?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
WIP FPC-III support
[linux/fpc-iii.git] / net / netlink / af_netlink.c
blobdaca50d6bb1283f3b04b585217f2aea6ba279b8b
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * NETLINK Kernel-user communication protocol.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * Patrick McHardy <kaber@trash.net>
8 *
9 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
10 * added netlink_proto_exit
11 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
12 * use nlk_sk, as sk->protinfo is on a diet 8)
13 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
14 * - inc module use count of module that owns
15 * the kernel socket in case userspace opens
16 * socket of same protocol
17 * - remove all module support, since netlink is
18 * mandatory if CONFIG_NET=y these days
19 */
20
21 #include <linux/module.h>
22
23 #include <linux/capability.h>
24 #include <linux/kernel.h>
25 #include <linux/init.h>
26 #include <linux/signal.h>
27 #include <linux/sched.h>
28 #include <linux/errno.h>
29 #include <linux/string.h>
30 #include <linux/stat.h>
31 #include <linux/socket.h>
32 #include <linux/un.h>
33 #include <linux/fcntl.h>
34 #include <linux/termios.h>
35 #include <linux/sockios.h>
36 #include <linux/net.h>
37 #include <linux/fs.h>
38 #include <linux/slab.h>
39 #include <linux/uaccess.h>
40 #include <linux/skbuff.h>
41 #include <linux/netdevice.h>
42 #include <linux/rtnetlink.h>
43 #include <linux/proc_fs.h>
44 #include <linux/seq_file.h>
45 #include <linux/notifier.h>
46 #include <linux/security.h>
47 #include <linux/jhash.h>
48 #include <linux/jiffies.h>
49 #include <linux/random.h>
50 #include <linux/bitops.h>
51 #include <linux/mm.h>
52 #include <linux/types.h>
53 #include <linux/audit.h>
54 #include <linux/mutex.h>
55 #include <linux/vmalloc.h>
56 #include <linux/if_arp.h>
57 #include <linux/rhashtable.h>
58 #include <asm/cacheflush.h>
59 #include <linux/hash.h>
60 #include <linux/genetlink.h>
61 #include <linux/net_namespace.h>
62 #include <linux/nospec.h>
63 #include <linux/btf_ids.h>
64
65 #include <net/net_namespace.h>
66 #include <net/netns/generic.h>
67 #include <net/sock.h>
68 #include <net/scm.h>
69 #include <net/netlink.h>
70
71 #include "af_netlink.h"
72
73 struct listeners {
74 struct rcu_head rcu;
75 unsigned long masks[];
76 };
77
78 /* state bits */
79 #define NETLINK_S_CONGESTED 0x0
80
81 static inline int netlink_is_kernel(struct sock *sk)
82 {
83 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
84 }
85
86 struct netlink_table *nl_table __read_mostly;
87 EXPORT_SYMBOL_GPL(nl_table);
88
89 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
90
91 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
92
93 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
94 "nlk_cb_mutex-ROUTE",
95 "nlk_cb_mutex-1",
96 "nlk_cb_mutex-USERSOCK",
97 "nlk_cb_mutex-FIREWALL",
98 "nlk_cb_mutex-SOCK_DIAG",
99 "nlk_cb_mutex-NFLOG",
100 "nlk_cb_mutex-XFRM",
101 "nlk_cb_mutex-SELINUX",
102 "nlk_cb_mutex-ISCSI",
103 "nlk_cb_mutex-AUDIT",
104 "nlk_cb_mutex-FIB_LOOKUP",
105 "nlk_cb_mutex-CONNECTOR",
106 "nlk_cb_mutex-NETFILTER",
107 "nlk_cb_mutex-IP6_FW",
108 "nlk_cb_mutex-DNRTMSG",
109 "nlk_cb_mutex-KOBJECT_UEVENT",
110 "nlk_cb_mutex-GENERIC",
111 "nlk_cb_mutex-17",
112 "nlk_cb_mutex-SCSITRANSPORT",
113 "nlk_cb_mutex-ECRYPTFS",
114 "nlk_cb_mutex-RDMA",
115 "nlk_cb_mutex-CRYPTO",
116 "nlk_cb_mutex-SMC",
117 "nlk_cb_mutex-23",
118 "nlk_cb_mutex-24",
119 "nlk_cb_mutex-25",
120 "nlk_cb_mutex-26",
121 "nlk_cb_mutex-27",
122 "nlk_cb_mutex-28",
123 "nlk_cb_mutex-29",
124 "nlk_cb_mutex-30",
125 "nlk_cb_mutex-31",
126 "nlk_cb_mutex-MAX_LINKS"
127 };
128
129 static int netlink_dump(struct sock *sk);
130
131 /* nl_table locking explained:
132 * Lookup and traversal are protected with an RCU read-side lock. Insertion
133 * and removal are protected with per bucket lock while using RCU list
134 * modification primitives and may run in parallel to RCU protected lookups.
135 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
136 * been acquired * either during or after the socket has been removed from
137 * the list and after an RCU grace period.
138 */
139 DEFINE_RWLOCK(nl_table_lock);
140 EXPORT_SYMBOL_GPL(nl_table_lock);
141 static atomic_t nl_table_users = ATOMIC_INIT(0);
142
143 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
144
145 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
146
147
148 static const struct rhashtable_params netlink_rhashtable_params;
149
150 static inline u32 netlink_group_mask(u32 group)
151 {
152 return group ? 1 << (group - 1) : 0;
153 }
154
155 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
156 gfp_t gfp_mask)
157 {
158 unsigned int len = skb_end_offset(skb);
159 struct sk_buff *new;
160
161 new = alloc_skb(len, gfp_mask);
162 if (new == NULL)
163 return NULL;
164
165 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
166 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
167 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
168
169 skb_put_data(new, skb->data, len);
170 return new;
171 }
172
173 static unsigned int netlink_tap_net_id;
174
175 struct netlink_tap_net {
176 struct list_head netlink_tap_all;
177 struct mutex netlink_tap_lock;
178 };
179
180 int netlink_add_tap(struct netlink_tap *nt)
181 {
182 struct net *net = dev_net(nt->dev);
183 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
184
185 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
186 return -EINVAL;
187
188 mutex_lock(&nn->netlink_tap_lock);
189 list_add_rcu(&nt->list, &nn->netlink_tap_all);
190 mutex_unlock(&nn->netlink_tap_lock);
191
192 __module_get(nt->module);
193
194 return 0;
195 }
196 EXPORT_SYMBOL_GPL(netlink_add_tap);
197
198 static int __netlink_remove_tap(struct netlink_tap *nt)
199 {
200 struct net *net = dev_net(nt->dev);
201 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
202 bool found = false;
203 struct netlink_tap *tmp;
204
205 mutex_lock(&nn->netlink_tap_lock);
206
207 list_for_each_entry(tmp, &nn->netlink_tap_all, list) {
208 if (nt == tmp) {
209 list_del_rcu(&nt->list);
210 found = true;
211 goto out;
212 }
213 }
214
215 pr_warn("__netlink_remove_tap: %p not found\n", nt);
216 out:
217 mutex_unlock(&nn->netlink_tap_lock);
218
219 if (found)
220 module_put(nt->module);
221
222 return found ? 0 : -ENODEV;
223 }
224
225 int netlink_remove_tap(struct netlink_tap *nt)
226 {
227 int ret;
228
229 ret = __netlink_remove_tap(nt);
230 synchronize_net();
231
232 return ret;
233 }
234 EXPORT_SYMBOL_GPL(netlink_remove_tap);
235
236 static __net_init int netlink_tap_init_net(struct net *net)
237 {
238 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
239
240 INIT_LIST_HEAD(&nn->netlink_tap_all);
241 mutex_init(&nn->netlink_tap_lock);
242 return 0;
243 }
244
245 static struct pernet_operations netlink_tap_net_ops = {
246 .init = netlink_tap_init_net,
247 .id = &netlink_tap_net_id,
248 .size = sizeof(struct netlink_tap_net),
249 };
250
251 static bool netlink_filter_tap(const struct sk_buff *skb)
252 {
253 struct sock *sk = skb->sk;
254
255 /* We take the more conservative approach and
256 * whitelist socket protocols that may pass.
257 */
258 switch (sk->sk_protocol) {
259 case NETLINK_ROUTE:
260 case NETLINK_USERSOCK:
261 case NETLINK_SOCK_DIAG:
262 case NETLINK_NFLOG:
263 case NETLINK_XFRM:
264 case NETLINK_FIB_LOOKUP:
265 case NETLINK_NETFILTER:
266 case NETLINK_GENERIC:
267 return true;
268 }
269
270 return false;
271 }
272
273 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
274 struct net_device *dev)
275 {
276 struct sk_buff *nskb;
277 struct sock *sk = skb->sk;
278 int ret = -ENOMEM;
279
280 if (!net_eq(dev_net(dev), sock_net(sk)))
281 return 0;
282
283 dev_hold(dev);
284
285 if (is_vmalloc_addr(skb->head))
286 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
287 else
288 nskb = skb_clone(skb, GFP_ATOMIC);
289 if (nskb) {
290 nskb->dev = dev;
291 nskb->protocol = htons((u16) sk->sk_protocol);
292 nskb->pkt_type = netlink_is_kernel(sk) ?
