?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
e1000e: bump the version number
[linux-2.6.git] / net / netlink / af_netlink.c
blob8b6bbb3032b0ca1718037ff0b3139e883c3d813a
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 #define NETLINK_BROADCAST_SEND_ERROR 0x4
89 #define NETLINK_RECV_NO_ENOBUFS 0x8
90
91 static inline struct netlink_sock *nlk_sk(struct sock *sk)
92 {
93 return container_of(sk, struct netlink_sock, sk);
94 }
95
96 static inline int netlink_is_kernel(struct sock *sk)
97 {
98 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
99 }
100
101 struct nl_pid_hash {
102 struct hlist_head *table;
103 unsigned long rehash_time;
104
105 unsigned int mask;
106 unsigned int shift;
107
108 unsigned int entries;
109 unsigned int max_shift;
110
111 u32 rnd;
112 };
113
114 struct netlink_table {
115 struct nl_pid_hash hash;
116 struct hlist_head mc_list;
117 unsigned long *listeners;
118 unsigned int nl_nonroot;
119 unsigned int groups;
120 struct mutex *cb_mutex;
121 struct module *module;
122 int registered;
123 };
124
125 static struct netlink_table *nl_table;
126
127 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
128
129 static int netlink_dump(struct sock *sk);
130 static void netlink_destroy_callback(struct netlink_callback *cb);
131
132 static DEFINE_RWLOCK(nl_table_lock);
133 static atomic_t nl_table_users = ATOMIC_INIT(0);
134
135 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
136
137 static u32 netlink_group_mask(u32 group)
138 {
139 return group ? 1 << (group - 1) : 0;
140 }
141
142 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
143 {
144 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
145 }
146
147 static void netlink_sock_destruct(struct sock *sk)
148 {
149 struct netlink_sock *nlk = nlk_sk(sk);
150
151 if (nlk->cb) {
152 if (nlk->cb->done)
153 nlk->cb->done(nlk->cb);
154 netlink_destroy_callback(nlk->cb);
155 }
156
157 skb_queue_purge(&sk->sk_receive_queue);
158
159 if (!sock_flag(sk, SOCK_DEAD)) {
160 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
161 return;
162 }
163
164 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
165 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
166 WARN_ON(nlk_sk(sk)->groups);
167 }
168
169 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
170 * SMP. Look, when several writers sleep and reader wakes them up, all but one
171 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
172 * this, _but_ remember, it adds useless work on UP machines.
173 */
174
175 static void netlink_table_grab(void)
176 __acquires(nl_table_lock)
177 {
178 write_lock_irq(&nl_table_lock);
179
180 if (atomic_read(&nl_table_users)) {
181 DECLARE_WAITQUEUE(wait, current);
182
183 add_wait_queue_exclusive(&nl_table_wait, &wait);
184 for (;;) {
185 set_current_state(TASK_UNINTERRUPTIBLE);
186 if (atomic_read(&nl_table_users) == 0)
187 break;
188 write_unlock_irq(&nl_table_lock);
189 schedule();
190 write_lock_irq(&nl_table_lock);
191 }
192
193 __set_current_state(TASK_RUNNING);
194 remove_wait_queue(&nl_table_wait, &wait);
195 }
196 }
197
198 static void netlink_table_ungrab(void)
199 __releases(nl_table_lock)
200 {
201 write_unlock_irq(&nl_table_lock);
202 wake_up(&nl_table_wait);
203 }
204
205 static inline void
206 netlink_lock_table(void)
207 {
208 /* read_lock() synchronizes us to netlink_table_grab */
209
210 read_lock(&nl_table_lock);
211 atomic_inc(&nl_table_users);
212 read_unlock(&nl_table_lock);
213 }
214
215 static inline void
216 netlink_unlock_table(void)
217 {
218 if (atomic_dec_and_test(&nl_table_users))
219 wake_up(&nl_table_wait);
220 }
221
222 static inline struct sock *netlink_lookup(struct net *net, int protocol,
223 u32 pid)
224 {
225 struct nl_pid_hash *hash = &nl_table[protocol].hash;
226 struct hlist_head *head;
227 struct sock *sk;
228 struct hlist_node *node;
229
230 read_lock(&nl_table_lock);
231 head = nl_pid_hashfn(hash, pid);
232 sk_for_each(sk, node, head) {
233 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
234 sock_hold(sk);
235 goto found;
236 }
237 }
238 sk = NULL;
239 found:
240 read_unlock(&nl_table_lock);
241 return sk;
242 }
243
244 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
245 {
246 if (size <= PAGE_SIZE)
247 return kzalloc(size, GFP_ATOMIC);
248 else
249 return (struct hlist_head *)
250 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
251 get_order(size));
252 }
253
254 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
255 {
256 if (size <= PAGE_SIZE)
257 kfree(table);
258 else
259 free_pages((unsigned long)table, get_order(size));
260 }
261
262 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
263 {
264 unsigned int omask, mask, shift;
265 size_t osize, size;
266 struct hlist_head *otable, *table;
267 int i;
268
269 omask = mask = hash->mask;
270 osize = size = (mask + 1) * sizeof(*table);
271 shift = hash->shift;
272
273 if (grow) {
274 if (++shift > hash->max_shift)
275 return 0;
276 mask = mask * 2 + 1;
277 size *= 2;
278 }
279
280 table = nl_pid_hash_zalloc(size);
281 if (!table)
282 return 0;
283
284 otable = hash->table;
285 hash->table = table;
286 hash->mask = mask;
287 hash->shift = shift;
288 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
289
290 for (i = 0; i <= omask; i++) {
291 struct sock *sk;
292 struct hlist_node *node, *tmp;
293
294 sk_for_each_safe(sk, node, tmp, &otable[i])
295 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
296 }
297
298 nl_pid_hash_free(otable, osize);
299 hash->rehash_time = jiffies + 10 * 60 * HZ;
300 return 1;
301 }
302
303 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
304 {
305 int avg = hash->entries >> hash->shift;
306
307 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
308 return 1;
309
310 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
311 nl_pid_hash_rehash(hash, 0);
312 return 1;
313 }
314
315 return 0;
316 }
317
318 static const struct proto_ops netlink_ops;
319
320 static void
321 netlink_update_listeners(struct sock *sk)
322 {
323 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
324 struct hlist_node *node;
325 unsigned long mask;
326 unsigned int i;
327
328 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
329 mask = 0;
330 sk_for_each_bound(sk, node, &tbl->mc_list) {
331 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
332 mask |= nlk_sk(sk)->groups[i];
333 }
334 tbl->listeners[i] = mask;
335 }
336 /* this function is only called with the netlink table "grabbed", which
337 * makes sure updates are visible before bind or setsockopt return. */
338 }
339
340 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
341 {
342 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
343 struct hlist_head *head;
344 int err = -EADDRINUSE;
345 struct sock *osk;
346 struct hlist_node *node;
347 int len;
348
349 netlink_table_grab();
350 head = nl_pid_hashfn(hash, pid);
351 len = 0;
352 sk_for_each(osk, node, head) {
353 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
354 break;
355 len++;
356 }
