?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held
[linux/fpc-iii.git] / net / netlink / af_netlink.c
blob24bc620b539aa74f4af32c631c1a57066af8b5f6
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 struct module *module;
84 };
85
86 struct listeners {
87 struct rcu_head rcu;
88 unsigned long masks[0];
89 };
90
91 #define NETLINK_KERNEL_SOCKET 0x1
92 #define NETLINK_RECV_PKTINFO 0x2
93 #define NETLINK_BROADCAST_SEND_ERROR 0x4
94 #define NETLINK_RECV_NO_ENOBUFS 0x8
95
96 static inline struct netlink_sock *nlk_sk(struct sock *sk)
97 {
98 return container_of(sk, struct netlink_sock, sk);
99 }
100
101 static inline int netlink_is_kernel(struct sock *sk)
102 {
103 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
104 }
105
106 struct nl_pid_hash {
107 struct hlist_head *table;
108 unsigned long rehash_time;
109
110 unsigned int mask;
111 unsigned int shift;
112
113 unsigned int entries;
114 unsigned int max_shift;
115
116 u32 rnd;
117 };
118
119 struct netlink_table {
120 struct nl_pid_hash hash;
121 struct hlist_head mc_list;
122 struct listeners __rcu *listeners;
123 unsigned int nl_nonroot;
124 unsigned int groups;
125 struct mutex *cb_mutex;
126 struct module *module;
127 int registered;
128 };
129
130 static struct netlink_table *nl_table;
131
132 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
133
134 static int netlink_dump(struct sock *sk);
135 static void netlink_destroy_callback(struct netlink_callback *cb);
136
137 static DEFINE_RWLOCK(nl_table_lock);
138 static atomic_t nl_table_users = ATOMIC_INIT(0);
139
140 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
141
142 static u32 netlink_group_mask(u32 group)
143 {
144 return group ? 1 << (group - 1) : 0;
145 }
146
147 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
148 {
149 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
150 }
151
152 static void netlink_sock_destruct(struct sock *sk)
153 {
154 struct netlink_sock *nlk = nlk_sk(sk);
155
156 if (nlk->cb) {
157 if (nlk->cb->done)
158 nlk->cb->done(nlk->cb);
159 netlink_destroy_callback(nlk->cb);
160 }
161
162 skb_queue_purge(&sk->sk_receive_queue);
163
164 if (!sock_flag(sk, SOCK_DEAD)) {
165 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
166 return;
167 }
168
169 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
170 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
171 WARN_ON(nlk_sk(sk)->groups);
172 }
173
174 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
175 * SMP. Look, when several writers sleep and reader wakes them up, all but one
176 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
177 * this, _but_ remember, it adds useless work on UP machines.
178 */
179
180 void netlink_table_grab(void)
181 __acquires(nl_table_lock)
182 {
183 might_sleep();
184
185 write_lock_irq(&nl_table_lock);
186
187 if (atomic_read(&nl_table_users)) {
188 DECLARE_WAITQUEUE(wait, current);
189
190 add_wait_queue_exclusive(&nl_table_wait, &wait);
191 for (;;) {
192 set_current_state(TASK_UNINTERRUPTIBLE);
193 if (atomic_read(&nl_table_users) == 0)
194 break;
195 write_unlock_irq(&nl_table_lock);
196 schedule();
197 write_lock_irq(&nl_table_lock);
198 }
199
200 __set_current_state(TASK_RUNNING);
201 remove_wait_queue(&nl_table_wait, &wait);
202 }
203 }
204
205 void netlink_table_ungrab(void)
206 __releases(nl_table_lock)
207 {
208 write_unlock_irq(&nl_table_lock);
209 wake_up(&nl_table_wait);
210 }
211
212 static inline void
213 netlink_lock_table(void)
214 {
215 /* read_lock() synchronizes us to netlink_table_grab */
216
217 read_lock(&nl_table_lock);
218 atomic_inc(&nl_table_users);
219 read_unlock(&nl_table_lock);
220 }
221
222 static inline void
223 netlink_unlock_table(void)
224 {
225 if (atomic_dec_and_test(&nl_table_users))
226 wake_up(&nl_table_wait);
227 }
228
229 static inline struct sock *netlink_lookup(struct net *net, int protocol,
230 u32 pid)
231 {
232 struct nl_pid_hash *hash = &nl_table[protocol].hash;
233 struct hlist_head *head;
234 struct sock *sk;
235 struct hlist_node *node;
236
237 read_lock(&nl_table_lock);
238 head = nl_pid_hashfn(hash, pid);
239 sk_for_each(sk, node, head) {
240 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
241 sock_hold(sk);
242 goto found;
243 }
244 }
245 sk = NULL;
246 found:
247 read_unlock(&nl_table_lock);
248 return sk;
249 }
250
251 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
252 {
253 if (size <= PAGE_SIZE)
254 return kzalloc(size, GFP_ATOMIC);
255 else
256 return (struct hlist_head *)
257 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
258 get_order(size));
259 }
260
261 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
262 {
263 if (size <= PAGE_SIZE)
264 kfree(table);
265 else
266 free_pages((unsigned long)table, get_order(size));
267 }
268
269 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
270 {
271 unsigned int omask, mask, shift;
272 size_t osize, size;
273 struct hlist_head *otable, *table;
274 int i;
275
276 omask = mask = hash->mask;
277 osize = size = (mask + 1) * sizeof(*table);
278 shift = hash->shift;
279
280 if (grow) {
281 if (++shift > hash->max_shift)
282 return 0;
283 mask = mask * 2 + 1;
284 size *= 2;
285 }
286
287 table = nl_pid_hash_zalloc(size);
288 if (!table)
289 return 0;
290
291 otable = hash->table;
292 hash->table = table;
293 hash->mask = mask;
294 hash->shift = shift;
295 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
296
297 for (i = 0; i <= omask; i++) {
298 struct sock *sk;
299 struct hlist_node *node, *tmp;
300
301 sk_for_each_safe(sk, node, tmp, &otable[i])
302 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
303 }
304
305 nl_pid_hash_free(otable, osize);
306 hash->rehash_time = jiffies + 10 * 60 * HZ;
307 return 1;
308 }
309
310 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
311 {
312 int avg = hash->entries >> hash->shift;
313
314 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
315 return 1;
316
317 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
318 nl_pid_hash_rehash(hash, 0);
319 return 1;
320 }
321
322 return 0;
323 }
324
325 static const struct proto_ops netlink_ops;
326
327 static void
328 netlink_update_listeners(struct sock *sk)
329 {
330 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
331 struct hlist_node *node;
332 unsigned long mask;
333 unsigned int i;
334
335 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
336 mask = 0;
337 sk_for_each_bound(sk, node, &tbl->mc_list) {
338 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
339 mask |= nlk_sk(sk)->groups[i];
340 }
341 tbl->listeners->masks[i] = mask;
342 }
343 /* this function is only called with the netlink table "grabbed", which
344 * makes sure updates are visible before bind or setsockopt return. */
345 }
346
347 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
348 {
349 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
350 struct hlist_head *head;
351 int err = -EADDRINUSE;
352 struct sock *osk;
353 struct hlist_node *node;
354 int len;
355
356 netlink_table_grab();
357 head = nl_pid_hashfn(hash, pid);
358 len = 0;
359 sk_for_each(osk, node, head) {
