]> git.ipfire.org Git - thirdparty/linux.git/blob - net/netlink/af_netlink.c
projects / thirdparty / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge remote-tracking branches 'asoc/topic/samsung', 'asoc/topic/sgtl5000', 'asoc...
[thirdparty/linux.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
66 #include <net/net_namespace.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[0];
76 };
77
78 /* state bits */
79 #define NETLINK_S_CONGESTED 0x0
80
81 /* flags */
82 #define NETLINK_F_KERNEL_SOCKET 0x1
83 #define NETLINK_F_RECV_PKTINFO 0x2
84 #define NETLINK_F_BROADCAST_SEND_ERROR 0x4
85 #define NETLINK_F_RECV_NO_ENOBUFS 0x8
86 #define NETLINK_F_LISTEN_ALL_NSID 0x10
87 #define NETLINK_F_CAP_ACK 0x20
88
89 static inline int netlink_is_kernel(struct sock *sk)
90 {
91 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
92 }
93
94 struct netlink_table *nl_table __read_mostly;
95 EXPORT_SYMBOL_GPL(nl_table);
96
97 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
98
99 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
100
101 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
102 "nlk_cb_mutex-ROUTE",
103 "nlk_cb_mutex-1",
104 "nlk_cb_mutex-USERSOCK",
105 "nlk_cb_mutex-FIREWALL",
106 "nlk_cb_mutex-SOCK_DIAG",
107 "nlk_cb_mutex-NFLOG",
108 "nlk_cb_mutex-XFRM",
109 "nlk_cb_mutex-SELINUX",
110 "nlk_cb_mutex-ISCSI",
111 "nlk_cb_mutex-AUDIT",
112 "nlk_cb_mutex-FIB_LOOKUP",
113 "nlk_cb_mutex-CONNECTOR",
114 "nlk_cb_mutex-NETFILTER",
115 "nlk_cb_mutex-IP6_FW",
116 "nlk_cb_mutex-DNRTMSG",
117 "nlk_cb_mutex-KOBJECT_UEVENT",
118 "nlk_cb_mutex-GENERIC",
119 "nlk_cb_mutex-17",
120 "nlk_cb_mutex-SCSITRANSPORT",
121 "nlk_cb_mutex-ECRYPTFS",
122 "nlk_cb_mutex-RDMA",
123 "nlk_cb_mutex-CRYPTO",
124 "nlk_cb_mutex-SMC",
125 "nlk_cb_mutex-23",
126 "nlk_cb_mutex-24",
127 "nlk_cb_mutex-25",
128 "nlk_cb_mutex-26",
129 "nlk_cb_mutex-27",
130 "nlk_cb_mutex-28",
131 "nlk_cb_mutex-29",
132 "nlk_cb_mutex-30",
133 "nlk_cb_mutex-31",
134 "nlk_cb_mutex-MAX_LINKS"
135 };
136
137 static int netlink_dump(struct sock *sk);
138 static void netlink_skb_destructor(struct sk_buff *skb);
139
140 /* nl_table locking explained:
141 * Lookup and traversal are protected with an RCU read-side lock. Insertion
142 * and removal are protected with per bucket lock while using RCU list
143 * modification primitives and may run in parallel to RCU protected lookups.
144 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
145 * been acquired * either during or after the socket has been removed from
146 * the list and after an RCU grace period.
147 */
148 DEFINE_RWLOCK(nl_table_lock);
149 EXPORT_SYMBOL_GPL(nl_table_lock);
150 static atomic_t nl_table_users = ATOMIC_INIT(0);
151
152 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
153
154 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
155
156 static DEFINE_SPINLOCK(netlink_tap_lock);
157 static struct list_head netlink_tap_all __read_mostly;
158
159 static const struct rhashtable_params netlink_rhashtable_params;
160
161 static inline u32 netlink_group_mask(u32 group)
162 {
163 return group ? 1 << (group - 1) : 0;
164 }
165
166 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
167 gfp_t gfp_mask)
168 {
169 unsigned int len = skb_end_offset(skb);
170 struct sk_buff *new;
171
172 new = alloc_skb(len, gfp_mask);
173 if (new == NULL)
174 return NULL;
175
176 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
177 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
178 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
179
180 memcpy(skb_put(new, len), skb->data, len);
181 return new;
182 }
183
184 int netlink_add_tap(struct netlink_tap *nt)
185 {
186 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
187 return -EINVAL;
188
189 spin_lock(&netlink_tap_lock);
190 list_add_rcu(&nt->list, &netlink_tap_all);
191 spin_unlock(&netlink_tap_lock);
192
193 __module_get(nt->module);
194
195 return 0;
196 }
197 EXPORT_SYMBOL_GPL(netlink_add_tap);
198
199 static int __netlink_remove_tap(struct netlink_tap *nt)
200 {
201 bool found = false;
202 struct netlink_tap *tmp;
203
204 spin_lock(&netlink_tap_lock);
205
206 list_for_each_entry(tmp, &netlink_tap_all, list) {
207 if (nt == tmp) {
208 list_del_rcu(&nt->list);
209 found = true;
210 goto out;
211 }
212 }
213
214 pr_warn("__netlink_remove_tap: %p not found\n", nt);
215 out:
216 spin_unlock(&netlink_tap_lock);
217
218 if (found)
219 module_put(nt->module);
220
221 return found ? 0 : -ENODEV;
222 }
223
224 int netlink_remove_tap(struct netlink_tap *nt)
225 {
226 int ret;
227
228 ret = __netlink_remove_tap(nt);
229 synchronize_net();
230
231 return ret;
232 }
233 EXPORT_SYMBOL_GPL(netlink_remove_tap);
234
235 static bool netlink_filter_tap(const struct sk_buff *skb)
236 {
237 struct sock *sk = skb->sk;
238
239 /* We take the more conservative approach and
240 * whitelist socket protocols that may pass.
241 */
242 switch (sk->sk_protocol) {
243 case NETLINK_ROUTE:
244 case NETLINK_USERSOCK:
245 case NETLINK_SOCK_DIAG:
246 case NETLINK_NFLOG:
247 case NETLINK_XFRM:
248 case NETLINK_FIB_LOOKUP:
249 case NETLINK_NETFILTER:
250 case NETLINK_GENERIC:
251 return true;
252 }
253
254 return false;
255 }
256
257 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
258 struct net_device *dev)
259 {
260 struct sk_buff *nskb;
261 struct sock *sk = skb->sk;
262 int ret = -ENOMEM;
263
264 dev_hold(dev);
265
266 if (is_vmalloc_addr(skb->head))
267 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
268 else
269 nskb = skb_clone(skb, GFP_ATOMIC);
270 if (nskb) {
271 nskb->dev = dev;
272 nskb->protocol = htons((u16) sk->sk_protocol);
273 nskb->pkt_type = netlink_is_kernel(sk) ?
274 PACKET_KERNEL : PACKET_USER;
275 skb_reset_network_header(nskb);
276 ret = dev_queue_xmit(nskb);
277 if (unlikely(ret > 0))
278 ret = net_xmit_errno(ret);
279 }
280
281 dev_put(dev);
282 return ret;
283 }
284
285 static void __netlink_deliver_tap(struct sk_buff *skb)
286 {
287 int ret;
288 struct netlink_tap *tmp;
289
290 if (!netlink_filter_tap(skb))
291 return;
292
293 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
294 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
295 if (unlikely(ret))
296 break;
297 }
298 }
299
300 static void netlink_deliver_tap(struct sk_buff *skb)
301 {
302 rcu_read_lock();
303
304 if (unlikely(!list_empty(&netlink_tap_all)))
305 __netlink_deliver_tap(skb);
306
307 rcu_read_unlock();
308 }
309
310 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
311 struct sk_buff *skb)
312 {
313 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
314 netlink_deliver_tap(skb);
315 }
316
317 static void netlink_overrun(struct sock *sk)
318 {
319 struct netlink_sock *nlk = nlk_sk(sk);
320
321 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
322 if (!test_and_set_bit(NETLINK_S_CONGESTED,
323 &nlk_sk(sk)->state)) {
324 sk->sk_err = ENOBUFS;
325 sk->sk_error_report(sk);
326 }
327 }
328 atomic_inc(&sk->sk_drops);
329 }
330
331 static void netlink_rcv_wake(struct sock *sk)
332 {
333 struct netlink_sock *nlk = nlk_sk(sk);
334
335 if (skb_queue_empty(&sk->sk_receive_queue))
336 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
337 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
338 wake_up_interruptible(&nlk->wait);
339 }
340
341 static void netlink_skb_destructor(struct sk_buff *skb)
342 {
343 if (is_vmalloc_addr(skb->head)) {
344 if (!skb->cloned ||
345 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
346 vfree(skb->head);
347
348 skb->head = NULL;
349 }
350 if (skb->sk != NULL)
351 sock_rfree(skb);
352 }
353
354 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
355 {
356 WARN_ON(skb->sk != NULL);
357 skb->sk = sk;
358 skb->destructor = netlink_skb_destructor;
359 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
360 sk_mem_charge(sk, skb->truesize);
361 }
362
363 static void netlink_sock_destruct(struct sock *sk)
364 {
365 struct netlink_sock *nlk = nlk_sk(sk);
366
367 if (nlk->cb_running) {
368 if (nlk->cb.done)
369 nlk->cb.done(&nlk->cb);
370 module_put(nlk->cb.module);
371 kfree_skb(nlk->cb.skb);
372 }
373
374 skb_queue_purge(&sk->sk_receive_queue);
375
376 if (!sock_flag(sk, SOCK_DEAD)) {
377 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
378 return;
379 }
380
381 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
382 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
383 WARN_ON(nlk_sk(sk)->groups);
384 }
385
386 static void netlink_sock_destruct_work(struct work_struct *work)
387 {
388 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
389 work);
390
391 sk_free(&nlk->sk);
392 }
393
394 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
395 * SMP. Look, when several writers sleep and reader wakes them up, all but one
396 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
397 * this, _but_ remember, it adds useless work on UP machines.
