?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
initial commit with v3.6.7
[linux-3.6.7-moxart.git] / net / netlink / af_netlink.c
blob0426b677dbb787d53b0644157455abcd998234fc
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 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
22 */
23
24 #include <linux/module.h>
25
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
35 #include <linux/un.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
40 #include <linux/fs.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
54 #include <linux/mm.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/mutex.h>
58
59 #include <net/net_namespace.h>
60 #include <net/sock.h>
61 #include <net/scm.h>
62 #include <net/netlink.h>
63
64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
65 #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
66
67 struct netlink_sock {
68 /* struct sock has to be the first member of netlink_sock */
69 struct sock sk;
70 u32 pid;
71 u32 dst_pid;
72 u32 dst_group;
73 u32 flags;
74 u32 subscriptions;
75 u32 ngroups;
76 unsigned long *groups;
77 unsigned long state;
78 wait_queue_head_t wait;
79 struct netlink_callback *cb;
80 struct mutex *cb_mutex;
81 struct mutex cb_def_mutex;
82 void (*netlink_rcv)(struct sk_buff *skb);
83 void (*netlink_bind)(int group);
84 struct module *module;
85 };
86
87 struct listeners {
88 struct rcu_head rcu;
89 unsigned long masks[0];
90 };
91
92 #define NETLINK_KERNEL_SOCKET 0x1
93 #define NETLINK_RECV_PKTINFO 0x2
94 #define NETLINK_BROADCAST_SEND_ERROR 0x4
95 #define NETLINK_RECV_NO_ENOBUFS 0x8
96
97 static inline struct netlink_sock *nlk_sk(struct sock *sk)
98 {
99 return container_of(sk, struct netlink_sock, sk);
100 }
101
102 static inline int netlink_is_kernel(struct sock *sk)
103 {
104 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
105 }
106
107 struct nl_pid_hash {
108 struct hlist_head *table;
109 unsigned long rehash_time;
110
111 unsigned int mask;
112 unsigned int shift;
113
114 unsigned int entries;
115 unsigned int max_shift;
116
117 u32 rnd;
118 };
119
120 struct netlink_table {
121 struct nl_pid_hash hash;
122 struct hlist_head mc_list;
123 struct listeners __rcu *listeners;
124 unsigned int nl_nonroot;
125 unsigned int groups;
126 struct mutex *cb_mutex;
127 struct module *module;
128 void (*bind)(int group);
129 int registered;
130 };
131
132 static struct netlink_table *nl_table;
133
134 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
135
136 static int netlink_dump(struct sock *sk);
137
138 static DEFINE_RWLOCK(nl_table_lock);
139 static atomic_t nl_table_users = ATOMIC_INIT(0);
140
141 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
142
143 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
144
145 static inline u32 netlink_group_mask(u32 group)
146 {
147 return group ? 1 << (group - 1) : 0;
148 }
149
150 static inline struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
151 {
152 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
153 }
154
155 static void netlink_destroy_callback(struct netlink_callback *cb)
156 {
157 kfree_skb(cb->skb);
158 kfree(cb);
159 }
160
161 static void netlink_consume_callback(struct netlink_callback *cb)
162 {
163 consume_skb(cb->skb);
164 kfree(cb);
165 }
166
167 static void netlink_sock_destruct(struct sock *sk)
168 {
169 struct netlink_sock *nlk = nlk_sk(sk);
170
171 if (nlk->cb) {
172 if (nlk->cb->done)
173 nlk->cb->done(nlk->cb);
174
175 module_put(nlk->cb->module);
176 netlink_destroy_callback(nlk->cb);
177 }
178
179 skb_queue_purge(&sk->sk_receive_queue);
180
181 if (!sock_flag(sk, SOCK_DEAD)) {
182 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
183 return;
184 }
185
186 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
187 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
188 WARN_ON(nlk_sk(sk)->groups);
189 }
190
191 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
192 * SMP. Look, when several writers sleep and reader wakes them up, all but one
193 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
194 * this, _but_ remember, it adds useless work on UP machines.
195 */
196
197 void netlink_table_grab(void)
198 __acquires(nl_table_lock)
199 {
200 might_sleep();
201
202 write_lock_irq(&nl_table_lock);
203
204 if (atomic_read(&nl_table_users)) {
205 DECLARE_WAITQUEUE(wait, current);
206
207 add_wait_queue_exclusive(&nl_table_wait, &wait);
208 for (;;) {
209 set_current_state(TASK_UNINTERRUPTIBLE);
210 if (atomic_read(&nl_table_users) == 0)
211 break;
212 write_unlock_irq(&nl_table_lock);
213 schedule();
214 write_lock_irq(&nl_table_lock);
215 }
216
217 __set_current_state(TASK_RUNNING);
218 remove_wait_queue(&nl_table_wait, &wait);
219 }
220 }
221
222 void netlink_table_ungrab(void)
223 __releases(nl_table_lock)
224 {
225 write_unlock_irq(&nl_table_lock);
226 wake_up(&nl_table_wait);
227 }
228
229 static inline void
230 netlink_lock_table(void)
231 {
232 /* read_lock() synchronizes us to netlink_table_grab */
233
234 read_lock(&nl_table_lock);
235 atomic_inc(&nl_table_users);
236 read_unlock(&nl_table_lock);
237 }
238
239 static inline void
240 netlink_unlock_table(void)
241 {
242 if (atomic_dec_and_test(&nl_table_users))
243 wake_up(&nl_table_wait);
244 }
245
246 static struct sock *netlink_lookup(struct net *net, int protocol, u32 pid)
247 {
248 struct nl_pid_hash *hash = &nl_table[protocol].hash;
249 struct hlist_head *head;
250 struct sock *sk;
251 struct hlist_node *node;
252
253 read_lock(&nl_table_lock);
254 head = nl_pid_hashfn(hash, pid);
255 sk_for_each(sk, node, head) {
256 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
257 sock_hold(sk);
258 goto found;
259 }
260 }
261 sk = NULL;
262 found:
263 read_unlock(&nl_table_lock);
264 return sk;
265 }
266
267 static struct hlist_head *nl_pid_hash_zalloc(size_t size)
268 {
269 if (size <= PAGE_SIZE)
270 return kzalloc(size, GFP_ATOMIC);
271 else
272 return (struct hlist_head *)
273 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
274 get_order(size));
275 }
276
277 static void nl_pid_hash_free(struct hlist_head *table, size_t size)
278 {
279 if (size <= PAGE_SIZE)
280 kfree(table);
281 else
282 free_pages((unsigned long)table, get_order(size));
283 }
284
285 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
286 {
287 unsigned int omask, mask, shift;
288 size_t osize, size;
289 struct hlist_head *otable, *table;
290 int i;
291
292 omask = mask = hash->mask;
293 osize = size = (mask + 1) * sizeof(*table);
294 shift = hash->shift;
295
296 if (grow) {
297 if (++shift > hash->max_shift)
298 return 0;
299 mask = mask * 2 + 1;
300 size *= 2;
301 }
302
303 table = nl_pid_hash_zalloc(size);
304 if (!table)
305 return 0;
306
307 otable = hash->table;
308 hash->table = table;
309 hash->mask = mask;
310 hash->shift = shift;
311 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
312
313 for (i = 0; i <= omask; i++) {
314 struct sock *sk;
315 struct hlist_node *node, *tmp;
316
317 sk_for_each_safe(sk, node, tmp, &otable[i])
318 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
319 }
320
321 nl_pid_hash_free(otable, osize);
322 hash->rehash_time = jiffies + 10 * 60 * HZ;
323 return 1;
324 }
325
326 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
327 {
328 int avg = hash->entries >> hash->shift;
329
330 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
331 return 1;
332
333 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
334 nl_pid_hash_rehash(hash, 0);
335 return 1;
336 }
337
338 return 0;
339 }
340
341 static const struct proto_ops netlink_ops;
342
343 static void
344 netlink_update_listeners(struct sock *sk)
345 {
346 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
347 struct hlist_node *node;
348 unsigned long mask;
349 unsigned int i;
350 struct listeners *listeners;
351
352 listeners = nl_deref_protected(tbl->listeners);
353 if (!listeners)
354 return;
355
356 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
357 mask = 0;
358 sk_for_each_bound(sk, node, &tbl->mc_list) {
