?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
netlink: add NETLINK_BROADCAST_ERROR socket option
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / netlink / af_netlink.c
blobed587be1e1c2b972770f37a167570b4d88b50861
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
90 static inline struct netlink_sock *nlk_sk(struct sock *sk)
91 {
92 return container_of(sk, struct netlink_sock, sk);
93 }
94
95 static inline int netlink_is_kernel(struct sock *sk)
96 {
97 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
98 }
99
100 struct nl_pid_hash {
101 struct hlist_head *table;
102 unsigned long rehash_time;
103
104 unsigned int mask;
105 unsigned int shift;
106
107 unsigned int entries;
108 unsigned int max_shift;
109
110 u32 rnd;
111 };
112
113 struct netlink_table {
114 struct nl_pid_hash hash;
115 struct hlist_head mc_list;
116 unsigned long *listeners;
117 unsigned int nl_nonroot;
118 unsigned int groups;
119 struct mutex *cb_mutex;
120 struct module *module;
121 int registered;
122 };
123
124 static struct netlink_table *nl_table;
125
126 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
127
128 static int netlink_dump(struct sock *sk);
129 static void netlink_destroy_callback(struct netlink_callback *cb);
130
131 static DEFINE_RWLOCK(nl_table_lock);
132 static atomic_t nl_table_users = ATOMIC_INIT(0);
133
134 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
135
136 static u32 netlink_group_mask(u32 group)
137 {
138 return group ? 1 << (group - 1) : 0;
139 }
140
141 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
142 {
143 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
144 }
145
146 static void netlink_sock_destruct(struct sock *sk)
147 {
148 struct netlink_sock *nlk = nlk_sk(sk);
149
150 if (nlk->cb) {
151 if (nlk->cb->done)
152 nlk->cb->done(nlk->cb);
153 netlink_destroy_callback(nlk->cb);
154 }
155
156 skb_queue_purge(&sk->sk_receive_queue);
157
158 if (!sock_flag(sk, SOCK_DEAD)) {
159 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
160 return;
161 }
162
163 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
164 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
165 WARN_ON(nlk_sk(sk)->groups);
166 }
167
168 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
169 * SMP. Look, when several writers sleep and reader wakes them up, all but one
170 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
171 * this, _but_ remember, it adds useless work on UP machines.
172 */
173
174 static void netlink_table_grab(void)
175 __acquires(nl_table_lock)
176 {
177 write_lock_irq(&nl_table_lock);
178
179 if (atomic_read(&nl_table_users)) {
180 DECLARE_WAITQUEUE(wait, current);
181
182 add_wait_queue_exclusive(&nl_table_wait, &wait);
183 for (;;) {
184 set_current_state(TASK_UNINTERRUPTIBLE);
185 if (atomic_read(&nl_table_users) == 0)
186 break;
187 write_unlock_irq(&nl_table_lock);
188 schedule();
189 write_lock_irq(&nl_table_lock);
190 }
191
192 __set_current_state(TASK_RUNNING);
193 remove_wait_queue(&nl_table_wait, &wait);
194 }
195 }
196
197 static void netlink_table_ungrab(void)
198 __releases(nl_table_lock)
199 {
200 write_unlock_irq(&nl_table_lock);
201 wake_up(&nl_table_wait);
202 }
203
204 static inline void
205 netlink_lock_table(void)
206 {
207 /* read_lock() synchronizes us to netlink_table_grab */
208
209 read_lock(&nl_table_lock);
210 atomic_inc(&nl_table_users);
211 read_unlock(&nl_table_lock);
212 }
213
214 static inline void
215 netlink_unlock_table(void)
216 {
217 if (atomic_dec_and_test(&nl_table_users))
218 wake_up(&nl_table_wait);
219 }
220
221 static inline struct sock *netlink_lookup(struct net *net, int protocol,
222 u32 pid)
223 {
224 struct nl_pid_hash *hash = &nl_table[protocol].hash;