357 if (node)
358 goto err;
359
360 err = -EBUSY;
361 if (nlk_sk(sk)->pid)
362 goto err;
363
364 err = -ENOMEM;
365 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
366 goto err;
367
368 if (len && nl_pid_hash_dilute(hash, len))
369 head = nl_pid_hashfn(hash, pid);
370 hash->entries++;
371 nlk_sk(sk)->pid = pid;
372 sk_add_node(sk, head);
373 err = 0;
374
375 err:
376 netlink_table_ungrab();
377 return err;
378 }
379
380 static void netlink_remove(struct sock *sk)
381 {
382 netlink_table_grab();
383 if (sk_del_node_init(sk))
384 nl_table[sk->sk_protocol].hash.entries--;
385 if (nlk_sk(sk)->subscriptions)
386 __sk_del_bind_node(sk);
387 netlink_table_ungrab();
388 }
389
390 static struct proto netlink_proto = {
391 .name = "NETLINK",
392 .owner = THIS_MODULE,
393 .obj_size = sizeof(struct netlink_sock),
394 };
395
396 static int __netlink_create(struct net *net, struct socket *sock,
397 struct mutex *cb_mutex, int protocol)
398 {
399 struct sock *sk;
400 struct netlink_sock *nlk;
401
402 sock->ops = &netlink_ops;
403
404 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
405 if (!sk)
406 return -ENOMEM;
407
408 sock_init_data(sock, sk);
409
410 nlk = nlk_sk(sk);
411 if (cb_mutex)
412 nlk->cb_mutex = cb_mutex;
413 else {
414 nlk->cb_mutex = &nlk->cb_def_mutex;
415 mutex_init(nlk->cb_mutex);
416 }
417 init_waitqueue_head(&nlk->wait);
418
419 sk->sk_destruct = netlink_sock_destruct;
420 sk->sk_protocol = protocol;
421 return 0;
422 }
423
424 static int netlink_create(struct net *net, struct socket *sock, int protocol)
425 {
426 struct module *module = NULL;
427 struct mutex *cb_mutex;
428 struct netlink_sock *nlk;
429 int err = 0;
430
431 sock->state = SS_UNCONNECTED;
432
433 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
434 return -ESOCKTNOSUPPORT;
435
436 if (protocol < 0 || protocol >= MAX_LINKS)
437 return -EPROTONOSUPPORT;
438
439 netlink_lock_table();
440 #ifdef CONFIG_MODULES
441 if (!nl_table[protocol].registered) {
442 netlink_unlock_table();
443 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
444 netlink_lock_table();
445 }
446 #endif
447 if (nl_table[protocol].registered &&
448 try_module_get(nl_table[protocol].module))
449 module = nl_table[protocol].module;
450 cb_mutex = nl_table[protocol].cb_mutex;
451 netlink_unlock_table();
452
453 err = __netlink_create(net, sock, cb_mutex, protocol);
454 if (err < 0)
455 goto out_module;
456
457 local_bh_disable();
458 sock_prot_inuse_add(net, &netlink_proto, 1);
459 local_bh_enable();
460
461 nlk = nlk_sk(sock->sk);
462 nlk->module = module;
463 out:
464 return err;
465
466 out_module:
467 module_put(module);
468 goto out;
469 }
470
471 static int netlink_release(struct socket *sock)
472 {
473 struct sock *sk = sock->sk;
474 struct netlink_sock *nlk;
475
476 if (!sk)
477 return 0;
478
479 netlink_remove(sk);
480 sock_orphan(sk);
481 nlk = nlk_sk(sk);
482
483 /*
484 * OK. Socket is unlinked, any packets that arrive now
485 * will be purged.
486 */
487
488 sock->sk = NULL;
489 wake_up_interruptible_all(&nlk->wait);
490
491 skb_queue_purge(&sk->sk_write_queue);
492
493 if (nlk->pid && !nlk->subscriptions) {
494 struct netlink_notify n = {
495 .net = sock_net(sk),
496 .protocol = sk->sk_protocol,
497 .pid = nlk->pid,
498 };
499 atomic_notifier_call_chain(&netlink_chain,
500 NETLINK_URELEASE, &n);
501 }
502
503 module_put(nlk->module);
504
505 netlink_table_grab();
506 if (netlink_is_kernel(sk)) {
507 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
508 if (--nl_table[sk->sk_protocol].registered == 0) {
509 kfree(nl_table[sk->sk_protocol].listeners);
510 nl_table[sk->sk_protocol].module = NULL;
511 nl_table[sk->sk_protocol].registered = 0;
512 }
513 } else if (nlk->subscriptions)
514 netlink_update_listeners(sk);
515 netlink_table_ungrab();
516
517 kfree(nlk->groups);
518 nlk->groups = NULL;
519
520 local_bh_disable();
521 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
522 local_bh_enable();
523 sock_put(sk);
524 return 0;
525 }
526
527 static int netlink_autobind(struct socket *sock)
528 {
529 struct sock *sk = sock->sk;
530 struct net *net = sock_net(sk);
531 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
532 struct hlist_head *head;
533 struct sock *osk;
534 struct hlist_node *node;
535 s32 pid = current->tgid;
536 int err;
537 static s32 rover = -4097;
538
539 retry:
540 cond_resched();
541 netlink_table_grab();
542 head = nl_pid_hashfn(hash, pid);
543 sk_for_each(osk, node, head) {
544 if (!net_eq(sock_net(osk), net))
545 continue;
546 if (nlk_sk(osk)->pid == pid) {
547 /* Bind collision, search negative pid values. */
548 pid = rover--;
549 if (rover > -4097)
550 rover = -4097;
551 netlink_table_ungrab();
552 goto retry;
553 }
554 }
555 netlink_table_ungrab();
556
557 err = netlink_insert(sk, net, pid);
558 if (err == -EADDRINUSE)
559 goto retry;
560
561 /* If 2 threads race to autobind, that is fine. */
562 if (err == -EBUSY)
563 err = 0;
564
565 return err;
566 }
567
568 static inline int netlink_capable(struct socket *sock, unsigned int flag)
569 {
570 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
571 capable(CAP_NET_ADMIN);
572 }
573
574 static void
575 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
576 {
577 struct netlink_sock *nlk = nlk_sk(sk);
578
579 if (nlk->subscriptions && !subscriptions)
580 __sk_del_bind_node(sk);
581 else if (!nlk->subscriptions && subscriptions)
582 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
583 nlk->subscriptions = subscriptions;
584 }
585
586 static int netlink_realloc_groups(struct sock *sk)
587 {
588 struct netlink_sock *nlk = nlk_sk(sk);
589 unsigned int groups;
590 unsigned long *new_groups;
591 int err = 0;
592
593 netlink_table_grab();
594
595 groups = nl_table[sk->sk_protocol].groups;
596 if (!nl_table[sk->sk_protocol].registered) {
597 err = -ENOENT;
598 goto out_unlock;
599 }
600
601 if (nlk->ngroups >= groups)
602 goto out_unlock;
603
604 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
605 if (new_groups == NULL) {
606 err = -ENOMEM;
607 goto out_unlock;
608 }
609 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
610 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
611
612 nlk->groups = new_groups;
613 nlk->ngroups = groups;
614 out_unlock:
615 netlink_table_ungrab();
616 return err;
617 }
618
619 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
620 int addr_len)
621 {
622 struct sock *sk = sock->sk;
623 struct net *net = sock_net(sk);
624 struct netlink_sock *nlk = nlk_sk(sk);
625 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
626 int err;
627
628 if (nladdr->nl_family != AF_NETLINK)
629 return -EINVAL;
630
631 /* Only superuser is allowed to listen multicasts */
632 if (nladdr->nl_groups) {
633 if (!netlink_capable(sock, NL_NONROOT_RECV))
634 return -EPERM;
635 err = netlink_realloc_groups(sk);
636 if (err)
637 return err;
638 }
639
640 if (nlk->pid) {
641 if (nladdr->nl_pid != nlk->pid)
642 return -EINVAL;
643 } else {
644 err = nladdr->nl_pid ?
