348
1349 return netlink_sendskb(sk, skb);
1350 }
1351 EXPORT_SYMBOL(netlink_unicast);
1352
1353 int netlink_has_listeners(struct sock *sk, unsigned int group)
1354 {
1355 int res = 0;
1356 struct listeners *listeners;
1357
1358 BUG_ON(!netlink_is_kernel(sk));
1359
1360 rcu_read_lock();
1361 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1362
1363 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1364 res = test_bit(group - 1, listeners->masks);
1365
1366 rcu_read_unlock();
1367
1368 return res;
1369 }
1370 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1371
1372 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1373 {
1374 struct netlink_sock *nlk = nlk_sk(sk);
1375
1376 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1377 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1378 netlink_skb_set_owner_r(skb, sk);
1379 __netlink_sendskb(sk, skb);
1380 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1381 }
1382 return -1;
1383 }
1384
1385 struct netlink_broadcast_data {
1386 struct sock *exclude_sk;
1387 struct net *net;
1388 u32 portid;
1389 u32 group;
1390 int failure;
1391 int delivery_failure;
1392 int congested;
1393 int delivered;
1394 gfp_t allocation;
1395 struct sk_buff *skb, *skb2;
1396 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1397 void *tx_data;
1398 };
1399
1400 static void do_one_broadcast(struct sock *sk,
1401 struct netlink_broadcast_data *p)
1402 {
1403 struct netlink_sock *nlk = nlk_sk(sk);
1404 int val;
1405
1406 if (p->exclude_sk == sk)
1407 return;
1408
1409 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1410 !test_bit(p->group - 1, nlk->groups))
1411 return;
1412
1413 if (!net_eq(sock_net(sk), p->net)) {
1414 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1415 return;
1416
1417 if (!peernet_has_id(sock_net(sk), p->net))
1418 return;
1419
1420 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1421 CAP_NET_BROADCAST))
1422 return;
1423 }
1424
1425 if (p->failure) {
1426 netlink_overrun(sk);
1427 return;
1428 }
1429
1430 sock_hold(sk);
1431 if (p->skb2 == NULL) {
1432 if (skb_shared(p->skb)) {
1433 p->skb2 = skb_clone(p->skb, p->allocation);
1434 } else {
1435 p->skb2 = skb_get(p->skb);
1436 /*
1437 * skb ownership may have been set when
1438 * delivered to a previous socket.
1439 */
1440 skb_orphan(p->skb2);
1441 }
1442 }
1443 if (p->skb2 == NULL) {
1444 netlink_overrun(sk);
1445 /* Clone failed. Notify ALL listeners. */
1446 p->failure = 1;
1447 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1448 p->delivery_failure = 1;
1449 goto out;
1450 }
1451 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1452 kfree_skb(p->skb2);
1453 p->skb2 = NULL;
1454 goto out;
1455 }
1456 if (sk_filter(sk, p->skb2)) {
1457 kfree_skb(p->skb2);
1458 p->skb2 = NULL;
1459 goto out;
1460 }
1461 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1462 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1463 NETLINK_CB(p->skb2).nsid_is_set = true;
1464 val = netlink_broadcast_deliver(sk, p->skb2);
1465 if (val < 0) {
1466 netlink_overrun(sk);
1467 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1468 p->delivery_failure = 1;
1469 } else {
1470 p->congested |= val;
1471 p->delivered = 1;
1472 p->skb2 = NULL;
1473 }
1474 out:
1475 sock_put(sk);
1476 }
1477
1478 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1479 u32 group, gfp_t allocation,
1480 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1481 void *filter_data)
1482 {
1483 struct net *net = sock_net(ssk);
1484 struct netlink_broadcast_data info;
1485 struct sock *sk;
1486
1487 skb = netlink_trim(skb, allocation);
1488
1489 info.exclude_sk = ssk;
1490 info.net = net;
1491 info.portid = portid;
1492 info.group = group;
1493 info.failure = 0;
1494 info.delivery_failure = 0;
1495 info.congested = 0;
1496 info.delivered = 0;
1497 info.allocation = allocation;
1498 info.skb = skb;
1499 info.skb2 = NULL;
1500 info.tx_filter = filter;
1501 info.tx_data = filter_data;
1502
1503 /* While we sleep in clone, do not allow to change socket list */
1504
1505 netlink_lock_table();
1506
1507 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1508 do_one_broadcast(sk, &info);
1509
1510 consume_skb(skb);
1511
1512 netlink_unlock_table();
1513
1514 if (info.delivery_failure) {
1515 kfree_skb(info.skb2);
1516 return -ENOBUFS;
1517 }
1518 consume_skb(info.skb2);
1519
1520 if (info.delivered) {
1521 if (info.congested && gfpflags_allow_blocking(allocation))
1522 yield();
1523 return 0;
1524 }
1525 return -ESRCH;
1526 }
1527 EXPORT_SYMBOL(netlink_broadcast_filtered);
1528
1529 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1530 u32 group, gfp_t allocation)
1531 {
1532 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1533 NULL, NULL);
1534 }
1535 EXPORT_SYMBOL(netlink_broadcast);
1536
1537 struct netlink_set_err_data {
1538 struct sock *exclude_sk;
1539 u32 portid;
1540 u32 group;
1541 int code;
1542 };
1543
1544 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1545 {
1546 struct netlink_sock *nlk = nlk_sk(sk);
1547 int ret = 0;
1548
1549 if (sk == p->exclude_sk)
1550 goto out;
1551
1552 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1553 goto out;
1554
1555 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1556 !test_bit(p->group - 1, nlk->groups))
1557 goto out;
1558
1559 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1560 ret = 1;
1561 goto out;
1562 }
1563
1564 sk->sk_err = p->code;
1565 sk->sk_error_report(sk);
1566 out:
1567 return ret;
1568 }
1569
1570 /**
1571 * netlink_set_err - report error to broadcast listeners
1572 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1573 * @portid: the PORTID of a process that we want to skip (if any)
1574 * @group: the broadcast group that will notice the error
1575 * @code: error code, must be negative (as usual in kernelspace)
1576 *
1577 * This function returns the number of broadcast listeners that have set the
1578 * NETLINK_NO_ENOBUFS socket option.