398 */
399
400 void netlink_table_grab(void)
401 __acquires(nl_table_lock)
402 {
403 might_sleep();
404
405 write_lock_irq(&nl_table_lock);
406
407 if (atomic_read(&nl_table_users)) {
408 DECLARE_WAITQUEUE(wait, current);
409
410 add_wait_queue_exclusive(&nl_table_wait, &wait);
411 for (;;) {
412 set_current_state(TASK_UNINTERRUPTIBLE);
413 if (atomic_read(&nl_table_users) == 0)
414 break;
415 write_unlock_irq(&nl_table_lock);
416 schedule();
417 write_lock_irq(&nl_table_lock);
418 }
419
420 __set_current_state(TASK_RUNNING);
421 remove_wait_queue(&nl_table_wait, &wait);
422 }
423 }
424
425 void netlink_table_ungrab(void)
426 __releases(nl_table_lock)
427 {
428 write_unlock_irq(&nl_table_lock);
429 wake_up(&nl_table_wait);
430 }
431
432 static inline void
433 netlink_lock_table(void)
434 {
435 /* read_lock() synchronizes us to netlink_table_grab */
436
437 read_lock(&nl_table_lock);
438 atomic_inc(&nl_table_users);
439 read_unlock(&nl_table_lock);
440 }
441
442 static inline void
443 netlink_unlock_table(void)
444 {
445 if (atomic_dec_and_test(&nl_table_users))
446 wake_up(&nl_table_wait);
447 }
448
449 struct netlink_compare_arg
450 {
451 possible_net_t pnet;
452 u32 portid;
453 };
454
455 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
456 #define netlink_compare_arg_len \
457 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
458
459 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
460 const void *ptr)
461 {
462 const struct netlink_compare_arg *x = arg->key;
463 const struct netlink_sock *nlk = ptr;
464
465 return nlk->portid != x->portid ||
466 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
467 }
468
469 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
470 struct net *net, u32 portid)
471 {
472 memset(arg, 0, sizeof(*arg));
473 write_pnet(&arg->pnet, net);
474 arg->portid = portid;
475 }
476
477 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
478 struct net *net)
479 {
480 struct netlink_compare_arg arg;
481
482 netlink_compare_arg_init(&arg, net, portid);
483 return rhashtable_lookup_fast(&table->hash, &arg,
484 netlink_rhashtable_params);
485 }
486
487 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
488 {
489 struct netlink_compare_arg arg;
490
491 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
492 return rhashtable_lookup_insert_key(&table->hash, &arg,
493 &nlk_sk(sk)->node,
494 netlink_rhashtable_params);
495 }
496
497 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
498 {
499 struct netlink_table *table = &nl_table[protocol];
500 struct sock *sk;
501
502 rcu_read_lock();
503 sk = __netlink_lookup(table, portid, net);
504 if (sk)
505 sock_hold(sk);
506 rcu_read_unlock();
507
508 return sk;
509 }
510
511 static const struct proto_ops netlink_ops;
512
513 static void
514 netlink_update_listeners(struct sock *sk)
515 {
516 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
517 unsigned long mask;
518 unsigned int i;
519 struct listeners *listeners;
520
521 listeners = nl_deref_protected(tbl->listeners);
522 if (!listeners)
523 return;
524
525 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
526 mask = 0;
527 sk_for_each_bound(sk, &tbl->mc_list) {
528 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
529 mask |= nlk_sk(sk)->groups[i];
530 }
531 listeners->masks[i] = mask;
532 }
533 /* this function is only called with the netlink table "grabbed", which
534 * makes sure updates are visible before bind or setsockopt return. */
535 }
536
537 static int netlink_insert(struct sock *sk, u32 portid)
538 {
539 struct netlink_table *table = &nl_table[sk->sk_protocol];
540 int err;
541
542 lock_sock(sk);
543
544 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
545 if (nlk_sk(sk)->bound)
546 goto err;
547
548 err = -ENOMEM;
549 if (BITS_PER_LONG > 32 &&
550 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
551 goto err;
552
553 nlk_sk(sk)->portid = portid;
554 sock_hold(sk);
555
556 err = __netlink_insert(table, sk);
557 if (err) {
558 /* In case the hashtable backend returns with -EBUSY
559 * from here, it must not escape to the caller.
560 */
561 if (unlikely(err == -EBUSY))
562 err = -EOVERFLOW;
563 if (err == -EEXIST)
564 err = -EADDRINUSE;
565 sock_put(sk);
566 goto err;
567 }
568
569 /* We need to ensure that the socket is hashed and visible. */
570 smp_wmb();
571 nlk_sk(sk)->bound = portid;
572
573 err:
574 release_sock(sk);
575 return err;
576 }
577
578 static void netlink_remove(struct sock *sk)
579 {
580 struct netlink_table *table;
581
582 table = &nl_table[sk->sk_protocol];
583 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
584 netlink_rhashtable_params)) {
585 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
586 __sock_put(sk);
587 }
588
589 netlink_table_grab();
590 if (nlk_sk(sk)->subscriptions) {
591 __sk_del_bind_node(sk);
592 netlink_update_listeners(sk);
593 }
594 if (sk->sk_protocol == NETLINK_GENERIC)
595 atomic_inc(&genl_sk_destructing_cnt);
596 netlink_table_ungrab();
597 }
598
599 static struct proto netlink_proto = {
600 .name = "NETLINK",
601 .owner = THIS_MODULE,
602 .obj_size = sizeof(struct netlink_sock),
603 };
604
605 static int __netlink_create(struct net *net, struct socket *sock,
606 struct mutex *cb_mutex, int protocol,
607 int kern)
608 {
609 struct sock *sk;
610 struct netlink_sock *nlk;
611
612 sock->ops = &netlink_ops;
613
614 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
615 if (!sk)
616 return -ENOMEM;
617
618 sock_init_data(sock, sk);
619
620 nlk = nlk_sk(sk);
621 if (cb_mutex) {
622 nlk->cb_mutex = cb_mutex;
623 } else {
624 nlk->cb_mutex = &nlk->cb_def_mutex;
625 mutex_init(nlk->cb_mutex);
626 lockdep_set_class_and_name(nlk->cb_mutex,
627 nlk_cb_mutex_keys + protocol,
628 nlk_cb_mutex_key_strings[protocol]);
629 }
630 init_waitqueue_head(&nlk->wait);
631
632 sk->sk_destruct = netlink_sock_destruct;
633 sk->sk_protocol = protocol;
634 return 0;
635 }
636
637 static int netlink_create(struct net *net, struct socket *sock, int protocol,
638 int kern)
639 {
640 struct module *module = NULL;
641 struct mutex *cb_mutex;
642 struct netlink_sock *nlk;
643 int (*bind)(struct net *net, int group);
644 void (*unbind)(struct net *net, int group);
645 int err = 0;
646
647 sock->state = SS_UNCONNECTED;
648
649 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
650 return -ESOCKTNOSUPPORT;
651
652 if (protocol < 0 || protocol >= MAX_LINKS)
653 return -EPROTONOSUPPORT;
654
655 netlink_lock_table();
656 #ifdef CONFIG_MODULES
657 if (!nl_table[protocol].registered) {
658 netlink_unlock_table();
659 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
660 netlink_lock_table();
661 }
662 #endif
663 if (nl_table[protocol].registered &&
664 try_module_get(nl_table[protocol].module))
665 module = nl_table[protocol].module;
666 else
667 err = -EPROTONOSUPPORT;
668 cb_mutex = nl_table[protocol].cb_mutex;
669 bind = nl_table[protocol].bind;
670 unbind = nl_table[protocol].unbind;
671 netlink_unlock_table();
672
673 if (err < 0)
674 goto out;
675
676 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
677 if (err < 0)
678 goto out_module;
679
680 local_bh_disable();
681 sock_prot_inuse_add(net, &netlink_proto, 1);
682 local_bh_enable();
683
684 nlk = nlk_sk(sock->sk);
685 nlk->module = module;
686 nlk->netlink_bind = bind;
687 nlk->netlink_unbind = unbind;
688 out:
689 return err;
690
691 out_module:
692 module_put(module);
693 goto out;
694 }
695
696 static void deferred_put_nlk_sk(struct rcu_head *head)
697 {
698 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
699 struct sock *sk = &nlk->sk;
700
701 if (!atomic_dec_and_test(&sk->sk_refcnt))
702 return;
703
704 if (nlk->cb_running && nlk->cb.done) {
705 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
706 schedule_work(&nlk->work);
707 return;
708 }
709
710 sk_free(sk);
711 }
712
713 static int netlink_release(struct socket *sock)
714 {
715 struct sock *sk = sock->sk;
716 struct netlink_sock *nlk;
717
718 if (!sk)
719 return 0;
720
721 netlink_remove(sk);
722 sock_orphan(sk);
723 nlk = nlk_sk(sk);
724
725 /*
726 * OK. Socket is unlinked, any packets that arrive now
727 * will be purged.