293 PACKET_KERNEL : PACKET_USER;
294 skb_reset_network_header(nskb);
295 ret = dev_queue_xmit(nskb);
296 if (unlikely(ret > 0))
297 ret = net_xmit_errno(ret);
298 }
299
300 dev_put(dev);
301 return ret;
302 }
303
304 static void __netlink_deliver_tap(struct sk_buff *skb, struct netlink_tap_net *nn)
305 {
306 int ret;
307 struct netlink_tap *tmp;
308
309 if (!netlink_filter_tap(skb))
310 return;
311
312 list_for_each_entry_rcu(tmp, &nn->netlink_tap_all, list) {
313 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
314 if (unlikely(ret))
315 break;
316 }
317 }
318
319 static void netlink_deliver_tap(struct net *net, struct sk_buff *skb)
320 {
321 struct netlink_tap_net *nn = net_generic(net, netlink_tap_net_id);
322
323 rcu_read_lock();
324
325 if (unlikely(!list_empty(&nn->netlink_tap_all)))
326 __netlink_deliver_tap(skb, nn);
327
328 rcu_read_unlock();
329 }
330
331 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
332 struct sk_buff *skb)
333 {
334 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
335 netlink_deliver_tap(sock_net(dst), skb);
336 }
337
338 static void netlink_overrun(struct sock *sk)
339 {
340 struct netlink_sock *nlk = nlk_sk(sk);
341
342 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
343 if (!test_and_set_bit(NETLINK_S_CONGESTED,
344 &nlk_sk(sk)->state)) {
345 sk->sk_err = ENOBUFS;
346 sk->sk_error_report(sk);
347 }
348 }
349 atomic_inc(&sk->sk_drops);
350 }
351
352 static void netlink_rcv_wake(struct sock *sk)
353 {
354 struct netlink_sock *nlk = nlk_sk(sk);
355
356 if (skb_queue_empty_lockless(&sk->sk_receive_queue))
357 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
358 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
359 wake_up_interruptible(&nlk->wait);
360 }
361
362 static void netlink_skb_destructor(struct sk_buff *skb)
363 {
364 if (is_vmalloc_addr(skb->head)) {
365 if (!skb->cloned ||
366 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
367 vfree(skb->head);
368
369 skb->head = NULL;
370 }
371 if (skb->sk != NULL)
372 sock_rfree(skb);
373 }
374
375 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
376 {
377 WARN_ON(skb->sk != NULL);
378 skb->sk = sk;
379 skb->destructor = netlink_skb_destructor;
380 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
381 sk_mem_charge(sk, skb->truesize);
382 }
383
384 static void netlink_sock_destruct(struct sock *sk)
385 {
386 struct netlink_sock *nlk = nlk_sk(sk);
387
388 if (nlk->cb_running) {
389 if (nlk->cb.done)
390 nlk->cb.done(&nlk->cb);
391 module_put(nlk->cb.module);
392 kfree_skb(nlk->cb.skb);
393 }
394
395 skb_queue_purge(&sk->sk_receive_queue);
396
397 if (!sock_flag(sk, SOCK_DEAD)) {
398 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
399 return;
400 }
401
402 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
403 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
404 WARN_ON(nlk_sk(sk)->groups);
405 }
406
407 static void netlink_sock_destruct_work(struct work_struct *work)
408 {
409 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
410 work);
411
412 sk_free(&nlk->sk);
413 }
414
415 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
416 * SMP. Look, when several writers sleep and reader wakes them up, all but one
417 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
418 * this, _but_ remember, it adds useless work on UP machines.
419 */
420
421 void netlink_table_grab(void)
422 __acquires(nl_table_lock)
423 {
424 might_sleep();
425
426 write_lock_irq(&nl_table_lock);
427
428 if (atomic_read(&nl_table_users)) {
429 DECLARE_WAITQUEUE(wait, current);
430
431 add_wait_queue_exclusive(&nl_table_wait, &wait);
432 for (;;) {
433 set_current_state(TASK_UNINTERRUPTIBLE);
434 if (atomic_read(&nl_table_users) == 0)
435 break;
436 write_unlock_irq(&nl_table_lock);
437 schedule();
438 write_lock_irq(&nl_table_lock);
439 }
440
441 __set_current_state(TASK_RUNNING);
442 remove_wait_queue(&nl_table_wait, &wait);
443 }
444 }
445
446 void netlink_table_ungrab(void)
447 __releases(nl_table_lock)
448 {
449 write_unlock_irq(&nl_table_lock);
450 wake_up(&nl_table_wait);
451 }
452
453 static inline void
454 netlink_lock_table(void)
455 {
456 /* read_lock() synchronizes us to netlink_table_grab */
457
458 read_lock(&nl_table_lock);
459 atomic_inc(&nl_table_users);
460 read_unlock(&nl_table_lock);
461 }
462
463 static inline void
464 netlink_unlock_table(void)
465 {
466 if (atomic_dec_and_test(&nl_table_users))
467 wake_up(&nl_table_wait);
468 }
469
470 struct netlink_compare_arg
471 {
472 possible_net_t pnet;
473 u32 portid;
474 };
475
476 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
477 #define netlink_compare_arg_len \
478 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
479
480 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
481 const void *ptr)
482 {
483 const struct netlink_compare_arg *x = arg->key;
484 const struct netlink_sock *nlk = ptr;
485
486 return nlk->portid != x->portid ||
487 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
488 }
489
490 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
491 struct net *net, u32 portid)
492 {
493 memset(arg, 0, sizeof(*arg));
494 write_pnet(&arg->pnet, net);
495 arg->portid = portid;
496 }
497
498 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
499 struct net *net)
500 {
501 struct netlink_compare_arg arg;
502
503 netlink_compare_arg_init(&arg, net, portid);
504 return rhashtable_lookup_fast(&table->hash, &arg,
505 netlink_rhashtable_params);
506 }
507
508 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
509 {
510 struct netlink_compare_arg arg;
511
512 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
513 return rhashtable_lookup_insert_key(&table->hash, &arg,
514 &nlk_sk(sk)->node,
515 netlink_rhashtable_params);
516 }
517
518 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
519 {
520 struct netlink_table *table = &nl_table[protocol];
521 struct sock *sk;
522
523 rcu_read_lock();
524 sk = __netlink_lookup(table, portid, net);
525 if (sk)
526 sock_hold(sk);
527 rcu_read_unlock();
528
529 return sk;
530 }
531
532 static const struct proto_ops netlink_ops;
533
534 static void
535 netlink_update_listeners(struct sock *sk)
536 {
537 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
538 unsigned long mask;
539 unsigned int i;
540 struct listeners *listeners;
541
542 listeners = nl_deref_protected(tbl->listeners);
543 if (!listeners)
544 return;
545
546 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
547 mask = 0;
548 sk_for_each_bound(sk, &tbl->mc_list) {
549 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
550 mask |= nlk_sk(sk)->groups[i];
551 }
552 listeners->masks[i] = mask;
553 }
554 /* this function is only called with the netlink table "grabbed", which
555 * makes sure updates are visible before bind or setsockopt return. */
556 }
557
558 static int netlink_insert(struct sock *sk, u32 portid)
559 {
560 struct netlink_table *table = &nl_table[sk->sk_protocol];
561 int err;
562
563 lock_sock(sk);
564
565 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
566 if (nlk_sk(sk)->bound)
567 goto err;
568
569 nlk_sk(sk)->portid = portid;
570 sock_hold(sk);
571
572 err = __netlink_insert(table, sk);
573 if (err) {
574 /* In case the hashtable backend returns with -EBUSY
575 * from here, it must not escape to the caller.
576 */
577 if (unlikely(err == -EBUSY))
578 err = -EOVERFLOW;
579 if (err == -EEXIST)
580 err = -EADDRINUSE;
581 sock_put(sk);
582 goto err;
583 }
584
585 /* We need to ensure that the socket is hashed and visible. */
586 smp_wmb();
587 nlk_sk(sk)->bound = portid;
588
589 err:
590 release_sock(sk);
591 return err;
592 }
593
594 static void netlink_remove(struct sock *sk)
595 {
596 struct netlink_table *table;
597
598 table = &nl_table[sk->sk_protocol];
599 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
600 netlink_rhashtable_params)) {
601 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
602 __sock_put(sk);
603 }
604
605 netlink_table_grab();
606 if (nlk_sk(sk)->subscriptions) {
607 __sk_del_bind_node(sk);
608 netlink_update_listeners(sk);
609 }
610 if (sk->sk_protocol == NETLINK_GENERIC)
611 atomic_inc(&genl_sk_destructing_cnt);
612 netlink_table_ungrab();
613 }
614
615 static struct proto netlink_proto = {
616 .name = "NETLINK",
617 .owner = THIS_MODULE,
618 .obj_size = sizeof(struct netlink_sock),
619 };
620
621 static int __netlink_create(struct net *net, struct socket *sock,
622 struct mutex *cb_mutex, int protocol,
623 int kern)
624 {
625 struct sock *sk;
626 struct netlink_sock *nlk;
627
628 sock->ops = &netlink_ops;
629
630 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
631 if (!sk)
632 return -ENOMEM;
633
634 sock_init_data(sock, sk);
635
636 nlk = nlk_sk(sk);
637 if (cb_mutex) {
638 nlk->cb_mutex = cb_mutex;
639 } else {
640 nlk->cb_mutex = &nlk->cb_def_mutex;
641 mutex_init(nlk->cb_mutex);
642 lockdep_set_class_and_name(nlk->cb_mutex,
643 nlk_cb_mutex_keys + protocol,
644 nlk_cb_mutex_key_strings[protocol]);
645 }
646 init_waitqueue_head(&nlk->wait);
647
648 sk->sk_destruct = netlink_sock_destruct;
649 sk->sk_protocol = protocol;
650 return 0;
651 }
652
653 static int netlink_create(struct net *net, struct socket *sock, int protocol,
654 int kern)
655 {
656 struct module *module = NULL;
657 struct mutex *cb_mutex;
658 struct netlink_sock *nlk;
659 int (*bind)(struct net *net, int group);
660 void (*unbind)(struct net *net, int group);
661 int err = 0;
662
663 sock->state = SS_UNCONNECTED;
664
665 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
666 return -ESOCKTNOSUPPORT;
667
668 if (protocol < 0 || protocol >= MAX_LINKS)
669 return -EPROTONOSUPPORT;
670 protocol = array_index_nospec(protocol, MAX_LINKS);
671
672 netlink_lock_table();
673 #ifdef CONFIG_MODULES
674 if (!nl_table[protocol].registered) {
675 netlink_unlock_table();
676 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
677 netlink_lock_table();
678 }
679 #endif
680 if (nl_table[protocol].registered &&
681 try_module_get(nl_table[protocol].module))
682 module = nl_table[protocol].module;
683 else
684 err = -EPROTONOSUPPORT;
685 cb_mutex = nl_table[protocol].cb_mutex;
686 bind = nl_table[protocol].bind;
687 unbind = nl_table[protocol].unbind;
688 netlink_unlock_table();
689
690 if (err < 0)
691 goto out;
692
693 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
694 if (err < 0)
695 goto out_module;
696
697 local_bh_disable();
698 sock_prot_inuse_add(net, &netlink_proto, 1);
699 local_bh_enable();
700
701 nlk = nlk_sk(sock->sk);
702 nlk->module = module;
703 nlk->netlink_bind = bind;
704 nlk->netlink_unbind = unbind;
705 out:
706 return err;
707
708 out_module:
709 module_put(module);
710 goto out;
711 }
712
713 static void deferred_put_nlk_sk(struct rcu_head *head)
714 {
715 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
716 struct sock *sk = &nlk->sk;
717
718 kfree(nlk->groups);
719 nlk->groups = NULL;
720
721 if (!refcount_dec_and_test(&sk->sk_refcnt))
722 return;
723
724 if (nlk->cb_running && nlk->cb.done) {
725 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
726 schedule_work(&nlk->work);
727 return;
728 }
729
730 sk_free(sk);
731 }
732
733 static int netlink_release(struct socket *sock)
734 {
735 struct sock *sk = sock->sk;
736 struct netlink_sock *nlk;
737
738 if (!sk)
739 return 0;
740
741 netlink_remove(sk);
742 sock_orphan(sk);
743 nlk = nlk_sk(sk);
744
745 /*
746 * OK. Socket is unlinked, any packets that arrive now
747 * will be purged.