1579 */
1580 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1581 {
1582 struct netlink_set_err_data info;
1583 struct sock *sk;
1584 int ret = 0;
1585
1586 info.exclude_sk = ssk;
1587 info.portid = portid;
1588 info.group = group;
1589 /* sk->sk_err wants a positive error value */
1590 info.code = -code;
1591
1592 read_lock(&nl_table_lock);
1593
1594 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1595 ret += do_one_set_err(sk, &info);
1596
1597 read_unlock(&nl_table_lock);
1598 return ret;
1599 }
1600 EXPORT_SYMBOL(netlink_set_err);
1601
1602 /* must be called with netlink table grabbed */
1603 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1604 unsigned int group,
1605 int is_new)
1606 {
1607 int old, new = !!is_new, subscriptions;
1608
1609 old = test_bit(group - 1, nlk->groups);
1610 subscriptions = nlk->subscriptions - old + new;
1611 if (new)
1612 __set_bit(group - 1, nlk->groups);
1613 else
1614 __clear_bit(group - 1, nlk->groups);
1615 netlink_update_subscriptions(&nlk->sk, subscriptions);
1616 netlink_update_listeners(&nlk->sk);
1617 }
1618
1619 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1620 char __user *optval, unsigned int optlen)
1621 {
1622 struct sock *sk = sock->sk;
1623 struct netlink_sock *nlk = nlk_sk(sk);
1624 unsigned int val = 0;
1625 int err;
1626
1627 if (level != SOL_NETLINK)
1628 return -ENOPROTOOPT;
1629
1630 if (optlen >= sizeof(int) &&
1631 get_user(val, (unsigned int __user *)optval))
1632 return -EFAULT;
1633
1634 switch (optname) {
1635 case NETLINK_PKTINFO:
1636 if (val)
1637 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1638 else
1639 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1640 err = 0;
1641 break;
1642 case NETLINK_ADD_MEMBERSHIP:
1643 case NETLINK_DROP_MEMBERSHIP: {
1644 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1645 return -EPERM;
1646 err = netlink_realloc_groups(sk);
1647 if (err)
1648 return err;
1649 if (!val || val - 1 >= nlk->ngroups)
1650 return -EINVAL;
1651 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1652 err = nlk->netlink_bind(sock_net(sk), val);
1653 if (err)
1654 return err;
1655 }
1656 netlink_table_grab();
1657 netlink_update_socket_mc(nlk, val,
1658 optname == NETLINK_ADD_MEMBERSHIP);
1659 netlink_table_ungrab();
1660 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1661 nlk->netlink_unbind(sock_net(sk), val);
1662
1663 err = 0;
1664 break;
1665 }
1666 case NETLINK_BROADCAST_ERROR:
1667 if (val)
1668 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1669 else
1670 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1671 err = 0;
1672 break;
1673 case NETLINK_NO_ENOBUFS:
1674 if (val) {
1675 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1676 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1677 wake_up_interruptible(&nlk->wait);
1678 } else {
1679 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1680 }
1681 err = 0;
1682 break;
1683 case NETLINK_LISTEN_ALL_NSID:
1684 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1685 return -EPERM;
1686
1687 if (val)
1688 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1689 else
1690 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1691 err = 0;
1692 break;
1693 case NETLINK_CAP_ACK:
1694 if (val)
1695 nlk->flags |= NETLINK_F_CAP_ACK;
1696 else
1697 nlk->flags &= ~NETLINK_F_CAP_ACK;
1698 err = 0;
1699 break;
1700 case NETLINK_EXT_ACK:
1701 if (val)
1702 nlk->flags |= NETLINK_F_EXT_ACK;
1703 else
1704 nlk->flags &= ~NETLINK_F_EXT_ACK;
1705 err = 0;
1706 break;
1707 default:
1708 err = -ENOPROTOOPT;
1709 }
1710 return err;
1711 }
1712
1713 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1714 char __user *optval, int __user *optlen)
1715 {
1716 struct sock *sk = sock->sk;
1717 struct netlink_sock *nlk = nlk_sk(sk);
1718 int len, val, err;
1719
1720 if (level != SOL_NETLINK)
1721 return -ENOPROTOOPT;
1722
1723 if (get_user(len, optlen))
1724 return -EFAULT;
1725 if (len < 0)
1726 return -EINVAL;
1727
1728 switch (optname) {
1729 case NETLINK_PKTINFO:
1730 if (len < sizeof(int))
1731 return -EINVAL;
1732 len = sizeof(int);
1733 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1734 if (put_user(len, optlen) ||
1735 put_user(val, optval))
1736 return -EFAULT;
1737 err = 0;
1738 break;
1739 case NETLINK_BROADCAST_ERROR:
1740 if (len < sizeof(int))
1741 return -EINVAL;
1742 len = sizeof(int);
1743 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1744 if (put_user(len, optlen) ||
1745 put_user(val, optval))
1746 return -EFAULT;
1747 err = 0;
1748 break;
1749 case NETLINK_NO_ENOBUFS:
1750 if (len < sizeof(int))
1751 return -EINVAL;