225 struct hlist_head *head;
226 struct sock *sk;
227 struct hlist_node *node;
228
229 read_lock(&nl_table_lock);
230 head = nl_pid_hashfn(hash, pid);
231 sk_for_each(sk, node, head) {
232 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
233 sock_hold(sk);
234 goto found;
235 }
236 }
237 sk = NULL;
238 found:
239 read_unlock(&nl_table_lock);
240 return sk;
241 }
242
243 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
244 {
245 if (size <= PAGE_SIZE)
246 return kzalloc(size, GFP_ATOMIC);
247 else
248 return (struct hlist_head *)
249 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
250 get_order(size));
251 }
252
253 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
254 {
255 if (size <= PAGE_SIZE)
256 kfree(table);
257 else
258 free_pages((unsigned long)table, get_order(size));
259 }
260
261 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
262 {
263 unsigned int omask, mask, shift;
264 size_t osize, size;
265 struct hlist_head *otable, *table;
266 int i;
267
268 omask = mask = hash->mask;
269 osize = size = (mask + 1) * sizeof(*table);
270 shift = hash->shift;
271
272 if (grow) {
273 if (++shift > hash->max_shift)
274 return 0;
275 mask = mask * 2 + 1;
276 size *= 2;
277 }
278
279 table = nl_pid_hash_zalloc(size);
280 if (!table)
281 return 0;
282
283 otable = hash->table;
284 hash->table = table;
285 hash->mask = mask;
286 hash->shift = shift;
287 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
288
289 for (i = 0; i <= omask; i++) {
290 struct sock *sk;
291 struct hlist_node *node, *tmp;
292
293 sk_for_each_safe(sk, node, tmp, &otable[i])
294 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
295 }
296
297 nl_pid_hash_free(otable, osize);
298 hash->rehash_time = jiffies + 10 * 60 * HZ;
299 return 1;
300 }
301
302 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
303 {
304 int avg = hash->entries >> hash->shift;
305
306 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
307 return 1;
308
309 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
310 nl_pid_hash_rehash(hash, 0);
311 return 1;
312 }
313
314 return 0;
315 }
316
317 static const struct proto_ops netlink_ops;
318
319 static void
320 netlink_update_listeners(struct sock *sk)
321 {
322 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
323 struct hlist_node *node;
324 unsigned long mask;
325 unsigned int i;
326
327 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
328 mask = 0;
329 sk_for_each_bound(sk, node, &tbl->mc_list) {
330 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
331 mask |= nlk_sk(sk)->groups[i];
332 }
333 tbl->listeners[i] = mask;
334 }
335 /* this function is only called with the netlink table "grabbed", which
336 * makes sure updates are visible before bind or setsockopt return. */
337 }
338
339 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
340 {
341 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
342 struct hlist_head *head;
343 int err = -EADDRINUSE;
344 struct sock *osk;
345 struct hlist_node *node;
346 int len;
347
348 netlink_table_grab();
349 head = nl_pid_hashfn(hash, pid);
350 len = 0;
351 sk_for_each(osk, node, head) {
352 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
353 break;
354 len++;
355 }
356 if (node)
357 goto err;
358
359 err = -EBUSY;
360 if (nlk_sk(sk)->pid)
361 goto err;
362
363 err = -ENOMEM;
364 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
365 goto err;
366
367 if (len && nl_pid_hash_dilute(hash, len))
368 head = nl_pid_hashfn(hash, pid);
369 hash->entries++;
370 nlk_sk(sk)->pid = pid;
371 sk_add_node(sk, head);
372 err = 0;
373
374 err:
375 netlink_table_ungrab();
376 return err;
377 }
378
379 static void netlink_remove(struct sock *sk)
380 {
381 netlink_table_grab();