359 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
360 mask |= nlk_sk(sk)->groups[i];
361 }
362 listeners->masks[i] = mask;
363 }
364 /* this function is only called with the netlink table "grabbed", which
365 * makes sure updates are visible before bind or setsockopt return. */
366 }
367
368 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
369 {
370 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
371 struct hlist_head *head;
372 int err = -EADDRINUSE;
373 struct sock *osk;
374 struct hlist_node *node;
375 int len;
376
377 netlink_table_grab();
378 head = nl_pid_hashfn(hash, pid);
379 len = 0;
380 sk_for_each(osk, node, head) {
381 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
382 break;
383 len++;
384 }
385 if (node)
386 goto err;
387
388 err = -EBUSY;
389 if (nlk_sk(sk)->pid)
390 goto err;
391
392 err = -ENOMEM;
393 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
394 goto err;
395
396 if (len && nl_pid_hash_dilute(hash, len))
397 head = nl_pid_hashfn(hash, pid);
398 hash->entries++;
399 nlk_sk(sk)->pid = pid;
400 sk_add_node(sk, head);
401 err = 0;
402
403 err:
404 netlink_table_ungrab();
405 return err;
406 }
407
408 static void netlink_remove(struct sock *sk)
409 {
410 netlink_table_grab();
411 if (sk_del_node_init(sk))
412 nl_table[sk->sk_protocol].hash.entries--;
413 if (nlk_sk(sk)->subscriptions)
414 __sk_del_bind_node(sk);
415 netlink_table_ungrab();
416 }
417
418 static struct proto netlink_proto = {
419 .name = "NETLINK",
420 .owner = THIS_MODULE,
421 .obj_size = sizeof(struct netlink_sock),
422 };
423
424 static int __netlink_create(struct net *net, struct socket *sock,
425 struct mutex *cb_mutex, int protocol)
426 {
427 struct sock *sk;
428 struct netlink_sock *nlk;
429
430 sock->ops = &netlink_ops;
431
432 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
433 if (!sk)
434 return -ENOMEM;
435
436 sock_init_data(sock, sk);
437
438 nlk = nlk_sk(sk);
439 if (cb_mutex) {
440 nlk->cb_mutex = cb_mutex;
441 } else {
442 nlk->cb_mutex = &nlk->cb_def_mutex;
443 mutex_init(nlk->cb_mutex);
444 }
445 init_waitqueue_head(&nlk->wait);
446
447 sk->sk_destruct = netlink_sock_destruct;
448 sk->sk_protocol = protocol;
449 return 0;
450 }
451
452 static int netlink_create(struct net *net, struct socket *sock, int protocol,
453 int kern)
454 {
455 struct module *module = NULL;
456 struct mutex *cb_mutex;
457 struct netlink_sock *nlk;
458 void (*bind)(int group);
459 int err = 0;
460
461 sock->state = SS_UNCONNECTED;
462
463 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
464 return -ESOCKTNOSUPPORT;
465
466 if (protocol < 0 || protocol >= MAX_LINKS)
467 return -EPROTONOSUPPORT;
468
469 netlink_lock_table();
470 #ifdef CONFIG_MODULES
471 if (!nl_table[protocol].registered) {
472 netlink_unlock_table();
473 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
474 netlink_lock_table();
475 }
476 #endif
477 if (nl_table[protocol].registered &&
478 try_module_get(nl_table[protocol].module))
479 module = nl_table[protocol].module;
480 else
481 err = -EPROTONOSUPPORT;
482 cb_mutex = nl_table[protocol].cb_mutex;
483 bind = nl_table[protocol].bind;
484 netlink_unlock_table();
485
486 if (err < 0)
487 goto out;
488
489 err = __netlink_create(net, sock, cb_mutex, protocol);
490 if (err < 0)
491 goto out_module;
492
493 local_bh_disable();
494 sock_prot_inuse_add(net, &netlink_proto, 1);
495 local_bh_enable();
496
497 nlk = nlk_sk(sock->sk);
498 nlk->module = module;
499 nlk->netlink_bind = bind;
500 out:
501 return err;
502
503 out_module:
504 module_put(module);
505 goto out;
506 }
507
508 static int netlink_release(struct socket *sock)
509 {
510 struct sock *sk = sock->sk;
511 struct netlink_sock *nlk;
512
513 if (!sk)
514 return 0;
515
516 netlink_remove(sk);
517 sock_orphan(sk);
518 nlk = nlk_sk(sk);
519
520 /*
521 * OK. Socket is unlinked, any packets that arrive now
522 * will be purged.
523 */
524
525 sock->sk = NULL;
526 wake_up_interruptible_all(&nlk->wait);
527
528 skb_queue_purge(&sk->sk_write_queue);
529
530 if (nlk->pid) {
531 struct netlink_notify n = {
532 .net = sock_net(sk),
533 .protocol = sk->sk_protocol,
534 .pid = nlk->pid,
535 };
536 atomic_notifier_call_chain(&netlink_chain,
537 NETLINK_URELEASE, &n);
538 }
539
540 module_put(nlk->module);
541
542 netlink_table_grab();
543 if (netlink_is_kernel(sk)) {
544 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
545 if (--nl_table[sk->sk_protocol].registered == 0) {
546 struct listeners *old;
547
548 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
549 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
550 kfree_rcu(old, rcu);
551 nl_table[sk->sk_protocol].module = NULL;
552 nl_table[sk->sk_protocol].registered = 0;
553 }
554 } else if (nlk->subscriptions) {
555 netlink_update_listeners(sk);
556 }
557 netlink_table_ungrab();
558
559 kfree(nlk->groups);
560 nlk->groups = NULL;
561
562 local_bh_disable();
563 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
564 local_bh_enable();
565 sock_put(sk);
566 return 0;
567 }
568
569 static int netlink_autobind(struct socket *sock)
570 {
571 struct sock *sk = sock->sk;
572 struct net *net = sock_net(sk);
573 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
574 struct hlist_head *head;
575 struct sock *osk;
576 struct hlist_node *node;
577 s32 pid = task_tgid_vnr(current);
578 int err;
579 static s32 rover = -4097;
580
581 retry:
582 cond_resched();
583 netlink_table_grab();
584 head = nl_pid_hashfn(hash, pid);
585 sk_for_each(osk, node, head) {
586 if (!net_eq(sock_net(osk), net))
587 continue;
588 if (nlk_sk(osk)->pid == pid) {
589 /* Bind collision, search negative pid values. */
590 pid = rover--;
591 if (rover > -4097)
592 rover = -4097;
593 netlink_table_ungrab();
594 goto retry;
595 }
596 }
597 netlink_table_ungrab();
598
599 err = netlink_insert(sk, net, pid);
600 if (err == -EADDRINUSE)
601 goto retry;
602
603 /* If 2 threads race to autobind, that is fine. */
604 if (err == -EBUSY)
605 err = 0;
606
607 return err;
608 }
609
610 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
611 {
612 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
613 capable(CAP_NET_ADMIN);
614 }
615
616 static void
617 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
618 {
619 struct netlink_sock *nlk = nlk_sk(sk);
620
621 if (nlk->subscriptions && !subscriptions)
622 __sk_del_bind_node(sk);
623 else if (!nlk->subscriptions && subscriptions)
624 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
625 nlk->subscriptions = subscriptions;
626 }
627
628 static int netlink_realloc_groups(struct sock *sk)
629 {
630 struct netlink_sock *nlk = nlk_sk(sk);
631 unsigned int groups;
632 unsigned long *new_groups;
633 int err = 0;
634
635 netlink_table_grab();
636
637 groups = nl_table[sk->sk_protocol].groups;
638 if (!nl_table[sk->sk_protocol].registered) {
639 err = -ENOENT;
640 goto out_unlock;
641 }
642
643 if (nlk->ngroups >= groups)
644 goto out_unlock;
645
646 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
647 if (new_groups == NULL) {
648 err = -ENOMEM;
649 goto out_unlock;
650 }
651 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
652 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
653
654 nlk->groups = new_groups;
655 nlk->ngroups = groups;
656 out_unlock:
657 netlink_table_ungrab();
658 return err;
659 }
660
661 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
662 int addr_len)
663 {
664 struct sock *sk = sock->sk;
665 struct net *net = sock_net(sk);
666 struct netlink_sock *nlk = nlk_sk(sk);
667 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
668 int err;
669
670 if (nladdr->nl_family != AF_NETLINK)
671 return -EINVAL;
672
673 /* Only superuser is allowed to listen multicasts */
674 if (nladdr->nl_groups) {
675 if (!netlink_capable(sock, NL_NONROOT_RECV))
676 return -EPERM;
677 err = netlink_realloc_groups(sk);
678 if (err)
679 return err;
680 }
681
682 if (nlk->pid) {
683 if (nladdr->nl_pid != nlk->pid)
684 return -EINVAL;
685 } else {
686 err = nladdr->nl_pid ?