1752 len = sizeof(int);
1753 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1754 if (put_user(len, optlen) ||
1755 put_user(val, optval))
1756 return -EFAULT;
1757 err = 0;
1758 break;
1759 case NETLINK_LIST_MEMBERSHIPS: {
1760 int pos, idx, shift;
1761
1762 err = 0;
1763 netlink_lock_table();
1764 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1765 if (len - pos < sizeof(u32))
1766 break;
1767
1768 idx = pos / sizeof(unsigned long);
1769 shift = (pos % sizeof(unsigned long)) * 8;
1770 if (put_user((u32)(nlk->groups[idx] >> shift),
1771 (u32 __user *)(optval + pos))) {
1772 err = -EFAULT;
1773 break;
1774 }
1775 }
1776 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1777 err = -EFAULT;
1778 netlink_unlock_table();
1779 break;
1780 }
1781 case NETLINK_CAP_ACK:
1782 if (len < sizeof(int))
1783 return -EINVAL;
1784 len = sizeof(int);
1785 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1786 if (put_user(len, optlen) ||
1787 put_user(val, optval))
1788 return -EFAULT;
1789 err = 0;
1790 break;
1791 case NETLINK_EXT_ACK:
1792 if (len < sizeof(int))
1793 return -EINVAL;
1794 len = sizeof(int);
1795 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1796 if (put_user(len, optlen) || put_user(val, optval))
1797 return -EFAULT;
1798 err = 0;
1799 break;
1800 default:
1801 err = -ENOPROTOOPT;
1802 }
1803 return err;
1804 }
1805
1806 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1807 {
1808 struct nl_pktinfo info;
1809
1810 info.group = NETLINK_CB(skb).dst_group;
1811 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1812 }
1813
1814 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1815 struct sk_buff *skb)
1816 {
1817 if (!NETLINK_CB(skb).nsid_is_set)
1818 return;
1819
1820 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1821 &NETLINK_CB(skb).nsid);
1822 }
1823
1824 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1825 {
1826 struct sock *sk = sock->sk;
1827 struct netlink_sock *nlk = nlk_sk(sk);
1828 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1829 u32 dst_portid;
1830 u32 dst_group;
1831 struct sk_buff *skb;
1832 int err;
1833 struct scm_cookie scm;
1834 u32 netlink_skb_flags = 0;
1835
1836 if (msg->msg_flags&MSG_OOB)
1837 return -EOPNOTSUPP;
1838
1839 err = scm_send(sock, msg, &scm, true);
1840 if (err < 0)
1841 return err;
1842
1843 if (msg->msg_namelen) {
1844 err = -EINVAL;
1845 if (addr->nl_family != AF_NETLINK)
1846 goto out;
1847 dst_portid = addr->nl_pid;
1848 dst_group = ffs(addr->nl_groups);
1849 err = -EPERM;
1850 if ((dst_group || dst_portid) &&
1851 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1852 goto out;
1853 netlink_skb_flags |= NETLINK_SKB_DST;
1854 } else {
1855 dst_portid = nlk->dst_portid;
1856 dst_group = nlk->dst_group;
1857 }
1858
1859 if (!nlk->bound) {
1860 err = netlink_autobind(sock);
1861 if (err)
1862 goto out;
1863 } else {
1864 /* Ensure nlk is hashed and visible. */
1865 smp_rmb();
1866 }
1867
1868 err = -EMSGSIZE;
1869 if (len > sk->sk_sndbuf - 32)
1870 goto out;
1871 err = -ENOBUFS;
1872 skb = netlink_alloc_large_skb(len, dst_group);
1873 if (skb == NULL)
1874 goto out;
1875
1876 NETLINK_CB(skb).portid = nlk->portid;
1877 NETLINK_CB(skb).dst_group = dst_group;
1878 NETLINK_CB(skb).creds = scm.creds;
1879 NETLINK_CB(skb).flags = netlink_skb_flags;
1880
1881 err = -EFAULT;
1882 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1883 kfree_skb(skb);
1884 goto out;
1885 }
1886
1887 err = security_netlink_send(sk, skb);
1888 if (err) {
1889 kfree_skb(skb);
1890 goto out;
1891 }
1892
1893 if (dst_group) {
1894 refcount_inc(&skb->users);
1895 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1896 }
1897 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1898
1899 out:
1900 scm_destroy(&scm);
1901 return err;
1902 }
1903
1904 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1905 int flags)
1906 {
1907 struct scm_cookie scm;
1908 struct sock *sk = sock->sk;
1909 struct netlink_sock *nlk = nlk_sk(sk);
1910 int noblock = flags&MSG_DONTWAIT;
1911 size_t copied;
1912 struct sk_buff *skb, *data_skb;
1913 int err, ret;
1914
1915 if (flags&MSG_OOB)
1916 return -EOPNOTSUPP;
1917
1918 copied = 0;
1919
1920 skb = skb_recv_datagram(sk, flags, noblock, &err);
1921 if (skb == NULL)
1922 goto out;
1923
1924 data_skb = skb;
1925
1926 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1927 if (unlikely(skb_shinfo(skb)->frag_list)) {
1928 /*
1929 * If this skb has a frag_list, then here that means that we
1930 * will have to use the frag_list skb's data for compat tasks
1931 * and the regular skb's data for normal (non-compat) tasks.