728 */
729
730 /* must not acquire netlink_table_lock in any way again before unbind
731 * and notifying genetlink is done as otherwise it might deadlock
732 */
733 if (nlk->netlink_unbind) {
734 int i;
735
736 for (i = 0; i < nlk->ngroups; i++)
737 if (test_bit(i, nlk->groups))
738 nlk->netlink_unbind(sock_net(sk), i + 1);
739 }
740 if (sk->sk_protocol == NETLINK_GENERIC &&
741 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
742 wake_up(&genl_sk_destructing_waitq);
743
744 sock->sk = NULL;
745 wake_up_interruptible_all(&nlk->wait);
746
747 skb_queue_purge(&sk->sk_write_queue);
748
749 if (nlk->portid && nlk->bound) {
750 struct netlink_notify n = {
751 .net = sock_net(sk),
752 .protocol = sk->sk_protocol,
753 .portid = nlk->portid,
754 };
755 blocking_notifier_call_chain(&netlink_chain,
756 NETLINK_URELEASE, &n);
757 }
758
759 module_put(nlk->module);
760
761 if (netlink_is_kernel(sk)) {
762 netlink_table_grab();
763 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
764 if (--nl_table[sk->sk_protocol].registered == 0) {
765 struct listeners *old;
766
767 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
768 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
769 kfree_rcu(old, rcu);
770 nl_table[sk->sk_protocol].module = NULL;
771 nl_table[sk->sk_protocol].bind = NULL;
772 nl_table[sk->sk_protocol].unbind = NULL;
773 nl_table[sk->sk_protocol].flags = 0;
774 nl_table[sk->sk_protocol].registered = 0;
775 }
776 netlink_table_ungrab();
777 }
778
779 kfree(nlk->groups);
780 nlk->groups = NULL;
781
782 local_bh_disable();
783 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
784 local_bh_enable();
785 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
786 return 0;
787 }
788
789 static int netlink_autobind(struct socket *sock)
790 {
791 struct sock *sk = sock->sk;
792 struct net *net = sock_net(sk);
793 struct netlink_table *table = &nl_table[sk->sk_protocol];
794 s32 portid = task_tgid_vnr(current);
795 int err;
796 s32 rover = -4096;
797 bool ok;
798
799 retry:
800 cond_resched();
801 rcu_read_lock();
802 ok = !__netlink_lookup(table, portid, net);
803 rcu_read_unlock();
804 if (!ok) {
805 /* Bind collision, search negative portid values. */
806 if (rover == -4096)
807 /* rover will be in range [S32_MIN, -4097] */
808 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
809 else if (rover >= -4096)
810 rover = -4097;
811 portid = rover--;
812 goto retry;
813 }
814
815 err = netlink_insert(sk, portid);
816 if (err == -EADDRINUSE)
817 goto retry;
818
819 /* If 2 threads race to autobind, that is fine. */
820 if (err == -EBUSY)
821 err = 0;
822
823 return err;
824 }
825
826 /**
827 * __netlink_ns_capable - General netlink message capability test
828 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
829 * @user_ns: The user namespace of the capability to use
830 * @cap: The capability to use
831 *
832 * Test to see if the opener of the socket we received the message
833 * from had when the netlink socket was created and the sender of the
834 * message has has the capability @cap in the user namespace @user_ns.
835 */
836 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
837 struct user_namespace *user_ns, int cap)
838 {
839 return ((nsp->flags & NETLINK_SKB_DST) ||
840 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
841 ns_capable(user_ns, cap);
842 }
843 EXPORT_SYMBOL(__netlink_ns_capable);
844
845 /**
846 * netlink_ns_capable - General netlink message capability test
847 * @skb: socket buffer holding a netlink command from userspace
848 * @user_ns: The user namespace of the capability to use
849 * @cap: The capability to use
850 *
851 * Test to see if the opener of the socket we received the message
852 * from had when the netlink socket was created and the sender of the
853 * message has has the capability @cap in the user namespace @user_ns.
854 */
855 bool netlink_ns_capable(const struct sk_buff *skb,
856 struct user_namespace *user_ns, int cap)
857 {
858 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
859 }
860 EXPORT_SYMBOL(netlink_ns_capable);
861
862 /**
863 * netlink_capable - Netlink global message capability test
864 * @skb: socket buffer holding a netlink command from userspace
865 * @cap: The capability to use
866 *
867 * Test to see if the opener of the socket we received the message
868 * from had when the netlink socket was created and the sender of the
869 * message has has the capability @cap in all user namespaces.
870 */
871 bool netlink_capable(const struct sk_buff *skb, int cap)
872 {
873 return netlink_ns_capable(skb, &init_user_ns, cap);
874 }
875 EXPORT_SYMBOL(netlink_capable);
876
877 /**
878 * netlink_net_capable - Netlink network namespace message capability test
879 * @skb: socket buffer holding a netlink command from userspace
880 * @cap: The capability to use
881 *
882 * Test to see if the opener of the socket we received the message
883 * from had when the netlink socket was created and the sender of the
884 * message has has the capability @cap over the network namespace of
885 * the socket we received the message from.
886 */
887 bool netlink_net_capable(const struct sk_buff *skb, int cap)
888 {
889 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
890 }
891 EXPORT_SYMBOL(netlink_net_capable);
892
893 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
894 {
895 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
896 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
897 }
898
899 static void
900 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
901 {
902 struct netlink_sock *nlk = nlk_sk(sk);
903
904 if (nlk->subscriptions && !subscriptions)
905 __sk_del_bind_node(sk);
906 else if (!nlk->subscriptions && subscriptions)
907 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
908 nlk->subscriptions = subscriptions;
909 }
910
911 static int netlink_realloc_groups(struct sock *sk)
912 {
913 struct netlink_sock *nlk = nlk_sk(sk);
914 unsigned int groups;
915 unsigned long *new_groups;
916 int err = 0;
917
918 netlink_table_grab();
919
920 groups = nl_table[sk->sk_protocol].groups;
921 if (!nl_table[sk->sk_protocol].registered) {
922 err = -ENOENT;
923 goto out_unlock;
924 }
925
926 if (nlk->ngroups >= groups)
927 goto out_unlock;
928
929 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
930 if (new_groups == NULL) {
931 err = -ENOMEM;
932 goto out_unlock;
933 }
934 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
935 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
936
937 nlk->groups = new_groups;
938 nlk->ngroups = groups;
939 out_unlock:
940 netlink_table_ungrab();
941 return err;
942 }
943
944 static void netlink_undo_bind(int group, long unsigned int groups,
945 struct sock *sk)
946 {
947 struct netlink_sock *nlk = nlk_sk(sk);
948 int undo;
949
950 if (!nlk->netlink_unbind)
951 return;
952
953 for (undo = 0; undo < group; undo++)
954 if (test_bit(undo, &groups))
955 nlk->netlink_unbind(sock_net(sk), undo + 1);
956 }
957
958 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
959 int addr_len)
960 {
961 struct sock *sk = sock->sk;
962 struct net *net = sock_net(sk);
963 struct netlink_sock *nlk = nlk_sk(sk);
964 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
965 int err;
966 long unsigned int groups = nladdr->nl_groups;
967 bool bound;
968
969 if (addr_len < sizeof(struct sockaddr_nl))
970 return -EINVAL;
971
972 if (nladdr->nl_family != AF_NETLINK)
973 return -EINVAL;
974
975 /* Only superuser is allowed to listen multicasts */
976 if (groups) {
977 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
978 return -EPERM;
979 err = netlink_realloc_groups(sk);
980 if (err)
981 return err;
982 }
983
984 bound = nlk->bound;
985 if (bound) {
986 /* Ensure nlk->portid is up-to-date. */
987 smp_rmb();
988
989 if (nladdr->nl_pid != nlk->portid)
990 return -EINVAL;
991 }
992
993 if (nlk->netlink_bind && groups) {
994 int group;
995
996 for (group = 0; group < nlk->ngroups; group++) {
997 if (!test_bit(group, &groups))
998 continue;
999 err = nlk->netlink_bind(net, group + 1);
1000 if (!err)
1001 continue;
1002 netlink_undo_bind(group, groups, sk);
1003 return err;
1004 }
1005 }
1006
1007 /* No need for barriers here as we return to user-space without
1008 * using any of the bound attributes.
1009 */
1010 if (!bound) {
1011 err = nladdr->nl_pid ?
1012 netlink_insert(sk, nladdr->nl_pid) :
1013 netlink_autobind(sock);
1014 if (err) {
1015 netlink_undo_bind(nlk->ngroups, groups, sk);
1016 return err;
1017 }
1018 }
1019
1020 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1021 return 0;
1022
1023 netlink_table_grab();
1024 netlink_update_subscriptions(sk, nlk->subscriptions +
1025 hweight32(groups) -
1026 hweight32(nlk->groups[0]));
1027 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1028 netlink_update_listeners(sk);
1029 netlink_table_ungrab();
1030
1031 return 0;
1032 }
1033
1034 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1035 int alen, int flags)
1036 {
1037 int err = 0;
1038 struct sock *sk = sock->sk;
1039 struct netlink_sock *nlk = nlk_sk(sk);
1040 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1041
1042 if (alen < sizeof(addr->sa_family))
1043 return -EINVAL;
1044
1045 if (addr->sa_family == AF_UNSPEC) {
1046 sk->sk_state = NETLINK_UNCONNECTED;
1047 nlk->dst_portid = 0;
1048 nlk->dst_group = 0;
1049 return 0;
1050 }
1051 if (addr->sa_family != AF_NETLINK)
1052 return -EINVAL;
1053
1054 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1055 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1056 return -EPERM;
1057
1058 /* No need for barriers here as we return to user-space without
1059 * using any of the bound attributes.