748 */
749
750 /* must not acquire netlink_table_lock in any way again before unbind
751 * and notifying genetlink is done as otherwise it might deadlock
752 */
753 if (nlk->netlink_unbind) {
754 int i;
755
756 for (i = 0; i < nlk->ngroups; i++)
757 if (test_bit(i, nlk->groups))
758 nlk->netlink_unbind(sock_net(sk), i + 1);
759 }
760 if (sk->sk_protocol == NETLINK_GENERIC &&
761 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
762 wake_up(&genl_sk_destructing_waitq);
763
764 sock->sk = NULL;
765 wake_up_interruptible_all(&nlk->wait);
766
767 skb_queue_purge(&sk->sk_write_queue);
768
769 if (nlk->portid && nlk->bound) {
770 struct netlink_notify n = {
771 .net = sock_net(sk),
772 .protocol = sk->sk_protocol,
773 .portid = nlk->portid,
774 };
775 blocking_notifier_call_chain(&netlink_chain,
776 NETLINK_URELEASE, &n);
777 }
778
779 module_put(nlk->module);
780
781 if (netlink_is_kernel(sk)) {
782 netlink_table_grab();
783 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
784 if (--nl_table[sk->sk_protocol].registered == 0) {
785 struct listeners *old;
786
787 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
788 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
789 kfree_rcu(old, rcu);
790 nl_table[sk->sk_protocol].module = NULL;
791 nl_table[sk->sk_protocol].bind = NULL;
792 nl_table[sk->sk_protocol].unbind = NULL;
793 nl_table[sk->sk_protocol].flags = 0;
794 nl_table[sk->sk_protocol].registered = 0;
795 }
796 netlink_table_ungrab();
797 }
798
799 local_bh_disable();
800 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
801 local_bh_enable();
802 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
803 return 0;
804 }
805
806 static int netlink_autobind(struct socket *sock)
807 {
808 struct sock *sk = sock->sk;
809 struct net *net = sock_net(sk);
810 struct netlink_table *table = &nl_table[sk->sk_protocol];
811 s32 portid = task_tgid_vnr(current);
812 int err;
813 s32 rover = -4096;
814 bool ok;
815
816 retry:
817 cond_resched();
818 rcu_read_lock();
819 ok = !__netlink_lookup(table, portid, net);
820 rcu_read_unlock();
821 if (!ok) {
822 /* Bind collision, search negative portid values. */
823 if (rover == -4096)
824 /* rover will be in range [S32_MIN, -4097] */
825 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
826 else if (rover >= -4096)
827 rover = -4097;
828 portid = rover--;
829 goto retry;
830 }
831
832 err = netlink_insert(sk, portid);
833 if (err == -EADDRINUSE)
834 goto retry;
835
836 /* If 2 threads race to autobind, that is fine. */
837 if (err == -EBUSY)
838 err = 0;
839
840 return err;
841 }
842
843 /**
844 * __netlink_ns_capable - General netlink message capability test
845 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
846 * @user_ns: The user namespace of the capability to use
847 * @cap: The capability to use
848 *
849 * Test to see if the opener of the socket we received the message
850 * from had when the netlink socket was created and the sender of the
851 * message has the capability @cap in the user namespace @user_ns.
852 */
853 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
854 struct user_namespace *user_ns, int cap)
855 {
856 return ((nsp->flags & NETLINK_SKB_DST) ||
857 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
858 ns_capable(user_ns, cap);
859 }
860 EXPORT_SYMBOL(__netlink_ns_capable);
861
862 /**
863 * netlink_ns_capable - General netlink message capability test
864 * @skb: socket buffer holding a netlink command from userspace
865 * @user_ns: The user namespace of the capability to use
866 * @cap: The capability to use
867 *
868 * Test to see if the opener of the socket we received the message
869 * from had when the netlink socket was created and the sender of the
870 * message has the capability @cap in the user namespace @user_ns.
871 */
872 bool netlink_ns_capable(const struct sk_buff *skb,
873 struct user_namespace *user_ns, int cap)
874 {
875 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
876 }
877 EXPORT_SYMBOL(netlink_ns_capable);
878
879 /**
880 * netlink_capable - Netlink global message capability test
881 * @skb: socket buffer holding a netlink command from userspace
882 * @cap: The capability to use
883 *
884 * Test to see if the opener of the socket we received the message
885 * from had when the netlink socket was created and the sender of the
886 * message has the capability @cap in all user namespaces.
887 */
888 bool netlink_capable(const struct sk_buff *skb, int cap)
889 {
890 return netlink_ns_capable(skb, &init_user_ns, cap);
891 }
892 EXPORT_SYMBOL(netlink_capable);
893
894 /**
895 * netlink_net_capable - Netlink network namespace message capability test
896 * @skb: socket buffer holding a netlink command from userspace
897 * @cap: The capability to use
898 *
899 * Test to see if the opener of the socket we received the message
900 * from had when the netlink socket was created and the sender of the
901 * message has the capability @cap over the network namespace of
902 * the socket we received the message from.
903 */
904 bool netlink_net_capable(const struct sk_buff *skb, int cap)
905 {
906 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
907 }
908 EXPORT_SYMBOL(netlink_net_capable);
909
910 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
911 {
912 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
913 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
914 }
915
916 static void
917 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
918 {
919 struct netlink_sock *nlk = nlk_sk(sk);
920
921 if (nlk->subscriptions && !subscriptions)
922 __sk_del_bind_node(sk);
923 else if (!nlk->subscriptions && subscriptions)
924 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
925 nlk->subscriptions = subscriptions;
926 }
927
928 static int netlink_realloc_groups(struct sock *sk)
929 {
930 struct netlink_sock *nlk = nlk_sk(sk);
931 unsigned int groups;
932 unsigned long *new_groups;
933 int err = 0;
934
935 netlink_table_grab();
936
937 groups = nl_table[sk->sk_protocol].groups;
938 if (!nl_table[sk->sk_protocol].registered) {
939 err = -ENOENT;
940 goto out_unlock;
941 }
942
943 if (nlk->ngroups >= groups)
944 goto out_unlock;
945
946 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
947 if (new_groups == NULL) {
948 err = -ENOMEM;
949 goto out_unlock;
950 }
951 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
952 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
953
954 nlk->groups = new_groups;
955 nlk->ngroups = groups;
956 out_unlock:
957 netlink_table_ungrab();
958 return err;
959 }
960
961 static void netlink_undo_bind(int group, long unsigned int groups,
962 struct sock *sk)
963 {
964 struct netlink_sock *nlk = nlk_sk(sk);
965 int undo;
966
967 if (!nlk->netlink_unbind)
968 return;
969
970 for (undo = 0; undo < group; undo++)
971 if (test_bit(undo, &groups))
972 nlk->netlink_unbind(sock_net(sk), undo + 1);
973 }
974
975 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
976 int addr_len)
977 {
978 struct sock *sk = sock->sk;
979 struct net *net = sock_net(sk);
980 struct netlink_sock *nlk = nlk_sk(sk);
981 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
982 int err = 0;
983 unsigned long groups;
984 bool bound;
985
986 if (addr_len < sizeof(struct sockaddr_nl))
987 return -EINVAL;
988
989 if (nladdr->nl_family != AF_NETLINK)
990 return -EINVAL;
991 groups = nladdr->nl_groups;
992
993 /* Only superuser is allowed to listen multicasts */
994 if (groups) {
995 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
996 return -EPERM;
997 err = netlink_realloc_groups(sk);
998 if (err)
999 return err;
1000 }
1001
1002 if (nlk->ngroups < BITS_PER_LONG)
1003 groups &= (1UL << nlk->ngroups) - 1;
1004
1005 bound = nlk->bound;
1006 if (bound) {
1007 /* Ensure nlk->portid is up-to-date. */
1008 smp_rmb();
1009
1010 if (nladdr->nl_pid != nlk->portid)
1011 return -EINVAL;
1012 }
1013
1014 netlink_lock_table();
1015 if (nlk->netlink_bind && groups) {
1016 int group;
1017
1018 /* nl_groups is a u32, so cap the maximum groups we can bind */
1019 for (group = 0; group < BITS_PER_TYPE(u32); group++) {
1020 if (!test_bit(group, &groups))
1021 continue;
1022 err = nlk->netlink_bind(net, group + 1);
1023 if (!err)
1024 continue;
1025 netlink_undo_bind(group, groups, sk);
1026 goto unlock;
1027 }
1028 }
1029
1030 /* No need for barriers here as we return to user-space without
1031 * using any of the bound attributes.
1032 */
1033 if (!bound) {
1034 err = nladdr->nl_pid ?
1035 netlink_insert(sk, nladdr->nl_pid) :
1036 netlink_autobind(sock);
1037 if (err) {
1038 netlink_undo_bind(BITS_PER_TYPE(u32), groups, sk);
1039 goto unlock;
1040 }
1041 }
1042
1043 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1044 goto unlock;
1045 netlink_unlock_table();
1046
1047 netlink_table_grab();
1048 netlink_update_subscriptions(sk, nlk->subscriptions +
1049 hweight32(groups) -
1050 hweight32(nlk->groups[0]));
1051 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1052 netlink_update_listeners(sk);
1053 netlink_table_ungrab();
1054
1055 return 0;
1056
1057 unlock:
1058 netlink_unlock_table();
1059 return err;
1060 }
1061
1062 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1063 int alen, int flags)
1064 {
1065 int err = 0;
1066 struct sock *sk = sock->sk;
1067 struct netlink_sock *nlk = nlk_sk(sk);
1068 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1069
1070 if (alen < sizeof(addr->sa_family))
1071 return -EINVAL;
1072
1073 if (addr->sa_family == AF_UNSPEC) {
1074 sk->sk_state = NETLINK_UNCONNECTED;
1075 nlk->dst_portid = 0;
1076 nlk->dst_group = 0;
1077 return 0;
1078 }
1079 if (addr->sa_family != AF_NETLINK)
1080 return -EINVAL;
1081
1082 if (alen < sizeof(struct sockaddr_nl))
1083 return -EINVAL;
1084
1085 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1086 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1087 return -EPERM;
1088
1089 /* No need for barriers here as we return to user-space without
1090 * using any of the bound attributes.