687 netlink_insert(sk, net, nladdr->nl_pid) :
688 netlink_autobind(sock);
689 if (err)
690 return err;
691 }
692
693 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
694 return 0;
695
696 netlink_table_grab();
697 netlink_update_subscriptions(sk, nlk->subscriptions +
698 hweight32(nladdr->nl_groups) -
699 hweight32(nlk->groups[0]));
700 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
701 netlink_update_listeners(sk);
702 netlink_table_ungrab();
703
704 if (nlk->netlink_bind && nlk->groups[0]) {
705 int i;
706
707 for (i=0; i<nlk->ngroups; i++) {
708 if (test_bit(i, nlk->groups))
709 nlk->netlink_bind(i);
710 }
711 }
712
713 return 0;
714 }
715
716 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
717 int alen, int flags)
718 {
719 int err = 0;
720 struct sock *sk = sock->sk;
721 struct netlink_sock *nlk = nlk_sk(sk);
722 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
723
724 if (alen < sizeof(addr->sa_family))
725 return -EINVAL;
726
727 if (addr->sa_family == AF_UNSPEC) {
728 sk->sk_state = NETLINK_UNCONNECTED;
729 nlk->dst_pid = 0;
730 nlk->dst_group = 0;
731 return 0;
732 }
733 if (addr->sa_family != AF_NETLINK)
734 return -EINVAL;
735
736 /* Only superuser is allowed to send multicasts */
737 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
738 return -EPERM;
739
740 if (!nlk->pid)
741 err = netlink_autobind(sock);
742
743 if (err == 0) {
744 sk->sk_state = NETLINK_CONNECTED;
745 nlk->dst_pid = nladdr->nl_pid;
746 nlk->dst_group = ffs(nladdr->nl_groups);
747 }
748
749 return err;
750 }
751
752 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
753 int *addr_len, int peer)
754 {
755 struct sock *sk = sock->sk;
756 struct netlink_sock *nlk = nlk_sk(sk);
757 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
758
759 nladdr->nl_family = AF_NETLINK;
760 nladdr->nl_pad = 0;
761 *addr_len = sizeof(*nladdr);
762
763 if (peer) {
764 nladdr->nl_pid = nlk->dst_pid;
765 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
766 } else {
767 nladdr->nl_pid = nlk->pid;
768 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
769 }
770 return 0;
771 }
772
773 static void netlink_overrun(struct sock *sk)
774 {
775 struct netlink_sock *nlk = nlk_sk(sk);
776
777 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
778 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
779 sk->sk_err = ENOBUFS;
780 sk->sk_error_report(sk);
781 }
782 }
783 atomic_inc(&sk->sk_drops);
784 }
785
786 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
787 {
788 struct sock *sock;
789 struct netlink_sock *nlk;
790
791 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
792 if (!sock)
793 return ERR_PTR(-ECONNREFUSED);
794
795 /* Don't bother queuing skb if kernel socket has no input function */
796 nlk = nlk_sk(sock);
797 if (sock->sk_state == NETLINK_CONNECTED &&
798 nlk->dst_pid != nlk_sk(ssk)->pid) {
799 sock_put(sock);
800 return ERR_PTR(-ECONNREFUSED);
801 }
802 return sock;
803 }
804
805 struct sock *netlink_getsockbyfilp(struct file *filp)
806 {
807 struct inode *inode = filp->f_path.dentry->d_inode;
808 struct sock *sock;
809
810 if (!S_ISSOCK(inode->i_mode))
811 return ERR_PTR(-ENOTSOCK);
812
813 sock = SOCKET_I(inode)->sk;
814 if (sock->sk_family != AF_NETLINK)
815 return ERR_PTR(-EINVAL);
816
817 sock_hold(sock);
818 return sock;
819 }
820
821 /*
822 * Attach a skb to a netlink socket.
823 * The caller must hold a reference to the destination socket. On error, the
824 * reference is dropped. The skb is not send to the destination, just all
825 * all error checks are performed and memory in the queue is reserved.
826 * Return values:
827 * < 0: error. skb freed, reference to sock dropped.
828 * 0: continue
829 * 1: repeat lookup - reference dropped while waiting for socket memory.