1060 */
1061 if (!nlk->bound)
1062 err = netlink_autobind(sock);
1063
1064 if (err == 0) {
1065 sk->sk_state = NETLINK_CONNECTED;
1066 nlk->dst_portid = nladdr->nl_pid;
1067 nlk->dst_group = ffs(nladdr->nl_groups);
1068 }
1069
1070 return err;
1071 }
1072
1073 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1074 int *addr_len, int peer)
1075 {
1076 struct sock *sk = sock->sk;
1077 struct netlink_sock *nlk = nlk_sk(sk);
1078 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1079
1080 nladdr->nl_family = AF_NETLINK;
1081 nladdr->nl_pad = 0;
1082 *addr_len = sizeof(*nladdr);
1083
1084 if (peer) {
1085 nladdr->nl_pid = nlk->dst_portid;
1086 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1087 } else {
1088 nladdr->nl_pid = nlk->portid;
1089 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1090 }
1091 return 0;
1092 }
1093
1094 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1095 unsigned long arg)
1096 {
1097 /* try to hand this ioctl down to the NIC drivers.
1098 */
1099 return -ENOIOCTLCMD;
1100 }
1101
1102 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1103 {
1104 struct sock *sock;
1105 struct netlink_sock *nlk;
1106
1107 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1108 if (!sock)
1109 return ERR_PTR(-ECONNREFUSED);
1110
1111 /* Don't bother queuing skb if kernel socket has no input function */
1112 nlk = nlk_sk(sock);
1113 if (sock->sk_state == NETLINK_CONNECTED &&
1114 nlk->dst_portid != nlk_sk(ssk)->portid) {
1115 sock_put(sock);
1116 return ERR_PTR(-ECONNREFUSED);
1117 }
1118 return sock;
1119 }
1120
1121 struct sock *netlink_getsockbyfilp(struct file *filp)
1122 {
1123 struct inode *inode = file_inode(filp);
1124 struct sock *sock;
1125
1126 if (!S_ISSOCK(inode->i_mode))
1127 return ERR_PTR(-ENOTSOCK);
1128
1129 sock = SOCKET_I(inode)->sk;
1130 if (sock->sk_family != AF_NETLINK)
1131 return ERR_PTR(-EINVAL);
1132
1133 sock_hold(sock);
1134 return sock;
1135 }
1136
1137 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1138 int broadcast)
1139 {
1140 struct sk_buff *skb;
1141 void *data;
1142
1143 if (size <= NLMSG_GOODSIZE || broadcast)
1144 return alloc_skb(size, GFP_KERNEL);
1145
1146 size = SKB_DATA_ALIGN(size) +
1147 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1148
1149 data = vmalloc(size);
1150 if (data == NULL)
1151 return NULL;
1152
1153 skb = __build_skb(data, size);
1154 if (skb == NULL)
1155 vfree(data);
1156 else
1157 skb->destructor = netlink_skb_destructor;
1158
1159 return skb;
1160 }
1161
1162 /*
1163 * Attach a skb to a netlink socket.
1164 * The caller must hold a reference to the destination socket. On error, the
1165 * reference is dropped. The skb is not send to the destination, just all
1166 * all error checks are performed and memory in the queue is reserved.
1167 * Return values:
1168 * < 0: error. skb freed, reference to sock dropped.
1169 * 0: continue
1170 * 1: repeat lookup - reference dropped while waiting for socket memory.
1171 */
1172 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1173 long *timeo, struct sock *ssk)
1174 {
1175 struct netlink_sock *nlk;
1176
1177 nlk = nlk_sk(sk);
1178
1179 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1180 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1181 DECLARE_WAITQUEUE(wait, current);
1182 if (!*timeo) {
1183 if (!ssk || netlink_is_kernel(ssk))
1184 netlink_overrun(sk);
1185 sock_put(sk);
1186 kfree_skb(skb);
1187 return -EAGAIN;
1188 }
1189
1190 __set_current_state(TASK_INTERRUPTIBLE);
1191 add_wait_queue(&nlk->wait, &wait);
1192
1193 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1194 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1195 !sock_flag(sk, SOCK_DEAD))
1196 *timeo = schedule_timeout(*timeo);
1197
1198 __set_current_state(TASK_RUNNING);
1199 remove_wait_queue(&nlk->wait, &wait);
1200 sock_put(sk);
1201
1202 if (signal_pending(current)) {
1203 kfree_skb(skb);
1204 return sock_intr_errno(*timeo);
1205 }
1206 return 1;
1207 }
1208 netlink_skb_set_owner_r(skb, sk);
1209 return 0;
1210 }
1211
1212 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1213 {
1214 int len = skb->len;
1215
1216 netlink_deliver_tap(skb);
1217
1218 skb_queue_tail(&sk->sk_receive_queue, skb);
1219 sk->sk_data_ready(sk);
1220 return len;
1221 }
1222
1223 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1224 {
1225 int len = __netlink_sendskb(sk, skb);
1226
1227 sock_put(sk);
1228 return len;
1229 }
1230
1231 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1232 {
1233 kfree_skb(skb);
1234 sock_put(sk);
1235 }
1236
1237 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1238 {
1239 int delta;
1240
1241 WARN_ON(skb->sk != NULL);
1242 delta = skb->end - skb->tail;
1243 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1244 return skb;
1245
1246 if (skb_shared(skb)) {
1247 struct sk_buff *nskb = skb_clone(skb, allocation);
1248 if (!nskb)
1249 return skb;
1250 consume_skb(skb);
1251 skb = nskb;
1252 }
1253
1254 pskb_expand_head(skb, 0, -delta,
1255 (allocation & ~__GFP_DIRECT_RECLAIM) |
1256 __GFP_NOWARN | __GFP_NORETRY);
1257 return skb;
1258 }
1259
1260 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1261 struct sock *ssk)
1262 {
1263 int ret;
1264 struct netlink_sock *nlk = nlk_sk(sk);
1265
1266 ret = -ECONNREFUSED;
1267 if (nlk->netlink_rcv != NULL) {
1268 ret = skb->len;
1269 netlink_skb_set_owner_r(skb, sk);
1270 NETLINK_CB(skb).sk = ssk;
1271 netlink_deliver_tap_kernel(sk, ssk, skb);
1272 nlk->netlink_rcv(skb);
1273 consume_skb(skb);
1274 } else {
1275 kfree_skb(skb);
1276 }
1277 sock_put(sk);
1278 return ret;
1279 }
1280
1281 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1282 u32 portid, int nonblock)
1283 {
1284 struct sock *sk;
1285 int err;
1286 long timeo;
1287
1288 skb = netlink_trim(skb, gfp_any());
1289
1290 timeo = sock_sndtimeo(ssk, nonblock);
1291 retry:
1292 sk = netlink_getsockbyportid(ssk, portid);
1293 if (IS_ERR(sk)) {
1294 kfree_skb(skb);
1295 return PTR_ERR(sk);
1296 }
1297 if (netlink_is_kernel(sk))
1298 return netlink_unicast_kernel(sk, skb, ssk);
1299
1300 if (sk_filter(sk, skb)) {
1301 err = skb->len;
1302 kfree_skb(skb);
1303 sock_put(sk);
1304 return err;
1305 }
1306
1307 err = netlink_attachskb(sk, skb, &timeo, ssk);
1308 if (err == 1)
1309 goto retry;
1310 if (err)
1311 return err;
1312
1313 return netlink_sendskb(sk, skb);
1314 }
1315 EXPORT_SYMBOL(netlink_unicast);
1316
1317 int netlink_has_listeners(struct sock *sk, unsigned int group)
1318 {
1319 int res = 0;
1320 struct listeners *listeners;
1321
1322 BUG_ON(!netlink_is_kernel(sk));
1323
1324 rcu_read_lock();
1325 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1326
1327 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1328 res = test_bit(group - 1, listeners->masks);
1329
1330 rcu_read_unlock();
1331
1332 return res;
1333 }
1334 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1335
1336 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1337 {
1338 struct netlink_sock *nlk = nlk_sk(sk);
1339
1340 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1341 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1342 netlink_skb_set_owner_r(skb, sk);
1343 __netlink_sendskb(sk, skb);
1344 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1345 }
1346 return -1;
1347 }
1348
1349 struct netlink_broadcast_data {
1350 struct sock *exclude_sk;
1351 struct net *net;
1352 u32 portid;
1353 u32 group;
1354 int failure;
1355 int delivery_failure;
1356 int congested;
1357 int delivered;
1358 gfp_t allocation;
1359 struct sk_buff *skb, *skb2;
1360 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1361 void *tx_data;
1362 };
1363
1364 static void do_one_broadcast(struct sock *sk,
1365 struct netlink_broadcast_data *p)
1366 {
1367 struct netlink_sock *nlk = nlk_sk(sk);
1368 int val;
1369
1370 if (p->exclude_sk == sk)
1371 return;
1372
1373 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1374 !test_bit(p->group - 1, nlk->groups))
1375 return;
1376
1377 if (!net_eq(sock_net(sk), p->net)) {
1378 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1379 return;
1380
1381 if (!peernet_has_id(sock_net(sk), p->net))
1382 return;
1383
1384 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1385 CAP_NET_BROADCAST))
1386 return;
1387 }
1388
1389 if (p->failure) {
1390 netlink_overrun(sk);
1391 return;
1392 }
1393
1394 sock_hold(sk);
1395 if (p->skb2 == NULL) {
1396 if (skb_shared(p->skb)) {
1397 p->skb2 = skb_clone(p->skb, p->allocation);
1398 } else {
1399 p->skb2 = skb_get(p->skb);
1400 /*
1401 * skb ownership may have been set when
1402 * delivered to a previous socket.