382 if (sk_del_node_init(sk))
383 nl_table[sk->sk_protocol].hash.entries--;
384 if (nlk_sk(sk)->subscriptions)
385 __sk_del_bind_node(sk);
386 netlink_table_ungrab();
387 }
388
389 static struct proto netlink_proto = {
390 .name = "NETLINK",
391 .owner = THIS_MODULE,
392 .obj_size = sizeof(struct netlink_sock),
393 };
394
395 static int __netlink_create(struct net *net, struct socket *sock,
396 struct mutex *cb_mutex, int protocol)
397 {
398 struct sock *sk;
399 struct netlink_sock *nlk;
400
401 sock->ops = &netlink_ops;
402
403 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
404 if (!sk)
405 return -ENOMEM;
406
407 sock_init_data(sock, sk);
408
409 nlk = nlk_sk(sk);
410 if (cb_mutex)
411 nlk->cb_mutex = cb_mutex;
412 else {
413 nlk->cb_mutex = &nlk->cb_def_mutex;
414 mutex_init(nlk->cb_mutex);
415 }
416 init_waitqueue_head(&nlk->wait);
417
418 sk->sk_destruct = netlink_sock_destruct;
419 sk->sk_protocol = protocol;
420 return 0;
421 }
422
423 static int netlink_create(struct net *net, struct socket *sock, int protocol)
424 {
425 struct module *module = NULL;
426 struct mutex *cb_mutex;
427 struct netlink_sock *nlk;
428 int err = 0;
429
430 sock->state = SS_UNCONNECTED;
431
432 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
433 return -ESOCKTNOSUPPORT;
434
435 if (protocol < 0 || protocol >= MAX_LINKS)
436 return -EPROTONOSUPPORT;
437
438 netlink_lock_table();
439 #ifdef CONFIG_MODULES
440 if (!nl_table[protocol].registered) {
441 netlink_unlock_table();
442 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
443 netlink_lock_table();
444 }
445 #endif
446 if (nl_table[protocol].registered &&
447 try_module_get(nl_table[protocol].module))
448 module = nl_table[protocol].module;
449 cb_mutex = nl_table[protocol].cb_mutex;
450 netlink_unlock_table();
451
452 err = __netlink_create(net, sock, cb_mutex, protocol);
453 if (err < 0)
454 goto out_module;
455
456 local_bh_disable();
457 sock_prot_inuse_add(net, &netlink_proto, 1);
458 local_bh_enable();
459
460 nlk = nlk_sk(sock->sk);
461 nlk->module = module;
462 out:
463 return err;
464
465 out_module:
466 module_put(module);
467 goto out;
468 }
469
470 static int netlink_release(struct socket *sock)
471 {
472 struct sock *sk = sock->sk;
473 struct netlink_sock *nlk;
474
475 if (!sk)
476 return 0;
477
478 netlink_remove(sk);
479 sock_orphan(sk);
480 nlk = nlk_sk(sk);
481
482 /*
483 * OK. Socket is unlinked, any packets that arrive now
484 * will be purged.
485 */
486
487 sock->sk = NULL;
488 wake_up_interruptible_all(&nlk->wait);
489
490 skb_queue_purge(&sk->sk_write_queue);
491
492 if (nlk->pid && !nlk->subscriptions) {
493 struct netlink_notify n = {
494 .net = sock_net(sk),
495 .protocol = sk->sk_protocol,
496 .pid = nlk->pid,
497 };
498 atomic_notifier_call_chain(&netlink_chain,
499 NETLINK_URELEASE, &n);
500 }
501
502 module_put(nlk->module);
503
504 netlink_table_grab();
505 if (netlink_is_kernel(sk)) {
506 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
507 if (--nl_table[sk->sk_protocol].registered == 0) {
508 kfree(nl_table[sk->sk_protocol].listeners);
509 nl_table[sk->sk_protocol].module = NULL;
510 nl_table[sk->sk_protocol].registered = 0;
511 }
512 } else if (nlk->subscriptions)
513 netlink_update_listeners(sk);
514 netlink_table_ungrab();
515
516 kfree(nlk->groups);
517 nlk->groups = NULL;
518
519 local_bh_disable();
520 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
521 local_bh_enable();
522 sock_put(sk);
523 return 0;
524 }
525
526 static int netlink_autobind(struct socket *sock)
527 {
528 struct sock *sk = sock->sk;
529 struct net *net = sock_net(sk);
530 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
531 struct hlist_head *head;
532 struct sock *osk;