830 */
831 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
832 long *timeo, struct sock *ssk)
833 {
834 struct netlink_sock *nlk;
835
836 nlk = nlk_sk(sk);
837
838 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
839 test_bit(0, &nlk->state)) {
840 DECLARE_WAITQUEUE(wait, current);
841 if (!*timeo) {
842 if (!ssk || netlink_is_kernel(ssk))
843 netlink_overrun(sk);
844 sock_put(sk);
845 kfree_skb(skb);
846 return -EAGAIN;
847 }
848
849 __set_current_state(TASK_INTERRUPTIBLE);
850 add_wait_queue(&nlk->wait, &wait);
851
852 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
853 test_bit(0, &nlk->state)) &&
854 !sock_flag(sk, SOCK_DEAD))
855 *timeo = schedule_timeout(*timeo);
856
857 __set_current_state(TASK_RUNNING);
858 remove_wait_queue(&nlk->wait, &wait);
859 sock_put(sk);
860
861 if (signal_pending(current)) {
862 kfree_skb(skb);
863 return sock_intr_errno(*timeo);
864 }
865 return 1;
866 }
867 skb_set_owner_r(skb, sk);
868 return 0;
869 }
870
871 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
872 {
873 int len = skb->len;
874
875 skb_queue_tail(&sk->sk_receive_queue, skb);
876 sk->sk_data_ready(sk, len);
877 return len;
878 }
879
880 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
881 {
882 int len = __netlink_sendskb(sk, skb);
883
884 sock_put(sk);
885 return len;
886 }
887
888 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
889 {
890 kfree_skb(skb);
891 sock_put(sk);
892 }
893
894 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
895 {
896 int delta;
897
898 skb_orphan(skb);
899
900 delta = skb->end - skb->tail;
901 if (delta * 2 < skb->truesize)
902 return skb;
903
904 if (skb_shared(skb)) {
905 struct sk_buff *nskb = skb_clone(skb, allocation);
906 if (!nskb)
907 return skb;
908 consume_skb(skb);
909 skb = nskb;
910 }
911
912 if (!pskb_expand_head(skb, 0, -delta, allocation))
913 skb->truesize -= delta;
914
915 return skb;
916 }
917
918 static void netlink_rcv_wake(struct sock *sk)
919 {
920 struct netlink_sock *nlk = nlk_sk(sk);
921
922 if (skb_queue_empty(&sk->sk_receive_queue))
923 clear_bit(0, &nlk->state);
924 if (!test_bit(0, &nlk->state))
925 wake_up_interruptible(&nlk->wait);
926 }
927
928 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
929 {
930 int ret;
931 struct netlink_sock *nlk = nlk_sk(sk);
932
933 ret = -ECONNREFUSED;
934 if (nlk->netlink_rcv != NULL) {
935 ret = skb->len;
936 skb_set_owner_r(skb, sk);
937 nlk->netlink_rcv(skb);
938 consume_skb(skb);
939 } else {
940 kfree_skb(skb);
941 }
942 sock_put(sk);
943 return ret;
944 }
945
946 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
947 u32 pid, int nonblock)
948 {
949 struct sock *sk;
950 int err;
951 long timeo;
952
953 skb = netlink_trim(skb, gfp_any());
954
955 timeo = sock_sndtimeo(ssk, nonblock);
956 retry:
957 sk = netlink_getsockbypid(ssk, pid);
958 if (IS_ERR(sk)) {
959 kfree_skb(skb);
960 return PTR_ERR(sk);
961 }
962 if (netlink_is_kernel(sk))
963 return netlink_unicast_kernel(sk, skb);
964
965 if (sk_filter(sk, skb)) {
966 err = skb->len;
967 kfree_skb(skb);
968 sock_put(sk);
969 return err;
970 }
971
972 err = netlink_attachskb(sk, skb, &timeo, ssk);
973 if (err == 1)
974 goto retry;
975 if (err)
976 return err;
977
978 return netlink_sendskb(sk, skb);
979 }
980 EXPORT_SYMBOL(netlink_unicast);
981
982 int netlink_has_listeners(struct sock *sk, unsigned int group)
983 {
984 int res = 0;
985 struct listeners *listeners;
986
987 BUG_ON(!netlink_is_kernel(sk));
988
989 rcu_read_lock();
990 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
991
992 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
993 res = test_bit(group - 1, listeners->masks);
994
995 rcu_read_unlock();
996
997 return res;
998 }
999 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1000
1001 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1002 {
1003 struct netlink_sock *nlk = nlk_sk(sk);
1004
1005 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1006 !test_bit(0, &nlk->state)) {
1007 skb_set_owner_r(skb, sk);
1008 __netlink_sendskb(sk, skb);
1009 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1010 }
1011 return -1;
1012 }
1013
1014 struct netlink_broadcast_data {
1015 struct sock *exclude_sk;
1016 struct net *net;
1017 u32 pid;
1018 u32 group;
1019 int failure;
1020 int delivery_failure;
1021 int congested;
1022 int delivered;
1023 gfp_t allocation;
1024 struct sk_buff *skb, *skb2;
1025 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1026 void *tx_data;
1027 };
1028
1029 static int do_one_broadcast(struct sock *sk,
1030 struct netlink_broadcast_data *p)
1031 {
1032 struct netlink_sock *nlk = nlk_sk(sk);
1033 int val;
1034
1035 if (p->exclude_sk == sk)
1036 goto out;
1037
1038 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1039 !test_bit(p->group - 1, nlk->groups))
1040 goto out;
1041
1042 if (!net_eq(sock_net(sk), p->net))
1043 goto out;
1044
1045 if (p->failure) {
1046 netlink_overrun(sk);
1047 goto out;
1048 }
1049
1050 sock_hold(sk);
1051 if (p->skb2 == NULL) {
1052 if (skb_shared(p->skb)) {
1053 p->skb2 = skb_clone(p->skb, p->allocation);
1054 } else {
1055 p->skb2 = skb_get(p->skb);
1056 /*
1057 * skb ownership may have been set when
1058 * delivered to a previous socket.
1059 */
1060 skb_orphan(p->skb2);
1061 }
1062 }
1063 if (p->skb2 == NULL) {
1064 netlink_overrun(sk);
1065 /* Clone failed. Notify ALL listeners. */
1066 p->failure = 1;
1067 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1068 p->delivery_failure = 1;
1069 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1070 kfree_skb(p->skb2);
1071 p->skb2 = NULL;
1072 } else if (sk_filter(sk, p->skb2)) {
1073 kfree_skb(p->skb2);
1074 p->skb2 = NULL;
1075 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1076 netlink_overrun(sk);
1077 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1078 p->delivery_failure = 1;
1079 } else {
1080 p->congested |= val;
1081 p->delivered = 1;
1082 p->skb2 = NULL;
1083 }
1084 sock_put(sk);
1085
1086 out:
1087 return 0;
1088 }
1089
1090 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid,
1091 u32 group, gfp_t allocation,
1092 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1093 void *filter_data)
1094 {
1095 struct net *net = sock_net(ssk);
1096 struct netlink_broadcast_data info;
1097 struct hlist_node *node;
1098 struct sock *sk;
1099
1100 skb = netlink_trim(skb, allocation);
1101
1102 info.exclude_sk = ssk;
1103 info.net = net;
1104 info.pid = pid;
1105 info.group = group;
1106 info.failure = 0;
1107 info.delivery_failure = 0;
1108 info.congested = 0;
1109 info.delivered = 0;
1110 info.allocation = allocation;
1111 info.skb = skb;
1112 info.skb2 = NULL;
1113 info.tx_filter = filter;
1114 info.tx_data = filter_data;
1115
1116 /* While we sleep in clone, do not allow to change socket list */
1117
1118 netlink_lock_table();
1119
1120 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1121 do_one_broadcast(sk, &info);
1122
1123 consume_skb(skb);
1124
1125 netlink_unlock_table();
1126
1127 if (info.delivery_failure) {
1128 kfree_skb(info.skb2);
1129 return -ENOBUFS;
1130 }
1131 consume_skb(info.skb2);
1132
1133 if (info.delivered) {
1134 if (info.congested && (allocation & __GFP_WAIT))
1135 yield();
1136 return 0;
1137 }
1138 return -ESRCH;
1139 }
1140 EXPORT_SYMBOL(netlink_broadcast_filtered);
1141
1142 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1143 u32 group, gfp_t allocation)
1144 {
1145 return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1146 NULL, NULL);
1147 }
1148 EXPORT_SYMBOL(netlink_broadcast);
1149
1150 struct netlink_set_err_data {
1151 struct sock *exclude_sk;
1152 u32 pid;
1153 u32 group;
1154 int code;
1155 };
1156
1157 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1158 {
1159 struct netlink_sock *nlk = nlk_sk(sk);
1160 int ret = 0;
1161
1162 if (sk == p->exclude_sk)
1163 goto out;
1164
1165 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1166 goto out;
1167
1168 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1169 !test_bit(p->group - 1, nlk->groups))
1170 goto out;
1171
1172 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1173 ret = 1;
1174 goto out;
1175 }
1176
1177 sk->sk_err = p->code;
1178 sk->sk_error_report(sk);
1179 out:
1180 return ret;
1181 }
1182
1183 /**
1184 * netlink_set_err - report error to broadcast listeners
1185 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1186 * @pid: the PID of a process that we want to skip (if any)
1187 * @groups: the broadcast group that will notice the error
1188 * @code: error code, must be negative (as usual in kernelspace)
1189 *
1190 * This function returns the number of broadcast listeners that have set the
1191 * NETLINK_RECV_NO_ENOBUFS socket option.
