?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
net: af_netlink should update its inuse counter
[linux-2.6/cjktty.git] / net / netlink / af_netlink.c
bloba2071dcfe9e8514cf9aad61aa6308bc596c8b176
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 #define NETLINK_KERNEL_SOCKET 0x1
87 #define NETLINK_RECV_PKTINFO 0x2
88
89 static inline struct netlink_sock *nlk_sk(struct sock *sk)
90 {
91 return container_of(sk, struct netlink_sock, sk);
92 }
93
94 static inline int netlink_is_kernel(struct sock *sk)
95 {
96 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
97 }
98
99 struct nl_pid_hash {
100 struct hlist_head *table;
101 unsigned long rehash_time;
102
103 unsigned int mask;
104 unsigned int shift;
105
106 unsigned int entries;
107 unsigned int max_shift;
108
109 u32 rnd;
110 };
111
112 struct netlink_table {
113 struct nl_pid_hash hash;
114 struct hlist_head mc_list;
115 unsigned long *listeners;
116 unsigned int nl_nonroot;
117 unsigned int groups;
118 struct mutex *cb_mutex;
119 struct module *module;
120 int registered;
121 };
122
123 static struct netlink_table *nl_table;
124
125 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
126
127 static int netlink_dump(struct sock *sk);
128 static void netlink_destroy_callback(struct netlink_callback *cb);
129
130 static DEFINE_RWLOCK(nl_table_lock);
131 static atomic_t nl_table_users = ATOMIC_INIT(0);
132
133 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
134
135 static u32 netlink_group_mask(u32 group)
136 {
137 return group ? 1 << (group - 1) : 0;
138 }
139
140 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
141 {
142 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
143 }
144
145 static void netlink_sock_destruct(struct sock *sk)
146 {
147 struct netlink_sock *nlk = nlk_sk(sk);
148
149 if (nlk->cb) {
150 if (nlk->cb->done)
151 nlk->cb->done(nlk->cb);
152 netlink_destroy_callback(nlk->cb);
153 }
154
155 skb_queue_purge(&sk->sk_receive_queue);
156
157 if (!sock_flag(sk, SOCK_DEAD)) {
158 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
159 return;
160 }
161
162 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
163 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
164 WARN_ON(nlk_sk(sk)->groups);
165 }
166
167 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
168 * SMP. Look, when several writers sleep and reader wakes them up, all but one
169 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
170 * this, _but_ remember, it adds useless work on UP machines.
171 */
172
173 static void netlink_table_grab(void)
174 __acquires(nl_table_lock)
175 {
176 write_lock_irq(&nl_table_lock);
177
178 if (atomic_read(&nl_table_users)) {
179 DECLARE_WAITQUEUE(wait, current);
180
181 add_wait_queue_exclusive(&nl_table_wait, &wait);
182 for (;;) {
183 set_current_state(TASK_UNINTERRUPTIBLE);
184 if (atomic_read(&nl_table_users) == 0)
185 break;
186 write_unlock_irq(&nl_table_lock);
187 schedule();
188 write_lock_irq(&nl_table_lock);
189 }
190
191 __set_current_state(TASK_RUNNING);
192 remove_wait_queue(&nl_table_wait, &wait);
193 }
194 }
195
196 static void netlink_table_ungrab(void)
197 __releases(nl_table_lock)
198 {
199 write_unlock_irq(&nl_table_lock);
200 wake_up(&nl_table_wait);
201 }
202
203 static inline void
204 netlink_lock_table(void)
205 {
206 /* read_lock() synchronizes us to netlink_table_grab */
207
208 read_lock(&nl_table_lock);
209 atomic_inc(&nl_table_users);
210 read_unlock(&nl_table_lock);
211 }
212
213 static inline void
214 netlink_unlock_table(void)
215 {
216 if (atomic_dec_and_test(&nl_table_users))
217 wake_up(&nl_table_wait);
218 }
219
220 static inline struct sock *netlink_lookup(struct net *net, int protocol,
221 u32 pid)
222 {
223 struct nl_pid_hash *hash = &nl_table[protocol].hash;
224 struct hlist_head *head;
225 struct sock *sk;
226 struct hlist_node *node;
227
228 read_lock(&nl_table_lock);
229 head = nl_pid_hashfn(hash, pid);
230 sk_for_each(sk, node, head) {
231 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
232 sock_hold(sk);
233 goto found;
234 }
235 }
236 sk = NULL;
237 found:
238 read_unlock(&nl_table_lock);
239 return sk;
240 }
241
242 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
243 {
244 if (size <= PAGE_SIZE)
245 return kzalloc(size, GFP_ATOMIC);
246 else
247 return (struct hlist_head *)
248 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
249 get_order(size));
250 }
251
252 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
253 {
254 if (size <= PAGE_SIZE)
255 kfree(table);
256 else
257 free_pages((unsigned long)table, get_order(size));
258 }
259
260 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
261 {
262 unsigned int omask, mask, shift;
263 size_t osize, size;
264 struct hlist_head *otable, *table;
265 int i;
266
267 omask = mask = hash->mask;
268 osize = size = (mask + 1) * sizeof(*table);
269 shift = hash->shift;
270
271 if (grow) {
272 if (++shift > hash->max_shift)
273 return 0;
274 mask = mask * 2 + 1;
275 size *= 2;
276 }
277
278 table = nl_pid_hash_zalloc(size);
279 if (!table)
280 return 0;
281
282 otable = hash->table;
283 hash->table = table;
284 hash->mask = mask;
285 hash->shift = shift;
286 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
287
288 for (i = 0; i <= omask; i++) {
289 struct sock *sk;
290 struct hlist_node *node, *tmp;
291
292 sk_for_each_safe(sk, node, tmp, &otable[i])
293 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
294 }
295
296 nl_pid_hash_free(otable, osize);
297 hash->rehash_time = jiffies + 10 * 60 * HZ;
298 return 1;
299 }
300
301 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
302 {
303 int avg = hash->entries >> hash->shift;
304
305 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
306 return 1;
307
308 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
309 nl_pid_hash_rehash(hash, 0);
310 return 1;
311 }
312
313 return 0;
314 }
315
316 static const struct proto_ops netlink_ops;
317
318 static void
319 netlink_update_listeners(struct sock *sk)
320 {
321 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
322 struct hlist_node *node;
323 unsigned long mask;
324 unsigned int i;
325
326 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
327 mask = 0;
328 sk_for_each_bound(sk, node, &tbl->mc_list) {
329 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
330 mask |= nlk_sk(sk)->groups[i];
331 }
332 tbl->listeners[i] = mask;
333 }
334 /* this function is only called with the netlink table "grabbed", which
335 * makes sure updates are visible before bind or setsockopt return. */
336 }
337
338 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
339 {
340 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
341 struct hlist_head *head;
342 int err = -EADDRINUSE;
343 struct sock *osk;
344 struct hlist_node *node;
345 int len;
346
347 netlink_table_grab();
348 head = nl_pid_hashfn(hash, pid);
349 len = 0;
350 sk_for_each(osk, node, head) {
351 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