360 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
361 break;
362 len++;
363 }
364 if (node)
365 goto err;
366
367 err = -EBUSY;
368 if (nlk_sk(sk)->pid)
369 goto err;
370
371 err = -ENOMEM;
372 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
373 goto err;
374
375 if (len && nl_pid_hash_dilute(hash, len))
376 head = nl_pid_hashfn(hash, pid);
377 hash->entries++;
378 nlk_sk(sk)->pid = pid;
379 sk_add_node(sk, head);
380 err = 0;
381
382 err:
383 netlink_table_ungrab();
384 return err;
385 }
386
387 static void netlink_remove(struct sock *sk)
388 {
389 netlink_table_grab();
390 if (sk_del_node_init(sk))
391 nl_table[sk->sk_protocol].hash.entries--;
392 if (nlk_sk(sk)->subscriptions)
393 __sk_del_bind_node(sk);
394 netlink_table_ungrab();
395 }
396
397 static struct proto netlink_proto = {
398 .name = "NETLINK",
399 .owner = THIS_MODULE,
400 .obj_size = sizeof(struct netlink_sock),
401 };
402
403 static int __netlink_create(struct net *net, struct socket *sock,
404 struct mutex *cb_mutex, int protocol)
405 {
406 struct sock *sk;
407 struct netlink_sock *nlk;
408
409 sock->ops = &netlink_ops;
410
411 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
412 if (!sk)
413 return -ENOMEM;
414
415 sock_init_data(sock, sk);
416
417 nlk = nlk_sk(sk);
418 if (cb_mutex)
419 nlk->cb_mutex = cb_mutex;
420 else {
421 nlk->cb_mutex = &nlk->cb_def_mutex;
422 mutex_init(nlk->cb_mutex);
423 }
424 init_waitqueue_head(&nlk->wait);
425
426 sk->sk_destruct = netlink_sock_destruct;
427 sk->sk_protocol = protocol;
428 return 0;
429 }
430
431 static int netlink_create(struct net *net, struct socket *sock, int protocol,
432 int kern)
433 {
434 struct module *module = NULL;
435 struct mutex *cb_mutex;
436 struct netlink_sock *nlk;
437 int err = 0;
438
439 sock->state = SS_UNCONNECTED;
440
441 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
442 return -ESOCKTNOSUPPORT;
443
444 if (protocol < 0 || protocol >= MAX_LINKS)
445 return -EPROTONOSUPPORT;
446
447 netlink_lock_table();
448 #ifdef CONFIG_MODULES
449 if (!nl_table[protocol].registered) {
450 netlink_unlock_table();
451 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
452 netlink_lock_table();
453 }
454 #endif
455 if (nl_table[protocol].registered &&
456 try_module_get(nl_table[protocol].module))
457 module = nl_table[protocol].module;
458 else
459 err = -EPROTONOSUPPORT;
460 cb_mutex = nl_table[protocol].cb_mutex;
461 netlink_unlock_table();
462
463 if (err < 0)
464 goto out;
465
466 err = __netlink_create(net, sock, cb_mutex, protocol);
467 if (err < 0)
468 goto out_module;
469
470 local_bh_disable();
471 sock_prot_inuse_add(net, &netlink_proto, 1);
472 local_bh_enable();
473
474 nlk = nlk_sk(sock->sk);
475 nlk->module = module;
476 out:
477 return err;
478
479 out_module:
480 module_put(module);
481 goto out;
482 }
483
484 static int netlink_release(struct socket *sock)
485 {
486 struct sock *sk = sock->sk;
487 struct netlink_sock *nlk;
488
489 if (!sk)
490 return 0;
491
492 netlink_remove(sk);
493 sock_orphan(sk);
494 nlk = nlk_sk(sk);
495
496 /*
497 * OK. Socket is unlinked, any packets that arrive now
498 * will be purged.
499 */
500
501 sock->sk = NULL;
502 wake_up_interruptible_all(&nlk->wait);
503
504 skb_queue_purge(&sk->sk_write_queue);
505
506 if (nlk->pid) {
507 struct netlink_notify n = {
508 .net = sock_net(sk),
509 .protocol = sk->sk_protocol,
510 .pid = nlk->pid,
511 };
512 atomic_notifier_call_chain(&netlink_chain,
513 NETLINK_URELEASE, &n);
514 }
515
516 module_put(nlk->module);
517
518 netlink_table_grab();
519 if (netlink_is_kernel(sk)) {
520 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
521 if (--nl_table[sk->sk_protocol].registered == 0) {
522 kfree(nl_table[sk->sk_protocol].listeners);
523 nl_table[sk->sk_protocol].module = NULL;
524 nl_table[sk->sk_protocol].registered = 0;
525 }
526 } else if (nlk->subscriptions)
527 netlink_update_listeners(sk);
528 netlink_table_ungrab();
529
530 kfree(nlk->groups);
531 nlk->groups = NULL;
532
533 local_bh_disable();
534 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
535 local_bh_enable();
536 sock_put(sk);
537 return 0;
538 }
539
540 static int netlink_autobind(struct socket *sock)
541 {
542 struct sock *sk = sock->sk;
543 struct net *net = sock_net(sk);
544 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
545 struct hlist_head *head;
546 struct sock *osk;
547 struct hlist_node *node;
548 s32 pid = task_tgid_vnr(current);
549 int err;
550 static s32 rover = -4097;
551
552 retry:
553 cond_resched();
554 netlink_table_grab();
555 head = nl_pid_hashfn(hash, pid);
556 sk_for_each(osk, node, head) {
557 if (!net_eq(sock_net(osk), net))
558 continue;
559 if (nlk_sk(osk)->pid == pid) {
560 /* Bind collision, search negative pid values. */
561 pid = rover--;
562 if (rover > -4097)
563 rover = -4097;
564 netlink_table_ungrab();
565 goto retry;
566 }
567 }
568 netlink_table_ungrab();
569
570 err = netlink_insert(sk, net, pid);
571 if (err == -EADDRINUSE)
572 goto retry;
573
574 /* If 2 threads race to autobind, that is fine. */
575 if (err == -EBUSY)
576 err = 0;
577
578 return err;
579 }
580
581 static inline int netlink_capable(struct socket *sock, unsigned int flag)
582 {
583 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
584 capable(CAP_NET_ADMIN);
585 }
586
587 static void
588 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
589 {
590 struct netlink_sock *nlk = nlk_sk(sk);
591
592 if (nlk->subscriptions && !subscriptions)
593 __sk_del_bind_node(sk);
594 else if (!nlk->subscriptions && subscriptions)
595 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
596 nlk->subscriptions = subscriptions;
597 }
598
599 static int netlink_realloc_groups(struct sock *sk)
600 {
601 struct netlink_sock *nlk = nlk_sk(sk);
602 unsigned int groups;
603 unsigned long *new_groups;
604 int err = 0;
605
606 netlink_table_grab();
607
608 groups = nl_table[sk->sk_protocol].groups;
609 if (!nl_table[sk->sk_protocol].registered) {
610 err = -ENOENT;
611 goto out_unlock;
612 }
613
614 if (nlk->ngroups >= groups)
615 goto out_unlock;
616
617 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
618 if (new_groups == NULL) {
619 err = -ENOMEM;
620 goto out_unlock;
621 }
622 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
623 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
624
625 nlk->groups = new_groups;
626 nlk->ngroups = groups;
627 out_unlock:
628 netlink_table_ungrab();
629 return err;
630 }
631
632 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
633 int addr_len)
634 {
635 struct sock *sk = sock->sk;
636 struct net *net = sock_net(sk);
637 struct netlink_sock *nlk = nlk_sk(sk);
638 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
639 int err;
640
641 if (nladdr->nl_family != AF_NETLINK)
642 return -EINVAL;
643
644 /* Only superuser is allowed to listen multicasts */
645 if (nladdr->nl_groups) {
646 if (!netlink_capable(sock, NL_NONROOT_RECV))
647 return -EPERM;
648 err = netlink_realloc_groups(sk);
649 if (err)
650 return err;
651 }
652
653 if (nlk->pid) {
654 if (nladdr->nl_pid != nlk->pid)
655 return -EINVAL;
656 } else {
657 err = nladdr->nl_pid ?