645 netlink_insert(sk, net, nladdr->nl_pid) :
646 netlink_autobind(sock);
647 if (err)
648 return err;
649 }
650
651 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
652 return 0;
653
654 netlink_table_grab();
655 netlink_update_subscriptions(sk, nlk->subscriptions +
656 hweight32(nladdr->nl_groups) -
657 hweight32(nlk->groups[0]));
658 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
659 netlink_update_listeners(sk);
660 netlink_table_ungrab();
661
662 return 0;
663 }
664
665 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
666 int alen, int flags)
667 {
668 int err = 0;
669 struct sock *sk = sock->sk;
670 struct netlink_sock *nlk = nlk_sk(sk);
671 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
672
673 if (addr->sa_family == AF_UNSPEC) {
674 sk->sk_state = NETLINK_UNCONNECTED;
675 nlk->dst_pid = 0;
676 nlk->dst_group = 0;
677 return 0;
678 }
679 if (addr->sa_family != AF_NETLINK)
680 return -EINVAL;
681
682 /* Only superuser is allowed to send multicasts */
683 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
684 return -EPERM;
685
686 if (!nlk->pid)
687 err = netlink_autobind(sock);
688
689 if (err == 0) {
690 sk->sk_state = NETLINK_CONNECTED;
691 nlk->dst_pid = nladdr->nl_pid;
692 nlk->dst_group = ffs(nladdr->nl_groups);
693 }
694
695 return err;
696 }
697
698 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
699 int *addr_len, int peer)
700 {
701 struct sock *sk = sock->sk;
702 struct netlink_sock *nlk = nlk_sk(sk);
703 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
704
705 nladdr->nl_family = AF_NETLINK;
706 nladdr->nl_pad = 0;
707 *addr_len = sizeof(*nladdr);
708
709 if (peer) {
710 nladdr->nl_pid = nlk->dst_pid;
711 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
712 } else {
713 nladdr->nl_pid = nlk->pid;
714 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
715 }
716 return 0;
717 }
718
719 static void netlink_overrun(struct sock *sk)
720 {
721 struct netlink_sock *nlk = nlk_sk(sk);
722
723 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
724 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
725 sk->sk_err = ENOBUFS;
726 sk->sk_error_report(sk);
727 }
728 }
729 atomic_inc(&sk->sk_drops);
730 }
731
732 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
733 {
734 struct sock *sock;
735 struct netlink_sock *nlk;
736
737 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
738 if (!sock)
739 return ERR_PTR(-ECONNREFUSED);
740
741 /* Don't bother queuing skb if kernel socket has no input function */
742 nlk = nlk_sk(sock);
743 if (sock->sk_state == NETLINK_CONNECTED &&
744 nlk->dst_pid != nlk_sk(ssk)->pid) {
745 sock_put(sock);
746 return ERR_PTR(-ECONNREFUSED);
747 }
748 return sock;
749 }
750
751 struct sock *netlink_getsockbyfilp(struct file *filp)
752 {
753 struct inode *inode = filp->f_path.dentry->d_inode;
754 struct sock *sock;
755
756 if (!S_ISSOCK(inode->i_mode))
757 return ERR_PTR(-ENOTSOCK);
758
759 sock = SOCKET_I(inode)->sk;
760 if (sock->sk_family != AF_NETLINK)
761 return ERR_PTR(-EINVAL);
762
763 sock_hold(sock);
764 return sock;
765 }
766
767 /*
768 * Attach a skb to a netlink socket.
769 * The caller must hold a reference to the destination socket. On error, the
770 * reference is dropped. The skb is not send to the destination, just all
771 * all error checks are performed and memory in the queue is reserved.
772 * Return values:
773 * < 0: error. skb freed, reference to sock dropped.
774 * 0: continue
775 * 1: repeat lookup - reference dropped while waiting for socket memory.
776 */
777 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
778 long *timeo, struct sock *ssk)
779 {
780 struct netlink_sock *nlk;
781
782 nlk = nlk_sk(sk);
783
784 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
785 test_bit(0, &nlk->state)) {
786 DECLARE_WAITQUEUE(wait, current);
787 if (!*timeo) {
788 if (!ssk || netlink_is_kernel(ssk))
789 netlink_overrun(sk);
790 sock_put(sk);
791 kfree_skb(skb);
792 return -EAGAIN;
793 }
794
795 __set_current_state(TASK_INTERRUPTIBLE);
796 add_wait_queue(&nlk->wait, &wait);
797
798 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
799 test_bit(0, &nlk->state)) &&
800 !sock_flag(sk, SOCK_DEAD))
801 *timeo = schedule_timeout(*timeo);
802
803 __set_current_state(TASK_RUNNING);
804 remove_wait_queue(&nlk->wait, &wait);
805 sock_put(sk);
806
807 if (signal_pending(current)) {
808 kfree_skb(skb);
809 return sock_intr_errno(*timeo);
810 }
811 return 1;
812 }
813 skb_set_owner_r(skb, sk);
814 return 0;
815 }
816
817 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
818 {
819 int len = skb->len;
820
821 skb_queue_tail(&sk->sk_receive_queue, skb);
822 sk->sk_data_ready(sk, len);
823 sock_put(sk);
824 return len;
825 }
826
827 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
828 {
829 kfree_skb(skb);
830 sock_put(sk);
831 }
832
833 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
834 gfp_t allocation)
835 {
836 int delta;
837
838 skb_orphan(skb);
839
840 delta = skb->end - skb->tail;
841 if (delta * 2 < skb->truesize)
842 return skb;
843
844 if (skb_shared(skb)) {
845 struct sk_buff *nskb = skb_clone(skb, allocation);
846 if (!nskb)
847 return skb;
848 kfree_skb(skb);
849 skb = nskb;
850 }
851
852 if (!pskb_expand_head(skb, 0, -delta, allocation))
853 skb->truesize -= delta;
854
855 return skb;
856 }
857
858 static inline void netlink_rcv_wake(struct sock *sk)
859 {
860 struct netlink_sock *nlk = nlk_sk(sk);
861
862 if (skb_queue_empty(&sk->sk_receive_queue))
863 clear_bit(0, &nlk->state);
864 if (!test_bit(0, &nlk->state))
865 wake_up_interruptible(&nlk->wait);
866 }
867
868 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
869 {
870 int ret;
871 struct netlink_sock *nlk = nlk_sk(sk);
872
873 ret = -ECONNREFUSED;
874 if (nlk->netlink_rcv != NULL) {
875 ret = skb->len;
876 skb_set_owner_r(skb, sk);
877 nlk->netlink_rcv(skb);
878 }
879 kfree_skb(skb);
880 sock_put(sk);
881 return ret;
882 }
883
884 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