352 break;
353 len++;
354 }
355 if (node)
356 goto err;
357
358 err = -EBUSY;
359 if (nlk_sk(sk)->pid)
360 goto err;
361
362 err = -ENOMEM;
363 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
364 goto err;
365
366 if (len && nl_pid_hash_dilute(hash, len))
367 head = nl_pid_hashfn(hash, pid);
368 hash->entries++;
369 nlk_sk(sk)->pid = pid;
370 sk_add_node(sk, head);
371 err = 0;
372
373 err:
374 netlink_table_ungrab();
375 return err;
376 }
377
378 static void netlink_remove(struct sock *sk)
379 {
380 netlink_table_grab();
381 if (sk_del_node_init(sk))
382 nl_table[sk->sk_protocol].hash.entries--;
383 if (nlk_sk(sk)->subscriptions)
384 __sk_del_bind_node(sk);
385 netlink_table_ungrab();
386 }
387
388 static struct proto netlink_proto = {
389 .name = "NETLINK",
390 .owner = THIS_MODULE,
391 .obj_size = sizeof(struct netlink_sock),
392 };
393
394 static int __netlink_create(struct net *net, struct socket *sock,
395 struct mutex *cb_mutex, int protocol)
396 {
397 struct sock *sk;
398 struct netlink_sock *nlk;
399
400 sock->ops = &netlink_ops;
401
402 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
403 if (!sk)
404 return -ENOMEM;
405
406 sock_init_data(sock, sk);
407
408 nlk = nlk_sk(sk);
409 if (cb_mutex)
410 nlk->cb_mutex = cb_mutex;
411 else {
412 nlk->cb_mutex = &nlk->cb_def_mutex;
413 mutex_init(nlk->cb_mutex);
414 }
415 init_waitqueue_head(&nlk->wait);
416
417 sk->sk_destruct = netlink_sock_destruct;
418 sk->sk_protocol = protocol;
419 return 0;
420 }
421
422 static int netlink_create(struct net *net, struct socket *sock, int protocol)
423 {
424 struct module *module = NULL;
425 struct mutex *cb_mutex;
426 struct netlink_sock *nlk;
427 int err = 0;
428
429 sock->state = SS_UNCONNECTED;
430
431 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
432 return -ESOCKTNOSUPPORT;
433
434 if (protocol < 0 || protocol >= MAX_LINKS)
435 return -EPROTONOSUPPORT;
436
437 netlink_lock_table();
438 #ifdef CONFIG_MODULES
439 if (!nl_table[protocol].registered) {
440 netlink_unlock_table();
441 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
442 netlink_lock_table();
443 }
444 #endif
445 if (nl_table[protocol].registered &&
446 try_module_get(nl_table[protocol].module))
447 module = nl_table[protocol].module;
448 cb_mutex = nl_table[protocol].cb_mutex;
449 netlink_unlock_table();
450
451 err = __netlink_create(net, sock, cb_mutex, protocol);
452 if (err < 0)
453 goto out_module;
454
455 sock_prot_inuse_add(net, &netlink_proto, 1);
456 nlk = nlk_sk(sock->sk);
457 nlk->module = module;
458 out:
459 return err;
460
461 out_module:
462 module_put(module);
463 goto out;
464 }
465
466 static int netlink_release(struct socket *sock)
467 {
468 struct sock *sk = sock->sk;
469 struct netlink_sock *nlk;
470
471 if (!sk)
472 return 0;
473
474 netlink_remove(sk);
475 sock_orphan(sk);
476 nlk = nlk_sk(sk);
477
478 /*
479 * OK. Socket is unlinked, any packets that arrive now
480 * will be purged.
481 */
482
483 sock->sk = NULL;
484 wake_up_interruptible_all(&nlk->wait);
485
486 skb_queue_purge(&sk->sk_write_queue);
487
488 if (nlk->pid && !nlk->subscriptions) {
489 struct netlink_notify n = {
490 .net = sock_net(sk),
491 .protocol = sk->sk_protocol,
492 .pid = nlk->pid,
493 };
494 atomic_notifier_call_chain(&netlink_chain,
495 NETLINK_URELEASE, &n);
496 }
497
498 module_put(nlk->module);
499
500 netlink_table_grab();
501 if (netlink_is_kernel(sk)) {
502 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
503 if (--nl_table[sk->sk_protocol].registered == 0) {
504 kfree(nl_table[sk->sk_protocol].listeners);
505 nl_table[sk->sk_protocol].module = NULL;
506 nl_table[sk->sk_protocol].registered = 0;
507 }
508 } else if (nlk->subscriptions)
509 netlink_update_listeners(sk);
510 netlink_table_ungrab();
511
512 kfree(nlk->groups);
513 nlk->groups = NULL;
514
515 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
516 sock_put(sk);
517 return 0;
518 }
519
520 static int netlink_autobind(struct socket *sock)
521 {
522 struct sock *sk = sock->sk;
523 struct net *net = sock_net(sk);
524 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
525 struct hlist_head *head;
526 struct sock *osk;
527 struct hlist_node *node;
528 s32 pid = current->tgid;
529 int err;
530 static s32 rover = -4097;
531
532 retry:
533 cond_resched();
534 netlink_table_grab();
535 head = nl_pid_hashfn(hash, pid);
536 sk_for_each(osk, node, head) {
537 if (!net_eq(sock_net(osk), net))
538 continue;
539 if (nlk_sk(osk)->pid == pid) {
540 /* Bind collision, search negative pid values. */
541 pid = rover--;
542 if (rover > -4097)
543 rover = -4097;
544 netlink_table_ungrab();
545 goto retry;
546 }
547 }
548 netlink_table_ungrab();
549
550 err = netlink_insert(sk, net, pid);
551 if (err == -EADDRINUSE)
552 goto retry;
553
554 /* If 2 threads race to autobind, that is fine. */
555 if (err == -EBUSY)
556 err = 0;
557
558 return err;
559 }
560
561 static inline int netlink_capable(struct socket *sock, unsigned int flag)
562 {
563 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
564 capable(CAP_NET_ADMIN);
565 }
566
567 static void
568 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
569 {
570 struct netlink_sock *nlk = nlk_sk(sk);
571
572 if (nlk->subscriptions && !subscriptions)
573 __sk_del_bind_node(sk);
574 else if (!nlk->subscriptions && subscriptions)
575 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
576 nlk->subscriptions = subscriptions;
577 }
578
579 static int netlink_realloc_groups(struct sock *sk)
580 {
581 struct netlink_sock *nlk = nlk_sk(sk);
582 unsigned int groups;
583 unsigned long *new_groups;
584 int err = 0;
585
586 netlink_table_grab();
587
588 groups = nl_table[sk->sk_protocol].groups;
589 if (!nl_table[sk->sk_protocol].registered) {
590 err = -ENOENT;
591 goto out_unlock;
592 }
593
594 if (nlk->ngroups >= groups)
595 goto out_unlock;
596
597 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
598 if (new_groups == NULL) {
599 err = -ENOMEM;
600 goto out_unlock;
601 }
602 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
603 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
604
605 nlk->groups = new_groups;
606 nlk->ngroups = groups;
607 out_unlock:
608 netlink_table_ungrab();
609 return err;
610 }
611
612 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
613 int addr_len)
614 {
615 struct sock *sk = sock->sk;
616 struct net *net = sock_net(sk);
617 struct netlink_sock *nlk = nlk_sk(sk);
618 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
619 int err;
620
621 if (nladdr->nl_family != AF_NETLINK)
622 return -EINVAL;
623
624 /* Only superuser is allowed to listen multicasts */
625 if (nladdr->nl_groups) {
626 if (!netlink_capable(sock, NL_NONROOT_RECV))
627 return -EPERM;
628 err = netlink_realloc_groups(sk);
629 if (err)
630 return err;
631 }
632
633 if (nlk->pid) {
634 if (nladdr->nl_pid != nlk->pid)
635 return -EINVAL;
636 } else {
637 err = nladdr->nl_pid ?