533 struct hlist_node *node;
534 s32 pid = current->tgid;
535 int err;
536 static s32 rover = -4097;
537
538 retry:
539 cond_resched();
540 netlink_table_grab();
541 head = nl_pid_hashfn(hash, pid);
542 sk_for_each(osk, node, head) {
543 if (!net_eq(sock_net(osk), net))
544 continue;
545 if (nlk_sk(osk)->pid == pid) {
546 /* Bind collision, search negative pid values. */
547 pid = rover--;
548 if (rover > -4097)
549 rover = -4097;
550 netlink_table_ungrab();
551 goto retry;
552 }
553 }
554 netlink_table_ungrab();
555
556 err = netlink_insert(sk, net, pid);
557 if (err == -EADDRINUSE)
558 goto retry;
559
560 /* If 2 threads race to autobind, that is fine. */
561 if (err == -EBUSY)
562 err = 0;
563
564 return err;
565 }
566
567 static inline int netlink_capable(struct socket *sock, unsigned int flag)
568 {
569 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
570 capable(CAP_NET_ADMIN);
571 }
572
573 static void
574 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
575 {
576 struct netlink_sock *nlk = nlk_sk(sk);
577
578 if (nlk->subscriptions && !subscriptions)
579 __sk_del_bind_node(sk);
580 else if (!nlk->subscriptions && subscriptions)
581 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
582 nlk->subscriptions = subscriptions;
583 }
584
585 static int netlink_realloc_groups(struct sock *sk)
586 {
587 struct netlink_sock *nlk = nlk_sk(sk);
588 unsigned int groups;
589 unsigned long *new_groups;
590 int err = 0;
591
592 netlink_table_grab();
593
594 groups = nl_table[sk->sk_protocol].groups;
595 if (!nl_table[sk->sk_protocol].registered) {
596 err = -ENOENT;
597 goto out_unlock;
598 }
599
600 if (nlk->ngroups >= groups)
601 goto out_unlock;
602
603 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
604 if (new_groups == NULL) {
605 err = -ENOMEM;
606 goto out_unlock;
607 }
608 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
609 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
610
611 nlk->groups = new_groups;
612 nlk->ngroups = groups;
613 out_unlock:
614 netlink_table_ungrab();
615 return err;
616 }
617
618 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
619 int addr_len)
620 {
621 struct sock *sk = sock->sk;
622 struct net *net = sock_net(sk);
623 struct netlink_sock *nlk = nlk_sk(sk);
624 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
625 int err;
626
627 if (nladdr->nl_family != AF_NETLINK)
628 return -EINVAL;
629
630 /* Only superuser is allowed to listen multicasts */
631 if (nladdr->nl_groups) {
632 if (!netlink_capable(sock, NL_NONROOT_RECV))
633 return -EPERM;
634 err = netlink_realloc_groups(sk);
635 if (err)
636 return err;
637 }
638
639 if (nlk->pid) {
640 if (nladdr->nl_pid != nlk->pid)
641 return -EINVAL;
642 } else {
643 err = nladdr->nl_pid ?
644 netlink_insert(sk, net, nladdr->nl_pid) :
645 netlink_autobind(sock);
646 if (err)
647 return err;
648 }
649
650 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
651 return 0;
652
653 netlink_table_grab();
654 netlink_update_subscriptions(sk, nlk->subscriptions +
655 hweight32(nladdr->nl_groups) -
656 hweight32(nlk->groups[0]));
657 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
658 netlink_update_listeners(sk);
659 netlink_table_ungrab();
660
661 return 0;
662 }
663
664 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
665 int alen, int flags)
666 {
667 int err = 0;
668 struct sock *sk = sock->sk;
669 struct netlink_sock *nlk = nlk_sk(sk);
670 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
671
672 if (addr->sa_family == AF_UNSPEC) {
673 sk->sk_state = NETLINK_UNCONNECTED;
674 nlk->dst_pid = 0;
675 nlk->dst_group = 0;
676 return 0;
677 }
678 if (addr->sa_family != AF_NETLINK)
679 return -EINVAL;
680