658 netlink_insert(sk, net, nladdr->nl_pid) :
659 netlink_autobind(sock);
660 if (err)
661 return err;
662 }
663
664 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
665 return 0;
666
667 netlink_table_grab();
668 netlink_update_subscriptions(sk, nlk->subscriptions +
669 hweight32(nladdr->nl_groups) -
670 hweight32(nlk->groups[0]));
671 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
672 netlink_update_listeners(sk);
673 netlink_table_ungrab();
674
675 return 0;
676 }
677
678 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
679 int alen, int flags)
680 {
681 int err = 0;
682 struct sock *sk = sock->sk;
683 struct netlink_sock *nlk = nlk_sk(sk);
684 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
685
686 if (alen < sizeof(addr->sa_family))
687 return -EINVAL;
688
689 if (addr->sa_family == AF_UNSPEC) {
690 sk->sk_state = NETLINK_UNCONNECTED;
691 nlk->dst_pid = 0;
692 nlk->dst_group = 0;
693 return 0;
694 }
695 if (addr->sa_family != AF_NETLINK)
696 return -EINVAL;
697
698 /* Only superuser is allowed to send multicasts */
699 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
700 return -EPERM;
701
702 if (!nlk->pid)
703 err = netlink_autobind(sock);
704
705 if (err == 0) {
706 sk->sk_state = NETLINK_CONNECTED;
707 nlk->dst_pid = nladdr->nl_pid;
708 nlk->dst_group = ffs(nladdr->nl_groups);
709 }
710
711 return err;
712 }
713
714 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
715 int *addr_len, int peer)
716 {
717 struct sock *sk = sock->sk;
718 struct netlink_sock *nlk = nlk_sk(sk);
719 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
720
721 nladdr->nl_family = AF_NETLINK;
722 nladdr->nl_pad = 0;
723 *addr_len = sizeof(*nladdr);
724
725 if (peer) {
726 nladdr->nl_pid = nlk->dst_pid;
727 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
728 } else {
729 nladdr->nl_pid = nlk->pid;
730 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
731 }
732 return 0;
733 }
734
735 static void netlink_overrun(struct sock *sk)
736 {
737 struct netlink_sock *nlk = nlk_sk(sk);
738
739 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
740 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
741 sk->sk_err = ENOBUFS;
742 sk->sk_error_report(sk);
743 }
744 }
745 atomic_inc(&sk->sk_drops);
746 }
747
748 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
749 {
750 struct sock *sock;
751 struct netlink_sock *nlk;
752
753 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
754 if (!sock)
755 return ERR_PTR(-ECONNREFUSED);
756
757 /* Don't bother queuing skb if kernel socket has no input function */
758 nlk = nlk_sk(sock);
759 if (sock->sk_state == NETLINK_CONNECTED &&
760 nlk->dst_pid != nlk_sk(ssk)->pid) {
761 sock_put(sock);
762 return ERR_PTR(-ECONNREFUSED);
763 }
764 return sock;
765 }
766
767 struct sock *netlink_getsockbyfilp(struct file *filp)
768 {
769 struct inode *inode = filp->f_path.dentry->d_inode;
770 struct sock *sock;
771
772 if (!S_ISSOCK(inode->i_mode))
773 return ERR_PTR(-ENOTSOCK);
774
775 sock = SOCKET_I(inode)->sk;
776 if (sock->sk_family != AF_NETLINK)
777 return ERR_PTR(-EINVAL);
778
779 sock_hold(sock);
780 return sock;
781 }
782
783 /*
784 * Attach a skb to a netlink socket.
785 * The caller must hold a reference to the destination socket. On error, the
786 * reference is dropped. The skb is not send to the destination, just all
787 * all error checks are performed and memory in the queue is reserved.
788 * Return values:
789 * < 0: error. skb freed, reference to sock dropped.
790 * 0: continue
791 * 1: repeat lookup - reference dropped while waiting for socket memory.
792 */
793 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
794 long *timeo, struct sock *ssk)
795 {
796 struct netlink_sock *nlk;
797
798 nlk = nlk_sk(sk);
799
800 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
801 test_bit(0, &nlk->state)) {
802 DECLARE_WAITQUEUE(wait, current);
803 if (!*timeo) {
804 if (!ssk || netlink_is_kernel(ssk))
805 netlink_overrun(sk);
806 sock_put(sk);
807 kfree_skb(skb);
808 return -EAGAIN;
809 }
810
811 __set_current_state(TASK_INTERRUPTIBLE);
812 add_wait_queue(&nlk->wait, &wait);
813
814 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
815 test_bit(0, &nlk->state)) &&
816 !sock_flag(sk, SOCK_DEAD))
817 *timeo = schedule_timeout(*timeo);
818
819 __set_current_state(TASK_RUNNING);
820 remove_wait_queue(&nlk->wait, &wait);
821 sock_put(sk);
822
823 if (signal_pending(current)) {
824 kfree_skb(skb);
825 return sock_intr_errno(*timeo);
826 }
827 return 1;
828 }
829 skb_set_owner_r(skb, sk);
830 return 0;
831 }
832
833 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
834 {
835 int len = skb->len;
836
837 skb_queue_tail(&sk->sk_receive_queue, skb);
838 sk->sk_data_ready(sk, len);
839 return len;
840 }
841
842 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
843 {
844 int len = __netlink_sendskb(sk, skb);
845
846 sock_put(sk);
847 return len;
848 }
849
850 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
851 {
852 kfree_skb(skb);
853 sock_put(sk);
854 }
855
856 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
857 gfp_t allocation)
858 {
859 int delta;
860
861 skb_orphan(skb);
862
863 delta = skb->end - skb->tail;
864 if (delta * 2 < skb->truesize)
865 return skb;
866
867 if (skb_shared(skb)) {
868 struct sk_buff *nskb = skb_clone(skb, allocation);
869 if (!nskb)
870 return skb;
871 kfree_skb(skb);
872 skb = nskb;
873 }
874
875 if (!pskb_expand_head(skb, 0, -delta, allocation))
876 skb->truesize -= delta;
877
878 return skb;
879 }
880
881 static inline void netlink_rcv_wake(struct sock *sk)
882 {
883 struct netlink_sock *nlk = nlk_sk(sk);
884
885 if (skb_queue_empty(&sk->sk_receive_queue))
886 clear_bit(0, &nlk->state);
887 if (!test_bit(0, &nlk->state))
888 wake_up_interruptible(&nlk->wait);
889 }
890
891 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
892 {
893 int ret;
894 struct netlink_sock *nlk = nlk_sk(sk);
895
896 ret = -ECONNREFUSED;
897 if (nlk->netlink_rcv != NULL) {
898 ret = skb->len;
899 skb_set_owner_r(skb, sk);
900 nlk->netlink_rcv(skb);
901 }
902 kfree_skb(skb);
903 sock_put(sk);
904 return ret;
905 }
906