1932 *
1933 * If we need to send the compat skb, assign it to the
1934 * 'data_skb' variable so that it will be used below for data
1935 * copying. We keep 'skb' for everything else, including
1936 * freeing both later.
1937 */
1938 if (flags & MSG_CMSG_COMPAT)
1939 data_skb = skb_shinfo(skb)->frag_list;
1940 }
1941 #endif
1942
1943 /* Record the max length of recvmsg() calls for future allocations */
1944 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1945 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1946 SKB_WITH_OVERHEAD(32768));
1947
1948 copied = data_skb->len;
1949 if (len < copied) {
1950 msg->msg_flags |= MSG_TRUNC;
1951 copied = len;
1952 }
1953
1954 skb_reset_transport_header(data_skb);
1955 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1956
1957 if (msg->msg_name) {
1958 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1959 addr->nl_family = AF_NETLINK;
1960 addr->nl_pad = 0;
1961 addr->nl_pid = NETLINK_CB(skb).portid;
1962 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1963 msg->msg_namelen = sizeof(*addr);
1964 }
1965
1966 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1967 netlink_cmsg_recv_pktinfo(msg, skb);
1968 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1969 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1970
1971 memset(&scm, 0, sizeof(scm));
1972 scm.creds = *NETLINK_CREDS(skb);
1973 if (flags & MSG_TRUNC)
1974 copied = data_skb->len;
1975
1976 skb_free_datagram(sk, skb);
1977
1978 if (nlk->cb_running &&
1979 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1980 ret = netlink_dump(sk);
1981 if (ret) {
1982 sk->sk_err = -ret;
1983 sk->sk_error_report(sk);
1984 }
1985 }
1986
1987 scm_recv(sock, msg, &scm, flags);
1988 out:
1989 netlink_rcv_wake(sk);
1990 return err ? : copied;
1991 }
1992
1993 static void netlink_data_ready(struct sock *sk)
1994 {
1995 BUG();
1996 }
1997
1998 /*
1999 * We export these functions to other modules. They provide a
2000 * complete set of kernel non-blocking support for message
2001 * queueing.
2002 */
2003
2004 struct sock *
2005 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2006 struct netlink_kernel_cfg *cfg)
2007 {
2008 struct socket *sock;
2009 struct sock *sk;
2010 struct netlink_sock *nlk;
2011 struct listeners *listeners = NULL;
2012 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2013 unsigned int groups;
2014
2015 BUG_ON(!nl_table);
2016
2017 if (unit < 0 || unit >= MAX_LINKS)
2018 return NULL;
2019
2020 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2021 return NULL;
2022
2023 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2024 goto out_sock_release_nosk;
2025
2026 sk = sock->sk;
2027
2028 if (!cfg || cfg->groups < 32)
2029 groups = 32;
2030 else
2031 groups = cfg->groups;
2032
2033 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2034 if (!listeners)
2035 goto out_sock_release;
2036
2037 sk->sk_data_ready = netlink_data_ready;
2038 if (cfg && cfg->input)
2039 nlk_sk(sk)->netlink_rcv = cfg->input;
2040
2041 if (netlink_insert(sk, 0))
2042 goto out_sock_release;
2043
2044 nlk = nlk_sk(sk);
2045 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2046
2047 netlink_table_grab();
2048 if (!nl_table[unit].registered) {
2049 nl_table[unit].groups = groups;
2050 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2051 nl_table[unit].cb_mutex = cb_mutex;
2052 nl_table[unit].module = module;
2053 if (cfg) {
2054 nl_table[unit].bind = cfg->bind;
2055 nl_table[unit].unbind = cfg->unbind;
2056 nl_table[unit].flags = cfg->flags;
2057 if (cfg->compare)
2058 nl_table[unit].compare = cfg->compare;
2059 }
2060 nl_table[unit].registered = 1;
2061 } else {
2062 kfree(listeners);
2063 nl_table[unit].registered++;
2064 }
2065 netlink_table_ungrab();
2066 return sk;
2067
2068 out_sock_release:
2069 kfree(listeners);
2070 netlink_kernel_release(sk);
2071 return NULL;
2072
2073 out_sock_release_nosk:
2074 sock_release(sock);
2075 return NULL;
2076 }
2077 EXPORT_SYMBOL(__netlink_kernel_create);
2078
2079 void
2080 netlink_kernel_release(struct sock *sk)
2081 {
2082 if (sk == NULL || sk->sk_socket == NULL)
2083 return;
2084
2085 sock_release(sk->sk_socket);
2086 }
2087 EXPORT_SYMBOL(netlink_kernel_release);
2088
2089 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2090 {
2091 struct listeners *new, *old;
2092 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2093
2094 if (groups < 32)
2095 groups = 32;
2096
2097 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2098 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2099 if (!new)
2100 return -ENOMEM;
2101 old = nl_deref_protected(tbl->listeners);
2102 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2103 rcu_assign_pointer(tbl->listeners, new);
2104
2105 kfree_rcu(old, rcu);
2106 }
2107 tbl->groups = groups;
2108
2109 return 0;
2110 }
2111
2112 /**
2113 * netlink_change_ngroups - change number of multicast groups
2114 *
2115 * This changes the number of multicast groups that are available
2116 * on a certain netlink family. Note that it is not possible to
2117 * change the number of groups to below 32. Also note that it does
2118 * not implicitly call netlink_clear_multicast_users() when the
2119 * number of groups is reduced.