1192 */
1193 int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1194 {
1195 struct netlink_set_err_data info;
1196 struct hlist_node *node;
1197 struct sock *sk;
1198 int ret = 0;
1199
1200 info.exclude_sk = ssk;
1201 info.pid = pid;
1202 info.group = group;
1203 /* sk->sk_err wants a positive error value */
1204 info.code = -code;
1205
1206 read_lock(&nl_table_lock);
1207
1208 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1209 ret += do_one_set_err(sk, &info);
1210
1211 read_unlock(&nl_table_lock);
1212 return ret;
1213 }
1214 EXPORT_SYMBOL(netlink_set_err);
1215
1216 /* must be called with netlink table grabbed */
1217 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1218 unsigned int group,
1219 int is_new)
1220 {
1221 int old, new = !!is_new, subscriptions;
1222
1223 old = test_bit(group - 1, nlk->groups);
1224 subscriptions = nlk->subscriptions - old + new;
1225 if (new)
1226 __set_bit(group - 1, nlk->groups);
1227 else
1228 __clear_bit(group - 1, nlk->groups);
1229 netlink_update_subscriptions(&nlk->sk, subscriptions);
1230 netlink_update_listeners(&nlk->sk);
1231 }
1232
1233 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1234 char __user *optval, unsigned int optlen)
1235 {
1236 struct sock *sk = sock->sk;
1237 struct netlink_sock *nlk = nlk_sk(sk);
1238 unsigned int val = 0;
1239 int err;
1240
1241 if (level != SOL_NETLINK)
1242 return -ENOPROTOOPT;
1243
1244 if (optlen >= sizeof(int) &&
1245 get_user(val, (unsigned int __user *)optval))
1246 return -EFAULT;
1247
1248 switch (optname) {
1249 case NETLINK_PKTINFO:
1250 if (val)
1251 nlk->flags |= NETLINK_RECV_PKTINFO;
1252 else
1253 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1254 err = 0;
1255 break;
1256 case NETLINK_ADD_MEMBERSHIP:
1257 case NETLINK_DROP_MEMBERSHIP: {
1258 if (!netlink_capable(sock, NL_NONROOT_RECV))
1259 return -EPERM;
1260 err = netlink_realloc_groups(sk);
1261 if (err)
1262 return err;
1263 if (!val || val - 1 >= nlk->ngroups)
1264 return -EINVAL;
1265 netlink_table_grab();
1266 netlink_update_socket_mc(nlk, val,
1267 optname == NETLINK_ADD_MEMBERSHIP);
1268 netlink_table_ungrab();
1269
1270 if (nlk->netlink_bind)
1271 nlk->netlink_bind(val);
1272
1273 err = 0;
1274 break;
1275 }
1276 case NETLINK_BROADCAST_ERROR:
1277 if (val)
1278 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1279 else
1280 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1281 err = 0;
1282 break;
1283 case NETLINK_NO_ENOBUFS:
1284 if (val) {
1285 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1286 clear_bit(0, &nlk->state);
1287 wake_up_interruptible(&nlk->wait);
1288 } else {
1289 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1290 }
1291 err = 0;
1292 break;
1293 default:
1294 err = -ENOPROTOOPT;
1295 }
1296 return err;
1297 }
1298
1299 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1300 char __user *optval, int __user *optlen)
1301 {
1302 struct sock *sk = sock->sk;
1303 struct netlink_sock *nlk = nlk_sk(sk);
1304 int len, val, err;
1305
1306 if (level != SOL_NETLINK)
1307 return -ENOPROTOOPT;
1308
1309 if (get_user(len, optlen))
1310 return -EFAULT;
1311 if (len < 0)
1312 return -EINVAL;
1313
1314 switch (optname) {
1315 case NETLINK_PKTINFO:
1316 if (len < sizeof(int))
1317 return -EINVAL;
1318 len = sizeof(int);
1319 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1320 if (put_user(len, optlen) ||
1321 put_user(val, optval))
1322 return -EFAULT;
1323 err = 0;
1324 break;
1325 case NETLINK_BROADCAST_ERROR:
1326 if (len < sizeof(int))
1327 return -EINVAL;
1328 len = sizeof(int);
1329 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1330 if (put_user(len, optlen) ||
1331 put_user(val, optval))
1332 return -EFAULT;
1333 err = 0;
1334 break;
1335 case NETLINK_NO_ENOBUFS:
1336 if (len < sizeof(int))
1337 return -EINVAL;
1338 len = sizeof(int);
1339 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1340 if (put_user(len, optlen) ||
1341 put_user(val, optval))
1342 return -EFAULT;
1343 err = 0;
1344 break;
1345 default:
1346 err = -ENOPROTOOPT;
1347 }
1348 return err;
1349 }
1350
1351 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1352 {
1353 struct nl_pktinfo info;
1354
1355 info.group = NETLINK_CB(skb).dst_group;
1356 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1357 }
1358
1359 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1360 struct msghdr *msg, size_t len)
1361 {
1362 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1363 struct sock *sk = sock->sk;
1364 struct netlink_sock *nlk = nlk_sk(sk);
1365 struct sockaddr_nl *addr = msg->msg_name;
1366 u32 dst_pid;
1367 u32 dst_group;
1368 struct sk_buff *skb;
1369 int err;
1370 struct scm_cookie scm;
1371
1372 if (msg->msg_flags&MSG_OOB)
1373 return -EOPNOTSUPP;
1374
1375 if (NULL == siocb->scm)
1376 siocb->scm = &scm;
1377
1378 err = scm_send(sock, msg, siocb->scm, true);
1379 if (err < 0)
1380 return err;
1381
1382 if (msg->msg_namelen) {
1383 err = -EINVAL;
1384 if (addr->nl_family != AF_NETLINK)
1385 goto out;
1386 dst_pid = addr->nl_pid;
1387 dst_group = ffs(addr->nl_groups);
1388 err = -EPERM;
1389 if ((dst_group || dst_pid) &&
1390 !netlink_capable(sock, NL_NONROOT_SEND))
1391 goto out;
1392 } else {
1393 dst_pid = nlk->dst_pid;
1394 dst_group = nlk->dst_group;
1395 }
1396
1397 if (!nlk->pid) {
1398 err = netlink_autobind(sock);
1399 if (err)
1400 goto out;
1401 }
1402
1403 err = -EMSGSIZE;
1404 if (len > sk->sk_sndbuf - 32)
1405 goto out;
1406 err = -ENOBUFS;
1407 skb = alloc_skb(len, GFP_KERNEL);
1408 if (skb == NULL)
1409 goto out;
1410
1411 NETLINK_CB(skb).pid = nlk->pid;
1412 NETLINK_CB(skb).dst_group = dst_group;
1413 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1414
1415 err = -EFAULT;
1416 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1417 kfree_skb(skb);
1418 goto out;
1419 }
1420
1421 err = security_netlink_send(sk, skb);
1422 if (err) {
1423 kfree_skb(skb);
1424 goto out;
1425 }
1426
1427 if (dst_group) {
1428 atomic_inc(&skb->users);
1429 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1430 }
1431 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1432
1433 out:
1434 scm_destroy(siocb->scm);
1435 return err;
1436 }
1437
1438 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1439 struct msghdr *msg, size_t len,
1440 int flags)
1441 {
1442 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1443 struct scm_cookie scm;
1444 struct sock *sk = sock->sk;
1445 struct netlink_sock *nlk = nlk_sk(sk);
1446 int noblock = flags&MSG_DONTWAIT;
1447 size_t copied;
1448 struct sk_buff *skb, *data_skb;
1449 int err, ret;
1450
1451 if (flags&MSG_OOB)
1452 return -EOPNOTSUPP;
1453
1454 copied = 0;
1455
1456 skb = skb_recv_datagram(sk, flags, noblock, &err);
1457 if (skb == NULL)
1458 goto out;
1459
1460 data_skb = skb;
1461
1462 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1463 if (unlikely(skb_shinfo(skb)->frag_list)) {
1464 /*
1465 * If this skb has a frag_list, then here that means that we
1466 * will have to use the frag_list skb's data for compat tasks
1467 * and the regular skb's data for normal (non-compat) tasks.