638 netlink_insert(sk, net, nladdr->nl_pid) :
639 netlink_autobind(sock);
640 if (err)
641 return err;
642 }
643
644 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
645 return 0;
646
647 netlink_table_grab();
648 netlink_update_subscriptions(sk, nlk->subscriptions +
649 hweight32(nladdr->nl_groups) -
650 hweight32(nlk->groups[0]));
651 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
652 netlink_update_listeners(sk);
653 netlink_table_ungrab();
654
655 return 0;
656 }
657
658 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
659 int alen, int flags)
660 {
661 int err = 0;
662 struct sock *sk = sock->sk;
663 struct netlink_sock *nlk = nlk_sk(sk);
664 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
665
666 if (addr->sa_family == AF_UNSPEC) {
667 sk->sk_state = NETLINK_UNCONNECTED;
668 nlk->dst_pid = 0;
669 nlk->dst_group = 0;
670 return 0;
671 }
672 if (addr->sa_family != AF_NETLINK)
673 return -EINVAL;
674
675 /* Only superuser is allowed to send multicasts */
676 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
677 return -EPERM;
678
679 if (!nlk->pid)
680 err = netlink_autobind(sock);
681
682 if (err == 0) {
683 sk->sk_state = NETLINK_CONNECTED;
684 nlk->dst_pid = nladdr->nl_pid;
685 nlk->dst_group = ffs(nladdr->nl_groups);
686 }
687
688 return err;
689 }
690
691 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
692 int *addr_len, int peer)
693 {
694 struct sock *sk = sock->sk;
695 struct netlink_sock *nlk = nlk_sk(sk);
696 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
697
698 nladdr->nl_family = AF_NETLINK;
699 nladdr->nl_pad = 0;
700 *addr_len = sizeof(*nladdr);
701
702 if (peer) {
703 nladdr->nl_pid = nlk->dst_pid;
704 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
705 } else {
706 nladdr->nl_pid = nlk->pid;
707 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
708 }
709 return 0;
710 }
711
712 static void netlink_overrun(struct sock *sk)
713 {
714 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
715 sk->sk_err = ENOBUFS;
716 sk->sk_error_report(sk);
717 }
718 }
719
720 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
721 {
722 struct sock *sock;
723 struct netlink_sock *nlk;
724
725 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
726 if (!sock)
727 return ERR_PTR(-ECONNREFUSED);
728
729 /* Don't bother queuing skb if kernel socket has no input function */
730 nlk = nlk_sk(sock);
731 if (sock->sk_state == NETLINK_CONNECTED &&
732 nlk->dst_pid != nlk_sk(ssk)->pid) {
733 sock_put(sock);
734 return ERR_PTR(-ECONNREFUSED);
735 }
736 return sock;
737 }
738
739 struct sock *netlink_getsockbyfilp(struct file *filp)
740 {
741 struct inode *inode = filp->f_path.dentry->d_inode;
742 struct sock *sock;
743
744 if (!S_ISSOCK(inode->i_mode))
745 return ERR_PTR(-ENOTSOCK);
746
747 sock = SOCKET_I(inode)->sk;
748 if (sock->sk_family != AF_NETLINK)
749 return ERR_PTR(-EINVAL);
750
751 sock_hold(sock);
752 return sock;
753 }
754
755 /*
756 * Attach a skb to a netlink socket.
757 * The caller must hold a reference to the destination socket. On error, the
758 * reference is dropped. The skb is not send to the destination, just all
759 * all error checks are performed and memory in the queue is reserved.
760 * Return values:
761 * < 0: error. skb freed, reference to sock dropped.
762 * 0: continue
763 * 1: repeat lookup - reference dropped while waiting for socket memory.