1403 */
1404 skb_orphan(p->skb2);
1405 }
1406 }
1407 if (p->skb2 == NULL) {
1408 netlink_overrun(sk);
1409 /* Clone failed. Notify ALL listeners. */
1410 p->failure = 1;
1411 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1412 p->delivery_failure = 1;
1413 goto out;
1414 }
1415 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1416 kfree_skb(p->skb2);
1417 p->skb2 = NULL;
1418 goto out;
1419 }
1420 if (sk_filter(sk, p->skb2)) {
1421 kfree_skb(p->skb2);
1422 p->skb2 = NULL;
1423 goto out;
1424 }
1425 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1426 NETLINK_CB(p->skb2).nsid_is_set = true;
1427 val = netlink_broadcast_deliver(sk, p->skb2);
1428 if (val < 0) {
1429 netlink_overrun(sk);
1430 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1431 p->delivery_failure = 1;
1432 } else {
1433 p->congested |= val;
1434 p->delivered = 1;
1435 p->skb2 = NULL;
1436 }
1437 out:
1438 sock_put(sk);
1439 }
1440
1441 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1442 u32 group, gfp_t allocation,
1443 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1444 void *filter_data)
1445 {
1446 struct net *net = sock_net(ssk);
1447 struct netlink_broadcast_data info;
1448 struct sock *sk;
1449
1450 skb = netlink_trim(skb, allocation);
1451
1452 info.exclude_sk = ssk;
1453 info.net = net;
1454 info.portid = portid;
1455 info.group = group;
1456 info.failure = 0;
1457 info.delivery_failure = 0;
1458 info.congested = 0;
1459 info.delivered = 0;
1460 info.allocation = allocation;
1461 info.skb = skb;
1462 info.skb2 = NULL;
1463 info.tx_filter = filter;
1464 info.tx_data = filter_data;
1465
1466 /* While we sleep in clone, do not allow to change socket list */
1467
1468 netlink_lock_table();
1469
1470 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1471 do_one_broadcast(sk, &info);
1472
1473 consume_skb(skb);
1474
1475 netlink_unlock_table();
1476
1477 if (info.delivery_failure) {
1478 kfree_skb(info.skb2);
1479 return -ENOBUFS;
1480 }
1481 consume_skb(info.skb2);
1482
1483 if (info.delivered) {
1484 if (info.congested && gfpflags_allow_blocking(allocation))
1485 yield();
1486 return 0;
1487 }
1488 return -ESRCH;
1489 }
1490 EXPORT_SYMBOL(netlink_broadcast_filtered);
1491
1492 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1493 u32 group, gfp_t allocation)
1494 {
1495 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1496 NULL, NULL);
1497 }
1498 EXPORT_SYMBOL(netlink_broadcast);
1499
1500 struct netlink_set_err_data {
1501 struct sock *exclude_sk;
1502 u32 portid;
1503 u32 group;
1504 int code;
1505 };
1506
1507 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1508 {
1509 struct netlink_sock *nlk = nlk_sk(sk);
1510 int ret = 0;
1511
1512 if (sk == p->exclude_sk)
1513 goto out;
1514
1515 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1516 goto out;
1517
1518 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1519 !test_bit(p->group - 1, nlk->groups))
1520 goto out;
1521
1522 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1523 ret = 1;
1524 goto out;
1525 }
1526
1527 sk->sk_err = p->code;
1528 sk->sk_error_report(sk);
1529 out:
1530 return ret;
1531 }
1532
1533 /**
1534 * netlink_set_err - report error to broadcast listeners
1535 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1536 * @portid: the PORTID of a process that we want to skip (if any)
1537 * @group: the broadcast group that will notice the error
1538 * @code: error code, must be negative (as usual in kernelspace)
1539 *
1540 * This function returns the number of broadcast listeners that have set the
1541 * NETLINK_NO_ENOBUFS socket option.
1542 */
1543 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1544 {
1545 struct netlink_set_err_data info;
1546 struct sock *sk;
1547 int ret = 0;
1548
1549 info.exclude_sk = ssk;
1550 info.portid = portid;
1551 info.group = group;
1552 /* sk->sk_err wants a positive error value */
1553 info.code = -code;
1554
1555 read_lock(&nl_table_lock);
1556
1557 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1558 ret += do_one_set_err(sk, &info);
1559
1560 read_unlock(&nl_table_lock);
1561 return ret;
1562 }
1563 EXPORT_SYMBOL(netlink_set_err);
1564
1565 /* must be called with netlink table grabbed */
1566 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1567 unsigned int group,
1568 int is_new)
1569 {
1570 int old, new = !!is_new, subscriptions;
1571
1572 old = test_bit(group - 1, nlk->groups);
1573 subscriptions = nlk->subscriptions - old + new;
1574 if (new)
1575 __set_bit(group - 1, nlk->groups);
1576 else
1577 __clear_bit(group - 1, nlk->groups);
1578 netlink_update_subscriptions(&nlk->sk, subscriptions);
1579 netlink_update_listeners(&nlk->sk);
1580 }
1581
1582 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1583 char __user *optval, unsigned int optlen)
1584 {
1585 struct sock *sk = sock->sk;
1586 struct netlink_sock *nlk = nlk_sk(sk);
1587 unsigned int val = 0;
1588 int err;
1589
1590 if (level != SOL_NETLINK)
1591 return -ENOPROTOOPT;
1592
1593 if (optlen >= sizeof(int) &&
1594 get_user(val, (unsigned int __user *)optval))
1595 return -EFAULT;
1596
1597 switch (optname) {
1598 case NETLINK_PKTINFO:
1599 if (val)
1600 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1601 else
1602 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1603 err = 0;
1604 break;
1605 case NETLINK_ADD_MEMBERSHIP:
1606 case NETLINK_DROP_MEMBERSHIP: {
1607 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1608 return -EPERM;
1609 err = netlink_realloc_groups(sk);
1610 if (err)
1611 return err;
1612 if (!val || val - 1 >= nlk->ngroups)
1613 return -EINVAL;
1614 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1615 err = nlk->netlink_bind(sock_net(sk), val);
1616 if (err)
1617 return err;
1618 }
1619 netlink_table_grab();
1620 netlink_update_socket_mc(nlk, val,
1621 optname == NETLINK_ADD_MEMBERSHIP);
1622 netlink_table_ungrab();
1623 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1624 nlk->netlink_unbind(sock_net(sk), val);
1625
1626 err = 0;
1627 break;
1628 }
1629 case NETLINK_BROADCAST_ERROR:
1630 if (val)
1631 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1632 else
1633 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1634 err = 0;
1635 break;
1636 case NETLINK_NO_ENOBUFS:
1637 if (val) {
1638 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1639 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1640 wake_up_interruptible(&nlk->wait);
1641 } else {
1642 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1643 }
1644 err = 0;
1645 break;
1646 case NETLINK_LISTEN_ALL_NSID:
1647 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1648 return -EPERM;
1649
1650 if (val)
1651 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1652 else
1653 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1654 err = 0;
1655 break;
1656 case NETLINK_CAP_ACK:
1657 if (val)
1658 nlk->flags |= NETLINK_F_CAP_ACK;
1659 else
1660 nlk->flags &= ~NETLINK_F_CAP_ACK;
1661 err = 0;
1662 break;
1663 default:
1664 err = -ENOPROTOOPT;
1665 }
1666 return err;
1667 }
1668
1669 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1670 char __user *optval, int __user *optlen)
1671 {
1672 struct sock *sk = sock->sk;
1673 struct netlink_sock *nlk = nlk_sk(sk);
1674 int len, val, err;
1675
1676 if (level != SOL_NETLINK)
1677 return -ENOPROTOOPT;
1678
1679 if (get_user(len, optlen))
1680 return -EFAULT;
1681 if (len < 0)
1682 return -EINVAL;
1683
1684 switch (optname) {
1685 case NETLINK_PKTINFO:
1686 if (len < sizeof(int))
1687 return -EINVAL;
1688 len = sizeof(int);
1689 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1690 if (put_user(len, optlen) ||
1691 put_user(val, optval))
1692 return -EFAULT;
1693 err = 0;
1694 break;
1695 case NETLINK_BROADCAST_ERROR:
1696 if (len < sizeof(int))
1697 return -EINVAL;
1698 len = sizeof(int);
1699 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1700 if (put_user(len, optlen) ||
1701 put_user(val, optval))
1702 return -EFAULT;
1703 err = 0;
1704 break;
1705 case NETLINK_NO_ENOBUFS:
1706 if (len < sizeof(int))
1707 return -EINVAL;
1708 len = sizeof(int);
1709 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1710 if (put_user(len, optlen) ||
1711 put_user(val, optval))
1712 return -EFAULT;
1713 err = 0;
1714 break;
1715 case NETLINK_LIST_MEMBERSHIPS: {
1716 int pos, idx, shift;
1717
1718 err = 0;
1719 netlink_lock_table();
1720 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1721 if (len - pos < sizeof(u32))
1722 break;
1723
1724 idx = pos / sizeof(unsigned long);
1725 shift = (pos % sizeof(unsigned long)) * 8;
1726 if (put_user((u32)(nlk->groups[idx] >> shift),
1727 (u32 __user *)(optval + pos))) {
1728 err = -EFAULT;
1729 break;
1730 }
1731 }
1732 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1733 err = -EFAULT;
1734 netlink_unlock_table();