1468 *
1469 * If we need to send the compat skb, assign it to the
1470 * 'data_skb' variable so that it will be used below for data
1471 * copying. We keep 'skb' for everything else, including
1472 * freeing both later.
1473 */
1474 if (flags & MSG_CMSG_COMPAT)
1475 data_skb = skb_shinfo(skb)->frag_list;
1476 }
1477 #endif
1478
1479 msg->msg_namelen = 0;
1480
1481 copied = data_skb->len;
1482 if (len < copied) {
1483 msg->msg_flags |= MSG_TRUNC;
1484 copied = len;
1485 }
1486
1487 skb_reset_transport_header(data_skb);
1488 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1489
1490 if (msg->msg_name) {
1491 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1492 addr->nl_family = AF_NETLINK;
1493 addr->nl_pad = 0;
1494 addr->nl_pid = NETLINK_CB(skb).pid;
1495 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1496 msg->msg_namelen = sizeof(*addr);
1497 }
1498
1499 if (nlk->flags & NETLINK_RECV_PKTINFO)
1500 netlink_cmsg_recv_pktinfo(msg, skb);
1501
1502 if (NULL == siocb->scm) {
1503 memset(&scm, 0, sizeof(scm));
1504 siocb->scm = &scm;
1505 }
1506 siocb->scm->creds = *NETLINK_CREDS(skb);
1507 if (flags & MSG_TRUNC)
1508 copied = data_skb->len;
1509
1510 skb_free_datagram(sk, skb);
1511
1512 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1513 ret = netlink_dump(sk);
1514 if (ret) {
1515 sk->sk_err = ret;
1516 sk->sk_error_report(sk);
1517 }
1518 }
1519
1520 scm_recv(sock, msg, siocb->scm, flags);
1521 out:
1522 netlink_rcv_wake(sk);
1523 return err ? : copied;
1524 }
1525
1526 static void netlink_data_ready(struct sock *sk, int len)
1527 {
1528 BUG();
1529 }
1530
1531 /*
1532 * We export these functions to other modules. They provide a
1533 * complete set of kernel non-blocking support for message
1534 * queueing.
1535 */
1536
1537 struct sock *
1538 netlink_kernel_create(struct net *net, int unit,
1539 struct module *module,
1540 struct netlink_kernel_cfg *cfg)
1541 {
1542 struct socket *sock;
1543 struct sock *sk;
1544 struct netlink_sock *nlk;
1545 struct listeners *listeners = NULL;
1546 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1547 unsigned int groups;
1548
1549 BUG_ON(!nl_table);
1550
1551 if (unit < 0 || unit >= MAX_LINKS)
1552 return NULL;
1553
1554 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1555 return NULL;
1556
1557 /*
1558 * We have to just have a reference on the net from sk, but don't
1559 * get_net it. Besides, we cannot get and then put the net here.
1560 * So we create one inside init_net and the move it to net.
1561 */
1562
1563 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1564 goto out_sock_release_nosk;
1565
1566 sk = sock->sk;
1567 sk_change_net(sk, net);
1568
1569 if (!cfg || cfg->groups < 32)
1570 groups = 32;
1571 else
1572 groups = cfg->groups;
1573
1574 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1575 if (!listeners)
1576 goto out_sock_release;
1577
1578 sk->sk_data_ready = netlink_data_ready;
1579 if (cfg && cfg->input)
1580 nlk_sk(sk)->netlink_rcv = cfg->input;
1581
1582 if (netlink_insert(sk, net, 0))
1583 goto out_sock_release;
1584
1585 nlk = nlk_sk(sk);
1586 nlk->flags |= NETLINK_KERNEL_SOCKET;
1587
1588 netlink_table_grab();
1589 if (!nl_table[unit].registered) {
1590 nl_table[unit].groups = groups;
1591 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1592 nl_table[unit].cb_mutex = cb_mutex;
1593 nl_table[unit].module = module;
1594 nl_table[unit].bind = cfg ? cfg->bind : NULL;
1595 nl_table[unit].registered = 1;
1596 } else {
1597 kfree(listeners);
1598 nl_table[unit].registered++;
1599 }
1600 netlink_table_ungrab();
1601 return sk;
1602
1603 out_sock_release:
1604 kfree(listeners);
1605 netlink_kernel_release(sk);
1606 return NULL;
1607
1608 out_sock_release_nosk:
1609 sock_release(sock);
1610 return NULL;
1611 }
1612 EXPORT_SYMBOL(netlink_kernel_create);
1613
1614
1615 void
1616 netlink_kernel_release(struct sock *sk)
1617 {
1618 sk_release_kernel(sk);
1619 }
1620 EXPORT_SYMBOL(netlink_kernel_release);
1621
1622 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1623 {
1624 struct listeners *new, *old;
1625 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1626
1627 if (groups < 32)
1628 groups = 32;
1629
1630 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1631 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1632 if (!new)
1633 return -ENOMEM;
1634 old = nl_deref_protected(tbl->listeners);
1635 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1636 rcu_assign_pointer(tbl->listeners, new);
1637
1638 kfree_rcu(old, rcu);
1639 }
1640 tbl->groups = groups;
1641
1642 return 0;
1643 }
1644
1645 /**
1646 * netlink_change_ngroups - change number of multicast groups
1647 *
1648 * This changes the number of multicast groups that are available
1649 * on a certain netlink family. Note that it is not possible to
1650 * change the number of groups to below 32. Also note that it does
1651 * not implicitly call netlink_clear_multicast_users() when the
1652 * number of groups is reduced.
1653 *
1654 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1655 * @groups: The new number of groups.
1656 */
1657 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1658 {
1659 int err;
1660
1661 netlink_table_grab();
1662 err = __netlink_change_ngroups(sk, groups);
1663 netlink_table_ungrab();
1664
1665 return err;
1666 }
1667
1668 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1669 {
1670 struct sock *sk;
1671 struct hlist_node *node;
1672 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1673
1674 sk_for_each_bound(sk, node, &tbl->mc_list)
1675 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1676 }
1677
1678 /**
1679 * netlink_clear_multicast_users - kick off multicast listeners
1680 *
1681 * This function removes all listeners from the given group.
1682 * @ksk: The kernel netlink socket, as returned by
1683 * netlink_kernel_create().
1684 * @group: The multicast group to clear.
1685 */
1686 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1687 {
1688 netlink_table_grab();
1689 __netlink_clear_multicast_users(ksk, group);
1690 netlink_table_ungrab();
1691 }
1692
1693 void netlink_set_nonroot(int protocol, unsigned int flags)
1694 {
1695 if ((unsigned int)protocol < MAX_LINKS)
1696 nl_table[protocol].nl_nonroot = flags;
1697 }
1698 EXPORT_SYMBOL(netlink_set_nonroot);
1699
1700 struct nlmsghdr *
1701 __nlmsg_put(struct sk_buff *skb, u32 pid, u32 seq, int type, int len, int flags)
1702 {
1703 struct nlmsghdr *nlh;
1704 int size = NLMSG_LENGTH(len);
1705
1706 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
1707 nlh->nlmsg_type = type;
1708 nlh->nlmsg_len = size;
1709 nlh->nlmsg_flags = flags;
1710 nlh->nlmsg_pid = pid;
1711 nlh->nlmsg_seq = seq;
1712 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
1713 memset(NLMSG_DATA(nlh) + len, 0, NLMSG_ALIGN(size) - size);
1714 return nlh;
1715 }
1716 EXPORT_SYMBOL(__nlmsg_put);
1717
1718 /*
1719 * It looks a bit ugly.