681 /* Only superuser is allowed to send multicasts */
682 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
683 return -EPERM;
684
685 if (!nlk->pid)
686 err = netlink_autobind(sock);
687
688 if (err == 0) {
689 sk->sk_state = NETLINK_CONNECTED;
690 nlk->dst_pid = nladdr->nl_pid;
691 nlk->dst_group = ffs(nladdr->nl_groups);
692 }
693
694 return err;
695 }
696
697 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
698 int *addr_len, int peer)
699 {
700 struct sock *sk = sock->sk;
701 struct netlink_sock *nlk = nlk_sk(sk);
702 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
703
704 nladdr->nl_family = AF_NETLINK;
705 nladdr->nl_pad = 0;
706 *addr_len = sizeof(*nladdr);
707
708 if (peer) {
709 nladdr->nl_pid = nlk->dst_pid;
710 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
711 } else {
712 nladdr->nl_pid = nlk->pid;
713 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
714 }
715 return 0;
716 }
717
718 static void netlink_overrun(struct sock *sk)
719 {
720 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
721 sk->sk_err = ENOBUFS;
722 sk->sk_error_report(sk);
723 }
724 }
725
726 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
727 {
728 struct sock *sock;
729 struct netlink_sock *nlk;
730
731 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
732 if (!sock)
733 return ERR_PTR(-ECONNREFUSED);
734
735 /* Don't bother queuing skb if kernel socket has no input function */
736 nlk = nlk_sk(sock);
737 if (sock->sk_state == NETLINK_CONNECTED &&
738 nlk->dst_pid != nlk_sk(ssk)->pid) {
739 sock_put(sock);
740 return ERR_PTR(-ECONNREFUSED);
741 }
742 return sock;
743 }
744
745 struct sock *netlink_getsockbyfilp(struct file *filp)
746 {
747 struct inode *inode = filp->f_path.dentry->d_inode;
748 struct sock *sock;
749
750 if (!S_ISSOCK(inode->i_mode))
751 return ERR_PTR(-ENOTSOCK);
752
753 sock = SOCKET_I(inode)->sk;
754 if (sock->sk_family != AF_NETLINK)
755 return ERR_PTR(-EINVAL);
756
757 sock_hold(sock);
758 return sock;
759 }
760
761 /*
762 * Attach a skb to a netlink socket.
763 * The caller must hold a reference to the destination socket. On error, the
764 * reference is dropped. The skb is not send to the destination, just all
765 * all error checks are performed and memory in the queue is reserved.
766 * Return values:
767 * < 0: error. skb freed, reference to sock dropped.
768 * 0: continue
769 * 1: repeat lookup - reference dropped while waiting for socket memory.
770 */
771 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
772 long *timeo, struct sock *ssk)
773 {
774 struct netlink_sock *nlk;
775
776 nlk = nlk_sk(sk);
777
778 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
779 test_bit(0, &nlk->state)) {
780 DECLARE_WAITQUEUE(wait, current);
781 if (!*timeo) {
782 if (!ssk || netlink_is_kernel(ssk))
783 netlink_overrun(sk);
784 sock_put(sk);
785 kfree_skb(skb);
786 return -EAGAIN;
787 }
788
789 __set_current_state(TASK_INTERRUPTIBLE);
790 add_wait_queue(&nlk->wait, &wait);
791
792 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
793 test_bit(0, &nlk->state)) &&
794 !sock_flag(sk, SOCK_DEAD))
795 *timeo = schedule_timeout(*timeo);
796
797 __set_current_state(TASK_RUNNING);
798 remove_wait_queue(&nlk->wait, &wait);
799 sock_put(sk);
800
801 if (signal_pending(current)) {
802 kfree_skb(skb);
803 return sock_intr_errno(*timeo);
804 }
805 return 1;
806 }
807 skb_set_owner_r(skb, sk);
808 return 0;
809 }
810
811 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
812 {
813 int len = skb->len;
814
815 skb_queue_tail(&sk->sk_receive_queue, skb);
816 sk->sk_data_ready(sk, len);
817 sock_put(sk);
818 return len;
819 }
820
821 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
822 {
823 kfree_skb(skb);
824 sock_put(sk);