907 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
908 u32 pid, int nonblock)
909 {
910 struct sock *sk;
911 int err;
912 long timeo;
913
914 skb = netlink_trim(skb, gfp_any());
915
916 timeo = sock_sndtimeo(ssk, nonblock);
917 retry:
918 sk = netlink_getsockbypid(ssk, pid);
919 if (IS_ERR(sk)) {
920 kfree_skb(skb);
921 return PTR_ERR(sk);
922 }
923 if (netlink_is_kernel(sk))
924 return netlink_unicast_kernel(sk, skb);
925
926 if (sk_filter(sk, skb)) {
927 err = skb->len;
928 kfree_skb(skb);
929 sock_put(sk);
930 return err;
931 }
932
933 err = netlink_attachskb(sk, skb, &timeo, ssk);
934 if (err == 1)
935 goto retry;
936 if (err)
937 return err;
938
939 return netlink_sendskb(sk, skb);
940 }
941 EXPORT_SYMBOL(netlink_unicast);
942
943 int netlink_has_listeners(struct sock *sk, unsigned int group)
944 {
945 int res = 0;
946 struct listeners *listeners;
947
948 BUG_ON(!netlink_is_kernel(sk));
949
950 rcu_read_lock();
951 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
952
953 if (group - 1 < nl_table[sk->sk_protocol].groups)
954 res = test_bit(group - 1, listeners->masks);
955
956 rcu_read_unlock();
957
958 return res;
959 }
960 EXPORT_SYMBOL_GPL(netlink_has_listeners);
961
962 static inline int netlink_broadcast_deliver(struct sock *sk,
963 struct sk_buff *skb)
964 {
965 struct netlink_sock *nlk = nlk_sk(sk);
966
967 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
968 !test_bit(0, &nlk->state)) {
969 skb_set_owner_r(skb, sk);
970 __netlink_sendskb(sk, skb);
971 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
972 }
973 return -1;
974 }
975
976 struct netlink_broadcast_data {
977 struct sock *exclude_sk;
978 struct net *net;
979 u32 pid;
980 u32 group;
981 int failure;
982 int delivery_failure;
983 int congested;
984 int delivered;
985 gfp_t allocation;
986 struct sk_buff *skb, *skb2;
987 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
988 void *tx_data;
989 };
990
991 static inline int do_one_broadcast(struct sock *sk,
992 struct netlink_broadcast_data *p)
993 {
994 struct netlink_sock *nlk = nlk_sk(sk);
995 int val;
996
997 if (p->exclude_sk == sk)
998 goto out;
999
1000 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1001 !test_bit(p->group - 1, nlk->groups))
1002 goto out;
1003
1004 if (!net_eq(sock_net(sk), p->net))
1005 goto out;
1006
1007 if (p->failure) {
1008 netlink_overrun(sk);
1009 goto out;
1010 }
1011
1012 sock_hold(sk);
1013 if (p->skb2 == NULL) {
1014 if (skb_shared(p->skb)) {
1015 p->skb2 = skb_clone(p->skb, p->allocation);
1016 } else {
1017 p->skb2 = skb_get(p->skb);
1018 /*
1019 * skb ownership may have been set when
1020 * delivered to a previous socket.
1021 */
1022 skb_orphan(p->skb2);
1023 }
1024 }
1025 if (p->skb2 == NULL) {
1026 netlink_overrun(sk);
1027 /* Clone failed. Notify ALL listeners. */
1028 p->failure = 1;
1029 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1030 p->delivery_failure = 1;
1031 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1032 kfree_skb(p->skb2);
1033 p->skb2 = NULL;
1034 } else if (sk_filter(sk, p->skb2)) {
1035 kfree_skb(p->skb2);
1036 p->skb2 = NULL;
1037 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1038 netlink_overrun(sk);
1039 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1040 p->delivery_failure = 1;
1041 } else {
1042 p->congested |= val;
1043 p->delivered = 1;
1044 p->skb2 = NULL;
1045 }
1046 sock_put(sk);
1047
1048 out:
1049 return 0;
1050 }
1051
1052 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid,
1053 u32 group, gfp_t allocation,
1054 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1055 void *filter_data)
1056 {
1057 struct net *net = sock_net(ssk);
1058 struct netlink_broadcast_data info;
1059 struct hlist_node *node;
1060 struct sock *sk;
1061
1062 skb = netlink_trim(skb, allocation);
1063
1064 info.exclude_sk = ssk;
1065 info.net = net;
1066 info.pid = pid;
1067 info.group = group;
1068 info.failure = 0;
1069 info.delivery_failure = 0;
1070 info.congested = 0;
1071 info.delivered = 0;
1072 info.allocation = allocation;
1073 info.skb = skb;
1074 info.skb2 = NULL;
1075 info.tx_filter = filter;
1076 info.tx_data = filter_data;
1077
1078 /* While we sleep in clone, do not allow to change socket list */
1079
1080 netlink_lock_table();
1081
1082 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1083 do_one_broadcast(sk, &info);
1084
1085 consume_skb(skb);
1086
1087 netlink_unlock_table();
1088
1089 if (info.delivery_failure) {
1090 kfree_skb(info.skb2);
1091 return -ENOBUFS;
1092 } else
1093 consume_skb(info.skb2);
1094
1095 if (info.delivered) {
1096 if (info.congested && (allocation & __GFP_WAIT))
1097 yield();
1098 return 0;
1099 }
1100 return -ESRCH;
1101 }
1102 EXPORT_SYMBOL(netlink_broadcast_filtered);
1103
1104 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1105 u32 group, gfp_t allocation)
1106 {
1107 return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1108 NULL, NULL);
1109 }
1110 EXPORT_SYMBOL(netlink_broadcast);
1111
1112 struct netlink_set_err_data {
1113 struct sock *exclude_sk;
1114 u32 pid;
1115 u32 group;
1116 int code;
1117 };
1118
1119 static inline int do_one_set_err(struct sock *sk,
1120 struct netlink_set_err_data *p)
1121 {
1122 struct netlink_sock *nlk = nlk_sk(sk);
1123 int ret = 0;
1124
1125 if (sk == p->exclude_sk)
1126 goto out;
1127
1128 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1129 goto out;
1130
1131 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1132 !test_bit(p->group - 1, nlk->groups))
1133 goto out;
1134
1135 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1136 ret = 1;
1137 goto out;
1138 }
1139
1140 sk->sk_err = p->code;
1141 sk->sk_error_report(sk);
1142 out:
1143 return ret;
1144 }
1145
1146 /**
1147 * netlink_set_err - report error to broadcast listeners
1148 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1149 * @pid: the PID of a process that we want to skip (if any)
1150 * @groups: the broadcast group that will notice the error
1151 * @code: error code, must be negative (as usual in kernelspace)
1152 *
1153 * This function returns the number of broadcast listeners that have set the
1154 * NETLINK_RECV_NO_ENOBUFS socket option.