885 u32 pid, int nonblock)
886 {
887 struct sock *sk;
888 int err;
889 long timeo;
890
891 skb = netlink_trim(skb, gfp_any());
892
893 timeo = sock_sndtimeo(ssk, nonblock);
894 retry:
895 sk = netlink_getsockbypid(ssk, pid);
896 if (IS_ERR(sk)) {
897 kfree_skb(skb);
898 return PTR_ERR(sk);
899 }
900 if (netlink_is_kernel(sk))
901 return netlink_unicast_kernel(sk, skb);
902
903 if (sk_filter(sk, skb)) {
904 err = skb->len;
905 kfree_skb(skb);
906 sock_put(sk);
907 return err;
908 }
909
910 err = netlink_attachskb(sk, skb, &timeo, ssk);
911 if (err == 1)
912 goto retry;
913 if (err)
914 return err;
915
916 return netlink_sendskb(sk, skb);
917 }
918 EXPORT_SYMBOL(netlink_unicast);
919
920 int netlink_has_listeners(struct sock *sk, unsigned int group)
921 {
922 int res = 0;
923 unsigned long *listeners;
924
925 BUG_ON(!netlink_is_kernel(sk));
926
927 rcu_read_lock();
928 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
929
930 if (group - 1 < nl_table[sk->sk_protocol].groups)
931 res = test_bit(group - 1, listeners);
932
933 rcu_read_unlock();
934
935 return res;
936 }
937 EXPORT_SYMBOL_GPL(netlink_has_listeners);
938
939 static inline int netlink_broadcast_deliver(struct sock *sk,
940 struct sk_buff *skb)
941 {
942 struct netlink_sock *nlk = nlk_sk(sk);
943
944 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
945 !test_bit(0, &nlk->state)) {
946 skb_set_owner_r(skb, sk);
947 skb_queue_tail(&sk->sk_receive_queue, skb);
948 sk->sk_data_ready(sk, skb->len);
949 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
950 }
951 return -1;
952 }
953
954 struct netlink_broadcast_data {
955 struct sock *exclude_sk;
956 struct net *net;
957 u32 pid;
958 u32 group;
959 int failure;
960 int delivery_failure;
961 int congested;
962 int delivered;
963 gfp_t allocation;
964 struct sk_buff *skb, *skb2;
965 };
966
967 static inline int do_one_broadcast(struct sock *sk,
968 struct netlink_broadcast_data *p)
969 {
970 struct netlink_sock *nlk = nlk_sk(sk);
971 int val;
972
973 if (p->exclude_sk == sk)
974 goto out;
975
976 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
977 !test_bit(p->group - 1, nlk->groups))
978 goto out;
979
980 if (!net_eq(sock_net(sk), p->net))
981 goto out;
982
983 if (p->failure) {
984 netlink_overrun(sk);
985 goto out;
986 }
987
988 sock_hold(sk);
989 if (p->skb2 == NULL) {
990 if (skb_shared(p->skb)) {
991 p->skb2 = skb_clone(p->skb, p->allocation);
992 } else {
993 p->skb2 = skb_get(p->skb);
994 /*
995 * skb ownership may have been set when
996 * delivered to a previous socket.
997 */
998 skb_orphan(p->skb2);
999 }
1000 }
1001 if (p->skb2 == NULL) {
1002 netlink_overrun(sk);
1003 /* Clone failed. Notify ALL listeners. */
1004 p->failure = 1;
1005 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1006 p->delivery_failure = 1;
1007 } else if (sk_filter(sk, p->skb2)) {
1008 kfree_skb(p->skb2);
1009 p->skb2 = NULL;
1010 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1011 netlink_overrun(sk);
1012 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1013 p->delivery_failure = 1;
1014 } else {
1015 p->congested |= val;
1016 p->delivered = 1;
1017 p->skb2 = NULL;
1018 }
1019 sock_put(sk);
1020
1021 out:
1022 return 0;
1023 }
1024
1025 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1026 u32 group, gfp_t allocation)
1027 {
1028 struct net *net = sock_net(ssk);
1029 struct netlink_broadcast_data info;
1030 struct hlist_node *node;
1031 struct sock *sk;
1032
1033 skb = netlink_trim(skb, allocation);
1034
1035 info.exclude_sk = ssk;
1036 info.net = net;
1037 info.pid = pid;
1038 info.group = group;
1039 info.failure = 0;
1040 info.delivery_failure = 0;
1041 info.congested = 0;
1042 info.delivered = 0;
1043 info.allocation = allocation;
1044 info.skb = skb;
1045 info.skb2 = NULL;
1046
1047 /* While we sleep in clone, do not allow to change socket list */
1048
1049 netlink_lock_table();
1050
1051 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1052 do_one_broadcast(sk, &info);
1053
1054 kfree_skb(skb);
1055
1056 netlink_unlock_table();
1057
1058 kfree_skb(info.skb2);
1059
1060 if (info.delivery_failure)
1061 return -ENOBUFS;
1062
1063 if (info.delivered) {
1064 if (info.congested && (allocation & __GFP_WAIT))
1065 yield();
1066 return 0;
1067 }
1068 return -ESRCH;
1069 }
1070 EXPORT_SYMBOL(netlink_broadcast);
1071
1072 struct netlink_set_err_data {
1073 struct sock *exclude_sk;
1074 u32 pid;
1075 u32 group;
1076 int code;
1077 };
1078
1079 static inline int do_one_set_err(struct sock *sk,
1080 struct netlink_set_err_data *p)
1081 {
1082 struct netlink_sock *nlk = nlk_sk(sk);
1083
1084 if (sk == p->exclude_sk)
1085 goto out;
1086
1087 if (sock_net(sk) != sock_net(p->exclude_sk))
1088 goto out;
1089
1090 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1091 !test_bit(p->group - 1, nlk->groups))
1092 goto out;
1093
1094 sk->sk_err = p->code;
1095 sk->sk_error_report(sk);
1096 out:
1097 return 0;
1098 }
1099
1100 /**
1101 * netlink_set_err - report error to broadcast listeners
1102 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1103 * @pid: the PID of a process that we want to skip (if any)
1104 * @groups: the broadcast group that will notice the error
1105 * @code: error code, must be negative (as usual in kernelspace)
1106 */
1107 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1108 {
1109 struct netlink_set_err_data info;
1110 struct hlist_node *node;
1111 struct sock *sk;
1112
1113 info.exclude_sk = ssk;
1114 info.pid = pid;
1115 info.group = group;
1116 /* sk->sk_err wants a positive error value */
1117 info.code = -code;
1118
1119 read_lock(&nl_table_lock);
1120
1121 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1122 do_one_set_err(sk, &info);
1123
1124 read_unlock(&nl_table_lock);
1125 }
1126 EXPORT_SYMBOL(netlink_set_err);
1127
1128 /* must be called with netlink table grabbed */
1129 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1130 unsigned int group,
1131 int is_new)
1132 {
1133 int old, new = !!is_new, subscriptions;