825 }
826
827 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
828 gfp_t allocation)
829 {
830 int delta;
831
832 skb_orphan(skb);
833
834 delta = skb->end - skb->tail;
835 if (delta * 2 < skb->truesize)
836 return skb;
837
838 if (skb_shared(skb)) {
839 struct sk_buff *nskb = skb_clone(skb, allocation);
840 if (!nskb)
841 return skb;
842 kfree_skb(skb);
843 skb = nskb;
844 }
845
846 if (!pskb_expand_head(skb, 0, -delta, allocation))
847 skb->truesize -= delta;
848
849 return skb;
850 }
851
852 static inline void netlink_rcv_wake(struct sock *sk)
853 {
854 struct netlink_sock *nlk = nlk_sk(sk);
855
856 if (skb_queue_empty(&sk->sk_receive_queue))
857 clear_bit(0, &nlk->state);
858 if (!test_bit(0, &nlk->state))
859 wake_up_interruptible(&nlk->wait);
860 }
861
862 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
863 {
864 int ret;
865 struct netlink_sock *nlk = nlk_sk(sk);
866
867 ret = -ECONNREFUSED;
868 if (nlk->netlink_rcv != NULL) {
869 ret = skb->len;
870 skb_set_owner_r(skb, sk);
871 nlk->netlink_rcv(skb);
872 }
873 kfree_skb(skb);
874 sock_put(sk);
875 return ret;
876 }
877
878 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
879 u32 pid, int nonblock)
880 {
881 struct sock *sk;
882 int err;
883 long timeo;
884
885 skb = netlink_trim(skb, gfp_any());
886
887 timeo = sock_sndtimeo(ssk, nonblock);
888 retry:
889 sk = netlink_getsockbypid(ssk, pid);
890 if (IS_ERR(sk)) {
891 kfree_skb(skb);
892 return PTR_ERR(sk);
893 }
894 if (netlink_is_kernel(sk))
895 return netlink_unicast_kernel(sk, skb);
896
897 if (sk_filter(sk, skb)) {
898 err = skb->len;
899 kfree_skb(skb);
900 sock_put(sk);
901 return err;
902 }
903
904 err = netlink_attachskb(sk, skb, &timeo, ssk);
905 if (err == 1)
906 goto retry;
907 if (err)
908 return err;
909
910 return netlink_sendskb(sk, skb);
911 }
912 EXPORT_SYMBOL(netlink_unicast);
913
914 int netlink_has_listeners(struct sock *sk, unsigned int group)
915 {
916 int res = 0;
917 unsigned long *listeners;
918
919 BUG_ON(!netlink_is_kernel(sk));
920
921 rcu_read_lock();
922 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
923
924 if (group - 1 < nl_table[sk->sk_protocol].groups)
925 res = test_bit(group - 1, listeners);
926
927 rcu_read_unlock();
928
929 return res;
930 }
931 EXPORT_SYMBOL_GPL(netlink_has_listeners);
932
933 static inline int netlink_broadcast_deliver(struct sock *sk,
934 struct sk_buff *skb)
935 {
936 struct netlink_sock *nlk = nlk_sk(sk);
937
938 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
939 !test_bit(0, &nlk->state)) {
940 skb_set_owner_r(skb, sk);
941 skb_queue_tail(&sk->sk_receive_queue, skb);
942 sk->sk_data_ready(sk, skb->len);
943 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
944 }
945 return -1;
946 }
947
948 struct netlink_broadcast_data {
949 struct sock *exclude_sk;
950 struct net *net;
951 u32 pid;
952 u32 group;
953 int failure;
954 int delivery_failure;
955 int congested;
956 int delivered;
957 gfp_t allocation;
958 struct sk_buff *skb, *skb2;
959 };
960
961 static inline int do_one_broadcast(struct sock *sk,
962 struct netlink_broadcast_data *p)
963 {
964 struct netlink_sock *nlk = nlk_sk(sk);
965 int val;
966
967 if (p->exclude_sk == sk)
968 goto out;
969
970 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
971 !test_bit(p->group - 1, nlk->groups))
972 goto out;
973
974 if (!net_eq(sock_net(sk), p->net))
975 goto out;
976
977 if (p->failure) {
978 netlink_overrun(sk);
979 goto out;
980 }
981
982 sock_hold(sk);
983 if (p->skb2 == NULL) {
984 if (skb_shared(p->skb)) {
985 p->skb2 = skb_clone(p->skb, p->allocation);
986 } else {
987 p->skb2 = skb_get(p->skb);
988 /*
989 * skb ownership may have been set when
990 * delivered to a previous socket.