2120 *
2121 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2122 * @groups: The new number of groups.
2123 */
2124 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2125 {
2126 int err;
2127
2128 netlink_table_grab();
2129 err = __netlink_change_ngroups(sk, groups);
2130 netlink_table_ungrab();
2131
2132 return err;
2133 }
2134
2135 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2136 {
2137 struct sock *sk;
2138 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2139
2140 sk_for_each_bound(sk, &tbl->mc_list)
2141 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2142 }
2143
2144 struct nlmsghdr *
2145 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2146 {
2147 struct nlmsghdr *nlh;
2148 int size = nlmsg_msg_size(len);
2149
2150 nlh = skb_put(skb, NLMSG_ALIGN(size));
2151 nlh->nlmsg_type = type;
2152 nlh->nlmsg_len = size;
2153 nlh->nlmsg_flags = flags;
2154 nlh->nlmsg_pid = portid;
2155 nlh->nlmsg_seq = seq;
2156 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2157 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2158 return nlh;
2159 }
2160 EXPORT_SYMBOL(__nlmsg_put);
2161
2162 /*
2163 * It looks a bit ugly.
2164 * It would be better to create kernel thread.
2165 */
2166
2167 static int netlink_dump(struct sock *sk)
2168 {
2169 struct netlink_sock *nlk = nlk_sk(sk);
2170 struct netlink_callback *cb;
2171 struct sk_buff *skb = NULL;
2172 struct nlmsghdr *nlh;
2173 struct module *module;
2174 int err = -ENOBUFS;
2175 int alloc_min_size;
2176 int alloc_size;
2177
2178 mutex_lock(nlk->cb_mutex);
2179 if (!nlk->cb_running) {
2180 err = -EINVAL;
2181 goto errout_skb;
2182 }
2183
2184 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2185 goto errout_skb;
2186
2187 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2188 * required, but it makes sense to _attempt_ a 16K bytes allocation
2189 * to reduce number of system calls on dump operations, if user
2190 * ever provided a big enough buffer.
2191 */
2192 cb = &nlk->cb;
2193 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2194
2195 if (alloc_min_size < nlk->max_recvmsg_len) {
2196 alloc_size = nlk->max_recvmsg_len;
2197 skb = alloc_skb(alloc_size,
2198 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2199 __GFP_NOWARN | __GFP_NORETRY);
2200 }
2201 if (!skb) {
2202 alloc_size = alloc_min_size;
2203 skb = alloc_skb(alloc_size, GFP_KERNEL);
2204 }
2205 if (!skb)
2206 goto errout_skb;
2207
2208 /* Trim skb to allocated size. User is expected to provide buffer as
2209 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2210 * netlink_recvmsg())). dump will pack as many smaller messages as
2211 * could fit within the allocated skb. skb is typically allocated
2212 * with larger space than required (could be as much as near 2x the
2213 * requested size with align to next power of 2 approach). Allowing
2214 * dump to use the excess space makes it difficult for a user to have a
2215 * reasonable static buffer based on the expected largest dump of a
2216 * single netdev. The outcome is MSG_TRUNC error.