1720 * It would be better to create kernel thread.
1721 */
1722
1723 static int netlink_dump(struct sock *sk)
1724 {
1725 struct netlink_sock *nlk = nlk_sk(sk);
1726 struct netlink_callback *cb;
1727 struct sk_buff *skb = NULL;
1728 struct nlmsghdr *nlh;
1729 int len, err = -ENOBUFS;
1730 int alloc_size;
1731
1732 mutex_lock(nlk->cb_mutex);
1733
1734 cb = nlk->cb;
1735 if (cb == NULL) {
1736 err = -EINVAL;
1737 goto errout_skb;
1738 }
1739
1740 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
1741
1742 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
1743 if (!skb)
1744 goto errout_skb;
1745
1746 len = cb->dump(skb, cb);
1747
1748 if (len > 0) {
1749 mutex_unlock(nlk->cb_mutex);
1750
1751 if (sk_filter(sk, skb))
1752 kfree_skb(skb);
1753 else
1754 __netlink_sendskb(sk, skb);
1755 return 0;
1756 }
1757
1758 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1759 if (!nlh)
1760 goto errout_skb;
1761
1762 nl_dump_check_consistent(cb, nlh);
1763
1764 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1765
1766 if (sk_filter(sk, skb))
1767 kfree_skb(skb);
1768 else
1769 __netlink_sendskb(sk, skb);
1770
1771 if (cb->done)
1772 cb->done(cb);
1773 nlk->cb = NULL;
1774 mutex_unlock(nlk->cb_mutex);
1775
1776 module_put(cb->module);
1777 netlink_consume_callback(cb);
1778 return 0;
1779
1780 errout_skb:
1781 mutex_unlock(nlk->cb_mutex);
1782 kfree_skb(skb);
1783 return err;
1784 }
1785
1786 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1787 const struct nlmsghdr *nlh,
1788 struct netlink_dump_control *control)
1789 {
1790 struct netlink_callback *cb;
1791 struct sock *sk;
1792 struct netlink_sock *nlk;
1793 int ret;
1794
1795 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1796 if (cb == NULL)
1797 return -ENOBUFS;
1798
1799 cb->dump = control->dump;
1800 cb->done = control->done;
1801 cb->nlh = nlh;
1802 cb->data = control->data;
1803 cb->module = control->module;
1804 cb->min_dump_alloc = control->min_dump_alloc;
1805 atomic_inc(&skb->users);
1806 cb->skb = skb;
1807
1808 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1809 if (sk == NULL) {
1810 netlink_destroy_callback(cb);
1811 return -ECONNREFUSED;
1812 }
1813 nlk = nlk_sk(sk);
1814
1815 mutex_lock(nlk->cb_mutex);
1816 /* A dump is in progress... */
1817 if (nlk->cb) {
1818 mutex_unlock(nlk->cb_mutex);
1819 netlink_destroy_callback(cb);
1820 ret = -EBUSY;
1821 goto out;
1822 }
1823 /* add reference of module which cb->dump belongs to */
1824 if (!try_module_get(cb->module)) {
1825 mutex_unlock(nlk->cb_mutex);
1826 netlink_destroy_callback(cb);
1827 ret = -EPROTONOSUPPORT;
1828 goto out;
1829 }
1830
1831 nlk->cb = cb;
1832 mutex_unlock(nlk->cb_mutex);
1833
1834 ret = netlink_dump(sk);
1835 out:
1836 sock_put(sk);
1837
1838 if (ret)
1839 return ret;
1840
1841 /* We successfully started a dump, by returning -EINTR we
1842 * signal not to send ACK even if it was requested.
1843 */
1844 return -EINTR;
1845 }
1846 EXPORT_SYMBOL(__netlink_dump_start);
1847
1848 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1849 {
1850 struct sk_buff *skb;
1851 struct nlmsghdr *rep;
1852 struct nlmsgerr *errmsg;
1853 size_t payload = sizeof(*errmsg);
1854
1855 /* error messages get the original request appened */
1856 if (err)
1857 payload += nlmsg_len(nlh);
1858
1859 skb = nlmsg_new(payload, GFP_KERNEL);
1860 if (!skb) {
1861 struct sock *sk;
1862
1863 sk = netlink_lookup(sock_net(in_skb->sk),
1864 in_skb->sk->sk_protocol,
1865 NETLINK_CB(in_skb).pid);
1866 if (sk) {
1867 sk->sk_err = ENOBUFS;
1868 sk->sk_error_report(sk);
1869 sock_put(sk);
1870 }
1871 return;
1872 }
1873
1874 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1875 NLMSG_ERROR, payload, 0);
1876 errmsg = nlmsg_data(rep);
1877 errmsg->error = err;
1878 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1879 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1880 }
1881 EXPORT_SYMBOL(netlink_ack);
1882
1883 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1884 struct nlmsghdr *))
1885 {
1886 struct nlmsghdr *nlh;
1887 int err;
1888
1889 while (skb->len >= nlmsg_total_size(0)) {
1890 int msglen;
1891
1892 nlh = nlmsg_hdr(skb);
1893 err = 0;
1894
1895 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1896 return 0;
1897
1898 /* Only requests are handled by the kernel */
1899 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1900 goto ack;
1901
1902 /* Skip control messages */
1903 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1904 goto ack;
1905
1906 err = cb(skb, nlh);
1907 if (err == -EINTR)
1908 goto skip;
1909
1910 ack:
1911 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1912 netlink_ack(skb, nlh, err);
1913
1914 skip:
1915 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1916 if (msglen > skb->len)
1917 msglen = skb->len;
1918 skb_pull(skb, msglen);
1919 }
1920
1921 return 0;
1922 }
1923 EXPORT_SYMBOL(netlink_rcv_skb);
1924
1925 /**
1926 * nlmsg_notify - send a notification netlink message
1927 * @sk: netlink socket to use
1928 * @skb: notification message
1929 * @pid: destination netlink pid for reports or 0
1930 * @group: destination multicast group or 0
1931 * @report: 1 to report back, 0 to disable
1932 * @flags: allocation flags
1933 */
1934 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1935 unsigned int group, int report, gfp_t flags)
1936 {
1937 int err = 0;
1938
1939 if (group) {
1940 int exclude_pid = 0;
1941
1942 if (report) {
1943 atomic_inc(&skb->users);
1944 exclude_pid = pid;
1945 }
1946
1947 /* errors reported via destination sk->sk_err, but propagate
1948 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1949 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1950 }
1951
1952 if (report) {
1953 int err2;
1954
1955 err2 = nlmsg_unicast(sk, skb, pid);
1956 if (!err || err == -ESRCH)
1957 err = err2;
1958 }
1959
1960 return err;
1961 }
1962 EXPORT_SYMBOL(nlmsg_notify);
1963
1964 #ifdef CONFIG_PROC_FS
1965 struct nl_seq_iter {
1966 struct seq_net_private p;
1967 int link;
1968 int hash_idx;
1969 };
1970
1971 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1972 {
1973 struct nl_seq_iter *iter = seq->private;
1974 int i, j;
1975 struct sock *s;
1976 struct hlist_node *node;
1977 loff_t off = 0;
1978
1979 for (i = 0; i < MAX_LINKS; i++) {
1980 struct nl_pid_hash *hash = &nl_table[i].hash;
1981
1982 for (j = 0; j <= hash->mask; j++) {
1983 sk_for_each(s, node, &hash->table[j]) {
1984 if (sock_net(s) != seq_file_net(seq))
1985 continue;
1986 if (off == pos) {
1987 iter->link = i;