991 */
992 skb_orphan(p->skb2);
993 }
994 }
995 if (p->skb2 == NULL) {
996 netlink_overrun(sk);
997 /* Clone failed. Notify ALL listeners. */
998 p->failure = 1;
999 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1000 p->delivery_failure = 1;
1001 } else if (sk_filter(sk, p->skb2)) {
1002 kfree_skb(p->skb2);
1003 p->skb2 = NULL;
1004 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1005 netlink_overrun(sk);
1006 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1007 p->delivery_failure = 1;
1008 } else {
1009 p->congested |= val;
1010 p->delivered = 1;
1011 p->skb2 = NULL;
1012 }
1013 sock_put(sk);
1014
1015 out:
1016 return 0;
1017 }
1018
1019 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1020 u32 group, gfp_t allocation)
1021 {
1022 struct net *net = sock_net(ssk);
1023 struct netlink_broadcast_data info;
1024 struct hlist_node *node;
1025 struct sock *sk;
1026
1027 skb = netlink_trim(skb, allocation);
1028
1029 info.exclude_sk = ssk;
1030 info.net = net;
1031 info.pid = pid;
1032 info.group = group;
1033 info.failure = 0;
1034 info.delivery_failure = 0;
1035 info.congested = 0;
1036 info.delivered = 0;
1037 info.allocation = allocation;
1038 info.skb = skb;
1039 info.skb2 = NULL;
1040
1041 /* While we sleep in clone, do not allow to change socket list */
1042
1043 netlink_lock_table();
1044
1045 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1046 do_one_broadcast(sk, &info);
1047
1048 kfree_skb(skb);
1049
1050 netlink_unlock_table();
1051
1052 if (info.skb2)
1053 kfree_skb(info.skb2);
1054
1055 if (info.delivery_failure)
1056 return -ENOBUFS;
1057
1058 if (info.delivered) {
1059 if (info.congested && (allocation & __GFP_WAIT))
1060 yield();
1061 return 0;
1062 }
1063 return -ESRCH;
1064 }
1065 EXPORT_SYMBOL(netlink_broadcast);
1066
1067 struct netlink_set_err_data {
1068 struct sock *exclude_sk;
1069 u32 pid;
1070 u32 group;
1071 int code;
1072 };
1073
1074 static inline int do_one_set_err(struct sock *sk,
1075 struct netlink_set_err_data *p)
1076 {
1077 struct netlink_sock *nlk = nlk_sk(sk);
1078
1079 if (sk == p->exclude_sk)
1080 goto out;
1081
1082 if (sock_net(sk) != sock_net(p->exclude_sk))
1083 goto out;
1084
1085 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1086 !test_bit(p->group - 1, nlk->groups))
1087 goto out;
1088
1089 sk->sk_err = p->code;
1090 sk->sk_error_report(sk);
1091 out:
1092 return 0;
1093 }
1094
1095 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1096 {
1097 struct netlink_set_err_data info;
1098 struct hlist_node *node;
1099 struct sock *sk;
1100
1101 info.exclude_sk = ssk;
1102 info.pid = pid;
1103 info.group = group;
1104 info.code = code;
1105
1106 read_lock(&nl_table_lock);
1107
1108 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1109 do_one_set_err(sk, &info);
1110
1111 read_unlock(&nl_table_lock);
1112 }
1113
1114 /* must be called with netlink table grabbed */
1115 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1116 unsigned int group,
1117 int is_new)
1118 {
1119 int old, new = !!is_new, subscriptions;
1120
1121 old = test_bit(group - 1, nlk->groups);
1122 subscriptions = nlk->subscriptions - old + new;
1123 if (new)
1124 __set_bit(group - 1, nlk->groups);
1125 else
1126 __clear_bit(group - 1, nlk->groups);
1127 netlink_update_subscriptions(&nlk->sk, subscriptions);
1128 netlink_update_listeners(&nlk->sk);
1129 }
1130
1131 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1132 char __user *optval, int optlen)
1133 {
1134 struct sock *sk = sock->sk;
1135 struct netlink_sock *nlk = nlk_sk(sk);
1136 unsigned int val = 0;
1137 int err;
1138
1139 if (level != SOL_NETLINK)
1140 return -ENOPROTOOPT;
1141
1142 if (optlen >= sizeof(int) &&
1143 get_user(val, (unsigned int __user *)optval))
1144 return -EFAULT;
1145
1146 switch (optname) {
1147 case NETLINK_PKTINFO:
1148 if (val)
1149 nlk->flags |= NETLINK_RECV_PKTINFO;
1150 else
1151 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1152 err = 0;
1153 break;
1154 case NETLINK_ADD_MEMBERSHIP:
1155 case NETLINK_DROP_MEMBERSHIP: {
1156 if (!netlink_capable(sock, NL_NONROOT_RECV))