764 */
765 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
766 long *timeo, struct sock *ssk)
767 {
768 struct netlink_sock *nlk;
769
770 nlk = nlk_sk(sk);
771
772 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
773 test_bit(0, &nlk->state)) {
774 DECLARE_WAITQUEUE(wait, current);
775 if (!*timeo) {
776 if (!ssk || netlink_is_kernel(ssk))
777 netlink_overrun(sk);
778 sock_put(sk);
779 kfree_skb(skb);
780 return -EAGAIN;
781 }
782
783 __set_current_state(TASK_INTERRUPTIBLE);
784 add_wait_queue(&nlk->wait, &wait);
785
786 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
787 test_bit(0, &nlk->state)) &&
788 !sock_flag(sk, SOCK_DEAD))
789 *timeo = schedule_timeout(*timeo);
790
791 __set_current_state(TASK_RUNNING);
792 remove_wait_queue(&nlk->wait, &wait);
793 sock_put(sk);
794
795 if (signal_pending(current)) {
796 kfree_skb(skb);
797 return sock_intr_errno(*timeo);
798 }
799 return 1;
800 }
801 skb_set_owner_r(skb, sk);
802 return 0;
803 }
804
805 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
806 {
807 int len = skb->len;
808
809 skb_queue_tail(&sk->sk_receive_queue, skb);
810 sk->sk_data_ready(sk, len);
811 sock_put(sk);
812 return len;
813 }
814
815 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
816 {
817 kfree_skb(skb);
818 sock_put(sk);
819 }
820
821 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
822 gfp_t allocation)
823 {
824 int delta;
825
826 skb_orphan(skb);
827
828 delta = skb->end - skb->tail;
829 if (delta * 2 < skb->truesize)
830 return skb;
831
832 if (skb_shared(skb)) {
833 struct sk_buff *nskb = skb_clone(skb, allocation);
834 if (!nskb)
835 return skb;
836 kfree_skb(skb);
837 skb = nskb;
838 }
839
840 if (!pskb_expand_head(skb, 0, -delta, allocation))
841 skb->truesize -= delta;
842
843 return skb;
844 }
845
846 static inline void netlink_rcv_wake(struct sock *sk)
847 {
848 struct netlink_sock *nlk = nlk_sk(sk);
849
850 if (skb_queue_empty(&sk->sk_receive_queue))
851 clear_bit(0, &nlk->state);
852 if (!test_bit(0, &nlk->state))
853 wake_up_interruptible(&nlk->wait);
854 }
855
856 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
857 {
858 int ret;
859 struct netlink_sock *nlk = nlk_sk(sk);
860
861 ret = -ECONNREFUSED;
862 if (nlk->netlink_rcv != NULL) {
863 ret = skb->len;
864 skb_set_owner_r(skb, sk);
865 nlk->netlink_rcv(skb);
866 }
867 kfree_skb(skb);
868 sock_put(sk);
869 return ret;
870 }
871
872 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
873 u32 pid, int nonblock)
874 {
875 struct sock *sk;
876 int err;
877 long timeo;
878
879 skb = netlink_trim(skb, gfp_any());
880
881 timeo = sock_sndtimeo(ssk, nonblock);
882 retry:
883 sk = netlink_getsockbypid(ssk, pid);
884 if (IS_ERR(sk)) {
885 kfree_skb(skb);
886 return PTR_ERR(sk);
887 }
888 if (netlink_is_kernel(sk))
889 return netlink_unicast_kernel(sk, skb);
890
891 if (sk_filter(sk, skb)) {
892 err = skb->len;
893 kfree_skb(skb);
894 sock_put(sk);
895 return err;
896 }
897
898 err = netlink_attachskb(sk, skb, &timeo, ssk);
899 if (err == 1)
900 goto retry;
901 if (err)
902 return err;
903
904 return netlink_sendskb(sk, skb);
905 }
906 EXPORT_SYMBOL(netlink_unicast);
907
908 int netlink_has_listeners(struct sock *sk, unsigned int group)
909 {
910 int res = 0;
911 unsigned long *listeners;
912
913 BUG_ON(!netlink_is_kernel(sk));
914
915 rcu_read_lock();
916 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
917
918 if (group - 1 < nl_table[sk->sk_protocol].groups)
919 res = test_bit(group - 1, listeners);
920
921 rcu_read_unlock();
922
923 return res;
924 }
925 EXPORT_SYMBOL_GPL(netlink_has_listeners);
926
927 static inline int netlink_broadcast_deliver(struct sock *sk,
928 struct sk_buff *skb)
929 {
930 struct netlink_sock *nlk = nlk_sk(sk);
931
932 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
933 !test_bit(0, &nlk->state)) {
934 skb_set_owner_r(skb, sk);
935 skb_queue_tail(&sk->sk_receive_queue, skb);
936 sk->sk_data_ready(sk, skb->len);
937 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
938 }
939 return -1;
940 }
941
942 struct netlink_broadcast_data {
943 struct sock *exclude_sk;
944 struct net *net;
945 u32 pid;
946 u32 group;
947 int failure;
948 int congested;
949 int delivered;
950 gfp_t allocation;
951 struct sk_buff *skb, *skb2;
952 };
953
954 static inline int do_one_broadcast(struct sock *sk,
955 struct netlink_broadcast_data *p)
956 {
957 struct netlink_sock *nlk = nlk_sk(sk);
958 int val;
959
960 if (p->exclude_sk == sk)
961 goto out;
962
963 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
964 !test_bit(p->group - 1, nlk->groups))
965 goto out;
966
967 if (!net_eq(sock_net(sk), p->net))
968 goto out;
969
970 if (p->failure) {
971 netlink_overrun(sk);
972 goto out;
973 }
974
975 sock_hold(sk);
976 if (p->skb2 == NULL) {
977 if (skb_shared(p->skb)) {
978 p->skb2 = skb_clone(p->skb, p->allocation);
979 } else {
980 p->skb2 = skb_get(p->skb);
981 /*
982 * skb ownership may have been set when
983 * delivered to a previous socket.
984 */
985 skb_orphan(p->skb2);
986 }
987 }
988 if (p->skb2 == NULL) {
989 netlink_overrun(sk);
990 /* Clone failed. Notify ALL listeners. */
991 p->failure = 1;
992 } else if (sk_filter(sk, p->skb2)) {
993 kfree_skb(p->skb2);
994 p->skb2 = NULL;
995 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
996 netlink_overrun(sk);
997 } else {
998 p->congested |= val;
999 p->delivered = 1;
1000 p->skb2 = NULL;
1001 }
1002 sock_put(sk);
1003
1004 out:
1005 return 0;
1006 }
1007
1008 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1009 u32 group, gfp_t allocation)
1010 {
1011 struct net *net = sock_net(ssk);
1012 struct netlink_broadcast_data info;
1013 struct hlist_node *node;
1014 struct sock *sk;
1015
1016 skb = netlink_trim(skb, allocation);
1017
1018 info.exclude_sk = ssk;
1019 info.net = net;
1020 info.pid = pid;
1021 info.group = group;
1022 info.failure = 0;
1023 info.congested = 0;
1024 info.delivered = 0;
1025 info.allocation = allocation;
1026 info.skb = skb;
1027 info.skb2 = NULL;
1028
1029 /* While we sleep in clone, do not allow to change socket list */
1030
1031 netlink_lock_table();
1032
1033 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1034 do_one_broadcast(sk, &info);
1035
1036 kfree_skb(skb);
1037
1038 netlink_unlock_table();
1039
1040 if (info.skb2)
1041 kfree_skb(info.skb2);
1042
1043 if (info.delivered) {
1044 if (info.congested && (allocation & __GFP_WAIT))
1045 yield();
1046 return 0;
1047 }
1048 if (info.failure)
1049 return -ENOBUFS;
1050 return -ESRCH;
1051 }
1052 EXPORT_SYMBOL(netlink_broadcast);
1053
1054 struct netlink_set_err_data {
1055 struct sock *exclude_sk;
1056 u32 pid;
1057 u32 group;
1058 int code;
1059 };
1060
1061 static inline int do_one_set_err(struct sock *sk,
1062 struct netlink_set_err_data *p)
1063 {
1064 struct netlink_sock *nlk = nlk_sk(sk);
1065
1066 if (sk == p->exclude_sk)
1067 goto out;
1068
1069 if (sock_net(sk) != sock_net(p->exclude_sk))
1070 goto out;
1071
1072 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1073 !test_bit(p->group - 1, nlk->groups))
1074 goto out;
1075
1076 sk->sk_err = p->code;
1077 sk->sk_error_report(sk);
1078 out:
1079 return 0;