1155 */
1156 int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1157 {
1158 struct netlink_set_err_data info;
1159 struct hlist_node *node;
1160 struct sock *sk;
1161 int ret = 0;
1162
1163 info.exclude_sk = ssk;
1164 info.pid = pid;
1165 info.group = group;
1166 /* sk->sk_err wants a positive error value */
1167 info.code = -code;
1168
1169 read_lock(&nl_table_lock);
1170
1171 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1172 ret += do_one_set_err(sk, &info);
1173
1174 read_unlock(&nl_table_lock);
1175 return ret;
1176 }
1177 EXPORT_SYMBOL(netlink_set_err);
1178
1179 /* must be called with netlink table grabbed */
1180 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1181 unsigned int group,
1182 int is_new)
1183 {
1184 int old, new = !!is_new, subscriptions;
1185
1186 old = test_bit(group - 1, nlk->groups);
1187 subscriptions = nlk->subscriptions - old + new;
1188 if (new)
1189 __set_bit(group - 1, nlk->groups);
1190 else
1191 __clear_bit(group - 1, nlk->groups);
1192 netlink_update_subscriptions(&nlk->sk, subscriptions);
1193 netlink_update_listeners(&nlk->sk);
1194 }
1195
1196 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1197 char __user *optval, unsigned int optlen)
1198 {
1199 struct sock *sk = sock->sk;
1200 struct netlink_sock *nlk = nlk_sk(sk);
1201 unsigned int val = 0;
1202 int err;
1203
1204 if (level != SOL_NETLINK)
1205 return -ENOPROTOOPT;
1206
1207 if (optlen >= sizeof(int) &&
1208 get_user(val, (unsigned int __user *)optval))
1209 return -EFAULT;
1210
1211 switch (optname) {
1212 case NETLINK_PKTINFO:
1213 if (val)
1214 nlk->flags |= NETLINK_RECV_PKTINFO;
1215 else
1216 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1217 err = 0;
1218 break;
1219 case NETLINK_ADD_MEMBERSHIP:
1220 case NETLINK_DROP_MEMBERSHIP: {
1221 if (!netlink_capable(sock, NL_NONROOT_RECV))
1222 return -EPERM;
1223 err = netlink_realloc_groups(sk);
1224 if (err)
1225 return err;
1226 if (!val || val - 1 >= nlk->ngroups)
1227 return -EINVAL;
1228 netlink_table_grab();
1229 netlink_update_socket_mc(nlk, val,
1230 optname == NETLINK_ADD_MEMBERSHIP);
1231 netlink_table_ungrab();
1232 err = 0;
1233 break;
1234 }
1235 case NETLINK_BROADCAST_ERROR:
1236 if (val)
1237 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1238 else
1239 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1240 err = 0;
1241 break;
1242 case NETLINK_NO_ENOBUFS:
1243 if (val) {
1244 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1245 clear_bit(0, &nlk->state);
1246 wake_up_interruptible(&nlk->wait);
1247 } else
1248 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1249 err = 0;
1250 break;
1251 default:
1252 err = -ENOPROTOOPT;
1253 }
1254 return err;
1255 }
1256
1257 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1258 char __user *optval, int __user *optlen)
1259 {
1260 struct sock *sk = sock->sk;
1261 struct netlink_sock *nlk = nlk_sk(sk);
1262 int len, val, err;
1263
1264 if (level != SOL_NETLINK)
1265 return -ENOPROTOOPT;
1266
1267 if (get_user(len, optlen))
1268 return -EFAULT;
1269 if (len < 0)
1270 return -EINVAL;
1271
1272 switch (optname) {
1273 case NETLINK_PKTINFO:
1274 if (len < sizeof(int))
1275 return -EINVAL;
1276 len = sizeof(int);
1277 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1278 if (put_user(len, optlen) ||
1279 put_user(val, optval))
1280 return -EFAULT;
1281 err = 0;
1282 break;
1283 case NETLINK_BROADCAST_ERROR:
1284 if (len < sizeof(int))
1285 return -EINVAL;
1286 len = sizeof(int);
1287 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1288 if (put_user(len, optlen) ||
1289 put_user(val, optval))
1290 return -EFAULT;
1291 err = 0;
1292 break;
1293 case NETLINK_NO_ENOBUFS:
1294 if (len < sizeof(int))
1295 return -EINVAL;
1296 len = sizeof(int);
1297 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1298 if (put_user(len, optlen) ||
1299 put_user(val, optval))
1300 return -EFAULT;
1301 err = 0;
1302 break;
1303 default:
1304 err = -ENOPROTOOPT;
1305 }
1306 return err;
1307 }
1308
1309 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1310 {
1311 struct nl_pktinfo info;
1312
1313 info.group = NETLINK_CB(skb).dst_group;
1314 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1315 }
1316
1317 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1318 struct msghdr *msg, size_t len)
1319 {
1320 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1321 struct sock *sk = sock->sk;
1322 struct netlink_sock *nlk = nlk_sk(sk);
1323 struct sockaddr_nl *addr = msg->msg_name;
1324 u32 dst_pid;
1325 u32 dst_group;
1326 struct sk_buff *skb;
1327 int err;
1328 struct scm_cookie scm;
1329
1330 if (msg->msg_flags&MSG_OOB)
1331 return -EOPNOTSUPP;
1332
1333 if (NULL == siocb->scm) {
1334 siocb->scm = &scm;
1335 memset(&scm, 0, sizeof(scm));
1336 }
1337 err = scm_send(sock, msg, siocb->scm);
1338 if (err < 0)
1339 return err;
1340
1341 if (msg->msg_namelen) {
1342 err = -EINVAL;
1343 if (addr->nl_family != AF_NETLINK)
1344 goto out;
1345 dst_pid = addr->nl_pid;
1346 dst_group = ffs(addr->nl_groups);
1347 err = -EPERM;
1348 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1349 goto out;
1350 } else {
1351 dst_pid = nlk->dst_pid;
1352 dst_group = nlk->dst_group;
1353 }
1354
1355 if (!nlk->pid) {
1356 err = netlink_autobind(sock);
1357 if (err)
1358 goto out;
1359 }
1360
1361 err = -EMSGSIZE;
1362 if (len > sk->sk_sndbuf - 32)
1363 goto out;
1364 err = -ENOBUFS;
1365 skb = alloc_skb(len, GFP_KERNEL);
1366 if (skb == NULL)
1367 goto out;
1368
1369 NETLINK_CB(skb).pid = nlk->pid;
1370 NETLINK_CB(skb).dst_group = dst_group;
1371 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1372
1373 err = -EFAULT;
1374 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1375 kfree_skb(skb);
1376 goto out;
1377 }
1378
1379 err = security_netlink_send(sk, skb);
1380 if (err) {
1381 kfree_skb(skb);
1382 goto out;
1383 }
1384
1385 if (dst_group) {
1386 atomic_inc(&skb->users);
1387 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1388 }
1389 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1390
1391 out:
1392 scm_destroy(siocb->scm);
1393 return err;
1394 }
1395
1396 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1397 struct msghdr *msg, size_t len,
1398 int flags)
1399 {
1400 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1401 struct scm_cookie scm;
1402 struct sock *sk = sock->sk;
1403 struct netlink_sock *nlk = nlk_sk(sk);
1404 int noblock = flags&MSG_DONTWAIT;
1405 size_t copied;
1406 struct sk_buff *skb, *data_skb;
1407 int err, ret;
1408
1409 if (flags&MSG_OOB)
1410 return -EOPNOTSUPP;
1411
1412 copied = 0;
1413
1414 skb = skb_recv_datagram(sk, flags, noblock, &err);
1415 if (skb == NULL)
1416 goto out;
1417
1418 data_skb = skb;
1419
1420 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1421 if (unlikely(skb_shinfo(skb)->frag_list)) {
1422 /*
1423 * If this skb has a frag_list, then here that means that we
1424 * will have to use the frag_list skb's data for compat tasks
1425 * and the regular skb's data for normal (non-compat) tasks.