1091 */
1092 if (!nlk->bound)
1093 err = netlink_autobind(sock);
1094
1095 if (err == 0) {
1096 sk->sk_state = NETLINK_CONNECTED;
1097 nlk->dst_portid = nladdr->nl_pid;
1098 nlk->dst_group = ffs(nladdr->nl_groups);
1099 }
1100
1101 return err;
1102 }
1103
1104 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1105 int peer)
1106 {
1107 struct sock *sk = sock->sk;
1108 struct netlink_sock *nlk = nlk_sk(sk);
1109 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1110
1111 nladdr->nl_family = AF_NETLINK;
1112 nladdr->nl_pad = 0;
1113
1114 if (peer) {
1115 nladdr->nl_pid = nlk->dst_portid;
1116 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1117 } else {
1118 nladdr->nl_pid = nlk->portid;
1119 netlink_lock_table();
1120 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1121 netlink_unlock_table();
1122 }
1123 return sizeof(*nladdr);
1124 }
1125
1126 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1127 unsigned long arg)
1128 {
1129 /* try to hand this ioctl down to the NIC drivers.
1130 */
1131 return -ENOIOCTLCMD;
1132 }
1133
1134 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1135 {
1136 struct sock *sock;
1137 struct netlink_sock *nlk;
1138
1139 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1140 if (!sock)
1141 return ERR_PTR(-ECONNREFUSED);
1142
1143 /* Don't bother queuing skb if kernel socket has no input function */
1144 nlk = nlk_sk(sock);
1145 if (sock->sk_state == NETLINK_CONNECTED &&
1146 nlk->dst_portid != nlk_sk(ssk)->portid) {
1147 sock_put(sock);
1148 return ERR_PTR(-ECONNREFUSED);
1149 }
1150 return sock;
1151 }
1152
1153 struct sock *netlink_getsockbyfilp(struct file *filp)
1154 {
1155 struct inode *inode = file_inode(filp);
1156 struct sock *sock;
1157
1158 if (!S_ISSOCK(inode->i_mode))
1159 return ERR_PTR(-ENOTSOCK);
1160
1161 sock = SOCKET_I(inode)->sk;
1162 if (sock->sk_family != AF_NETLINK)
1163 return ERR_PTR(-EINVAL);
1164
1165 sock_hold(sock);
1166 return sock;
1167 }
1168
1169 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1170 int broadcast)
1171 {
1172 struct sk_buff *skb;
1173 void *data;
1174
1175 if (size <= NLMSG_GOODSIZE || broadcast)
1176 return alloc_skb(size, GFP_KERNEL);
1177
1178 size = SKB_DATA_ALIGN(size) +
1179 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1180
1181 data = vmalloc(size);
1182 if (data == NULL)
1183 return NULL;
1184
1185 skb = __build_skb(data, size);
1186 if (skb == NULL)
1187 vfree(data);
1188 else
1189 skb->destructor = netlink_skb_destructor;
1190
1191 return skb;
1192 }
1193
1194 /*
1195 * Attach a skb to a netlink socket.
1196 * The caller must hold a reference to the destination socket. On error, the
1197 * reference is dropped. The skb is not send to the destination, just all
1198 * all error checks are performed and memory in the queue is reserved.
1199 * Return values:
1200 * < 0: error. skb freed, reference to sock dropped.
1201 * 0: continue
1202 * 1: repeat lookup - reference dropped while waiting for socket memory.
1203 */
1204 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1205 long *timeo, struct sock *ssk)
1206 {
1207 struct netlink_sock *nlk;
1208
1209 nlk = nlk_sk(sk);
1210
1211 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1212 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1213 DECLARE_WAITQUEUE(wait, current);
1214 if (!*timeo) {
1215 if (!ssk || netlink_is_kernel(ssk))
1216 netlink_overrun(sk);
1217 sock_put(sk);
1218 kfree_skb(skb);
1219 return -EAGAIN;
1220 }
1221
1222 __set_current_state(TASK_INTERRUPTIBLE);
1223 add_wait_queue(&nlk->wait, &wait);
1224
1225 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1226 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1227 !sock_flag(sk, SOCK_DEAD))
1228 *timeo = schedule_timeout(*timeo);
1229
1230 __set_current_state(TASK_RUNNING);
1231 remove_wait_queue(&nlk->wait, &wait);
1232 sock_put(sk);
1233
1234 if (signal_pending(current)) {
1235 kfree_skb(skb);
1236 return sock_intr_errno(*timeo);
1237 }
1238 return 1;
1239 }
1240 netlink_skb_set_owner_r(skb, sk);
1241 return 0;
1242 }
1243
1244 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1245 {
1246 int len = skb->len;
1247
1248 netlink_deliver_tap(sock_net(sk), skb);
1249
1250 skb_queue_tail(&sk->sk_receive_queue, skb);
1251 sk->sk_data_ready(sk);
1252 return len;
1253 }
1254
1255 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1256 {
1257 int len = __netlink_sendskb(sk, skb);
1258
1259 sock_put(sk);
1260 return len;
1261 }
1262
1263 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1264 {
1265 kfree_skb(skb);
1266 sock_put(sk);
1267 }
1268
1269 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1270 {
1271 int delta;
1272
1273 WARN_ON(skb->sk != NULL);
1274 delta = skb->end - skb->tail;
1275 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1276 return skb;
1277
1278 if (skb_shared(skb)) {
1279 struct sk_buff *nskb = skb_clone(skb, allocation);
1280 if (!nskb)
1281 return skb;
1282 consume_skb(skb);
1283 skb = nskb;
1284 }
1285
1286 pskb_expand_head(skb, 0, -delta,
1287 (allocation & ~__GFP_DIRECT_RECLAIM) |
1288 __GFP_NOWARN | __GFP_NORETRY);
1289 return skb;
1290 }
1291
1292 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1293 struct sock *ssk)
1294 {
1295 int ret;
1296 struct netlink_sock *nlk = nlk_sk(sk);
1297
1298 ret = -ECONNREFUSED;
1299 if (nlk->netlink_rcv != NULL) {
1300 ret = skb->len;
1301 netlink_skb_set_owner_r(skb, sk);
1302 NETLINK_CB(skb).sk = ssk;
1303 netlink_deliver_tap_kernel(sk, ssk, skb);
1304 nlk->netlink_rcv(skb);
1305 consume_skb(skb);
1306 } else {
1307 kfree_skb(skb);
1308 }
1309 sock_put(sk);
1310 return ret;
1311 }
1312
1313 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1314 u32 portid, int nonblock)
1315 {
1316 struct sock *sk;
1317 int err;
1318 long timeo;
1319
1320 skb = netlink_trim(skb, gfp_any());
1321
1322 timeo = sock_sndtimeo(ssk, nonblock);
1323 retry:
1324 sk = netlink_getsockbyportid(ssk, portid);
1325 if (IS_ERR(sk)) {
1326 kfree_skb(skb);
1327 return PTR_ERR(sk);
1328 }
1329 if (netlink_is_kernel(sk))
1330 return netlink_unicast_kernel(sk, skb, ssk);
1331
1332 if (sk_filter(sk, skb)) {
1333 err = skb->len;
1334 kfree_skb(skb);
1335 sock_put(sk);
1336 return err;
1337 }
1338
1339 err = netlink_attachskb(sk, skb, &timeo, ssk);
1340 if (err == 1)
1341 goto retry;
1342 if (err)
1343 return err;
1344
1345 return netlink_sendskb(sk, skb);
1346 }
1347 EXPORT_SYMBOL(netlink_unicast);
1348
1349 int netlink_has_listeners(struct sock *sk, unsigned int group)
1350 {
1351 int res = 0;
1352 struct listeners *listeners;
1353
1354 BUG_ON(!netlink_is_kernel(sk));
1355
1356 rcu_read_lock();
1357 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1358
1359 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1360 res = test_bit(group - 1, listeners->masks);
1361
1362 rcu_read_unlock();
1363
1364 return res;
1365 }
1366 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1367
1368 bool netlink_strict_get_check(struct sk_buff *skb)
1369 {
1370 const struct netlink_sock *nlk = nlk_sk(NETLINK_CB(skb).sk);
1371
1372 return nlk->flags & NETLINK_F_STRICT_CHK;
1373 }
1374 EXPORT_SYMBOL_GPL(netlink_strict_get_check);
1375
1376 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1377 {
1378 struct netlink_sock *nlk = nlk_sk(sk);
1379
1380 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1381 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1382 netlink_skb_set_owner_r(skb, sk);
1383 __netlink_sendskb(sk, skb);
1384 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1385 }
1386 return -1;
1387 }
1388
1389 struct netlink_broadcast_data {
1390 struct sock *exclude_sk;
1391 struct net *net;
1392 u32 portid;
1393 u32 group;
1394 int failure;
1395 int delivery_failure;
1396 int congested;
1397 int delivered;
1398 gfp_t allocation;
1399 struct sk_buff *skb, *skb2;
1400 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1401 void *tx_data;
1402 };
1403
1404 static void do_one_broadcast(struct sock *sk,
1405 struct netlink_broadcast_data *p)
1406 {
1407 struct netlink_sock *nlk = nlk_sk(sk);
1408 int val;
1409
1410 if (p->exclude_sk == sk)
1411 return;
1412
1413 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1414 !test_bit(p->group - 1, nlk->groups))
1415 return;
1416
1417 if (!net_eq(sock_net(sk), p->net)) {
1418 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1419 return;
1420
1421 if (!peernet_has_id(sock_net(sk), p->net))
1422 return;
1423
1424 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1425 CAP_NET_BROADCAST))
1426 return;
1427 }
1428
1429 if (p->failure) {
1430 netlink_overrun(sk);
1431 return;
1432 }
1433
1434 sock_hold(sk);
1435 if (p->skb2 == NULL) {
1436 if (skb_shared(p->skb)) {
1437 p->skb2 = skb_clone(p->skb, p->allocation);
1438 } else {
1439 p->skb2 = skb_get(p->skb);
1440 /*
1441 * skb ownership may have been set when
1442 * delivered to a previous socket.