1080 }
1081
1082 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1083 {
1084 struct netlink_set_err_data info;
1085 struct hlist_node *node;
1086 struct sock *sk;
1087
1088 info.exclude_sk = ssk;
1089 info.pid = pid;
1090 info.group = group;
1091 info.code = code;
1092
1093 read_lock(&nl_table_lock);
1094
1095 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1096 do_one_set_err(sk, &info);
1097
1098 read_unlock(&nl_table_lock);
1099 }
1100
1101 /* must be called with netlink table grabbed */
1102 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1103 unsigned int group,
1104 int is_new)
1105 {
1106 int old, new = !!is_new, subscriptions;
1107
1108 old = test_bit(group - 1, nlk->groups);
1109 subscriptions = nlk->subscriptions - old + new;
1110 if (new)
1111 __set_bit(group - 1, nlk->groups);
1112 else
1113 __clear_bit(group - 1, nlk->groups);
1114 netlink_update_subscriptions(&nlk->sk, subscriptions);
1115 netlink_update_listeners(&nlk->sk);
1116 }
1117
1118 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1119 char __user *optval, int optlen)
1120 {
1121 struct sock *sk = sock->sk;
1122 struct netlink_sock *nlk = nlk_sk(sk);
1123 unsigned int val = 0;
1124 int err;
1125
1126 if (level != SOL_NETLINK)
1127 return -ENOPROTOOPT;
1128
1129 if (optlen >= sizeof(int) &&
1130 get_user(val, (unsigned int __user *)optval))
1131 return -EFAULT;
1132
1133 switch (optname) {
1134 case NETLINK_PKTINFO:
1135 if (val)
1136 nlk->flags |= NETLINK_RECV_PKTINFO;
1137 else
1138 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1139 err = 0;
1140 break;
1141 case NETLINK_ADD_MEMBERSHIP:
1142 case NETLINK_DROP_MEMBERSHIP: {
1143 if (!netlink_capable(sock, NL_NONROOT_RECV))
1144 return -EPERM;
1145 err = netlink_realloc_groups(sk);
1146 if (err)
1147 return err;
1148 if (!val || val - 1 >= nlk->ngroups)
1149 return -EINVAL;
1150 netlink_table_grab();
1151 netlink_update_socket_mc(nlk, val,
1152 optname == NETLINK_ADD_MEMBERSHIP);
1153 netlink_table_ungrab();
1154 err = 0;
1155 break;
1156 }
1157 default:
1158 err = -ENOPROTOOPT;
1159 }
1160 return err;
1161 }
1162
1163 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1164 char __user *optval, int __user *optlen)
1165 {
1166 struct sock *sk = sock->sk;
1167 struct netlink_sock *nlk = nlk_sk(sk);
1168 int len, val, err;
1169
1170 if (level != SOL_NETLINK)
1171 return -ENOPROTOOPT;
1172
1173 if (get_user(len, optlen))
1174 return -EFAULT;
1175 if (len < 0)
1176 return -EINVAL;
1177
1178 switch (optname) {
1179 case NETLINK_PKTINFO:
1180 if (len < sizeof(int))
1181 return -EINVAL;
1182 len = sizeof(int);
1183 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1184 if (put_user(len, optlen) ||
1185 put_user(val, optval))
1186 return -EFAULT;
1187 err = 0;
1188 break;
1189 default:
1190 err = -ENOPROTOOPT;
1191 }
1192 return err;
1193 }
1194
1195 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1196 {
1197 struct nl_pktinfo info;
1198
1199 info.group = NETLINK_CB(skb).dst_group;
1200 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1201 }
1202
1203 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1204 struct msghdr *msg, size_t len)
1205 {
1206 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1207 struct sock *sk = sock->sk;
1208 struct netlink_sock *nlk = nlk_sk(sk);
1209 struct sockaddr_nl *addr = msg->msg_name;
1210 u32 dst_pid;
1211 u32 dst_group;
1212 struct sk_buff *skb;
1213 int err;
1214 struct scm_cookie scm;
1215
1216 if (msg->msg_flags&MSG_OOB)
1217 return -EOPNOTSUPP;
1218
1219 if (NULL == siocb->scm)
1220 siocb->scm = &scm;
1221 err = scm_send(sock, msg, siocb->scm);
1222 if (err < 0)
1223 return err;
1224
1225 if (msg->msg_namelen) {
1226 if (addr->nl_family != AF_NETLINK)
1227 return -EINVAL;
1228 dst_pid = addr->nl_pid;
1229 dst_group = ffs(addr->nl_groups);
1230 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1231 return -EPERM;
1232 } else {
1233 dst_pid = nlk->dst_pid;
1234 dst_group = nlk->dst_group;
1235 }
1236
1237 if (!nlk->pid) {
1238 err = netlink_autobind(sock);
1239 if (err)
1240 goto out;
1241 }
1242
1243 err = -EMSGSIZE;
1244 if (len > sk->sk_sndbuf - 32)
1245 goto out;
1246 err = -ENOBUFS;
1247 skb = alloc_skb(len, GFP_KERNEL);
1248 if (skb == NULL)
1249 goto out;
1250
1251 NETLINK_CB(skb).pid = nlk->pid;
1252 NETLINK_CB(skb).dst_group = dst_group;
1253 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1254 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1255 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1256 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1257
1258 /* What can I do? Netlink is asynchronous, so that
1259 we will have to save current capabilities to
1260 check them, when this message will be delivered
1261 to corresponding kernel module. --ANK (980802)
1262 */
1263
1264 err = -EFAULT;
1265 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1266 kfree_skb(skb);
1267 goto out;
1268 }
1269
1270 err = security_netlink_send(sk, skb);
1271 if (err) {
1272 kfree_skb(skb);
1273 goto out;
1274 }
1275
1276 if (dst_group) {
1277 atomic_inc(&skb->users);
1278 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1279 }
1280 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1281
1282 out:
1283 return err;
1284 }
1285
1286 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1287 struct msghdr *msg, size_t len,
1288 int flags)
1289 {
1290 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1291 struct scm_cookie scm;
1292 struct sock *sk = sock->sk;
1293 struct netlink_sock *nlk = nlk_sk(sk);
1294 int noblock = flags&MSG_DONTWAIT;
1295 size_t copied;
1296 struct sk_buff *skb;
1297 int err;
1298
1299 if (flags&MSG_OOB)
1300 return -EOPNOTSUPP;
1301
1302 copied = 0;
1303
1304 skb = skb_recv_datagram(sk, flags, noblock, &err);
1305 if (skb == NULL)
1306 goto out;
1307
1308 msg->msg_namelen = 0;
1309
1310 copied = skb->len;
1311 if (len < copied) {
1312 msg->msg_flags |= MSG_TRUNC;
1313 copied = len;
1314 }
1315
1316 skb_reset_transport_header(skb);
1317 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1318
1319 if (msg->msg_name) {
1320 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1321 addr->nl_family = AF_NETLINK;
1322 addr->nl_pad = 0;
1323 addr->nl_pid = NETLINK_CB(skb).pid;
1324 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1325 msg->msg_namelen = sizeof(*addr);
1326 }
1327
1328 if (nlk->flags & NETLINK_RECV_PKTINFO)
1329 netlink_cmsg_recv_pktinfo(msg, skb);
1330
1331 if (NULL == siocb->scm) {
1332 memset(&scm, 0, sizeof(scm));
1333 siocb->scm = &scm;
1334 }
1335 siocb->scm->creds = *NETLINK_CREDS(skb);
1336 if (flags & MSG_TRUNC)
1337 copied = skb->len;
1338 skb_free_datagram(sk, skb);
1339
1340 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1341 netlink_dump(sk);
1342
1343 scm_recv(sock, msg, siocb->scm, flags);
1344 out:
1345 netlink_rcv_wake(sk);
1346 return err ? : copied;
1347 }
1348
1349 static void netlink_data_ready(struct sock *sk, int len)
1350 {
1351 BUG();
1352 }
1353
1354 /*
1355 * We export these functions to other modules. They provide a
1356 * complete set of kernel non-blocking support for message
1357 * queueing.