2217 */
2218 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2219 netlink_skb_set_owner_r(skb, sk);
2220
2221 if (nlk->dump_done_errno > 0)
2222 nlk->dump_done_errno = cb->dump(skb, cb);
2223
2224 if (nlk->dump_done_errno > 0 ||
2225 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2226 mutex_unlock(nlk->cb_mutex);
2227
2228 if (sk_filter(sk, skb))
2229 kfree_skb(skb);
2230 else
2231 __netlink_sendskb(sk, skb);
2232 return 0;
2233 }
2234
2235 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2236 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2237 if (WARN_ON(!nlh))
2238 goto errout_skb;
2239
2240 nl_dump_check_consistent(cb, nlh);
2241
2242 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2243 sizeof(nlk->dump_done_errno));
2244
2245 if (sk_filter(sk, skb))
2246 kfree_skb(skb);
2247 else
2248 __netlink_sendskb(sk, skb);
2249
2250 if (cb->done)
2251 cb->done(cb);
2252
2253 nlk->cb_running = false;
2254 module = cb->module;
2255 skb = cb->skb;
2256 mutex_unlock(nlk->cb_mutex);
2257 module_put(module);
2258 consume_skb(skb);
2259 return 0;
2260
2261 errout_skb:
2262 mutex_unlock(nlk->cb_mutex);
2263 kfree_skb(skb);
2264 return err;
2265 }
2266
2267 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2268 const struct nlmsghdr *nlh,
2269 struct netlink_dump_control *control)
2270 {
2271 struct netlink_callback *cb;
2272 struct sock *sk;
2273 struct netlink_sock *nlk;
2274 int ret;
2275
2276 refcount_inc(&skb->users);
2277
2278 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2279 if (sk == NULL) {
2280 ret = -ECONNREFUSED;
2281 goto error_free;
2282 }
2283
2284 nlk = nlk_sk(sk);
2285 mutex_lock(nlk->cb_mutex);
2286 /* A dump is in progress... */
2287 if (nlk->cb_running) {
2288 ret = -EBUSY;
2289 goto error_unlock;
2290 }
2291 /* add reference of module which cb->dump belongs to */
2292 if (!try_module_get(control->module)) {
2293 ret = -EPROTONOSUPPORT;
2294 goto error_unlock;
2295 }
2296
2297 cb = &nlk->cb;
2298 memset(cb, 0, sizeof(*cb));
2299 cb->start = control->start;
2300 cb->dump = control->dump;
2301 cb->done = control->done;
2302 cb->nlh = nlh;
2303 cb->data = control->data;
2304 cb->module = control->module;
2305 cb->min_dump_alloc = control->min_dump_alloc;
2306 cb->skb = skb;
2307
2308 if (cb->start) {
2309 ret = cb->start(cb);
2310 if (ret)
2311 goto error_unlock;
2312 }
2313
2314 nlk->cb_running = true;
2315 nlk->dump_done_errno = INT_MAX;
2316
2317 mutex_unlock(nlk->cb_mutex);
2318
2319 ret = netlink_dump(sk);
2320
2321 sock_put(sk);
2322
2323 if (ret)
2324 return ret;
2325
2326 /* We successfully started a dump, by returning -EINTR we
2327 * signal not to send ACK even if it was requested.
2328 */
2329 return -EINTR;
2330
2331 error_unlock:
2332 sock_put(sk);
2333 mutex_unlock(nlk->cb_mutex);
2334 error_free:
2335 kfree_skb(skb);
2336 return ret;
2337 }
2338 EXPORT_SYMBOL(__netlink_dump_start);
2339
2340 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2341 const struct netlink_ext_ack *extack)
2342 {
2343 struct sk_buff *skb;
2344 struct nlmsghdr *rep;
2345 struct nlmsgerr *errmsg;
2346 size_t payload = sizeof(*errmsg);
2347 size_t tlvlen = 0;
2348 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2349 unsigned int flags = 0;
2350 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2351
2352 /* Error messages get the original request appened, unless the user
2353 * requests to cap the error message, and get extra error data if
2354 * requested.
2355 */
2356 if (nlk_has_extack && extack && extack->_msg)
2357 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2358
2359 if (err) {
2360 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2361 payload += nlmsg_len(nlh);
2362 else
2363 flags |= NLM_F_CAPPED;
2364 if (nlk_has_extack && extack && extack->bad_attr)
2365 tlvlen += nla_total_size(sizeof(u32));
2366 } else {
2367 flags |= NLM_F_CAPPED;
2368
2369 if (nlk_has_extack && extack && extack->cookie_len)
2370 tlvlen += nla_total_size(extack->cookie_len);
2371 }
2372
2373 if (tlvlen)
2374 flags |= NLM_F_ACK_TLVS;
2375
2376 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2377 if (!skb) {
2378 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2379 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2380 return;
2381 }
2382
2383 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2384 NLMSG_ERROR, payload, flags);
2385 errmsg = nlmsg_data(rep);
2386 errmsg->error = err;
2387 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2388
2389 if (nlk_has_extack && extack) {
2390 if (extack->_msg) {
2391 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2392 extack->_msg));
2393 }
2394 if (err) {
2395 if (extack->bad_attr &&
2396 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2397 (u8 *)extack->bad_attr >= in_skb->data +
2398 in_skb->len))
2399 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2400 (u8 *)extack->bad_attr -