1988 iter->hash_idx = j;
1989 return s;
1990 }
1991 ++off;
1992 }
1993 }
1994 }
1995 return NULL;
1996 }
1997
1998 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1999 __acquires(nl_table_lock)
2000 {
2001 read_lock(&nl_table_lock);
2002 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2003 }
2004
2005 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2006 {
2007 struct sock *s;
2008 struct nl_seq_iter *iter;
2009 int i, j;
2010
2011 ++*pos;
2012
2013 if (v == SEQ_START_TOKEN)
2014 return netlink_seq_socket_idx(seq, 0);
2015
2016 iter = seq->private;
2017 s = v;
2018 do {
2019 s = sk_next(s);
2020 } while (s && sock_net(s) != seq_file_net(seq));
2021 if (s)
2022 return s;
2023
2024 i = iter->link;
2025 j = iter->hash_idx + 1;
2026
2027 do {
2028 struct nl_pid_hash *hash = &nl_table[i].hash;
2029
2030 for (; j <= hash->mask; j++) {
2031 s = sk_head(&hash->table[j]);
2032 while (s && sock_net(s) != seq_file_net(seq))
2033 s = sk_next(s);
2034 if (s) {
2035 iter->link = i;
2036 iter->hash_idx = j;
2037 return s;
2038 }
2039 }
2040
2041 j = 0;
2042 } while (++i < MAX_LINKS);
2043
2044 return NULL;
2045 }
2046
2047 static void netlink_seq_stop(struct seq_file *seq, void *v)
2048 __releases(nl_table_lock)
2049 {
2050 read_unlock(&nl_table_lock);
2051 }
2052
2053
2054 static int netlink_seq_show(struct seq_file *seq, void *v)
2055 {
2056 if (v == SEQ_START_TOKEN) {
2057 seq_puts(seq,
2058 "sk Eth Pid Groups "
2059 "Rmem Wmem Dump Locks Drops Inode\n");
2060 } else {
2061 struct sock *s = v;
2062 struct netlink_sock *nlk = nlk_sk(s);
2063
2064 seq_printf(seq, "%pK %-3d %-6d %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2065 s,
2066 s->sk_protocol,
2067 nlk->pid,
2068 nlk->groups ? (u32)nlk->groups[0] : 0,
2069 sk_rmem_alloc_get(s),
2070 sk_wmem_alloc_get(s),
2071 nlk->cb,
2072 atomic_read(&s->sk_refcnt),
2073 atomic_read(&s->sk_drops),
2074 sock_i_ino(s)
2075 );
2076
2077 }
2078 return 0;
2079 }
2080
2081 static const struct seq_operations netlink_seq_ops = {
2082 .start = netlink_seq_start,
2083 .next = netlink_seq_next,
2084 .stop = netlink_seq_stop,
2085 .show = netlink_seq_show,
2086 };
2087
2088
2089 static int netlink_seq_open(struct inode *inode, struct file *file)
2090 {
2091 return seq_open_net(inode, file, &netlink_seq_ops,
2092 sizeof(struct nl_seq_iter));
2093 }
2094
2095 static const struct file_operations netlink_seq_fops = {
2096 .owner = THIS_MODULE,
2097 .open = netlink_seq_open,
2098 .read = seq_read,
2099 .llseek = seq_lseek,
2100 .release = seq_release_net,
2101 };
2102
2103 #endif
2104
2105 int netlink_register_notifier(struct notifier_block *nb)
2106 {
2107 return atomic_notifier_chain_register(&netlink_chain, nb);
2108 }
2109 EXPORT_SYMBOL(netlink_register_notifier);
2110
2111 int netlink_unregister_notifier(struct notifier_block *nb)
2112 {
2113 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2114 }
2115 EXPORT_SYMBOL(netlink_unregister_notifier);
2116
2117 static const struct proto_ops netlink_ops = {
2118 .family = PF_NETLINK,
2119 .owner = THIS_MODULE,
2120 .release = netlink_release,
2121 .bind = netlink_bind,
2122 .connect = netlink_connect,
2123 .socketpair = sock_no_socketpair,
2124 .accept = sock_no_accept,
2125 .getname = netlink_getname,
2126 .poll = datagram_poll,
2127 .ioctl = sock_no_ioctl,
2128 .listen = sock_no_listen,
2129 .shutdown = sock_no_shutdown,
2130 .setsockopt = netlink_setsockopt,
2131 .getsockopt = netlink_getsockopt,
2132 .sendmsg = netlink_sendmsg,
2133 .recvmsg = netlink_recvmsg,
2134 .mmap = sock_no_mmap,
2135 .sendpage = sock_no_sendpage,
2136 };
2137
2138 static const struct net_proto_family netlink_family_ops = {
2139 .family = PF_NETLINK,
2140 .create = netlink_create,
2141 .owner = THIS_MODULE, /* for consistency 8) */
2142 };
2143
2144 static int __net_init netlink_net_init(struct net *net)
2145 {
2146 #ifdef CONFIG_PROC_FS
2147 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2148 return -ENOMEM;
2149 #endif
2150 return 0;
2151 }
2152
2153 static void __net_exit netlink_net_exit(struct net *net)
2154 {
2155 #ifdef CONFIG_PROC_FS
2156 proc_net_remove(net, "netlink");
2157 #endif
2158 }
2159
2160 static void __init netlink_add_usersock_entry(void)
2161 {
2162 struct listeners *listeners;
2163 int groups = 32;
2164
2165 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2166 if (!listeners)
2167 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2168
2169 netlink_table_grab();
2170
2171 nl_table[NETLINK_USERSOCK].groups = groups;
2172 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2173 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2174 nl_table[NETLINK_USERSOCK].registered = 1;
2175 nl_table[NETLINK_USERSOCK].nl_nonroot = NL_NONROOT_SEND;
2176
2177 netlink_table_ungrab();
2178 }
2179
2180 static struct pernet_operations __net_initdata netlink_net_ops = {
2181 .init = netlink_net_init,
2182 .exit = netlink_net_exit,
2183 };
2184
2185 static int __init netlink_proto_init(void)
2186 {
2187 struct sk_buff *dummy_skb;
2188 int i;
2189 unsigned long limit;
2190 unsigned int order;
2191 int err = proto_register(&netlink_proto, 0);
2192
2193 if (err != 0)
2194 goto out;
2195
2196 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2197
2198 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2199 if (!nl_table)
2200 goto panic;
2201
2202 if (totalram_pages >= (128 * 1024))
2203 limit = totalram_pages >> (21 - PAGE_SHIFT);
2204 else
2205 limit = totalram_pages >> (23 - PAGE_SHIFT);
2206
2207 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2208 limit = (1UL << order) / sizeof(struct hlist_head);
2209 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2210
2211 for (i = 0; i < MAX_LINKS; i++) {
2212 struct nl_pid_hash *hash = &nl_table[i].hash;
2213
2214 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2215 if (!hash->table) {
2216 while (i-- > 0)
2217 nl_pid_hash_free(nl_table[i].hash.table,
2218 1 * sizeof(*hash->table));
2219 kfree(nl_table);
2220 goto panic;
2221 }
2222 hash->max_shift = order;
2223 hash->shift = 0;
2224 hash->mask = 0;
2225 hash->rehash_time = jiffies;
2226 }
2227
2228 netlink_add_usersock_entry();
2229
2230 sock_register(&netlink_family_ops);
2231 register_pernet_subsys(&netlink_net_ops);
2232 /* The netlink device handler may be needed early. */
2233 rtnetlink_init();
2234 out:
2235 return err;
2236 panic:
2237 panic("netlink_init: Cannot allocate nl_table\n");
2238 }
2239
2240 core_initcall(netlink_proto_init);
Linux ARM platform port for MOXA ART SoC