1735 break;
1736 }
1737 case NETLINK_CAP_ACK:
1738 if (len < sizeof(int))
1739 return -EINVAL;
1740 len = sizeof(int);
1741 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1742 if (put_user(len, optlen) ||
1743 put_user(val, optval))
1744 return -EFAULT;
1745 err = 0;
1746 break;
1747 default:
1748 err = -ENOPROTOOPT;
1749 }
1750 return err;
1751 }
1752
1753 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1754 {
1755 struct nl_pktinfo info;
1756
1757 info.group = NETLINK_CB(skb).dst_group;
1758 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1759 }
1760
1761 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1762 struct sk_buff *skb)
1763 {
1764 if (!NETLINK_CB(skb).nsid_is_set)
1765 return;
1766
1767 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1768 &NETLINK_CB(skb).nsid);
1769 }
1770
1771 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1772 {
1773 struct sock *sk = sock->sk;
1774 struct netlink_sock *nlk = nlk_sk(sk);
1775 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1776 u32 dst_portid;
1777 u32 dst_group;
1778 struct sk_buff *skb;
1779 int err;
1780 struct scm_cookie scm;
1781 u32 netlink_skb_flags = 0;
1782
1783 if (msg->msg_flags&MSG_OOB)
1784 return -EOPNOTSUPP;
1785
1786 err = scm_send(sock, msg, &scm, true);
1787 if (err < 0)
1788 return err;
1789
1790 if (msg->msg_namelen) {
1791 err = -EINVAL;
1792 if (addr->nl_family != AF_NETLINK)
1793 goto out;
1794 dst_portid = addr->nl_pid;
1795 dst_group = ffs(addr->nl_groups);
1796 err = -EPERM;
1797 if ((dst_group || dst_portid) &&
1798 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1799 goto out;
1800 netlink_skb_flags |= NETLINK_SKB_DST;
1801 } else {
1802 dst_portid = nlk->dst_portid;
1803 dst_group = nlk->dst_group;
1804 }
1805
1806 if (!nlk->bound) {
1807 err = netlink_autobind(sock);
1808 if (err)
1809 goto out;
1810 } else {
1811 /* Ensure nlk is hashed and visible. */
1812 smp_rmb();
1813 }
1814
1815 err = -EMSGSIZE;
1816 if (len > sk->sk_sndbuf - 32)
1817 goto out;
1818 err = -ENOBUFS;
1819 skb = netlink_alloc_large_skb(len, dst_group);
1820 if (skb == NULL)
1821 goto out;
1822
1823 NETLINK_CB(skb).portid = nlk->portid;
1824 NETLINK_CB(skb).dst_group = dst_group;
1825 NETLINK_CB(skb).creds = scm.creds;
1826 NETLINK_CB(skb).flags = netlink_skb_flags;
1827
1828 err = -EFAULT;
1829 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1830 kfree_skb(skb);
1831 goto out;
1832 }
1833
1834 err = security_netlink_send(sk, skb);
1835 if (err) {
1836 kfree_skb(skb);
1837 goto out;
1838 }
1839
1840 if (dst_group) {
1841 atomic_inc(&skb->users);
1842 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1843 }
1844 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1845
1846 out:
1847 scm_destroy(&scm);
1848 return err;
1849 }
1850
1851 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1852 int flags)
1853 {
1854 struct scm_cookie scm;
1855 struct sock *sk = sock->sk;
1856 struct netlink_sock *nlk = nlk_sk(sk);
1857 int noblock = flags&MSG_DONTWAIT;
1858 size_t copied;
1859 struct sk_buff *skb, *data_skb;
1860 int err, ret;
1861
1862 if (flags&MSG_OOB)
1863 return -EOPNOTSUPP;
1864
1865 copied = 0;
1866
1867 skb = skb_recv_datagram(sk, flags, noblock, &err);
1868 if (skb == NULL)
1869 goto out;
1870
1871 data_skb = skb;
1872
1873 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1874 if (unlikely(skb_shinfo(skb)->frag_list)) {
1875 /*
1876 * If this skb has a frag_list, then here that means that we
1877 * will have to use the frag_list skb's data for compat tasks
1878 * and the regular skb's data for normal (non-compat) tasks.
1879 *
1880 * If we need to send the compat skb, assign it to the
1881 * 'data_skb' variable so that it will be used below for data
1882 * copying. We keep 'skb' for everything else, including
1883 * freeing both later.
1884 */
1885 if (flags & MSG_CMSG_COMPAT)
1886 data_skb = skb_shinfo(skb)->frag_list;
1887 }
1888 #endif
1889
1890 /* Record the max length of recvmsg() calls for future allocations */
1891 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1892 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1893 SKB_WITH_OVERHEAD(32768));
1894
1895 copied = data_skb->len;
1896 if (len < copied) {
1897 msg->msg_flags |= MSG_TRUNC;
1898 copied = len;
1899 }
1900
1901 skb_reset_transport_header(data_skb);
1902 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1903
1904 if (msg->msg_name) {
1905 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1906 addr->nl_family = AF_NETLINK;
1907 addr->nl_pad = 0;
1908 addr->nl_pid = NETLINK_CB(skb).portid;
1909 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1910 msg->msg_namelen = sizeof(*addr);
1911 }
1912
1913 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1914 netlink_cmsg_recv_pktinfo(msg, skb);
1915 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1916 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1917
1918 memset(&scm, 0, sizeof(scm));
1919 scm.creds = *NETLINK_CREDS(skb);
1920 if (flags & MSG_TRUNC)
1921 copied = data_skb->len;
1922
1923 skb_free_datagram(sk, skb);
1924
1925 if (nlk->cb_running &&
1926 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1927 ret = netlink_dump(sk);
1928 if (ret) {
1929 sk->sk_err = -ret;
1930 sk->sk_error_report(sk);
1931 }
1932 }
1933
1934 scm_recv(sock, msg, &scm, flags);
1935 out:
1936 netlink_rcv_wake(sk);
1937 return err ? : copied;
1938 }
1939
1940 static void netlink_data_ready(struct sock *sk)
1941 {
1942 BUG();
1943 }
1944
1945 /*
1946 * We export these functions to other modules. They provide a
1947 * complete set of kernel non-blocking support for message
1948 * queueing.
1949 */
1950
1951 struct sock *
1952 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1953 struct netlink_kernel_cfg *cfg)
1954 {
1955 struct socket *sock;
1956 struct sock *sk;
1957 struct netlink_sock *nlk;
1958 struct listeners *listeners = NULL;
1959 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1960 unsigned int groups;
1961
1962 BUG_ON(!nl_table);
1963
1964 if (unit < 0 || unit >= MAX_LINKS)
1965 return NULL;
1966
1967 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1968 return NULL;
1969
1970 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1971 goto out_sock_release_nosk;
1972
1973 sk = sock->sk;
1974
1975 if (!cfg || cfg->groups < 32)
1976 groups = 32;
1977 else
1978 groups = cfg->groups;
1979
1980 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1981 if (!listeners)
1982 goto out_sock_release;
1983
1984 sk->sk_data_ready = netlink_data_ready;
1985 if (cfg && cfg->input)
1986 nlk_sk(sk)->netlink_rcv = cfg->input;
1987
1988 if (netlink_insert(sk, 0))
1989 goto out_sock_release;
1990
1991 nlk = nlk_sk(sk);
1992 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
1993
1994 netlink_table_grab();
1995 if (!nl_table[unit].registered) {
1996 nl_table[unit].groups = groups;
1997 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1998 nl_table[unit].cb_mutex = cb_mutex;
1999 nl_table[unit].module = module;
2000 if (cfg) {
2001 nl_table[unit].bind = cfg->bind;
2002 nl_table[unit].unbind = cfg->unbind;
2003 nl_table[unit].flags = cfg->flags;
2004 if (cfg->compare)
2005 nl_table[unit].compare = cfg->compare;
2006 }
2007 nl_table[unit].registered = 1;
2008 } else {
2009 kfree(listeners);
2010 nl_table[unit].registered++;
2011 }
2012 netlink_table_ungrab();
2013 return sk;
2014
2015 out_sock_release:
2016 kfree(listeners);
2017 netlink_kernel_release(sk);
2018 return NULL;
2019
2020 out_sock_release_nosk:
2021 sock_release(sock);
2022 return NULL;
2023 }
2024 EXPORT_SYMBOL(__netlink_kernel_create);
2025
2026 void
2027 netlink_kernel_release(struct sock *sk)
2028 {
2029 if (sk == NULL || sk->sk_socket == NULL)
2030 return;
2031
2032 sock_release(sk->sk_socket);
2033 }
2034 EXPORT_SYMBOL(netlink_kernel_release);
2035
2036 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2037 {
2038 struct listeners *new, *old;
2039 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2040
2041 if (groups < 32)
2042 groups = 32;
2043
2044 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2045 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2046 if (!new)
2047 return -ENOMEM;
2048 old = nl_deref_protected(tbl->listeners);
2049 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2050 rcu_assign_pointer(tbl->listeners, new);
2051
2052 kfree_rcu(old, rcu);
2053 }
2054 tbl->groups = groups;
2055
2056 return 0;
2057 }
2058
2059 /**
2060 * netlink_change_ngroups - change number of multicast groups
2061 *
2062 * This changes the number of multicast groups that are available
2063 * on a certain netlink family. Note that it is not possible to
2064 * change the number of groups to below 32. Also note that it does
2065 * not implicitly call netlink_clear_multicast_users() when the
2066 * number of groups is reduced.