2401 in_skb->data));
2402 } else {
2403 if (extack->cookie_len)
2404 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2405 extack->cookie_len,
2406 extack->cookie));
2407 }
2408 }
2409
2410 nlmsg_end(skb, rep);
2411
2412 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2413 }
2414 EXPORT_SYMBOL(netlink_ack);
2415
2416 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2417 struct nlmsghdr *,
2418 struct netlink_ext_ack *))
2419 {
2420 struct netlink_ext_ack extack;
2421 struct nlmsghdr *nlh;
2422 int err;
2423
2424 while (skb->len >= nlmsg_total_size(0)) {
2425 int msglen;
2426
2427 memset(&extack, 0, sizeof(extack));
2428 nlh = nlmsg_hdr(skb);
2429 err = 0;
2430
2431 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2432 return 0;
2433
2434 /* Only requests are handled by the kernel */
2435 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2436 goto ack;
2437
2438 /* Skip control messages */
2439 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2440 goto ack;
2441
2442 err = cb(skb, nlh, &extack);
2443 if (err == -EINTR)
2444 goto skip;
2445
2446 ack:
2447 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2448 netlink_ack(skb, nlh, err, &extack);
2449
2450 skip:
2451 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2452 if (msglen > skb->len)
2453 msglen = skb->len;
2454 skb_pull(skb, msglen);
2455 }
2456
2457 return 0;
2458 }
2459 EXPORT_SYMBOL(netlink_rcv_skb);
2460
2461 /**
2462 * nlmsg_notify - send a notification netlink message
2463 * @sk: netlink socket to use
2464 * @skb: notification message
2465 * @portid: destination netlink portid for reports or 0
2466 * @group: destination multicast group or 0
2467 * @report: 1 to report back, 0 to disable
2468 * @flags: allocation flags
2469 */
2470 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2471 unsigned int group, int report, gfp_t flags)
2472 {
2473 int err = 0;
2474
2475 if (group) {
2476 int exclude_portid = 0;
2477
2478 if (report) {
2479 refcount_inc(&skb->users);
2480 exclude_portid = portid;
2481 }
2482
2483 /* errors reported via destination sk->sk_err, but propagate
2484 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2485 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2486 }
2487
2488 if (report) {
2489 int err2;
2490
2491 err2 = nlmsg_unicast(sk, skb, portid);
2492 if (!err || err == -ESRCH)
2493 err = err2;
2494 }
2495
2496 return err;
2497 }
2498 EXPORT_SYMBOL(nlmsg_notify);
2499
2500 #ifdef CONFIG_PROC_FS
2501 struct nl_seq_iter {
2502 struct seq_net_private p;
2503 struct rhashtable_iter hti;
2504 int link;
2505 };
2506
2507 static int netlink_walk_start(struct nl_seq_iter *iter)
2508 {
2509 int err;
2510
2511 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2512 GFP_KERNEL);
2513 if (err) {
2514 iter->link = MAX_LINKS;
2515 return err;
2516 }
2517
2518 rhashtable_walk_start(&iter->hti);
2519
2520 return 0;
2521 }
2522
2523 static void netlink_walk_stop(struct nl_seq_iter *iter)
2524 {
2525 rhashtable_walk_stop(&iter->hti);
2526 rhashtable_walk_exit(&iter->hti);
2527 }
2528
2529 static void *__netlink_seq_next(struct seq_file *seq)
2530 {
2531 struct nl_seq_iter *iter = seq->private;
2532 struct netlink_sock *nlk;
2533
2534 do {
2535 for (;;) {
2536 int err;
2537
2538 nlk = rhashtable_walk_next(&iter->hti);
2539
2540 if (IS_ERR(nlk)) {
2541 if (PTR_ERR(nlk) == -EAGAIN)
2542 continue;
2543
2544 return nlk;
2545 }
2546
2547 if (nlk)
2548 break;
2549
2550 netlink_walk_stop(iter);
2551 if (++iter->link >= MAX_LINKS)
2552 return NULL;
2553
2554 err = netlink_walk_start(iter);
2555 if (err)
2556 return ERR_PTR(err);
2557 }
2558 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2559
2560 return nlk;
2561 }
2562
2563 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2564 {
2565 struct nl_seq_iter *iter = seq->private;
2566 void *obj = SEQ_START_TOKEN;
2567 loff_t pos;
2568 int err;
2569
2570 iter->link = 0;
2571
2572 err = netlink_walk_start(iter);
2573 if (err)
2574 return ERR_PTR(err);
2575
2576 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2577 obj = __netlink_seq_next(seq);
2578
2579 return obj;
2580 }
2581
2582 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2583 {
2584 ++*pos;
2585 return __netlink_seq_next(seq);
2586 }
2587
2588 static void netlink_seq_stop(struct seq_file *seq, void *v)
2589 {
2590 struct nl_seq_iter *iter = seq->private;
2591
2592 if (iter->link >= MAX_LINKS)
2593 return;
2594
2595 netlink_walk_stop(iter);
2596 }
2597
2598
2599 static int netlink_seq_show(struct seq_file *seq, void *v)
2600 {
2601 if (v == SEQ_START_TOKEN) {
2602 seq_puts(seq,
2603 "sk Eth Pid Groups "
2604 "Rmem Wmem Dump Locks Drops Inode\n");
2605 } else {
2606 struct sock *s = v;
2607 struct netlink_sock *nlk = nlk_sk(s);
2608
2609 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2610 s,