1426 *
1427 * If we need to send the compat skb, assign it to the
1428 * 'data_skb' variable so that it will be used below for data
1429 * copying. We keep 'skb' for everything else, including
1430 * freeing both later.
1431 */
1432 if (flags & MSG_CMSG_COMPAT)
1433 data_skb = skb_shinfo(skb)->frag_list;
1434 }
1435 #endif
1436
1437 msg->msg_namelen = 0;
1438
1439 copied = data_skb->len;
1440 if (len < copied) {
1441 msg->msg_flags |= MSG_TRUNC;
1442 copied = len;
1443 }
1444
1445 skb_reset_transport_header(data_skb);
1446 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1447
1448 if (msg->msg_name) {
1449 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1450 addr->nl_family = AF_NETLINK;
1451 addr->nl_pad = 0;
1452 addr->nl_pid = NETLINK_CB(skb).pid;
1453 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1454 msg->msg_namelen = sizeof(*addr);
1455 }
1456
1457 if (nlk->flags & NETLINK_RECV_PKTINFO)
1458 netlink_cmsg_recv_pktinfo(msg, skb);
1459
1460 if (NULL == siocb->scm) {
1461 memset(&scm, 0, sizeof(scm));
1462 siocb->scm = &scm;
1463 }
1464 siocb->scm->creds = *NETLINK_CREDS(skb);
1465 if (flags & MSG_TRUNC)
1466 copied = data_skb->len;
1467
1468 skb_free_datagram(sk, skb);
1469
1470 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1471 ret = netlink_dump(sk);
1472 if (ret) {
1473 sk->sk_err = ret;
1474 sk->sk_error_report(sk);
1475 }
1476 }
1477
1478 scm_recv(sock, msg, siocb->scm, flags);
1479 out:
1480 netlink_rcv_wake(sk);
1481 return err ? : copied;
1482 }
1483
1484 static void netlink_data_ready(struct sock *sk, int len)
1485 {
1486 BUG();
1487 }
1488
1489 /*
1490 * We export these functions to other modules. They provide a
1491 * complete set of kernel non-blocking support for message
1492 * queueing.
1493 */
1494
1495 struct sock *
1496 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1497 void (*input)(struct sk_buff *skb),
1498 struct mutex *cb_mutex, struct module *module)
1499 {
1500 struct socket *sock;
1501 struct sock *sk;
1502 struct netlink_sock *nlk;
1503 struct listeners *listeners = NULL;
1504
1505 BUG_ON(!nl_table);
1506
1507 if (unit < 0 || unit >= MAX_LINKS)
1508 return NULL;
1509
1510 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1511 return NULL;
1512
1513 /*
1514 * We have to just have a reference on the net from sk, but don't
1515 * get_net it. Besides, we cannot get and then put the net here.
1516 * So we create one inside init_net and the move it to net.
1517 */
1518
1519 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1520 goto out_sock_release_nosk;
1521
1522 sk = sock->sk;
1523 sk_change_net(sk, net);
1524
1525 if (groups < 32)
1526 groups = 32;
1527
1528 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1529 if (!listeners)
1530 goto out_sock_release;
1531
1532 sk->sk_data_ready = netlink_data_ready;
1533 if (input)
1534 nlk_sk(sk)->netlink_rcv = input;
1535
1536 if (netlink_insert(sk, net, 0))
1537 goto out_sock_release;
1538
1539 nlk = nlk_sk(sk);
1540 nlk->flags |= NETLINK_KERNEL_SOCKET;
1541
1542 netlink_table_grab();
1543 if (!nl_table[unit].registered) {
1544 nl_table[unit].groups = groups;
1545 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1546 nl_table[unit].cb_mutex = cb_mutex;
1547 nl_table[unit].module = module;
1548 nl_table[unit].registered = 1;
1549 } else {
1550 kfree(listeners);
1551 nl_table[unit].registered++;
1552 }
1553 netlink_table_ungrab();
1554 return sk;
1555
1556 out_sock_release:
1557 kfree(listeners);
1558 netlink_kernel_release(sk);
1559 return NULL;
1560
1561 out_sock_release_nosk:
1562 sock_release(sock);
1563 return NULL;
1564 }
1565 EXPORT_SYMBOL(netlink_kernel_create);
1566
1567
1568 void
1569 netlink_kernel_release(struct sock *sk)
1570 {
1571 sk_release_kernel(sk);
1572 }
1573 EXPORT_SYMBOL(netlink_kernel_release);
1574
1575 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1576 {
1577 struct listeners *new, *old;
1578 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1579
1580 if (groups < 32)
1581 groups = 32;
1582
1583 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1584 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1585 if (!new)
1586 return -ENOMEM;
1587 old = rcu_dereference_raw(tbl->listeners);
1588 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1589 rcu_assign_pointer(tbl->listeners, new);
1590
1591 kfree_rcu(old, rcu);
1592 }
1593 tbl->groups = groups;
1594
1595 return 0;
1596 }
1597
1598 /**
1599 * netlink_change_ngroups - change number of multicast groups
1600 *
1601 * This changes the number of multicast groups that are available
1602 * on a certain netlink family. Note that it is not possible to
1603 * change the number of groups to below 32. Also note that it does
1604 * not implicitly call netlink_clear_multicast_users() when the
1605 * number of groups is reduced.
1606 *
1607 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1608 * @groups: The new number of groups.
1609 */
1610 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1611 {
1612 int err;
1613
1614 netlink_table_grab();
1615 err = __netlink_change_ngroups(sk, groups);
1616 netlink_table_ungrab();
1617
1618 return err;
1619 }
1620
1621 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1622 {
1623 struct sock *sk;
1624 struct hlist_node *node;
1625 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1626
1627 sk_for_each_bound(sk, node, &tbl->mc_list)
1628 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1629 }
1630
1631 /**
1632 * netlink_clear_multicast_users - kick off multicast listeners
1633 *
1634 * This function removes all listeners from the given group.
1635 * @ksk: The kernel netlink socket, as returned by
1636 * netlink_kernel_create().
1637 * @group: The multicast group to clear.
1638 */
1639 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1640 {
1641 netlink_table_grab();
1642 __netlink_clear_multicast_users(ksk, group);
1643 netlink_table_ungrab();
1644 }
1645
1646 void netlink_set_nonroot(int protocol, unsigned int flags)
1647 {
1648 if ((unsigned int)protocol < MAX_LINKS)
1649 nl_table[protocol].nl_nonroot = flags;
1650 }
1651 EXPORT_SYMBOL(netlink_set_nonroot);
1652
1653 static void netlink_destroy_callback(struct netlink_callback *cb)
1654 {
1655 kfree_skb(cb->skb);
1656 kfree(cb);
1657 }
1658
1659 /*
1660 * It looks a bit ugly.