2067 *
2068 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2069 * @groups: The new number of groups.
2070 */
2071 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2072 {
2073 int err;
2074
2075 netlink_table_grab();
2076 err = __netlink_change_ngroups(sk, groups);
2077 netlink_table_ungrab();
2078
2079 return err;
2080 }
2081
2082 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2083 {
2084 struct sock *sk;
2085 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2086
2087 sk_for_each_bound(sk, &tbl->mc_list)
2088 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2089 }
2090
2091 struct nlmsghdr *
2092 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2093 {
2094 struct nlmsghdr *nlh;
2095 int size = nlmsg_msg_size(len);
2096
2097 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2098 nlh->nlmsg_type = type;
2099 nlh->nlmsg_len = size;
2100 nlh->nlmsg_flags = flags;
2101 nlh->nlmsg_pid = portid;
2102 nlh->nlmsg_seq = seq;
2103 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2104 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2105 return nlh;
2106 }
2107 EXPORT_SYMBOL(__nlmsg_put);
2108
2109 /*
2110 * It looks a bit ugly.
2111 * It would be better to create kernel thread.
2112 */
2113
2114 static int netlink_dump(struct sock *sk)
2115 {
2116 struct netlink_sock *nlk = nlk_sk(sk);
2117 struct netlink_callback *cb;
2118 struct sk_buff *skb = NULL;
2119 struct nlmsghdr *nlh;
2120 struct module *module;
2121 int len, err = -ENOBUFS;
2122 int alloc_min_size;
2123 int alloc_size;
2124
2125 mutex_lock(nlk->cb_mutex);
2126 if (!nlk->cb_running) {
2127 err = -EINVAL;
2128 goto errout_skb;
2129 }
2130
2131 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2132 goto errout_skb;
2133
2134 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2135 * required, but it makes sense to _attempt_ a 16K bytes allocation
2136 * to reduce number of system calls on dump operations, if user
2137 * ever provided a big enough buffer.
2138 */
2139 cb = &nlk->cb;
2140 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2141
2142 if (alloc_min_size < nlk->max_recvmsg_len) {
2143 alloc_size = nlk->max_recvmsg_len;
2144 skb = alloc_skb(alloc_size,
2145 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2146 __GFP_NOWARN | __GFP_NORETRY);
2147 }
2148 if (!skb) {
2149 alloc_size = alloc_min_size;
2150 skb = alloc_skb(alloc_size, GFP_KERNEL);
2151 }
2152 if (!skb)
2153 goto errout_skb;
2154
2155 /* Trim skb to allocated size. User is expected to provide buffer as
2156 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2157 * netlink_recvmsg())). dump will pack as many smaller messages as
2158 * could fit within the allocated skb. skb is typically allocated
2159 * with larger space than required (could be as much as near 2x the
2160 * requested size with align to next power of 2 approach). Allowing
2161 * dump to use the excess space makes it difficult for a user to have a
2162 * reasonable static buffer based on the expected largest dump of a
2163 * single netdev. The outcome is MSG_TRUNC error.
2164 */
2165 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2166 netlink_skb_set_owner_r(skb, sk);
2167
2168 len = cb->dump(skb, cb);
2169
2170 if (len > 0) {
2171 mutex_unlock(nlk->cb_mutex);
2172
2173 if (sk_filter(sk, skb))
2174 kfree_skb(skb);
2175 else
2176 __netlink_sendskb(sk, skb);
2177 return 0;
2178 }
2179
2180 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2181 if (!nlh)
2182 goto errout_skb;
2183
2184 nl_dump_check_consistent(cb, nlh);
2185
2186 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2187
2188 if (sk_filter(sk, skb))
2189 kfree_skb(skb);
2190 else
2191 __netlink_sendskb(sk, skb);
2192
2193 if (cb->done)
2194 cb->done(cb);
2195
2196 nlk->cb_running = false;
2197 module = cb->module;
2198 skb = cb->skb;
2199 mutex_unlock(nlk->cb_mutex);
2200 module_put(module);
2201 consume_skb(skb);
2202 return 0;
2203
2204 errout_skb:
2205 mutex_unlock(nlk->cb_mutex);
2206 kfree_skb(skb);
2207 return err;
2208 }
2209
2210 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2211 const struct nlmsghdr *nlh,
2212 struct netlink_dump_control *control)
2213 {
2214 struct netlink_callback *cb;
2215 struct sock *sk;
2216 struct netlink_sock *nlk;
2217 int ret;
2218
2219 atomic_inc(&skb->users);
2220
2221 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2222 if (sk == NULL) {
2223 ret = -ECONNREFUSED;
2224 goto error_free;
2225 }
2226
2227 nlk = nlk_sk(sk);
2228 mutex_lock(nlk->cb_mutex);
2229 /* A dump is in progress... */
2230 if (nlk->cb_running) {
2231 ret = -EBUSY;
2232 goto error_unlock;
2233 }
2234 /* add reference of module which cb->dump belongs to */
2235 if (!try_module_get(control->module)) {
2236 ret = -EPROTONOSUPPORT;
2237 goto error_unlock;
2238 }
2239
2240 cb = &nlk->cb;
2241 memset(cb, 0, sizeof(*cb));
2242 cb->start = control->start;
2243 cb->dump = control->dump;
2244 cb->done = control->done;
2245 cb->nlh = nlh;
2246 cb->data = control->data;
2247 cb->module = control->module;
2248 cb->min_dump_alloc = control->min_dump_alloc;
2249 cb->skb = skb;
2250
2251 nlk->cb_running = true;
2252
2253 mutex_unlock(nlk->cb_mutex);
2254
2255 if (cb->start)
2256 cb->start(cb);
2257
2258 ret = netlink_dump(sk);
2259 sock_put(sk);
2260
2261 if (ret)
2262 return ret;
2263
2264 /* We successfully started a dump, by returning -EINTR we
2265 * signal not to send ACK even if it was requested.
2266 */
2267 return -EINTR;
2268
2269 error_unlock:
2270 sock_put(sk);
2271 mutex_unlock(nlk->cb_mutex);
2272 error_free:
2273 kfree_skb(skb);
2274 return ret;
2275 }
2276 EXPORT_SYMBOL(__netlink_dump_start);
2277
2278 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
2279 {
2280 struct sk_buff *skb;
2281 struct nlmsghdr *rep;
2282 struct nlmsgerr *errmsg;
2283 size_t payload = sizeof(*errmsg);
2284 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2285
2286 /* Error messages get the original request appened, unless the user
2287 * requests to cap the error message.