2611 s->sk_protocol,
2612 nlk->portid,
2613 nlk->groups ? (u32)nlk->groups[0] : 0,
2614 sk_rmem_alloc_get(s),
2615 sk_wmem_alloc_get(s),
2616 nlk->cb_running,
2617 refcount_read(&s->sk_refcnt),
2618 atomic_read(&s->sk_drops),
2619 sock_i_ino(s)
2620 );
2621
2622 }
2623 return 0;
2624 }
2625
2626 static const struct seq_operations netlink_seq_ops = {
2627 .start = netlink_seq_start,
2628 .next = netlink_seq_next,
2629 .stop = netlink_seq_stop,
2630 .show = netlink_seq_show,
2631 };
2632
2633
2634 static int netlink_seq_open(struct inode *inode, struct file *file)
2635 {
2636 return seq_open_net(inode, file, &netlink_seq_ops,
2637 sizeof(struct nl_seq_iter));
2638 }
2639
2640 static const struct file_operations netlink_seq_fops = {
2641 .open = netlink_seq_open,
2642 .read = seq_read,
2643 .llseek = seq_lseek,
2644 .release = seq_release_net,
2645 };
2646
2647 #endif
2648
2649 int netlink_register_notifier(struct notifier_block *nb)
2650 {
2651 return blocking_notifier_chain_register(&netlink_chain, nb);
2652 }
2653 EXPORT_SYMBOL(netlink_register_notifier);
2654
2655 int netlink_unregister_notifier(struct notifier_block *nb)
2656 {
2657 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2658 }
2659 EXPORT_SYMBOL(netlink_unregister_notifier);
2660
2661 static const struct proto_ops netlink_ops = {
2662 .family = PF_NETLINK,
2663 .owner = THIS_MODULE,
2664 .release = netlink_release,
2665 .bind = netlink_bind,
2666 .connect = netlink_connect,
2667 .socketpair = sock_no_socketpair,
2668 .accept = sock_no_accept,
2669 .getname = netlink_getname,
2670 .poll = datagram_poll,
2671 .ioctl = netlink_ioctl,
2672 .listen = sock_no_listen,
2673 .shutdown = sock_no_shutdown,
2674 .setsockopt = netlink_setsockopt,
2675 .getsockopt = netlink_getsockopt,
2676 .sendmsg = netlink_sendmsg,
2677 .recvmsg = netlink_recvmsg,
2678 .mmap = sock_no_mmap,
2679 .sendpage = sock_no_sendpage,
2680 };
2681
2682 static const struct net_proto_family netlink_family_ops = {
2683 .family = PF_NETLINK,
2684 .create = netlink_create,
2685 .owner = THIS_MODULE, /* for consistency 8) */
2686 };
2687
2688 static int __net_init netlink_net_init(struct net *net)
2689 {
2690 #ifdef CONFIG_PROC_FS
2691 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2692 return -ENOMEM;
2693 #endif
2694 return 0;
2695 }
2696
2697 static void __net_exit netlink_net_exit(struct net *net)
2698 {
2699 #ifdef CONFIG_PROC_FS
2700 remove_proc_entry("netlink", net->proc_net);
2701 #endif
2702 }
2703
2704 static void __init netlink_add_usersock_entry(void)
2705 {
2706 struct listeners *listeners;
2707 int groups = 32;
2708
2709 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2710 if (!listeners)
2711 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2712
2713 netlink_table_grab();
2714
2715 nl_table[NETLINK_USERSOCK].groups = groups;
2716 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2717 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2718 nl_table[NETLINK_USERSOCK].registered = 1;
2719 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2720
2721 netlink_table_ungrab();
2722 }
2723
2724 static struct pernet_operations __net_initdata netlink_net_ops = {
2725 .init = netlink_net_init,
2726 .exit = netlink_net_exit,
2727 };
2728
2729 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2730 {
2731 const struct netlink_sock *nlk = data;
2732 struct netlink_compare_arg arg;
2733
2734 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2735 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2736 }
2737
2738 static const struct rhashtable_params netlink_rhashtable_params = {
2739 .head_offset = offsetof(struct netlink_sock, node),
2740 .key_len = netlink_compare_arg_len,
2741 .obj_hashfn = netlink_hash,
2742 .obj_cmpfn = netlink_compare,
2743 .automatic_shrinking = true,
2744 };
2745
2746 static int __init netlink_proto_init(void)
2747 {
2748 int i;
2749 int err = proto_register(&netlink_proto, 0);
2750
2751 if (err != 0)
2752 goto out;
2753
2754 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2755
2756 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2757 if (!nl_table)
2758 goto panic;
2759
2760 for (i = 0; i < MAX_LINKS; i++) {
2761 if (rhashtable_init(&nl_table[i].hash,
2762 &netlink_rhashtable_params) < 0) {
2763 while (--i > 0)
2764 rhashtable_destroy(&nl_table[i].hash);
2765 kfree(nl_table);
2766 goto panic;
2767 }
2768 }
2769
2770 netlink_add_usersock_entry();
2771
2772 sock_register(&netlink_family_ops);
2773 register_pernet_subsys(&netlink_net_ops);
2774 register_pernet_subsys(&netlink_tap_net_ops);
2775 /* The netlink device handler may be needed early. */
2776 rtnetlink_init();
2777 out:
2778 return err;
2779 panic:
2780 panic("netlink_init: Cannot allocate nl_table\n");
2781 }
2782
2783 core_initcall(netlink_proto_init);
Ubuntu Hirsute Linux kernel mirror