1661 * It would be better to create kernel thread.
1662 */
1663
1664 static int netlink_dump(struct sock *sk)
1665 {
1666 struct netlink_sock *nlk = nlk_sk(sk);
1667 struct netlink_callback *cb;
1668 struct sk_buff *skb;
1669 struct nlmsghdr *nlh;
1670 int len, err = -ENOBUFS;
1671
1672 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1673 if (!skb)
1674 goto errout;
1675
1676 mutex_lock(nlk->cb_mutex);
1677
1678 cb = nlk->cb;
1679 if (cb == NULL) {
1680 err = -EINVAL;
1681 goto errout_skb;
1682 }
1683
1684 len = cb->dump(skb, cb);
1685
1686 if (len > 0) {
1687 mutex_unlock(nlk->cb_mutex);
1688
1689 if (sk_filter(sk, skb))
1690 kfree_skb(skb);
1691 else
1692 __netlink_sendskb(sk, skb);
1693 return 0;
1694 }
1695
1696 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1697 if (!nlh)
1698 goto errout_skb;
1699
1700 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1701
1702 if (sk_filter(sk, skb))
1703 kfree_skb(skb);
1704 else
1705 __netlink_sendskb(sk, skb);
1706
1707 if (cb->done)
1708 cb->done(cb);
1709 nlk->cb = NULL;
1710 mutex_unlock(nlk->cb_mutex);
1711
1712 netlink_destroy_callback(cb);
1713 return 0;
1714
1715 errout_skb:
1716 mutex_unlock(nlk->cb_mutex);
1717 kfree_skb(skb);
1718 errout:
1719 return err;
1720 }
1721
1722 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1723 const struct nlmsghdr *nlh,
1724 int (*dump)(struct sk_buff *skb,
1725 struct netlink_callback *),
1726 int (*done)(struct netlink_callback *))
1727 {
1728 struct netlink_callback *cb;
1729 struct sock *sk;
1730 struct netlink_sock *nlk;
1731 int ret;
1732
1733 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1734 if (cb == NULL)
1735 return -ENOBUFS;
1736
1737 cb->dump = dump;
1738 cb->done = done;
1739 cb->nlh = nlh;
1740 atomic_inc(&skb->users);
1741 cb->skb = skb;
1742
1743 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1744 if (sk == NULL) {
1745 netlink_destroy_callback(cb);
1746 return -ECONNREFUSED;
1747 }
1748 nlk = nlk_sk(sk);
1749 /* A dump is in progress... */
1750 mutex_lock(nlk->cb_mutex);
1751 if (nlk->cb) {
1752 mutex_unlock(nlk->cb_mutex);
1753 netlink_destroy_callback(cb);
1754 sock_put(sk);
1755 return -EBUSY;
1756 }
1757 nlk->cb = cb;
1758 mutex_unlock(nlk->cb_mutex);
1759
1760 ret = netlink_dump(sk);
1761
1762 sock_put(sk);
1763
1764 if (ret)
1765 return ret;
1766
1767 /* We successfully started a dump, by returning -EINTR we
1768 * signal not to send ACK even if it was requested.
1769 */
1770 return -EINTR;
1771 }
1772 EXPORT_SYMBOL(netlink_dump_start);
1773
1774 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1775 {
1776 struct sk_buff *skb;
1777 struct nlmsghdr *rep;
1778 struct nlmsgerr *errmsg;
1779 size_t payload = sizeof(*errmsg);
1780
1781 /* error messages get the original request appened */
1782 if (err)
1783 payload += nlmsg_len(nlh);
1784
1785 skb = nlmsg_new(payload, GFP_KERNEL);
1786 if (!skb) {
1787 struct sock *sk;
1788
1789 sk = netlink_lookup(sock_net(in_skb->sk),
1790 in_skb->sk->sk_protocol,
1791 NETLINK_CB(in_skb).pid);
1792 if (sk) {
1793 sk->sk_err = ENOBUFS;
1794 sk->sk_error_report(sk);
1795 sock_put(sk);
1796 }
1797 return;
1798 }
1799
1800 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1801 NLMSG_ERROR, payload, 0);
1802 errmsg = nlmsg_data(rep);
1803 errmsg->error = err;
1804 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1805 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1806 }
1807 EXPORT_SYMBOL(netlink_ack);
1808
1809 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1810 struct nlmsghdr *))
1811 {
1812 struct nlmsghdr *nlh;
1813 int err;
1814
1815 while (skb->len >= nlmsg_total_size(0)) {
1816 int msglen;
1817
1818 nlh = nlmsg_hdr(skb);
1819 err = 0;
1820
1821 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1822 return 0;
1823
1824 /* Only requests are handled by the kernel */
1825 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1826 goto ack;
1827
1828 /* Skip control messages */
1829 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1830 goto ack;
1831
1832 err = cb(skb, nlh);
1833 if (err == -EINTR)
1834 goto skip;
1835
1836 ack:
1837 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1838 netlink_ack(skb, nlh, err);
1839
1840 skip:
1841 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1842 if (msglen > skb->len)
1843 msglen = skb->len;
1844 skb_pull(skb, msglen);
1845 }
1846
1847 return 0;
1848 }
1849 EXPORT_SYMBOL(netlink_rcv_skb);
1850
1851 /**
1852 * nlmsg_notify - send a notification netlink message
1853 * @sk: netlink socket to use
1854 * @skb: notification message
1855 * @pid: destination netlink pid for reports or 0
1856 * @group: destination multicast group or 0
1857 * @report: 1 to report back, 0 to disable
1858 * @flags: allocation flags
1859 */
1860 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1861 unsigned int group, int report, gfp_t flags)
1862 {
1863 int err = 0;
1864
1865 if (group) {
1866 int exclude_pid = 0;
1867
1868 if (report) {
1869 atomic_inc(&skb->users);
1870 exclude_pid = pid;
1871 }
1872
1873 /* errors reported via destination sk->sk_err, but propagate
1874 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1875 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1876 }
1877
1878 if (report) {
1879 int err2;
1880
1881 err2 = nlmsg_unicast(sk, skb, pid);
1882 if (!err || err == -ESRCH)
1883 err = err2;
1884 }
1885
1886 return err;
1887 }
1888 EXPORT_SYMBOL(nlmsg_notify);
1889
1890 #ifdef CONFIG_PROC_FS
1891 struct nl_seq_iter {
1892 struct seq_net_private p;
1893 int link;
1894 int hash_idx;
1895 };
1896
1897 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1898 {
1899 struct nl_seq_iter *iter = seq->private;
1900 int i, j;
1901 struct sock *s;
1902 struct hlist_node *node;
1903 loff_t off = 0;
1904
1905 for (i = 0; i < MAX_LINKS; i++) {
1906 struct nl_pid_hash *hash = &nl_table[i].hash;
1907
1908 for (j = 0; j <= hash->mask; j++) {
1909 sk_for_each(s, node, &hash->table[j]) {
1910 if (sock_net(s) != seq_file_net(seq))
1911 continue;
1912 if (off == pos) {
1913 iter->link = i;
1914 iter->hash_idx = j;
1915 return s;
1916 }
1917 ++off;
1918 }
1919 }
1920 }
1921 return NULL;
1922 }
1923
1924 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1925 __acquires(nl_table_lock)