1134
1135 old = test_bit(group - 1, nlk->groups);
1136 subscriptions = nlk->subscriptions - old + new;
1137 if (new)
1138 __set_bit(group - 1, nlk->groups);
1139 else
1140 __clear_bit(group - 1, nlk->groups);
1141 netlink_update_subscriptions(&nlk->sk, subscriptions);
1142 netlink_update_listeners(&nlk->sk);
1143 }
1144
1145 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1146 char __user *optval, int optlen)
1147 {
1148 struct sock *sk = sock->sk;
1149 struct netlink_sock *nlk = nlk_sk(sk);
1150 unsigned int val = 0;
1151 int err;
1152
1153 if (level != SOL_NETLINK)
1154 return -ENOPROTOOPT;
1155
1156 if (optlen >= sizeof(int) &&
1157 get_user(val, (unsigned int __user *)optval))
1158 return -EFAULT;
1159
1160 switch (optname) {
1161 case NETLINK_PKTINFO:
1162 if (val)
1163 nlk->flags |= NETLINK_RECV_PKTINFO;
1164 else
1165 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1166 err = 0;
1167 break;
1168 case NETLINK_ADD_MEMBERSHIP:
1169 case NETLINK_DROP_MEMBERSHIP: {
1170 if (!netlink_capable(sock, NL_NONROOT_RECV))
1171 return -EPERM;
1172 err = netlink_realloc_groups(sk);
1173 if (err)
1174 return err;
1175 if (!val || val - 1 >= nlk->ngroups)
1176 return -EINVAL;
1177 netlink_table_grab();
1178 netlink_update_socket_mc(nlk, val,
1179 optname == NETLINK_ADD_MEMBERSHIP);
1180 netlink_table_ungrab();
1181 err = 0;
1182 break;
1183 }
1184 case NETLINK_BROADCAST_ERROR:
1185 if (val)
1186 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1187 else
1188 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1189 err = 0;
1190 break;
1191 case NETLINK_NO_ENOBUFS:
1192 if (val) {
1193 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1194 clear_bit(0, &nlk->state);
1195 wake_up_interruptible(&nlk->wait);
1196 } else
1197 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1198 err = 0;
1199 break;
1200 default:
1201 err = -ENOPROTOOPT;
1202 }
1203 return err;
1204 }
1205
1206 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1207 char __user *optval, int __user *optlen)
1208 {
1209 struct sock *sk = sock->sk;
1210 struct netlink_sock *nlk = nlk_sk(sk);
1211 int len, val, err;
1212
1213 if (level != SOL_NETLINK)
1214 return -ENOPROTOOPT;
1215
1216 if (get_user(len, optlen))
1217 return -EFAULT;
1218 if (len < 0)
1219 return -EINVAL;
1220
1221 switch (optname) {
1222 case NETLINK_PKTINFO:
1223 if (len < sizeof(int))
1224 return -EINVAL;
1225 len = sizeof(int);
1226 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1227 if (put_user(len, optlen) ||
1228 put_user(val, optval))
1229 return -EFAULT;
1230 err = 0;
1231 break;
1232 case NETLINK_BROADCAST_ERROR:
1233 if (len < sizeof(int))
1234 return -EINVAL;
1235 len = sizeof(int);
1236 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1237 if (put_user(len, optlen) ||
1238 put_user(val, optval))
1239 return -EFAULT;
1240 err = 0;
1241 break;
1242 case NETLINK_NO_ENOBUFS:
1243 if (len < sizeof(int))
1244 return -EINVAL;
1245 len = sizeof(int);
1246 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1247 if (put_user(len, optlen) ||
1248 put_user(val, optval))
1249 return -EFAULT;
1250 err = 0;
1251 break;
1252 default:
1253 err = -ENOPROTOOPT;
1254 }
1255 return err;
1256 }
1257
1258 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1259 {
1260 struct nl_pktinfo info;
1261
1262 info.group = NETLINK_CB(skb).dst_group;
1263 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1264 }
1265
1266 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1267 struct msghdr *msg, size_t len)
1268 {
1269 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1270 struct sock *sk = sock->sk;
1271 struct netlink_sock *nlk = nlk_sk(sk);
1272 struct sockaddr_nl *addr = msg->msg_name;
1273 u32 dst_pid;
1274 u32 dst_group;
1275 struct sk_buff *skb;
1276 int err;
1277 struct scm_cookie scm;
1278
1279 if (msg->msg_flags&MSG_OOB)
1280 return -EOPNOTSUPP;
1281
1282 if (NULL == siocb->scm)
1283 siocb->scm = &scm;
1284 err = scm_send(sock, msg, siocb->scm);
1285 if (err < 0)
1286 return err;
1287
1288 if (msg->msg_namelen) {
1289 if (addr->nl_family != AF_NETLINK)
1290 return -EINVAL;
1291 dst_pid = addr->nl_pid;
1292 dst_group = ffs(addr->nl_groups);
1293 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1294 return -EPERM;
1295 } else {
1296 dst_pid = nlk->dst_pid;
1297 dst_group = nlk->dst_group;
1298 }
1299
1300 if (!nlk->pid) {
1301 err = netlink_autobind(sock);
1302 if (err)
1303 goto out;
1304 }
1305
1306 err = -EMSGSIZE;
1307 if (len > sk->sk_sndbuf - 32)
1308 goto out;
1309 err = -ENOBUFS;
1310 skb = alloc_skb(len, GFP_KERNEL);
1311 if (skb == NULL)
1312 goto out;
1313
1314 NETLINK_CB(skb).pid = nlk->pid;
1315 NETLINK_CB(skb).dst_group = dst_group;
1316 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1317 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1318 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1319 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1320
1321 /* What can I do? Netlink is asynchronous, so that
1322 we will have to save current capabilities to
1323 check them, when this message will be delivered
1324 to corresponding kernel module. --ANK (980802)
1325 */
1326
1327 err = -EFAULT;
1328 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1329 kfree_skb(skb);
1330 goto out;
1331 }
1332
1333 err = security_netlink_send(sk, skb);
1334 if (err) {
1335 kfree_skb(skb);
1336 goto out;
1337 }
1338
1339 if (dst_group) {
1340 atomic_inc(&skb->users);
1341 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1342 }
1343 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1344
1345 out:
1346 return err;
1347 }
1348
1349 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1350 struct msghdr *msg, size_t len,
1351 int flags)
1352 {
1353 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1354 struct scm_cookie scm;
1355 struct sock *sk = sock->sk;
1356 struct netlink_sock *nlk = nlk_sk(sk);
1357 int noblock = flags&MSG_DONTWAIT;
1358 size_t copied;
1359 struct sk_buff *skb;