1443 */
1444 skb_orphan(p->skb2);
1445 }
1446 }
1447 if (p->skb2 == NULL) {
1448 netlink_overrun(sk);
1449 /* Clone failed. Notify ALL listeners. */
1450 p->failure = 1;
1451 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1452 p->delivery_failure = 1;
1453 goto out;
1454 }
1455 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1456 kfree_skb(p->skb2);
1457 p->skb2 = NULL;
1458 goto out;
1459 }
1460 if (sk_filter(sk, p->skb2)) {
1461 kfree_skb(p->skb2);
1462 p->skb2 = NULL;
1463 goto out;
1464 }
1465 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1466 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1467 NETLINK_CB(p->skb2).nsid_is_set = true;
1468 val = netlink_broadcast_deliver(sk, p->skb2);
1469 if (val < 0) {
1470 netlink_overrun(sk);
1471 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1472 p->delivery_failure = 1;
1473 } else {
1474 p->congested |= val;
1475 p->delivered = 1;
1476 p->skb2 = NULL;
1477 }
1478 out:
1479 sock_put(sk);
1480 }
1481
1482 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1483 u32 group, gfp_t allocation,
1484 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1485 void *filter_data)
1486 {
1487 struct net *net = sock_net(ssk);
1488 struct netlink_broadcast_data info;
1489 struct sock *sk;
1490
1491 skb = netlink_trim(skb, allocation);
1492
1493 info.exclude_sk = ssk;
1494 info.net = net;
1495 info.portid = portid;
1496 info.group = group;
1497 info.failure = 0;
1498 info.delivery_failure = 0;
1499 info.congested = 0;
1500 info.delivered = 0;
1501 info.allocation = allocation;
1502 info.skb = skb;
1503 info.skb2 = NULL;
1504 info.tx_filter = filter;
1505 info.tx_data = filter_data;
1506
1507 /* While we sleep in clone, do not allow to change socket list */
1508
1509 netlink_lock_table();
1510
1511 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1512 do_one_broadcast(sk, &info);
1513
1514 consume_skb(skb);
1515
1516 netlink_unlock_table();
1517
1518 if (info.delivery_failure) {
1519 kfree_skb(info.skb2);
1520 return -ENOBUFS;
1521 }
1522 consume_skb(info.skb2);
1523
1524 if (info.delivered) {
1525 if (info.congested && gfpflags_allow_blocking(allocation))
1526 yield();
1527 return 0;
1528 }
1529 return -ESRCH;
1530 }
1531 EXPORT_SYMBOL(netlink_broadcast_filtered);
1532
1533 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1534 u32 group, gfp_t allocation)
1535 {
1536 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1537 NULL, NULL);
1538 }
1539 EXPORT_SYMBOL(netlink_broadcast);
1540
1541 struct netlink_set_err_data {
1542 struct sock *exclude_sk;
1543 u32 portid;
1544 u32 group;
1545 int code;
1546 };
1547
1548 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1549 {
1550 struct netlink_sock *nlk = nlk_sk(sk);
1551 int ret = 0;
1552
1553 if (sk == p->exclude_sk)
1554 goto out;
1555
1556 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1557 goto out;
1558
1559 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1560 !test_bit(p->group - 1, nlk->groups))
1561 goto out;
1562
1563 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1564 ret = 1;
1565 goto out;
1566 }
1567
1568 sk->sk_err = p->code;
1569 sk->sk_error_report(sk);
1570 out:
1571 return ret;
1572 }
1573
1574 /**
1575 * netlink_set_err - report error to broadcast listeners
1576 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1577 * @portid: the PORTID of a process that we want to skip (if any)
1578 * @group: the broadcast group that will notice the error
1579 * @code: error code, must be negative (as usual in kernelspace)
1580 *
1581 * This function returns the number of broadcast listeners that have set the
1582 * NETLINK_NO_ENOBUFS socket option.
1583 */
1584 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1585 {
1586 struct netlink_set_err_data info;
1587 struct sock *sk;
1588 int ret = 0;
1589
1590 info.exclude_sk = ssk;
1591 info.portid = portid;
1592 info.group = group;
1593 /* sk->sk_err wants a positive error value */
1594 info.code = -code;
1595
1596 read_lock(&nl_table_lock);
1597
1598 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1599 ret += do_one_set_err(sk, &info);
1600
1601 read_unlock(&nl_table_lock);
1602 return ret;
1603 }
1604 EXPORT_SYMBOL(netlink_set_err);
1605
1606 /* must be called with netlink table grabbed */
1607 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1608 unsigned int group,
1609 int is_new)
1610 {
1611 int old, new = !!is_new, subscriptions;
1612
1613 old = test_bit(group - 1, nlk->groups);
1614 subscriptions = nlk->subscriptions - old + new;
1615 if (new)
1616 __set_bit(group - 1, nlk->groups);
1617 else
1618 __clear_bit(group - 1, nlk->groups);
1619 netlink_update_subscriptions(&nlk->sk, subscriptions);
1620 netlink_update_listeners(&nlk->sk);
1621 }
1622
1623 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1624 sockptr_t optval, unsigned int optlen)
1625 {
1626 struct sock *sk = sock->sk;
1627 struct netlink_sock *nlk = nlk_sk(sk);
1628 unsigned int val = 0;
1629 int err;
1630
1631 if (level != SOL_NETLINK)
1632 return -ENOPROTOOPT;
1633
1634 if (optlen >= sizeof(int) &&
1635 copy_from_sockptr(&val, optval, sizeof(val)))
1636 return -EFAULT;
1637
1638 switch (optname) {
1639 case NETLINK_PKTINFO:
1640 if (val)
1641 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1642 else
1643 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1644 err = 0;
1645 break;
1646 case NETLINK_ADD_MEMBERSHIP:
1647 case NETLINK_DROP_MEMBERSHIP: {
1648 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1649 return -EPERM;
1650 err = netlink_realloc_groups(sk);
1651 if (err)
1652 return err;
1653 if (!val || val - 1 >= nlk->ngroups)
1654 return -EINVAL;
1655 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1656 err = nlk->netlink_bind(sock_net(sk), val);
1657 if (err)
1658 return err;
1659 }
1660 netlink_table_grab();
1661 netlink_update_socket_mc(nlk, val,
1662 optname == NETLINK_ADD_MEMBERSHIP);
1663 netlink_table_ungrab();
1664 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1665 nlk->netlink_unbind(sock_net(sk), val);
1666
1667 err = 0;
1668 break;
1669 }
1670 case NETLINK_BROADCAST_ERROR:
1671 if (val)
1672 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1673 else
1674 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1675 err = 0;
1676 break;
1677 case NETLINK_NO_ENOBUFS:
1678 if (val) {
1679 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1680 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1681 wake_up_interruptible(&nlk->wait);
1682 } else {
1683 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1684 }
1685 err = 0;
1686 break;
1687 case NETLINK_LISTEN_ALL_NSID:
1688 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1689 return -EPERM;
1690
1691 if (val)
1692 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1693 else
1694 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1695 err = 0;
1696 break;
1697 case NETLINK_CAP_ACK:
1698 if (val)
1699 nlk->flags |= NETLINK_F_CAP_ACK;
1700 else
1701 nlk->flags &= ~NETLINK_F_CAP_ACK;
1702 err = 0;
1703 break;
1704 case NETLINK_EXT_ACK:
1705 if (val)
1706 nlk->flags |= NETLINK_F_EXT_ACK;
1707 else
1708 nlk->flags &= ~NETLINK_F_EXT_ACK;
1709 err = 0;
1710 break;
1711 case NETLINK_GET_STRICT_CHK:
1712 if (val)
1713 nlk->flags |= NETLINK_F_STRICT_CHK;
1714 else
1715 nlk->flags &= ~NETLINK_F_STRICT_CHK;
1716 err = 0;
1717 break;
1718 default:
1719 err = -ENOPROTOOPT;
1720 }
1721 return err;
1722 }
1723
1724 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1725 char __user *optval, int __user *optlen)
1726 {
1727 struct sock *sk = sock->sk;
1728 struct netlink_sock *nlk = nlk_sk(sk);
1729 int len, val, err;
1730
1731 if (level != SOL_NETLINK)
1732 return -ENOPROTOOPT;
1733
1734 if (get_user(len, optlen))
1735 return -EFAULT;
1736 if (len < 0)
1737 return -EINVAL;
1738
1739 switch (optname) {
1740 case NETLINK_PKTINFO:
1741 if (len < sizeof(int))
1742 return -EINVAL;
1743 len = sizeof(int);
1744 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1745 if (put_user(len, optlen) ||
1746 put_user(val, optval))
1747 return -EFAULT;
1748 err = 0;
1749 break;
1750 case NETLINK_BROADCAST_ERROR:
1751 if (len < sizeof(int))
1752 return -EINVAL;
1753 len = sizeof(int);
1754 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1755 if (put_user(len, optlen) ||
1756 put_user(val, optval))
1757 return -EFAULT;
1758 err = 0;
1759 break;
1760 case NETLINK_NO_ENOBUFS:
1761 if (len < sizeof(int))
1762 return -EINVAL;
1763 len = sizeof(int);
1764 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1765 if (put_user(len, optlen) ||
1766 put_user(val, optval))
1767 return -EFAULT;
1768 err = 0;
1769 break;
1770 case NETLINK_LIST_MEMBERSHIPS: {
1771 int pos, idx, shift;
1772
1773 err = 0;
1774 netlink_lock_table();
1775 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1776 if (len - pos < sizeof(u32))
1777 break;
1778
1779 idx = pos / sizeof(unsigned long);
1780 shift = (pos % sizeof(unsigned long)) * 8;
1781 if (put_user((u32)(nlk->groups[idx] >> shift),
1782 (u32 __user *)(optval + pos))) {
1783 err = -EFAULT;
1784 break;
1785 }
1786 }
1787 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1788 err = -EFAULT;
1789 netlink_unlock_table();
1790 break;
1791 }
1792 case NETLINK_CAP_ACK:
1793 if (len < sizeof(int))
1794 return -EINVAL;
1795 len = sizeof(int);
1796 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1797 if (put_user(len, optlen) ||
1798 put_user(val, optval))
1799 return -EFAULT;
1800 err = 0;
1801 break;
1802 case NETLINK_EXT_ACK:
1803 if (len < sizeof(int))
1804 return -EINVAL;
1805 len = sizeof(int);
1806 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1807 if (put_user(len, optlen) || put_user(val, optval))
1808 return -EFAULT;
1809 err = 0;
1810 break;
1811 case NETLINK_GET_STRICT_CHK:
1812 if (len < sizeof(int))
1813 return -EINVAL;
1814 len = sizeof(int);
1815 val = nlk->flags & NETLINK_F_STRICT_CHK ? 1 : 0;
1816 if (put_user(len, optlen) || put_user(val, optval))
1817 return -EFAULT;
1818 err = 0;
1819 break;
1820 default:
1821 err = -ENOPROTOOPT;
1822 }
1823 return err;
1824 }
1825
1826 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1827 {
1828 struct nl_pktinfo info;
1829
1830 info.group = NETLINK_CB(skb).dst_group;
1831 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1832 }
1833
1834 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1835 struct sk_buff *skb)
1836 {
1837 if (!NETLINK_CB(skb).nsid_is_set)
1838 return;
1839
1840 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1841 &NETLINK_CB(skb).nsid);
1842 }
1843
1844 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1845 {
1846 struct sock *sk = sock->sk;
1847 struct netlink_sock *nlk = nlk_sk(sk);
1848 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1849 u32 dst_portid;
1850 u32 dst_group;
1851 struct sk_buff *skb;
1852 int err;
1853 struct scm_cookie scm;
1854 u32 netlink_skb_flags = 0;
1855
1856 if (msg->msg_flags & MSG_OOB)
1857 return -EOPNOTSUPP;
1858
1859 err = scm_send(sock, msg, &scm, true);
1860 if (err < 0)
1861 return err;
1862
1863 if (msg->msg_namelen) {
1864 err = -EINVAL;
1865 if (msg->msg_namelen < sizeof(struct sockaddr_nl))
1866 goto out;
1867 if (addr->nl_family != AF_NETLINK)
1868 goto out;
1869 dst_portid = addr->nl_pid;
1870 dst_group = ffs(addr->nl_groups);
1871 err = -EPERM;
1872 if ((dst_group || dst_portid) &&
1873 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1874 goto out;
1875 netlink_skb_flags |= NETLINK_SKB_DST;
1876 } else {
1877 dst_portid = nlk->dst_portid;
1878 dst_group = nlk->dst_group;
1879 }
1880
1881 if (!nlk->bound) {
1882 err = netlink_autobind(sock);
1883 if (err)
1884 goto out;
1885 } else {
1886 /* Ensure nlk is hashed and visible. */
1887 smp_rmb();
1888 }
1889
1890 err = -EMSGSIZE;
1891 if (len > sk->sk_sndbuf - 32)
1892 goto out;
1893 err = -ENOBUFS;
1894 skb = netlink_alloc_large_skb(len, dst_group);
1895 if (skb == NULL)
1896 goto out;
1897
1898 NETLINK_CB(skb).portid = nlk->portid;
1899 NETLINK_CB(skb).dst_group = dst_group;
1900 NETLINK_CB(skb).creds = scm.creds;
1901 NETLINK_CB(skb).flags = netlink_skb_flags;
1902
1903 err = -EFAULT;
1904 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1905 kfree_skb(skb);
1906 goto out;
1907 }
1908
1909 err = security_netlink_send(sk, skb);
1910 if (err) {
1911 kfree_skb(skb);
1912 goto out;
1913 }
1914
1915 if (dst_group) {
1916 refcount_inc(&skb->users);
1917 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1918 }
1919 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags & MSG_DONTWAIT);
1920
1921 out:
1922 scm_destroy(&scm);
1923 return err;
1924 }
1925
1926 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1927 int flags)
1928 {
1929 struct scm_cookie scm;
1930 struct sock *sk = sock->sk;
1931 struct netlink_sock *nlk = nlk_sk(sk);
1932 int noblock = flags & MSG_DONTWAIT;
1933 size_t copied;
1934 struct sk_buff *skb, *data_skb;
1935 int err, ret;
1936
1937 if (flags & MSG_OOB)
1938 return -EOPNOTSUPP;
1939
1940 copied = 0;
1941
1942 skb = skb_recv_datagram(sk, flags, noblock, &err);
1943 if (skb == NULL)
1944 goto out;
1945
1946 data_skb = skb;
1947
1948 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1949 if (unlikely(skb_shinfo(skb)->frag_list)) {
1950 /*
1951 * If this skb has a frag_list, then here that means that we
1952 * will have to use the frag_list skb's data for compat tasks
1953 * and the regular skb's data for normal (non-compat) tasks.