1358 */
1359
1360 struct sock *
1361 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1362 void (*input)(struct sk_buff *skb),
1363 struct mutex *cb_mutex, struct module *module)
1364 {
1365 struct socket *sock;
1366 struct sock *sk;
1367 struct netlink_sock *nlk;
1368 unsigned long *listeners = NULL;
1369
1370 BUG_ON(!nl_table);
1371
1372 if (unit < 0 || unit >= MAX_LINKS)
1373 return NULL;
1374
1375 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1376 return NULL;
1377
1378 /*
1379 * We have to just have a reference on the net from sk, but don't
1380 * get_net it. Besides, we cannot get and then put the net here.
1381 * So we create one inside init_net and the move it to net.
1382 */
1383
1384 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1385 goto out_sock_release_nosk;
1386
1387 sk = sock->sk;
1388 sk_change_net(sk, net);
1389
1390 if (groups < 32)
1391 groups = 32;
1392
1393 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1394 if (!listeners)
1395 goto out_sock_release;
1396
1397 sk->sk_data_ready = netlink_data_ready;
1398 if (input)
1399 nlk_sk(sk)->netlink_rcv = input;
1400
1401 if (netlink_insert(sk, net, 0))
1402 goto out_sock_release;
1403
1404 nlk = nlk_sk(sk);
1405 nlk->flags |= NETLINK_KERNEL_SOCKET;
1406
1407 netlink_table_grab();
1408 if (!nl_table[unit].registered) {
1409 nl_table[unit].groups = groups;
1410 nl_table[unit].listeners = listeners;
1411 nl_table[unit].cb_mutex = cb_mutex;
1412 nl_table[unit].module = module;
1413 nl_table[unit].registered = 1;
1414 } else {
1415 kfree(listeners);
1416 nl_table[unit].registered++;
1417 }
1418 netlink_table_ungrab();
1419 return sk;
1420
1421 out_sock_release:
1422 kfree(listeners);
1423 netlink_kernel_release(sk);
1424 return NULL;
1425
1426 out_sock_release_nosk:
1427 sock_release(sock);
1428 return NULL;
1429 }
1430 EXPORT_SYMBOL(netlink_kernel_create);
1431
1432
1433 void
1434 netlink_kernel_release(struct sock *sk)
1435 {
1436 sk_release_kernel(sk);
1437 }
1438 EXPORT_SYMBOL(netlink_kernel_release);
1439
1440
1441 /**
1442 * netlink_change_ngroups - change number of multicast groups
1443 *
1444 * This changes the number of multicast groups that are available
1445 * on a certain netlink family. Note that it is not possible to
1446 * change the number of groups to below 32. Also note that it does
1447 * not implicitly call netlink_clear_multicast_users() when the
1448 * number of groups is reduced.
1449 *
1450 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1451 * @groups: The new number of groups.
1452 */
1453 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1454 {
1455 unsigned long *listeners, *old = NULL;
1456 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1457 int err = 0;
1458
1459 if (groups < 32)
1460 groups = 32;
1461
1462 netlink_table_grab();
1463 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1464 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1465 if (!listeners) {
1466 err = -ENOMEM;
1467 goto out_ungrab;
1468 }
1469 old = tbl->listeners;
1470 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1471 rcu_assign_pointer(tbl->listeners, listeners);
1472 }
1473 tbl->groups = groups;
1474
1475 out_ungrab:
1476 netlink_table_ungrab();
1477 synchronize_rcu();
1478 kfree(old);
1479 return err;
1480 }
1481 EXPORT_SYMBOL(netlink_change_ngroups);
1482
1483 /**
1484 * netlink_clear_multicast_users - kick off multicast listeners
1485 *
1486 * This function removes all listeners from the given group.
1487 * @ksk: The kernel netlink socket, as returned by
1488 * netlink_kernel_create().
1489 * @group: The multicast group to clear.
1490 */
1491 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1492 {
1493 struct sock *sk;
1494 struct hlist_node *node;
1495 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1496
1497 netlink_table_grab();
1498
1499 sk_for_each_bound(sk, node, &tbl->mc_list)
1500 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1501
1502 netlink_table_ungrab();
1503 }
1504 EXPORT_SYMBOL(netlink_clear_multicast_users);
1505
1506 void netlink_set_nonroot(int protocol, unsigned int flags)
1507 {
1508 if ((unsigned int)protocol < MAX_LINKS)
1509 nl_table[protocol].nl_nonroot = flags;
1510 }
1511 EXPORT_SYMBOL(netlink_set_nonroot);
1512
1513 static void netlink_destroy_callback(struct netlink_callback *cb)
1514 {
1515 if (cb->skb)
1516 kfree_skb(cb->skb);
1517 kfree(cb);
1518 }
1519
1520 /*
1521 * It looks a bit ugly.
1522 * It would be better to create kernel thread.