2288 */
2289 if (!(nlk->flags & NETLINK_F_CAP_ACK) && err)
2290 payload += nlmsg_len(nlh);
2291
2292 skb = nlmsg_new(payload, GFP_KERNEL);
2293 if (!skb) {
2294 struct sock *sk;
2295
2296 sk = netlink_lookup(sock_net(in_skb->sk),
2297 in_skb->sk->sk_protocol,
2298 NETLINK_CB(in_skb).portid);
2299 if (sk) {
2300 sk->sk_err = ENOBUFS;
2301 sk->sk_error_report(sk);
2302 sock_put(sk);
2303 }
2304 return;
2305 }
2306
2307 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2308 NLMSG_ERROR, payload, 0);
2309 errmsg = nlmsg_data(rep);
2310 errmsg->error = err;
2311 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2312 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2313 }
2314 EXPORT_SYMBOL(netlink_ack);
2315
2316 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2317 struct nlmsghdr *))
2318 {
2319 struct nlmsghdr *nlh;
2320 int err;
2321
2322 while (skb->len >= nlmsg_total_size(0)) {
2323 int msglen;
2324
2325 nlh = nlmsg_hdr(skb);
2326 err = 0;
2327
2328 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2329 return 0;
2330
2331 /* Only requests are handled by the kernel */
2332 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2333 goto ack;
2334
2335 /* Skip control messages */
2336 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2337 goto ack;
2338
2339 err = cb(skb, nlh);
2340 if (err == -EINTR)
2341 goto skip;
2342
2343 ack:
2344 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2345 netlink_ack(skb, nlh, err);
2346
2347 skip:
2348 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2349 if (msglen > skb->len)
2350 msglen = skb->len;
2351 skb_pull(skb, msglen);
2352 }
2353
2354 return 0;
2355 }
2356 EXPORT_SYMBOL(netlink_rcv_skb);
2357
2358 /**
2359 * nlmsg_notify - send a notification netlink message
2360 * @sk: netlink socket to use
2361 * @skb: notification message
2362 * @portid: destination netlink portid for reports or 0
2363 * @group: destination multicast group or 0
2364 * @report: 1 to report back, 0 to disable
2365 * @flags: allocation flags
2366 */
2367 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2368 unsigned int group, int report, gfp_t flags)
2369 {
2370 int err = 0;
2371
2372 if (group) {
2373 int exclude_portid = 0;
2374
2375 if (report) {
2376 atomic_inc(&skb->users);
2377 exclude_portid = portid;
2378 }
2379
2380 /* errors reported via destination sk->sk_err, but propagate
2381 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2382 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2383 }
2384
2385 if (report) {
2386 int err2;
2387
2388 err2 = nlmsg_unicast(sk, skb, portid);
2389 if (!err || err == -ESRCH)
2390 err = err2;
2391 }
2392
2393 return err;
2394 }
2395 EXPORT_SYMBOL(nlmsg_notify);
2396
2397 #ifdef CONFIG_PROC_FS
2398 struct nl_seq_iter {
2399 struct seq_net_private p;
2400 struct rhashtable_iter hti;
2401 int link;
2402 };
2403
2404 static int netlink_walk_start(struct nl_seq_iter *iter)
2405 {
2406 int err;
2407
2408 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2409 GFP_KERNEL);
2410 if (err) {
2411 iter->link = MAX_LINKS;
2412 return err;
2413 }
2414
2415 err = rhashtable_walk_start(&iter->hti);
2416 return err == -EAGAIN ? 0 : err;
2417 }
2418
2419 static void netlink_walk_stop(struct nl_seq_iter *iter)
2420 {
2421 rhashtable_walk_stop(&iter->hti);
2422 rhashtable_walk_exit(&iter->hti);
2423 }
2424
2425 static void *__netlink_seq_next(struct seq_file *seq)
2426 {
2427 struct nl_seq_iter *iter = seq->private;
2428 struct netlink_sock *nlk;
2429
2430 do {
2431 for (;;) {
2432 int err;
2433
2434 nlk = rhashtable_walk_next(&iter->hti);
2435
2436 if (IS_ERR(nlk)) {
2437 if (PTR_ERR(nlk) == -EAGAIN)
2438 continue;
2439
2440 return nlk;
2441 }
2442
2443 if (nlk)
2444 break;
2445
2446 netlink_walk_stop(iter);
2447 if (++iter->link >= MAX_LINKS)
2448 return NULL;
2449
2450 err = netlink_walk_start(iter);
2451 if (err)
2452 return ERR_PTR(err);
2453 }
2454 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2455
2456 return nlk;
2457 }
2458
2459 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2460 {
2461 struct nl_seq_iter *iter = seq->private;
2462 void *obj = SEQ_START_TOKEN;
2463 loff_t pos;
2464 int err;
2465
2466 iter->link = 0;
2467
2468 err = netlink_walk_start(iter);
2469 if (err)
2470 return ERR_PTR(err);
2471
2472 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2473 obj = __netlink_seq_next(seq);
2474
2475 return obj;
2476 }
2477
2478 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2479 {
2480 ++*pos;
2481 return __netlink_seq_next(seq);
2482 }
2483
2484 static void netlink_seq_stop(struct seq_file *seq, void *v)
2485 {
2486 struct nl_seq_iter *iter = seq->private;
2487
2488 if (iter->link >= MAX_LINKS)
2489 return;
2490
2491 netlink_walk_stop(iter);
2492 }
2493
2494
2495 static int netlink_seq_show(struct seq_file *seq, void *v)
2496 {
2497 if (v == SEQ_START_TOKEN) {
2498 seq_puts(seq,
2499 "sk Eth Pid Groups "
2500 "Rmem Wmem Dump Locks Drops Inode\n");
2501 } else {
2502 struct sock *s = v;
2503 struct netlink_sock *nlk = nlk_sk(s);
2504
2505 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2506 s,
2507 s->sk_protocol,
2508 nlk->portid,
2509 nlk->groups ? (u32)nlk->groups[0] : 0,
2510 sk_rmem_alloc_get(s),
2511 sk_wmem_alloc_get(s),
2512 nlk->cb_running,
2513 atomic_read(&s->sk_refcnt),
2514 atomic_read(&s->sk_drops),
2515 sock_i_ino(s)
2516 );
2517
2518 }
2519 return 0;
2520 }
2521
2522 static const struct seq_operations netlink_seq_ops = {
2523 .start = netlink_seq_start,
2524 .next = netlink_seq_next,
2525 .stop = netlink_seq_stop,
2526 .show = netlink_seq_show,
2527 };
2528
2529
2530 static int netlink_seq_open(struct inode *inode, struct file *file)
2531 {
2532 return seq_open_net(inode, file, &netlink_seq_ops,
2533 sizeof(struct nl_seq_iter));
2534 }
2535
2536 static const struct file_operations netlink_seq_fops = {
2537 .owner = THIS_MODULE,
2538 .open = netlink_seq_open,
2539 .read = seq_read,
2540 .llseek = seq_lseek,
2541 .release = seq_release_net,
2542 };
2543
2544 #endif
2545
2546 int netlink_register_notifier(struct notifier_block *nb)
2547 {
2548 return blocking_notifier_chain_register(&netlink_chain, nb);
2549 }
2550 EXPORT_SYMBOL(netlink_register_notifier);
2551
2552 int netlink_unregister_notifier(struct notifier_block *nb)
2553 {
2554 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2555 }
2556 EXPORT_SYMBOL(netlink_unregister_notifier);
2557
2558 static const struct proto_ops netlink_ops = {
2559 .family = PF_NETLINK,
2560 .owner = THIS_MODULE,
2561 .release = netlink_release,
2562 .bind = netlink_bind,
2563 .connect = netlink_connect,
2564 .socketpair = sock_no_socketpair,
2565 .accept = sock_no_accept,
2566 .getname = netlink_getname,
2567 .poll = datagram_poll,
2568 .ioctl = netlink_ioctl,
2569 .listen = sock_no_listen,
2570 .shutdown = sock_no_shutdown,
2571 .setsockopt = netlink_setsockopt,
2572 .getsockopt = netlink_getsockopt,
2573 .sendmsg = netlink_sendmsg,
2574 .recvmsg = netlink_recvmsg,
2575 .mmap = sock_no_mmap,
2576 .sendpage = sock_no_sendpage,
2577 };
2578
2579 static const struct net_proto_family netlink_family_ops = {
2580 .family = PF_NETLINK,
2581 .create = netlink_create,
2582 .owner = THIS_MODULE, /* for consistency 8) */
2583 };
2584
2585 static int __net_init netlink_net_init(struct net *net)
2586 {
2587 #ifdef CONFIG_PROC_FS
2588 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2589 return -ENOMEM;
2590 #endif
2591 return 0;
2592 }
2593
2594 static void __net_exit netlink_net_exit(struct net *net)
2595 {
2596 #ifdef CONFIG_PROC_FS
2597 remove_proc_entry("netlink", net->proc_net);
2598 #endif
2599 }
2600
2601 static void __init netlink_add_usersock_entry(void)
2602 {
2603 struct listeners *listeners;
2604 int groups = 32;
2605
2606 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2607 if (!listeners)
2608 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2609
2610 netlink_table_grab();
2611
2612 nl_table[NETLINK_USERSOCK].groups = groups;
2613 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2614 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2615 nl_table[NETLINK_USERSOCK].registered = 1;
2616 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2617
2618 netlink_table_ungrab();
2619 }
2620
2621 static struct pernet_operations __net_initdata netlink_net_ops = {
2622 .init = netlink_net_init,
2623 .exit = netlink_net_exit,
2624 };
2625
2626 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2627 {
2628 const struct netlink_sock *nlk = data;
2629 struct netlink_compare_arg arg;
2630
2631 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2632 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2633 }
2634
2635 static const struct rhashtable_params netlink_rhashtable_params = {
2636 .head_offset = offsetof(struct netlink_sock, node),
2637 .key_len = netlink_compare_arg_len,
2638 .obj_hashfn = netlink_hash,
2639 .obj_cmpfn = netlink_compare,
2640 .automatic_shrinking = true,
2641 };
2642
2643 static int __init netlink_proto_init(void)
2644 {
2645 int i;
2646 int err = proto_register(&netlink_proto, 0);
2647
2648 if (err != 0)
2649 goto out;
2650
2651 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2652
2653 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2654 if (!nl_table)
2655 goto panic;
2656
2657 for (i = 0; i < MAX_LINKS; i++) {
2658 if (rhashtable_init(&nl_table[i].hash,
2659 &netlink_rhashtable_params) < 0) {
2660 while (--i > 0)
2661 rhashtable_destroy(&nl_table[i].hash);
2662 kfree(nl_table);
2663 goto panic;
2664 }
2665 }
2666
2667 INIT_LIST_HEAD(&netlink_tap_all);
2668
2669 netlink_add_usersock_entry();
2670
2671 sock_register(&netlink_family_ops);
2672 register_pernet_subsys(&netlink_net_ops);
2673 /* The netlink device handler may be needed early. */
2674 rtnetlink_init();
2675 out:
2676 return err;
2677 panic:
2678 panic("netlink_init: Cannot allocate nl_table\n");
2679 }
2680
2681 core_initcall(netlink_proto_init);
A mirror of Linus' kernel repository