1954 *
1955 * If we need to send the compat skb, assign it to the
1956 * 'data_skb' variable so that it will be used below for data
1957 * copying. We keep 'skb' for everything else, including
1958 * freeing both later.
1959 */
1960 if (flags & MSG_CMSG_COMPAT)
1961 data_skb = skb_shinfo(skb)->frag_list;
1962 }
1963 #endif
1964
1965 /* Record the max length of recvmsg() calls for future allocations */
1966 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1967 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1968 SKB_WITH_OVERHEAD(32768));
1969
1970 copied = data_skb->len;
1971 if (len < copied) {
1972 msg->msg_flags |= MSG_TRUNC;
1973 copied = len;
1974 }
1975
1976 skb_reset_transport_header(data_skb);
1977 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1978
1979 if (msg->msg_name) {
1980 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1981 addr->nl_family = AF_NETLINK;
1982 addr->nl_pad = 0;
1983 addr->nl_pid = NETLINK_CB(skb).portid;
1984 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1985 msg->msg_namelen = sizeof(*addr);
1986 }
1987
1988 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1989 netlink_cmsg_recv_pktinfo(msg, skb);
1990 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1991 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1992
1993 memset(&scm, 0, sizeof(scm));
1994 scm.creds = *NETLINK_CREDS(skb);
1995 if (flags & MSG_TRUNC)
1996 copied = data_skb->len;
1997
1998 skb_free_datagram(sk, skb);
1999
2000 if (nlk->cb_running &&
2001 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
2002 ret = netlink_dump(sk);
2003 if (ret) {
2004 sk->sk_err = -ret;
2005 sk->sk_error_report(sk);
2006 }
2007 }
2008
2009 scm_recv(sock, msg, &scm, flags);
2010 out:
2011 netlink_rcv_wake(sk);
2012 return err ? : copied;
2013 }
2014
2015 static void netlink_data_ready(struct sock *sk)
2016 {
2017 BUG();
2018 }
2019
2020 /*
2021 * We export these functions to other modules. They provide a
2022 * complete set of kernel non-blocking support for message
2023 * queueing.
2024 */
2025
2026 struct sock *
2027 __netlink_kernel_create(struct net *net, int unit, struct module *module,
2028 struct netlink_kernel_cfg *cfg)
2029 {
2030 struct socket *sock;
2031 struct sock *sk;
2032 struct netlink_sock *nlk;
2033 struct listeners *listeners = NULL;
2034 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
2035 unsigned int groups;
2036
2037 BUG_ON(!nl_table);
2038
2039 if (unit < 0 || unit >= MAX_LINKS)
2040 return NULL;
2041
2042 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
2043 return NULL;
2044
2045 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
2046 goto out_sock_release_nosk;
2047
2048 sk = sock->sk;
2049
2050 if (!cfg || cfg->groups < 32)
2051 groups = 32;
2052 else
2053 groups = cfg->groups;
2054
2055 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2056 if (!listeners)
2057 goto out_sock_release;
2058
2059 sk->sk_data_ready = netlink_data_ready;
2060 if (cfg && cfg->input)
2061 nlk_sk(sk)->netlink_rcv = cfg->input;
2062
2063 if (netlink_insert(sk, 0))
2064 goto out_sock_release;
2065
2066 nlk = nlk_sk(sk);
2067 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2068
2069 netlink_table_grab();
2070 if (!nl_table[unit].registered) {
2071 nl_table[unit].groups = groups;
2072 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2073 nl_table[unit].cb_mutex = cb_mutex;
2074 nl_table[unit].module = module;
2075 if (cfg) {
2076 nl_table[unit].bind = cfg->bind;
2077 nl_table[unit].unbind = cfg->unbind;
2078 nl_table[unit].flags = cfg->flags;
2079 if (cfg->compare)
2080 nl_table[unit].compare = cfg->compare;
2081 }
2082 nl_table[unit].registered = 1;
2083 } else {
2084 kfree(listeners);
2085 nl_table[unit].registered++;
2086 }
2087 netlink_table_ungrab();
2088 return sk;
2089
2090 out_sock_release:
2091 kfree(listeners);
2092 netlink_kernel_release(sk);
2093 return NULL;
2094
2095 out_sock_release_nosk:
2096 sock_release(sock);
2097 return NULL;
2098 }
2099 EXPORT_SYMBOL(__netlink_kernel_create);
2100
2101 void
2102 netlink_kernel_release(struct sock *sk)
2103 {
2104 if (sk == NULL || sk->sk_socket == NULL)
2105 return;
2106
2107 sock_release(sk->sk_socket);
2108 }
2109 EXPORT_SYMBOL(netlink_kernel_release);
2110
2111 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2112 {
2113 struct listeners *new, *old;
2114 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2115
2116 if (groups < 32)
2117 groups = 32;
2118
2119 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2120 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2121 if (!new)
2122 return -ENOMEM;
2123 old = nl_deref_protected(tbl->listeners);
2124 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2125 rcu_assign_pointer(tbl->listeners, new);
2126
2127 kfree_rcu(old, rcu);
2128 }
2129 tbl->groups = groups;
2130
2131 return 0;
2132 }
2133
2134 /**
2135 * netlink_change_ngroups - change number of multicast groups
2136 *
2137 * This changes the number of multicast groups that are available
2138 * on a certain netlink family. Note that it is not possible to
2139 * change the number of groups to below 32. Also note that it does
2140 * not implicitly call netlink_clear_multicast_users() when the
2141 * number of groups is reduced.
2142 *
2143 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2144 * @groups: The new number of groups.
2145 */
2146 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2147 {
2148 int err;
2149
2150 netlink_table_grab();
2151 err = __netlink_change_ngroups(sk, groups);
2152 netlink_table_ungrab();
2153
2154 return err;
2155 }
2156
2157 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2158 {
2159 struct sock *sk;
2160 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2161
2162 sk_for_each_bound(sk, &tbl->mc_list)
2163 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2164 }
2165
2166 struct nlmsghdr *
2167 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2168 {
2169 struct nlmsghdr *nlh;
2170 int size = nlmsg_msg_size(len);
2171
2172 nlh = skb_put(skb, NLMSG_ALIGN(size));
2173 nlh->nlmsg_type = type;
2174 nlh->nlmsg_len = size;
2175 nlh->nlmsg_flags = flags;
2176 nlh->nlmsg_pid = portid;
2177 nlh->nlmsg_seq = seq;
2178 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2179 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2180 return nlh;
2181 }
2182 EXPORT_SYMBOL(__nlmsg_put);
2183
2184 /*
2185 * It looks a bit ugly.
2186 * It would be better to create kernel thread.
2187 */
2188
2189 static int netlink_dump_done(struct netlink_sock *nlk, struct sk_buff *skb,
2190 struct netlink_callback *cb,
2191 struct netlink_ext_ack *extack)
2192 {
2193 struct nlmsghdr *nlh;
2194
2195 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(nlk->dump_done_errno),
2196 NLM_F_MULTI | cb->answer_flags);
2197 if (WARN_ON(!nlh))
2198 return -ENOBUFS;
2199
2200 nl_dump_check_consistent(cb, nlh);
2201 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno, sizeof(nlk->dump_done_errno));
2202
2203 if (extack->_msg && nlk->flags & NETLINK_F_EXT_ACK) {
2204 nlh->nlmsg_flags |= NLM_F_ACK_TLVS;
2205 if (!nla_put_string(skb, NLMSGERR_ATTR_MSG, extack->_msg))
2206 nlmsg_end(skb, nlh);
2207 }
2208
2209 return 0;
2210 }
2211
2212 static int netlink_dump(struct sock *sk)
2213 {
2214 struct netlink_sock *nlk = nlk_sk(sk);
2215 struct netlink_ext_ack extack = {};
2216 struct netlink_callback *cb;
2217 struct sk_buff *skb = NULL;
2218 struct module *module;
2219 int err = -ENOBUFS;
2220 int alloc_min_size;
2221 int alloc_size;
2222
2223 mutex_lock(nlk->cb_mutex);
2224 if (!nlk->cb_running) {
2225 err = -EINVAL;
2226 goto errout_skb;
2227 }
2228
2229 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2230 goto errout_skb;
2231
2232 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2233 * required, but it makes sense to _attempt_ a 16K bytes allocation
2234 * to reduce number of system calls on dump operations, if user
2235 * ever provided a big enough buffer.