1926 {
1927 read_lock(&nl_table_lock);
1928 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1929 }
1930
1931 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1932 {
1933 struct sock *s;
1934 struct nl_seq_iter *iter;
1935 int i, j;
1936
1937 ++*pos;
1938
1939 if (v == SEQ_START_TOKEN)
1940 return netlink_seq_socket_idx(seq, 0);
1941
1942 iter = seq->private;
1943 s = v;
1944 do {
1945 s = sk_next(s);
1946 } while (s && sock_net(s) != seq_file_net(seq));
1947 if (s)
1948 return s;
1949
1950 i = iter->link;
1951 j = iter->hash_idx + 1;
1952
1953 do {
1954 struct nl_pid_hash *hash = &nl_table[i].hash;
1955
1956 for (; j <= hash->mask; j++) {
1957 s = sk_head(&hash->table[j]);
1958 while (s && sock_net(s) != seq_file_net(seq))
1959 s = sk_next(s);
1960 if (s) {
1961 iter->link = i;
1962 iter->hash_idx = j;
1963 return s;
1964 }
1965 }
1966
1967 j = 0;
1968 } while (++i < MAX_LINKS);
1969
1970 return NULL;
1971 }
1972
1973 static void netlink_seq_stop(struct seq_file *seq, void *v)
1974 __releases(nl_table_lock)
1975 {
1976 read_unlock(&nl_table_lock);
1977 }
1978
1979
1980 static int netlink_seq_show(struct seq_file *seq, void *v)
1981 {
1982 if (v == SEQ_START_TOKEN)
1983 seq_puts(seq,
1984 "sk Eth Pid Groups "
1985 "Rmem Wmem Dump Locks Drops Inode\n");
1986 else {
1987 struct sock *s = v;
1988 struct netlink_sock *nlk = nlk_sk(s);
1989
1990 seq_printf(seq, "%pK %-3d %-6d %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
1991 s,
1992 s->sk_protocol,
1993 nlk->pid,
1994 nlk->groups ? (u32)nlk->groups[0] : 0,
1995 sk_rmem_alloc_get(s),
1996 sk_wmem_alloc_get(s),
1997 nlk->cb,
1998 atomic_read(&s->sk_refcnt),
1999 atomic_read(&s->sk_drops),
2000 sock_i_ino(s)
2001 );
2002
2003 }
2004 return 0;
2005 }
2006
2007 static const struct seq_operations netlink_seq_ops = {
2008 .start = netlink_seq_start,
2009 .next = netlink_seq_next,
2010 .stop = netlink_seq_stop,
2011 .show = netlink_seq_show,
2012 };
2013
2014
2015 static int netlink_seq_open(struct inode *inode, struct file *file)
2016 {
2017 return seq_open_net(inode, file, &netlink_seq_ops,
2018 sizeof(struct nl_seq_iter));
2019 }
2020
2021 static const struct file_operations netlink_seq_fops = {
2022 .owner = THIS_MODULE,
2023 .open = netlink_seq_open,
2024 .read = seq_read,
2025 .llseek = seq_lseek,
2026 .release = seq_release_net,
2027 };
2028
2029 #endif
2030
2031 int netlink_register_notifier(struct notifier_block *nb)
2032 {
2033 return atomic_notifier_chain_register(&netlink_chain, nb);
2034 }
2035 EXPORT_SYMBOL(netlink_register_notifier);
2036
2037 int netlink_unregister_notifier(struct notifier_block *nb)
2038 {
2039 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2040 }
2041 EXPORT_SYMBOL(netlink_unregister_notifier);
2042
2043 static const struct proto_ops netlink_ops = {
2044 .family = PF_NETLINK,
2045 .owner = THIS_MODULE,
2046 .release = netlink_release,
2047 .bind = netlink_bind,
2048 .connect = netlink_connect,
2049 .socketpair = sock_no_socketpair,
2050 .accept = sock_no_accept,
2051 .getname = netlink_getname,
2052 .poll = datagram_poll,
2053 .ioctl = sock_no_ioctl,
2054 .listen = sock_no_listen,
2055 .shutdown = sock_no_shutdown,
2056 .setsockopt = netlink_setsockopt,
2057 .getsockopt = netlink_getsockopt,
2058 .sendmsg = netlink_sendmsg,
2059 .recvmsg = netlink_recvmsg,
2060 .mmap = sock_no_mmap,
2061 .sendpage = sock_no_sendpage,
2062 };
2063
2064 static const struct net_proto_family netlink_family_ops = {
2065 .family = PF_NETLINK,
2066 .create = netlink_create,
2067 .owner = THIS_MODULE, /* for consistency 8) */
2068 };
2069
2070 static int __net_init netlink_net_init(struct net *net)
2071 {
2072 #ifdef CONFIG_PROC_FS
2073 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2074 return -ENOMEM;
2075 #endif
2076 return 0;
2077 }
2078
2079 static void __net_exit netlink_net_exit(struct net *net)
2080 {
2081 #ifdef CONFIG_PROC_FS
2082 proc_net_remove(net, "netlink");
2083 #endif
2084 }
2085
2086 static void __init netlink_add_usersock_entry(void)
2087 {
2088 struct listeners *listeners;
2089 int groups = 32;
2090
2091 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2092 if (!listeners)
2093 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2094
2095 netlink_table_grab();
2096
2097 nl_table[NETLINK_USERSOCK].groups = groups;
2098 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2099 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2100 nl_table[NETLINK_USERSOCK].registered = 1;
2101
2102 netlink_table_ungrab();
2103 }
2104
2105 static struct pernet_operations __net_initdata netlink_net_ops = {
2106 .init = netlink_net_init,
2107 .exit = netlink_net_exit,
2108 };
2109
2110 static int __init netlink_proto_init(void)
2111 {
2112 struct sk_buff *dummy_skb;
2113 int i;
2114 unsigned long limit;
2115 unsigned int order;
2116 int err = proto_register(&netlink_proto, 0);
2117
2118 if (err != 0)
2119 goto out;
2120
2121 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2122
2123 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2124 if (!nl_table)
2125 goto panic;
2126
2127 if (totalram_pages >= (128 * 1024))
2128 limit = totalram_pages >> (21 - PAGE_SHIFT);
2129 else
2130 limit = totalram_pages >> (23 - PAGE_SHIFT);
2131
2132 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2133 limit = (1UL << order) / sizeof(struct hlist_head);
2134 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2135
2136 for (i = 0; i < MAX_LINKS; i++) {
2137 struct nl_pid_hash *hash = &nl_table[i].hash;
2138
2139 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2140 if (!hash->table) {
2141 while (i-- > 0)
2142 nl_pid_hash_free(nl_table[i].hash.table,
2143 1 * sizeof(*hash->table));
2144 kfree(nl_table);
2145 goto panic;
2146 }
2147 hash->max_shift = order;
2148 hash->shift = 0;
2149 hash->mask = 0;
2150 hash->rehash_time = jiffies;
2151 }
2152
2153 netlink_add_usersock_entry();
2154
2155 sock_register(&netlink_family_ops);
2156 register_pernet_subsys(&netlink_net_ops);
2157 /* The netlink device handler may be needed early. */
2158 rtnetlink_init();
2159 out:
2160 return err;
2161 panic:
2162 panic("netlink_init: Cannot allocate nl_table\n");
2163 }
2164
2165 core_initcall(netlink_proto_init);
Linux with added FPC-III support