1360 int err;
1361
1362 if (flags&MSG_OOB)
1363 return -EOPNOTSUPP;
1364
1365 copied = 0;
1366
1367 skb = skb_recv_datagram(sk, flags, noblock, &err);
1368 if (skb == NULL)
1369 goto out;
1370
1371 msg->msg_namelen = 0;
1372
1373 copied = skb->len;
1374 if (len < copied) {
1375 msg->msg_flags |= MSG_TRUNC;
1376 copied = len;
1377 }
1378
1379 skb_reset_transport_header(skb);
1380 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1381
1382 if (msg->msg_name) {
1383 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1384 addr->nl_family = AF_NETLINK;
1385 addr->nl_pad = 0;
1386 addr->nl_pid = NETLINK_CB(skb).pid;
1387 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1388 msg->msg_namelen = sizeof(*addr);
1389 }
1390
1391 if (nlk->flags & NETLINK_RECV_PKTINFO)
1392 netlink_cmsg_recv_pktinfo(msg, skb);
1393
1394 if (NULL == siocb->scm) {
1395 memset(&scm, 0, sizeof(scm));
1396 siocb->scm = &scm;
1397 }
1398 siocb->scm->creds = *NETLINK_CREDS(skb);
1399 if (flags & MSG_TRUNC)
1400 copied = skb->len;
1401 skb_free_datagram(sk, skb);
1402
1403 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1404 netlink_dump(sk);
1405
1406 scm_recv(sock, msg, siocb->scm, flags);
1407 out:
1408 netlink_rcv_wake(sk);
1409 return err ? : copied;
1410 }
1411
1412 static void netlink_data_ready(struct sock *sk, int len)
1413 {
1414 BUG();
1415 }
1416
1417 /*
1418 * We export these functions to other modules. They provide a
1419 * complete set of kernel non-blocking support for message
1420 * queueing.
1421 */
1422
1423 struct sock *
1424 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1425 void (*input)(struct sk_buff *skb),
1426 struct mutex *cb_mutex, struct module *module)
1427 {
1428 struct socket *sock;
1429 struct sock *sk;
1430 struct netlink_sock *nlk;
1431 unsigned long *listeners = NULL;
1432
1433 BUG_ON(!nl_table);
1434
1435 if (unit < 0 || unit >= MAX_LINKS)
1436 return NULL;
1437
1438 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1439 return NULL;
1440
1441 /*
1442 * We have to just have a reference on the net from sk, but don't
1443 * get_net it. Besides, we cannot get and then put the net here.
1444 * So we create one inside init_net and the move it to net.
1445 */
1446
1447 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1448 goto out_sock_release_nosk;
1449
1450 sk = sock->sk;
1451 sk_change_net(sk, net);
1452
1453 if (groups < 32)
1454 groups = 32;
1455
1456 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1457 if (!listeners)
1458 goto out_sock_release;
1459
1460 sk->sk_data_ready = netlink_data_ready;
1461 if (input)
1462 nlk_sk(sk)->netlink_rcv = input;
1463
1464 if (netlink_insert(sk, net, 0))
1465 goto out_sock_release;
1466
1467 nlk = nlk_sk(sk);
1468 nlk->flags |= NETLINK_KERNEL_SOCKET;
1469
1470 netlink_table_grab();
1471 if (!nl_table[unit].registered) {
1472 nl_table[unit].groups = groups;
1473 nl_table[unit].listeners = listeners;
1474 nl_table[unit].cb_mutex = cb_mutex;
1475 nl_table[unit].module = module;
1476 nl_table[unit].registered = 1;
1477 } else {
1478 kfree(listeners);
1479 nl_table[unit].registered++;
1480 }
1481 netlink_table_ungrab();
1482 return sk;
1483
1484 out_sock_release:
1485 kfree(listeners);
1486 netlink_kernel_release(sk);
1487 return NULL;
1488
1489 out_sock_release_nosk:
1490 sock_release(sock);
1491 return NULL;
1492 }
1493 EXPORT_SYMBOL(netlink_kernel_create);
1494
1495
1496 void
1497 netlink_kernel_release(struct sock *sk)
1498 {
1499 sk_release_kernel(sk);
1500 }
1501 EXPORT_SYMBOL(netlink_kernel_release);
1502
1503
1504 /**
1505 * netlink_change_ngroups - change number of multicast groups
1506 *
1507 * This changes the number of multicast groups that are available
1508 * on a certain netlink family. Note that it is not possible to
1509 * change the number of groups to below 32. Also note that it does
1510 * not implicitly call netlink_clear_multicast_users() when the
1511 * number of groups is reduced.
1512 *
1513 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1514 * @groups: The new number of groups.
1515 */
1516 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1517 {
1518 unsigned long *listeners, *old = NULL;
1519 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1520 int err = 0;
1521
1522 if (groups < 32)
1523 groups = 32;
1524
1525 netlink_table_grab();
1526 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1527 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1528 if (!listeners) {
1529 err = -ENOMEM;
1530 goto out_ungrab;
1531 }
1532 old = tbl->listeners;
1533 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1534 rcu_assign_pointer(tbl->listeners, listeners);
1535 }
1536 tbl->groups = groups;
1537
1538 out_ungrab:
1539 netlink_table_ungrab();
1540 synchronize_rcu();
1541 kfree(old);
1542 return err;
1543 }
1544 EXPORT_SYMBOL(netlink_change_ngroups);
1545
1546 /**
1547 * netlink_clear_multicast_users - kick off multicast listeners
1548 *
1549 * This function removes all listeners from the given group.
1550 * @ksk: The kernel netlink socket, as returned by
1551 * netlink_kernel_create().
1552 * @group: The multicast group to clear.
1553 */
1554 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1555 {
1556 struct sock *sk;
1557 struct hlist_node *node;
1558 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1559
1560 netlink_table_grab();
1561
1562 sk_for_each_bound(sk, node, &tbl->mc_list)
1563 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1564
1565 netlink_table_ungrab();
1566 }
1567 EXPORT_SYMBOL(netlink_clear_multicast_users);
1568
1569 void netlink_set_nonroot(int protocol, unsigned int flags)
1570 {
1571 if ((unsigned int)protocol < MAX_LINKS)
1572 nl_table[protocol].nl_nonroot = flags;
1573 }
1574 EXPORT_SYMBOL(netlink_set_nonroot);
1575
1576 static void netlink_destroy_callback(struct netlink_callback *cb)
1577 {
1578 kfree_skb(cb->skb);
1579 kfree(cb);
1580 }
1581
1582 /*
1583 * It looks a bit ugly.
1584 * It would be better to create kernel thread.