2236 */
2237 cb = &nlk->cb;
2238 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2239
2240 if (alloc_min_size < nlk->max_recvmsg_len) {
2241 alloc_size = nlk->max_recvmsg_len;
2242 skb = alloc_skb(alloc_size,
2243 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2244 __GFP_NOWARN | __GFP_NORETRY);
2245 }
2246 if (!skb) {
2247 alloc_size = alloc_min_size;
2248 skb = alloc_skb(alloc_size, GFP_KERNEL);
2249 }
2250 if (!skb)
2251 goto errout_skb;
2252
2253 /* Trim skb to allocated size. User is expected to provide buffer as
2254 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2255 * netlink_recvmsg())). dump will pack as many smaller messages as
2256 * could fit within the allocated skb. skb is typically allocated
2257 * with larger space than required (could be as much as near 2x the
2258 * requested size with align to next power of 2 approach). Allowing
2259 * dump to use the excess space makes it difficult for a user to have a
2260 * reasonable static buffer based on the expected largest dump of a
2261 * single netdev. The outcome is MSG_TRUNC error.
2262 */
2263 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2264 netlink_skb_set_owner_r(skb, sk);
2265
2266 if (nlk->dump_done_errno > 0) {
2267 cb->extack = &extack;
2268 nlk->dump_done_errno = cb->dump(skb, cb);
2269 cb->extack = NULL;
2270 }
2271
2272 if (nlk->dump_done_errno > 0 ||
2273 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2274 mutex_unlock(nlk->cb_mutex);
2275
2276 if (sk_filter(sk, skb))
2277 kfree_skb(skb);
2278 else
2279 __netlink_sendskb(sk, skb);
2280 return 0;
2281 }
2282
2283 if (netlink_dump_done(nlk, skb, cb, &extack))
2284 goto errout_skb;
2285
2286 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
2287 /* frag_list skb's data is used for compat tasks
2288 * and the regular skb's data for normal (non-compat) tasks.
2289 * See netlink_recvmsg().
2290 */
2291 if (unlikely(skb_shinfo(skb)->frag_list)) {
2292 if (netlink_dump_done(nlk, skb_shinfo(skb)->frag_list, cb, &extack))
2293 goto errout_skb;
2294 }
2295 #endif
2296
2297 if (sk_filter(sk, skb))
2298 kfree_skb(skb);
2299 else
2300 __netlink_sendskb(sk, skb);
2301
2302 if (cb->done)
2303 cb->done(cb);
2304
2305 nlk->cb_running = false;
2306 module = cb->module;
2307 skb = cb->skb;
2308 mutex_unlock(nlk->cb_mutex);
2309 module_put(module);
2310 consume_skb(skb);
2311 return 0;
2312
2313 errout_skb:
2314 mutex_unlock(nlk->cb_mutex);
2315 kfree_skb(skb);
2316 return err;
2317 }
2318
2319 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2320 const struct nlmsghdr *nlh,
2321 struct netlink_dump_control *control)
2322 {
2323 struct netlink_sock *nlk, *nlk2;
2324 struct netlink_callback *cb;
2325 struct sock *sk;
2326 int ret;
2327
2328 refcount_inc(&skb->users);
2329
2330 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2331 if (sk == NULL) {
2332 ret = -ECONNREFUSED;
2333 goto error_free;
2334 }
2335
2336 nlk = nlk_sk(sk);
2337 mutex_lock(nlk->cb_mutex);
2338 /* A dump is in progress... */
2339 if (nlk->cb_running) {
2340 ret = -EBUSY;
2341 goto error_unlock;
2342 }
2343 /* add reference of module which cb->dump belongs to */
2344 if (!try_module_get(control->module)) {
2345 ret = -EPROTONOSUPPORT;
2346 goto error_unlock;
2347 }
2348
2349 cb = &nlk->cb;
2350 memset(cb, 0, sizeof(*cb));
2351 cb->dump = control->dump;
2352 cb->done = control->done;
2353 cb->nlh = nlh;
2354 cb->data = control->data;
2355 cb->module = control->module;
2356 cb->min_dump_alloc = control->min_dump_alloc;
2357 cb->skb = skb;
2358
2359 nlk2 = nlk_sk(NETLINK_CB(skb).sk);
2360 cb->strict_check = !!(nlk2->flags & NETLINK_F_STRICT_CHK);
2361
2362 if (control->start) {
2363 ret = control->start(cb);
2364 if (ret)
2365 goto error_put;
2366 }
2367
2368 nlk->cb_running = true;
2369 nlk->dump_done_errno = INT_MAX;
2370
2371 mutex_unlock(nlk->cb_mutex);
2372
2373 ret = netlink_dump(sk);
2374
2375 sock_put(sk);
2376
2377 if (ret)
2378 return ret;
2379
2380 /* We successfully started a dump, by returning -EINTR we
2381 * signal not to send ACK even if it was requested.
2382 */
2383 return -EINTR;
2384
2385 error_put:
2386 module_put(control->module);
2387 error_unlock:
2388 sock_put(sk);
2389 mutex_unlock(nlk->cb_mutex);
2390 error_free:
2391 kfree_skb(skb);
2392 return ret;
2393 }
2394 EXPORT_SYMBOL(__netlink_dump_start);
2395
2396 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2397 const struct netlink_ext_ack *extack)
2398 {
2399 struct sk_buff *skb;
2400 struct nlmsghdr *rep;
2401 struct nlmsgerr *errmsg;
2402 size_t payload = sizeof(*errmsg);
2403 size_t tlvlen = 0;
2404 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2405 unsigned int flags = 0;
2406 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2407
2408 /* Error messages get the original request appened, unless the user
2409 * requests to cap the error message, and get extra error data if
2410 * requested.
2411 */
2412 if (nlk_has_extack && extack && extack->_msg)
2413 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2414
2415 if (err && !(nlk->flags & NETLINK_F_CAP_ACK))
2416 payload += nlmsg_len(nlh);
2417 else
2418 flags |= NLM_F_CAPPED;
2419 if (err && nlk_has_extack && extack && extack->bad_attr)
2420 tlvlen += nla_total_size(sizeof(u32));
2421 if (nlk_has_extack && extack && extack->cookie_len)
2422 tlvlen += nla_total_size(extack->cookie_len);
2423 if (err && nlk_has_extack && extack && extack->policy)
2424 tlvlen += netlink_policy_dump_attr_size_estimate(extack->policy);
2425
2426 if (tlvlen)
2427 flags |= NLM_F_ACK_TLVS;
2428
2429 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2430 if (!skb) {
2431 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2432 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2433 return;
2434 }
2435
2436 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2437 NLMSG_ERROR, payload, flags);
2438 errmsg = nlmsg_data(rep);
2439 errmsg->error = err;
2440 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2441
2442 if (nlk_has_extack && extack) {
2443 if (extack->_msg) {
2444 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2445 extack->_msg));
2446 }
2447 if (err && extack->bad_attr &&
2448 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2449 (u8 *)extack->bad_attr >= in_skb->data +
2450 in_skb->len))
2451 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2452 (u8 *)extack->bad_attr -
2453 (u8 *)nlh));
2454 if (extack->cookie_len)
2455 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2456 extack->cookie_len, extack->cookie));
2457 if (extack->policy)
2458 netlink_policy_dump_write_attr(skb, extack->policy,
2459 NLMSGERR_ATTR_POLICY);
2460 }
2461
2462 nlmsg_end(skb, rep);
2463
2464 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2465 }
2466 EXPORT_SYMBOL(netlink_ack);
2467
2468 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2469 struct nlmsghdr *,
2470 struct netlink_ext_ack *))
2471 {
2472 struct netlink_ext_ack extack;
2473 struct nlmsghdr *nlh;
2474 int err;
2475
2476 while (skb->len >= nlmsg_total_size(0)) {
2477 int msglen;
2478
2479 memset(&extack, 0, sizeof(extack));
2480 nlh = nlmsg_hdr(skb);
2481 err = 0;
2482
2483 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2484 return 0;
2485
2486 /* Only requests are handled by the kernel */
2487 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2488 goto ack;
2489
2490 /* Skip control messages */
2491 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2492 goto ack;
2493
2494 err = cb(skb, nlh, &extack);
2495 if (err == -EINTR)
2496 goto skip;
2497
2498 ack:
2499 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2500 netlink_ack(skb, nlh, err, &extack);
2501
2502 skip:
2503 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2504 if (msglen > skb->len)
2505 msglen = skb->len;
2506 skb_pull(skb, msglen);
2507 }
2508
2509 return 0;
2510 }
2511 EXPORT_SYMBOL(netlink_rcv_skb);
2512
2513 /**
2514 * nlmsg_notify - send a notification netlink message
2515 * @sk: netlink socket to use
2516 * @skb: notification message
2517 * @portid: destination netlink portid for reports or 0
2518 * @group: destination multicast group or 0
2519 * @report: 1 to report back, 0 to disable
2520 * @flags: allocation flags
2521 */
2522 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2523 unsigned int group, int report, gfp_t flags)
2524 {
2525 int err = 0;
2526
2527 if (group) {
2528 int exclude_portid = 0;
2529
2530 if (report) {
2531 refcount_inc(&skb->users);
2532 exclude_portid = portid;
2533 }
2534
2535 /* errors reported via destination sk->sk_err, but propagate
2536 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2537 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2538 }
2539
2540 if (report) {
2541 int err2;
2542
2543 err2 = nlmsg_unicast(sk, skb, portid);
2544 if (!err || err == -ESRCH)
2545 err = err2;
2546 }
2547
2548 return err;
2549 }
2550 EXPORT_SYMBOL(nlmsg_notify);
2551
2552 #ifdef CONFIG_PROC_FS
2553 struct nl_seq_iter {
2554 struct seq_net_private p;
2555 struct rhashtable_iter hti;
2556 int link;
2557 };
2558
2559 static void netlink_walk_start(struct nl_seq_iter *iter)