1523 */
1524
1525 static int netlink_dump(struct sock *sk)
1526 {
1527 struct netlink_sock *nlk = nlk_sk(sk);
1528 struct netlink_callback *cb;
1529 struct sk_buff *skb;
1530 struct nlmsghdr *nlh;
1531 int len, err = -ENOBUFS;
1532
1533 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1534 if (!skb)
1535 goto errout;
1536
1537 mutex_lock(nlk->cb_mutex);
1538
1539 cb = nlk->cb;
1540 if (cb == NULL) {
1541 err = -EINVAL;
1542 goto errout_skb;
1543 }
1544
1545 len = cb->dump(skb, cb);
1546
1547 if (len > 0) {
1548 mutex_unlock(nlk->cb_mutex);
1549
1550 if (sk_filter(sk, skb))
1551 kfree_skb(skb);
1552 else {
1553 skb_queue_tail(&sk->sk_receive_queue, skb);
1554 sk->sk_data_ready(sk, skb->len);
1555 }
1556 return 0;
1557 }
1558
1559 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1560 if (!nlh)
1561 goto errout_skb;
1562
1563 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1564
1565 if (sk_filter(sk, skb))
1566 kfree_skb(skb);
1567 else {
1568 skb_queue_tail(&sk->sk_receive_queue, skb);
1569 sk->sk_data_ready(sk, skb->len);
1570 }
1571
1572 if (cb->done)
1573 cb->done(cb);
1574 nlk->cb = NULL;
1575 mutex_unlock(nlk->cb_mutex);
1576
1577 netlink_destroy_callback(cb);
1578 return 0;
1579
1580 errout_skb:
1581 mutex_unlock(nlk->cb_mutex);
1582 kfree_skb(skb);
1583 errout:
1584 return err;
1585 }
1586
1587 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1588 struct nlmsghdr *nlh,
1589 int (*dump)(struct sk_buff *skb,
1590 struct netlink_callback *),
1591 int (*done)(struct netlink_callback *))
1592 {
1593 struct netlink_callback *cb;
1594 struct sock *sk;
1595 struct netlink_sock *nlk;
1596
1597 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1598 if (cb == NULL)
1599 return -ENOBUFS;
1600
1601 cb->dump = dump;
1602 cb->done = done;
1603 cb->nlh = nlh;
1604 atomic_inc(&skb->users);
1605 cb->skb = skb;
1606
1607 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1608 if (sk == NULL) {
1609 netlink_destroy_callback(cb);
1610 return -ECONNREFUSED;
1611 }
1612 nlk = nlk_sk(sk);
1613 /* A dump is in progress... */
1614 mutex_lock(nlk->cb_mutex);
1615 if (nlk->cb) {
1616 mutex_unlock(nlk->cb_mutex);
1617 netlink_destroy_callback(cb);
1618 sock_put(sk);
1619 return -EBUSY;
1620 }
1621 nlk->cb = cb;
1622 mutex_unlock(nlk->cb_mutex);
1623
1624 netlink_dump(sk);
1625 sock_put(sk);
1626
1627 /* We successfully started a dump, by returning -EINTR we
1628 * signal not to send ACK even if it was requested.
1629 */
1630 return -EINTR;
1631 }
1632 EXPORT_SYMBOL(netlink_dump_start);
1633
1634 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1635 {
1636 struct sk_buff *skb;
1637 struct nlmsghdr *rep;
1638 struct nlmsgerr *errmsg;
1639 size_t payload = sizeof(*errmsg);
1640
1641 /* error messages get the original request appened */
1642 if (err)
1643 payload += nlmsg_len(nlh);
1644
1645 skb = nlmsg_new(payload, GFP_KERNEL);
1646 if (!skb) {
1647 struct sock *sk;
1648
1649 sk = netlink_lookup(sock_net(in_skb->sk),
1650 in_skb->sk->sk_protocol,
1651 NETLINK_CB(in_skb).pid);
1652 if (sk) {
1653 sk->sk_err = ENOBUFS;
1654 sk->sk_error_report(sk);
1655 sock_put(sk);
1656 }
1657 return;
1658 }
1659
1660 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1661 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1662 errmsg = nlmsg_data(rep);
1663 errmsg->error = err;
1664 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1665 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1666 }
1667 EXPORT_SYMBOL(netlink_ack);
1668
1669 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1670 struct nlmsghdr *))
1671 {
1672 struct nlmsghdr *nlh;
1673 int err;
1674
1675 while (skb->len >= nlmsg_total_size(0)) {
1676 int msglen;
1677
1678 nlh = nlmsg_hdr(skb);
1679 err = 0;
1680
1681 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1682 return 0;
1683
1684 /* Only requests are handled by the kernel */
1685 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1686 goto ack;
1687
1688 /* Skip control messages */
1689 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1690 goto ack;
1691
1692 err = cb(skb, nlh);
1693 if (err == -EINTR)
1694 goto skip;
1695
1696 ack:
1697 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1698 netlink_ack(skb, nlh, err);
1699
1700 skip:
1701 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1702 if (msglen > skb->len)
1703 msglen = skb->len;
1704 skb_pull(skb, msglen);
1705 }
1706
1707 return 0;
1708 }
1709 EXPORT_SYMBOL(netlink_rcv_skb);
1710
1711 /**
1712 * nlmsg_notify - send a notification netlink message
1713 * @sk: netlink socket to use
1714 * @skb: notification message
1715 * @pid: destination netlink pid for reports or 0
1716 * @group: destination multicast group or 0
1717 * @report: 1 to report back, 0 to disable
1718 * @flags: allocation flags
1719 */
1720 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1721 unsigned int group, int report, gfp_t flags)
1722 {
1723 int err = 0;
1724
1725 if (group) {
1726 int exclude_pid = 0;
1727
1728 if (report) {
1729 atomic_inc(&skb->users);
1730 exclude_pid = pid;
1731 }
1732
1733 /* errors reported via destination sk->sk_err */
1734 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1735 }
1736
1737 if (report)
1738 err = nlmsg_unicast(sk, skb, pid);
1739
1740 return err;
1741 }
1742 EXPORT_SYMBOL(nlmsg_notify);
1743
1744 #ifdef CONFIG_PROC_FS
1745 struct nl_seq_iter {
1746 struct seq_net_private p;
1747 int link;
1748 int hash_idx;
1749 };
1750
1751 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1752 {
1753 struct nl_seq_iter *iter = seq->private;
1754 int i, j;
1755 struct sock *s;
1756 struct hlist_node *node;
1757 loff_t off = 0;
1758
1759 for (i = 0; i < MAX_LINKS; i++) {
1760 struct nl_pid_hash *hash = &nl_table[i].hash;
1761
1762 for (j = 0; j <= hash->mask; j++) {
1763 sk_for_each(s, node, &hash->table[j]) {