1585 */
1586
1587 static int netlink_dump(struct sock *sk)
1588 {
1589 struct netlink_sock *nlk = nlk_sk(sk);
1590 struct netlink_callback *cb;
1591 struct sk_buff *skb;
1592 struct nlmsghdr *nlh;
1593 int len, err = -ENOBUFS;
1594
1595 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1596 if (!skb)
1597 goto errout;
1598
1599 mutex_lock(nlk->cb_mutex);
1600
1601 cb = nlk->cb;
1602 if (cb == NULL) {
1603 err = -EINVAL;
1604 goto errout_skb;
1605 }
1606
1607 len = cb->dump(skb, cb);
1608
1609 if (len > 0) {
1610 mutex_unlock(nlk->cb_mutex);
1611
1612 if (sk_filter(sk, skb))
1613 kfree_skb(skb);
1614 else {
1615 skb_queue_tail(&sk->sk_receive_queue, skb);
1616 sk->sk_data_ready(sk, skb->len);
1617 }
1618 return 0;
1619 }
1620
1621 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1622 if (!nlh)
1623 goto errout_skb;
1624
1625 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1626
1627 if (sk_filter(sk, skb))
1628 kfree_skb(skb);
1629 else {
1630 skb_queue_tail(&sk->sk_receive_queue, skb);
1631 sk->sk_data_ready(sk, skb->len);
1632 }
1633
1634 if (cb->done)
1635 cb->done(cb);
1636 nlk->cb = NULL;
1637 mutex_unlock(nlk->cb_mutex);
1638
1639 netlink_destroy_callback(cb);
1640 return 0;
1641
1642 errout_skb:
1643 mutex_unlock(nlk->cb_mutex);
1644 kfree_skb(skb);
1645 errout:
1646 return err;
1647 }
1648
1649 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1650 struct nlmsghdr *nlh,
1651 int (*dump)(struct sk_buff *skb,
1652 struct netlink_callback *),
1653 int (*done)(struct netlink_callback *))
1654 {
1655 struct netlink_callback *cb;
1656 struct sock *sk;
1657 struct netlink_sock *nlk;
1658
1659 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1660 if (cb == NULL)
1661 return -ENOBUFS;
1662
1663 cb->dump = dump;
1664 cb->done = done;
1665 cb->nlh = nlh;
1666 atomic_inc(&skb->users);
1667 cb->skb = skb;
1668
1669 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1670 if (sk == NULL) {
1671 netlink_destroy_callback(cb);
1672 return -ECONNREFUSED;
1673 }
1674 nlk = nlk_sk(sk);
1675 /* A dump is in progress... */
1676 mutex_lock(nlk->cb_mutex);
1677 if (nlk->cb) {
1678 mutex_unlock(nlk->cb_mutex);
1679 netlink_destroy_callback(cb);
1680 sock_put(sk);
1681 return -EBUSY;
1682 }
1683 nlk->cb = cb;
1684 mutex_unlock(nlk->cb_mutex);
1685
1686 netlink_dump(sk);
1687 sock_put(sk);
1688
1689 /* We successfully started a dump, by returning -EINTR we
1690 * signal not to send ACK even if it was requested.
1691 */
1692 return -EINTR;
1693 }
1694 EXPORT_SYMBOL(netlink_dump_start);
1695
1696 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1697 {
1698 struct sk_buff *skb;
1699 struct nlmsghdr *rep;
1700 struct nlmsgerr *errmsg;
1701 size_t payload = sizeof(*errmsg);
1702
1703 /* error messages get the original request appened */
1704 if (err)
1705 payload += nlmsg_len(nlh);
1706
1707 skb = nlmsg_new(payload, GFP_KERNEL);
1708 if (!skb) {
1709 struct sock *sk;
1710
1711 sk = netlink_lookup(sock_net(in_skb->sk),
1712 in_skb->sk->sk_protocol,
1713 NETLINK_CB(in_skb).pid);
1714 if (sk) {
1715 sk->sk_err = ENOBUFS;
1716 sk->sk_error_report(sk);
1717 sock_put(sk);
1718 }
1719 return;
1720 }
1721
1722 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1723 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1724 errmsg = nlmsg_data(rep);
1725 errmsg->error = err;
1726 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1727 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1728 }
1729 EXPORT_SYMBOL(netlink_ack);
1730
1731 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1732 struct nlmsghdr *))
1733 {
1734 struct nlmsghdr *nlh;
1735 int err;
1736
1737 while (skb->len >= nlmsg_total_size(0)) {
1738 int msglen;
1739
1740 nlh = nlmsg_hdr(skb);
1741 err = 0;
1742
1743 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1744 return 0;
1745
1746 /* Only requests are handled by the kernel */
1747 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1748 goto ack;
1749
1750 /* Skip control messages */
1751 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1752 goto ack;
1753
1754 err = cb(skb, nlh);
1755 if (err == -EINTR)
1756 goto skip;
1757
1758 ack:
1759 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1760 netlink_ack(skb, nlh, err);
1761
1762 skip:
1763 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1764 if (msglen > skb->len)
1765 msglen = skb->len;
1766 skb_pull(skb, msglen);
1767 }
1768
1769 return 0;
1770 }
1771 EXPORT_SYMBOL(netlink_rcv_skb);
1772
1773 /**
1774 * nlmsg_notify - send a notification netlink message
1775 * @sk: netlink socket to use
1776 * @skb: notification message
1777 * @pid: destination netlink pid for reports or 0
1778 * @group: destination multicast group or 0
1779 * @report: 1 to report back, 0 to disable
1780 * @flags: allocation flags
1781 */
1782 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1783 unsigned int group, int report, gfp_t flags)
1784 {
1785 int err = 0;
1786
1787 if (group) {
1788 int exclude_pid = 0;
1789
1790 if (report) {
1791 atomic_inc(&skb->users);
1792 exclude_pid = pid;
1793 }
1794
1795 /* errors reported via destination sk->sk_err, but propagate
1796 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1797 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1798 }
1799
1800 if (report) {
1801 int err2;
1802
1803 err2 = nlmsg_unicast(sk, skb, pid);
1804 if (!err || err == -ESRCH)
1805 err = err2;
1806 }
1807
1808 return err;
1809 }
1810 EXPORT_SYMBOL(nlmsg_notify);
1811
1812 #ifdef CONFIG_PROC_FS
1813 struct nl_seq_iter {
1814 struct seq_net_private p;
1815 int link;
1816 int hash_idx;
1817 };
1818
1819 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1820 {
1821 struct nl_seq_iter *iter = seq->private;
1822 int i, j;
1823 struct sock *s;
1824 struct hlist_node *node;
1825 loff_t off = 0;
1826
1827 for (i = 0; i < MAX_LINKS; i++) {
1828 struct nl_pid_hash *hash = &nl_table[i].hash;
1829
1830 for (j = 0; j <= hash->mask; j++) {