2560 {
2561 rhashtable_walk_enter(&nl_table[iter->link].hash, &iter->hti);
2562 rhashtable_walk_start(&iter->hti);
2563 }
2564
2565 static void netlink_walk_stop(struct nl_seq_iter *iter)
2566 {
2567 rhashtable_walk_stop(&iter->hti);
2568 rhashtable_walk_exit(&iter->hti);
2569 }
2570
2571 static void *__netlink_seq_next(struct seq_file *seq)
2572 {
2573 struct nl_seq_iter *iter = seq->private;
2574 struct netlink_sock *nlk;
2575
2576 do {
2577 for (;;) {
2578 nlk = rhashtable_walk_next(&iter->hti);
2579
2580 if (IS_ERR(nlk)) {
2581 if (PTR_ERR(nlk) == -EAGAIN)
2582 continue;
2583
2584 return nlk;
2585 }
2586
2587 if (nlk)
2588 break;
2589
2590 netlink_walk_stop(iter);
2591 if (++iter->link >= MAX_LINKS)
2592 return NULL;
2593
2594 netlink_walk_start(iter);
2595 }
2596 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2597
2598 return nlk;
2599 }
2600
2601 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2602 __acquires(RCU)
2603 {
2604 struct nl_seq_iter *iter = seq->private;
2605 void *obj = SEQ_START_TOKEN;
2606 loff_t pos;
2607
2608 iter->link = 0;
2609
2610 netlink_walk_start(iter);
2611
2612 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2613 obj = __netlink_seq_next(seq);
2614
2615 return obj;
2616 }
2617
2618 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2619 {
2620 ++*pos;
2621 return __netlink_seq_next(seq);
2622 }
2623
2624 static void netlink_native_seq_stop(struct seq_file *seq, void *v)
2625 {
2626 struct nl_seq_iter *iter = seq->private;
2627
2628 if (iter->link >= MAX_LINKS)
2629 return;
2630
2631 netlink_walk_stop(iter);
2632 }
2633
2634
2635 static int netlink_native_seq_show(struct seq_file *seq, void *v)
2636 {
2637 if (v == SEQ_START_TOKEN) {
2638 seq_puts(seq,
2639 "sk Eth Pid Groups "
2640 "Rmem Wmem Dump Locks Drops Inode\n");
2641 } else {
2642 struct sock *s = v;
2643 struct netlink_sock *nlk = nlk_sk(s);
2644
2645 seq_printf(seq, "%pK %-3d %-10u %08x %-8d %-8d %-5d %-8d %-8u %-8lu\n",
2646 s,
2647 s->sk_protocol,
2648 nlk->portid,
2649 nlk->groups ? (u32)nlk->groups[0] : 0,
2650 sk_rmem_alloc_get(s),
2651 sk_wmem_alloc_get(s),
2652 nlk->cb_running,
2653 refcount_read(&s->sk_refcnt),
2654 atomic_read(&s->sk_drops),
2655 sock_i_ino(s)
2656 );
2657
2658 }
2659 return 0;
2660 }
2661
2662 #ifdef CONFIG_BPF_SYSCALL
2663 struct bpf_iter__netlink {
2664 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2665 __bpf_md_ptr(struct netlink_sock *, sk);
2666 };
2667
2668 DEFINE_BPF_ITER_FUNC(netlink, struct bpf_iter_meta *meta, struct netlink_sock *sk)
2669
2670 static int netlink_prog_seq_show(struct bpf_prog *prog,
2671 struct bpf_iter_meta *meta,
2672 void *v)
2673 {
2674 struct bpf_iter__netlink ctx;
2675
2676 meta->seq_num--; /* skip SEQ_START_TOKEN */
2677 ctx.meta = meta;
2678 ctx.sk = nlk_sk((struct sock *)v);
2679 return bpf_iter_run_prog(prog, &ctx);
2680 }
2681
2682 static int netlink_seq_show(struct seq_file *seq, void *v)
2683 {
2684 struct bpf_iter_meta meta;
2685 struct bpf_prog *prog;
2686
2687 meta.seq = seq;
2688 prog = bpf_iter_get_info(&meta, false);
2689 if (!prog)
2690 return netlink_native_seq_show(seq, v);
2691
2692 if (v != SEQ_START_TOKEN)
2693 return netlink_prog_seq_show(prog, &meta, v);
2694
2695 return 0;
2696 }
2697
2698 static void netlink_seq_stop(struct seq_file *seq, void *v)
2699 {
2700 struct bpf_iter_meta meta;
2701 struct bpf_prog *prog;
2702
2703 if (!v) {
2704 meta.seq = seq;
2705 prog = bpf_iter_get_info(&meta, true);
2706 if (prog)
2707 (void)netlink_prog_seq_show(prog, &meta, v);
2708 }
2709
2710 netlink_native_seq_stop(seq, v);
2711 }
2712 #else
2713 static int netlink_seq_show(struct seq_file *seq, void *v)
2714 {
2715 return netlink_native_seq_show(seq, v);
2716 }
2717
2718 static void netlink_seq_stop(struct seq_file *seq, void *v)
2719 {
2720 netlink_native_seq_stop(seq, v);
2721 }
2722 #endif
2723
2724 static const struct seq_operations netlink_seq_ops = {
2725 .start = netlink_seq_start,
2726 .next = netlink_seq_next,
2727 .stop = netlink_seq_stop,
2728 .show = netlink_seq_show,
2729 };
2730 #endif
2731
2732 int netlink_register_notifier(struct notifier_block *nb)
2733 {
2734 return blocking_notifier_chain_register(&netlink_chain, nb);
2735 }
2736 EXPORT_SYMBOL(netlink_register_notifier);
2737
2738 int netlink_unregister_notifier(struct notifier_block *nb)
2739 {
2740 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2741 }
2742 EXPORT_SYMBOL(netlink_unregister_notifier);
2743
2744 static const struct proto_ops netlink_ops = {
2745 .family = PF_NETLINK,
2746 .owner = THIS_MODULE,
2747 .release = netlink_release,
2748 .bind = netlink_bind,
2749 .connect = netlink_connect,
2750 .socketpair = sock_no_socketpair,
2751 .accept = sock_no_accept,
2752 .getname = netlink_getname,
2753 .poll = datagram_poll,
2754 .ioctl = netlink_ioctl,
2755 .listen = sock_no_listen,
2756 .shutdown = sock_no_shutdown,
2757 .setsockopt = netlink_setsockopt,
2758 .getsockopt = netlink_getsockopt,
2759 .sendmsg = netlink_sendmsg,
2760 .recvmsg = netlink_recvmsg,
2761 .mmap = sock_no_mmap,
2762 .sendpage = sock_no_sendpage,
2763 };
2764
2765 static const struct net_proto_family netlink_family_ops = {
2766 .family = PF_NETLINK,
2767 .create = netlink_create,
2768 .owner = THIS_MODULE, /* for consistency 8) */
2769 };
2770
2771 static int __net_init netlink_net_init(struct net *net)
2772 {
2773 #ifdef CONFIG_PROC_FS
2774 if (!proc_create_net("netlink", 0, net->proc_net, &netlink_seq_ops,
2775 sizeof(struct nl_seq_iter)))
2776 return -ENOMEM;
2777 #endif
2778 return 0;
2779 }
2780
2781 static void __net_exit netlink_net_exit(struct net *net)
2782 {
2783 #ifdef CONFIG_PROC_FS
2784 remove_proc_entry("netlink", net->proc_net);
2785 #endif
2786 }
2787
2788 static void __init netlink_add_usersock_entry(void)
2789 {
2790 struct listeners *listeners;
2791 int groups = 32;
2792
2793 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2794 if (!listeners)
2795 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2796
2797 netlink_table_grab();
2798
2799 nl_table[NETLINK_USERSOCK].groups = groups;
2800 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2801 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2802 nl_table[NETLINK_USERSOCK].registered = 1;
2803 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2804
2805 netlink_table_ungrab();
2806 }
2807
2808 static struct pernet_operations __net_initdata netlink_net_ops = {
2809 .init = netlink_net_init,
2810 .exit = netlink_net_exit,
2811 };
2812
2813 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2814 {
2815 const struct netlink_sock *nlk = data;
2816 struct netlink_compare_arg arg;
2817
2818 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2819 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2820 }
2821
2822 static const struct rhashtable_params netlink_rhashtable_params = {
2823 .head_offset = offsetof(struct netlink_sock, node),
2824 .key_len = netlink_compare_arg_len,
2825 .obj_hashfn = netlink_hash,
2826 .obj_cmpfn = netlink_compare,
2827 .automatic_shrinking = true,
2828 };
2829
2830 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2831 BTF_ID_LIST(btf_netlink_sock_id)
2832 BTF_ID(struct, netlink_sock)
2833
2834 static const struct bpf_iter_seq_info netlink_seq_info = {
2835 .seq_ops = &netlink_seq_ops,
2836 .init_seq_private = bpf_iter_init_seq_net,
2837 .fini_seq_private = bpf_iter_fini_seq_net,
2838 .seq_priv_size = sizeof(struct nl_seq_iter),
2839 };
2840
2841 static struct bpf_iter_reg netlink_reg_info = {
2842 .target = "netlink",
2843 .ctx_arg_info_size = 1,
2844 .ctx_arg_info = {
2845 { offsetof(struct bpf_iter__netlink, sk),
2846 PTR_TO_BTF_ID_OR_NULL },
2847 },
2848 .seq_info = &netlink_seq_info,
2849 };
2850
2851 static int __init bpf_iter_register(void)
2852 {
2853 netlink_reg_info.ctx_arg_info[0].btf_id = *btf_netlink_sock_id;
2854 return bpf_iter_reg_target(&netlink_reg_info);
2855 }
2856 #endif
2857
2858 static int __init netlink_proto_init(void)
2859 {
2860 int i;
2861 int err = proto_register(&netlink_proto, 0);
2862
2863 if (err != 0)
2864 goto out;
2865
2866 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2867 err = bpf_iter_register();
2868 if (err)
2869 goto out;
2870 #endif
2871
2872 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof_field(struct sk_buff, cb));
2873
2874 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2875 if (!nl_table)
2876 goto panic;
2877
2878 for (i = 0; i < MAX_LINKS; i++) {
2879 if (rhashtable_init(&nl_table[i].hash,
2880 &netlink_rhashtable_params) < 0) {
2881 while (--i > 0)
2882 rhashtable_destroy(&nl_table[i].hash);
2883 kfree(nl_table);
2884 goto panic;
2885 }
2886 }
2887
2888 netlink_add_usersock_entry();
2889
2890 sock_register(&netlink_family_ops);
2891 register_pernet_subsys(&netlink_net_ops);
2892 register_pernet_subsys(&netlink_tap_net_ops);
2893 /* The netlink device handler may be needed early. */
2894 rtnetlink_init();
2895 out:
2896 return err;
2897 panic:
2898 panic("netlink_init: Cannot allocate nl_table\n");
2899 }
2900
2901 core_initcall(netlink_proto_init);
Linux with added FPC-III support