1764 if (sock_net(s) != seq_file_net(seq))
1765 continue;
1766 if (off == pos) {
1767 iter->link = i;
1768 iter->hash_idx = j;
1769 return s;
1770 }
1771 ++off;
1772 }
1773 }
1774 }
1775 return NULL;
1776 }
1777
1778 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1779 __acquires(nl_table_lock)
1780 {
1781 read_lock(&nl_table_lock);
1782 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1783 }
1784
1785 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1786 {
1787 struct sock *s;
1788 struct nl_seq_iter *iter;
1789 int i, j;
1790
1791 ++*pos;
1792
1793 if (v == SEQ_START_TOKEN)
1794 return netlink_seq_socket_idx(seq, 0);
1795
1796 iter = seq->private;
1797 s = v;
1798 do {
1799 s = sk_next(s);
1800 } while (s && sock_net(s) != seq_file_net(seq));
1801 if (s)
1802 return s;
1803
1804 i = iter->link;
1805 j = iter->hash_idx + 1;
1806
1807 do {
1808 struct nl_pid_hash *hash = &nl_table[i].hash;
1809
1810 for (; j <= hash->mask; j++) {
1811 s = sk_head(&hash->table[j]);
1812 while (s && sock_net(s) != seq_file_net(seq))
1813 s = sk_next(s);
1814 if (s) {
1815 iter->link = i;
1816 iter->hash_idx = j;
1817 return s;
1818 }
1819 }
1820
1821 j = 0;
1822 } while (++i < MAX_LINKS);
1823
1824 return NULL;
1825 }
1826
1827 static void netlink_seq_stop(struct seq_file *seq, void *v)
1828 __releases(nl_table_lock)
1829 {
1830 read_unlock(&nl_table_lock);
1831 }
1832
1833
1834 static int netlink_seq_show(struct seq_file *seq, void *v)
1835 {
1836 if (v == SEQ_START_TOKEN)
1837 seq_puts(seq,
1838 "sk Eth Pid Groups "
1839 "Rmem Wmem Dump Locks\n");
1840 else {
1841 struct sock *s = v;
1842 struct netlink_sock *nlk = nlk_sk(s);
1843
1844 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1845 s,
1846 s->sk_protocol,
1847 nlk->pid,
1848 nlk->groups ? (u32)nlk->groups[0] : 0,
1849 atomic_read(&s->sk_rmem_alloc),
1850 atomic_read(&s->sk_wmem_alloc),
1851 nlk->cb,
1852 atomic_read(&s->sk_refcnt)
1853 );
1854
1855 }
1856 return 0;
1857 }
1858
1859 static const struct seq_operations netlink_seq_ops = {
1860 .start = netlink_seq_start,
1861 .next = netlink_seq_next,
1862 .stop = netlink_seq_stop,
1863 .show = netlink_seq_show,
1864 };
1865
1866
1867 static int netlink_seq_open(struct inode *inode, struct file *file)
1868 {
1869 return seq_open_net(inode, file, &netlink_seq_ops,
1870 sizeof(struct nl_seq_iter));
1871 }
1872
1873 static const struct file_operations netlink_seq_fops = {
1874 .owner = THIS_MODULE,
1875 .open = netlink_seq_open,
1876 .read = seq_read,
1877 .llseek = seq_lseek,
1878 .release = seq_release_net,
1879 };
1880
1881 #endif
1882
1883 int netlink_register_notifier(struct notifier_block *nb)
1884 {
1885 return atomic_notifier_chain_register(&netlink_chain, nb);
1886 }
1887 EXPORT_SYMBOL(netlink_register_notifier);
1888
1889 int netlink_unregister_notifier(struct notifier_block *nb)
1890 {
1891 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1892 }
1893 EXPORT_SYMBOL(netlink_unregister_notifier);
1894
1895 static const struct proto_ops netlink_ops = {
1896 .family = PF_NETLINK,
1897 .owner = THIS_MODULE,
1898 .release = netlink_release,
1899 .bind = netlink_bind,
1900 .connect = netlink_connect,
1901 .socketpair = sock_no_socketpair,
1902 .accept = sock_no_accept,
1903 .getname = netlink_getname,
1904 .poll = datagram_poll,
1905 .ioctl = sock_no_ioctl,
1906 .listen = sock_no_listen,
1907 .shutdown = sock_no_shutdown,
1908 .setsockopt = netlink_setsockopt,
1909 .getsockopt = netlink_getsockopt,
1910 .sendmsg = netlink_sendmsg,
1911 .recvmsg = netlink_recvmsg,
1912 .mmap = sock_no_mmap,
1913 .sendpage = sock_no_sendpage,
1914 };
1915
1916 static struct net_proto_family netlink_family_ops = {
1917 .family = PF_NETLINK,
1918 .create = netlink_create,
1919 .owner = THIS_MODULE, /* for consistency 8) */
1920 };
1921
1922 static int __net_init netlink_net_init(struct net *net)
1923 {
1924 #ifdef CONFIG_PROC_FS
1925 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1926 return -ENOMEM;
1927 #endif
1928 return 0;
1929 }
1930
1931 static void __net_exit netlink_net_exit(struct net *net)
1932 {
1933 #ifdef CONFIG_PROC_FS
1934 proc_net_remove(net, "netlink");
1935 #endif
1936 }
1937
1938 static struct pernet_operations __net_initdata netlink_net_ops = {
1939 .init = netlink_net_init,
1940 .exit = netlink_net_exit,
1941 };
1942
1943 static int __init netlink_proto_init(void)
1944 {
1945 struct sk_buff *dummy_skb;
1946 int i;
1947 unsigned long limit;
1948 unsigned int order;
1949 int err = proto_register(&netlink_proto, 0);
1950
1951 if (err != 0)
1952 goto out;
1953
1954 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1955
1956 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1957 if (!nl_table)
1958 goto panic;
1959
1960 if (num_physpages >= (128 * 1024))
1961 limit = num_physpages >> (21 - PAGE_SHIFT);
1962 else
1963 limit = num_physpages >> (23 - PAGE_SHIFT);
1964
1965 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1966 limit = (1UL << order) / sizeof(struct hlist_head);
1967 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1968
1969 for (i = 0; i < MAX_LINKS; i++) {
1970 struct nl_pid_hash *hash = &nl_table[i].hash;
1971
1972 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1973 if (!hash->table) {
1974 while (i-- > 0)
1975 nl_pid_hash_free(nl_table[i].hash.table,
1976 1 * sizeof(*hash->table));
1977 kfree(nl_table);
1978 goto panic;
1979 }
1980 hash->max_shift = order;
1981 hash->shift = 0;
1982 hash->mask = 0;
1983 hash->rehash_time = jiffies;
1984 }
1985
1986 sock_register(&netlink_family_ops);
1987 register_pernet_subsys(&netlink_net_ops);
1988 /* The netlink device handler may be needed early. */
1989 rtnetlink_init();
1990 out:
1991 return err;
1992 panic:
1993 panic("netlink_init: Cannot allocate nl_table\n");
1994 }
1995
1996 core_initcall(netlink_proto_init);
CJK support for tty framebuffer vt