1831 sk_for_each(s, node, &hash->table[j]) {
1832 if (sock_net(s) != seq_file_net(seq))
1833 continue;
1834 if (off == pos) {
1835 iter->link = i;
1836 iter->hash_idx = j;
1837 return s;
1838 }
1839 ++off;
1840 }
1841 }
1842 }
1843 return NULL;
1844 }
1845
1846 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1847 __acquires(nl_table_lock)
1848 {
1849 read_lock(&nl_table_lock);
1850 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1851 }
1852
1853 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1854 {
1855 struct sock *s;
1856 struct nl_seq_iter *iter;
1857 int i, j;
1858
1859 ++*pos;
1860
1861 if (v == SEQ_START_TOKEN)
1862 return netlink_seq_socket_idx(seq, 0);
1863
1864 iter = seq->private;
1865 s = v;
1866 do {
1867 s = sk_next(s);
1868 } while (s && sock_net(s) != seq_file_net(seq));
1869 if (s)
1870 return s;
1871
1872 i = iter->link;
1873 j = iter->hash_idx + 1;
1874
1875 do {
1876 struct nl_pid_hash *hash = &nl_table[i].hash;
1877
1878 for (; j <= hash->mask; j++) {
1879 s = sk_head(&hash->table[j]);
1880 while (s && sock_net(s) != seq_file_net(seq))
1881 s = sk_next(s);
1882 if (s) {
1883 iter->link = i;
1884 iter->hash_idx = j;
1885 return s;
1886 }
1887 }
1888
1889 j = 0;
1890 } while (++i < MAX_LINKS);
1891
1892 return NULL;
1893 }
1894
1895 static void netlink_seq_stop(struct seq_file *seq, void *v)
1896 __releases(nl_table_lock)
1897 {
1898 read_unlock(&nl_table_lock);
1899 }
1900
1901
1902 static int netlink_seq_show(struct seq_file *seq, void *v)
1903 {
1904 if (v == SEQ_START_TOKEN)
1905 seq_puts(seq,
1906 "sk Eth Pid Groups "
1907 "Rmem Wmem Dump Locks Drops\n");
1908 else {
1909 struct sock *s = v;
1910 struct netlink_sock *nlk = nlk_sk(s);
1911
1912 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %-8d %-8d\n",
1913 s,
1914 s->sk_protocol,
1915 nlk->pid,
1916 nlk->groups ? (u32)nlk->groups[0] : 0,
1917 atomic_read(&s->sk_rmem_alloc),
1918 atomic_read(&s->sk_wmem_alloc),
1919 nlk->cb,
1920 atomic_read(&s->sk_refcnt),
1921 atomic_read(&s->sk_drops)
1922 );
1923
1924 }
1925 return 0;
1926 }
1927
1928 static const struct seq_operations netlink_seq_ops = {
1929 .start = netlink_seq_start,
1930 .next = netlink_seq_next,
1931 .stop = netlink_seq_stop,
1932 .show = netlink_seq_show,
1933 };
1934
1935
1936 static int netlink_seq_open(struct inode *inode, struct file *file)
1937 {
1938 return seq_open_net(inode, file, &netlink_seq_ops,
1939 sizeof(struct nl_seq_iter));
1940 }
1941
1942 static const struct file_operations netlink_seq_fops = {
1943 .owner = THIS_MODULE,
1944 .open = netlink_seq_open,
1945 .read = seq_read,
1946 .llseek = seq_lseek,
1947 .release = seq_release_net,
1948 };
1949
1950 #endif
1951
1952 int netlink_register_notifier(struct notifier_block *nb)
1953 {
1954 return atomic_notifier_chain_register(&netlink_chain, nb);
1955 }
1956 EXPORT_SYMBOL(netlink_register_notifier);
1957
1958 int netlink_unregister_notifier(struct notifier_block *nb)
1959 {
1960 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1961 }
1962 EXPORT_SYMBOL(netlink_unregister_notifier);
1963
1964 static const struct proto_ops netlink_ops = {
1965 .family = PF_NETLINK,
1966 .owner = THIS_MODULE,
1967 .release = netlink_release,
1968 .bind = netlink_bind,
1969 .connect = netlink_connect,
1970 .socketpair = sock_no_socketpair,
1971 .accept = sock_no_accept,
1972 .getname = netlink_getname,
1973 .poll = datagram_poll,
1974 .ioctl = sock_no_ioctl,
1975 .listen = sock_no_listen,
1976 .shutdown = sock_no_shutdown,
1977 .setsockopt = netlink_setsockopt,
1978 .getsockopt = netlink_getsockopt,
1979 .sendmsg = netlink_sendmsg,
1980 .recvmsg = netlink_recvmsg,
1981 .mmap = sock_no_mmap,
1982 .sendpage = sock_no_sendpage,
1983 };
1984
1985 static struct net_proto_family netlink_family_ops = {
1986 .family = PF_NETLINK,
1987 .create = netlink_create,
1988 .owner = THIS_MODULE, /* for consistency 8) */
1989 };
1990
1991 static int __net_init netlink_net_init(struct net *net)
1992 {
1993 #ifdef CONFIG_PROC_FS
1994 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1995 return -ENOMEM;
1996 #endif
1997 return 0;
1998 }
1999
2000 static void __net_exit netlink_net_exit(struct net *net)
2001 {
2002 #ifdef CONFIG_PROC_FS
2003 proc_net_remove(net, "netlink");
2004 #endif
2005 }
2006
2007 static struct pernet_operations __net_initdata netlink_net_ops = {
2008 .init = netlink_net_init,
2009 .exit = netlink_net_exit,
2010 };
2011
2012 static int __init netlink_proto_init(void)
2013 {
2014 struct sk_buff *dummy_skb;
2015 int i;
2016 unsigned long limit;
2017 unsigned int order;
2018 int err = proto_register(&netlink_proto, 0);
2019
2020 if (err != 0)
2021 goto out;
2022
2023 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2024
2025 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2026 if (!nl_table)
2027 goto panic;
2028
2029 if (num_physpages >= (128 * 1024))
2030 limit = num_physpages >> (21 - PAGE_SHIFT);
2031 else
2032 limit = num_physpages >> (23 - PAGE_SHIFT);
2033
2034 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2035 limit = (1UL << order) / sizeof(struct hlist_head);
2036 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2037
2038 for (i = 0; i < MAX_LINKS; i++) {
2039 struct nl_pid_hash *hash = &nl_table[i].hash;
2040
2041 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2042 if (!hash->table) {
2043 while (i-- > 0)
2044 nl_pid_hash_free(nl_table[i].hash.table,
2045 1 * sizeof(*hash->table));
2046 kfree(nl_table);
2047 goto panic;
2048 }
2049 hash->max_shift = order;
2050 hash->shift = 0;
2051 hash->mask = 0;
2052 hash->rehash_time = jiffies;
2053 }
2054
2055 sock_register(&netlink_family_ops);
2056 register_pernet_subsys(&netlink_net_ops);
2057 /* The netlink device handler may be needed early. */
2058 rtnetlink_init();
2059 out:
2060 return err;
2061 panic:
2062 panic("netlink_init: Cannot allocate nl_table\n");
2063 }
2064
2065 core_initcall(netlink_proto_init);
Linus' kernel tree