1157 return -EPERM;
1158 err = netlink_realloc_groups(sk);
1159 if (err)
1160 return err;
1161 if (!val || val - 1 >= nlk->ngroups)
1162 return -EINVAL;
1163 netlink_table_grab();
1164 netlink_update_socket_mc(nlk, val,
1165 optname == NETLINK_ADD_MEMBERSHIP);
1166 netlink_table_ungrab();
1167 err = 0;
1168 break;
1169 }
1170 case NETLINK_BROADCAST_ERROR:
1171 if (val)
1172 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1173 else
1174 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1175 err = 0;
1176 break;
1177 default:
1178 err = -ENOPROTOOPT;
1179 }
1180 return err;
1181 }
1182
1183 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1184 char __user *optval, int __user *optlen)
1185 {
1186 struct sock *sk = sock->sk;
1187 struct netlink_sock *nlk = nlk_sk(sk);
1188 int len, val, err;
1189
1190 if (level != SOL_NETLINK)
1191 return -ENOPROTOOPT;
1192
1193 if (get_user(len, optlen))
1194 return -EFAULT;
1195 if (len < 0)
1196 return -EINVAL;
1197
1198 switch (optname) {
1199 case NETLINK_PKTINFO:
1200 if (len < sizeof(int))
1201 return -EINVAL;
1202 len = sizeof(int);
1203 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1204 if (put_user(len, optlen) ||
1205 put_user(val, optval))
1206 return -EFAULT;
1207 err = 0;
1208 break;
1209 case NETLINK_BROADCAST_ERROR:
1210 if (len < sizeof(int))
1211 return -EINVAL;
1212 len = sizeof(int);
1213 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1214 if (put_user(len, optlen) ||
1215 put_user(val, optval))
1216 return -EFAULT;
1217 err = 0;
1218 break;
1219 default:
1220 err = -ENOPROTOOPT;
1221 }
1222 return err;
1223 }
1224
1225 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1226 {
1227 struct nl_pktinfo info;
1228
1229 info.group = NETLINK_CB(skb).dst_group;
1230 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1231 }
1232
1233 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1234 struct msghdr *msg, size_t len)
1235 {
1236 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1237 struct sock *sk = sock->sk;
1238 struct netlink_sock *nlk = nlk_sk(sk);
1239 struct sockaddr_nl *addr = msg->msg_name;
1240 u32 dst_pid;
1241 u32 dst_group;
1242 struct sk_buff *skb;
1243 int err;
1244 struct scm_cookie scm;
1245
1246 if (msg->msg_flags&MSG_OOB)
1247 return -EOPNOTSUPP;
1248
1249 if (NULL == siocb->scm)
1250 siocb->scm = &scm;
1251 err = scm_send(sock, msg, siocb->scm);
1252 if (err < 0)
1253 return err;
1254
1255 if (msg->msg_namelen) {
1256 if (addr->nl_family != AF_NETLINK)
1257 return -EINVAL;
1258 dst_pid = addr->nl_pid;
1259 dst_group = ffs(addr->nl_groups);
1260 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1261 return -EPERM;
1262 } else {
1263 dst_pid = nlk->dst_pid;
1264 dst_group = nlk->dst_group;
1265 }
1266
1267 if (!nlk->pid) {
1268 err = netlink_autobind(sock);
1269 if (err)
1270 goto out;
1271 }
1272
1273 err = -EMSGSIZE;
1274 if (len > sk->sk_sndbuf - 32)
1275 goto out;
1276 err = -ENOBUFS;
1277 skb = alloc_skb(len, GFP_KERNEL);
1278 if (skb == NULL)
1279 goto out;
1280
1281 NETLINK_CB(skb).pid = nlk->pid;
1282 NETLINK_CB(skb).dst_group = dst_group;
1283 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1284 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1285 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1286 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1287
1288 /* What can I do? Netlink is asynchronous, so that
1289 we will have to save current capabilities to
1290 check them, when this message will be delivered
1291 to corresponding kernel module. --ANK (980802)
1292 */
1293
1294 err = -EFAULT;
1295 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1296 kfree_skb(skb);
1297 goto out;
1298 }
1299
1300 err = security_netlink_send(sk, skb);
1301 if (err) {
1302 kfree_skb(skb);
1303 goto out;
1304 }
1305
1306 if (dst_group) {
1307 atomic_inc(&skb->users);
1308 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1309 }
1310 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1311
1312 out:
1313 return err;
1314 }
1315
1316 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1317 struct msghdr *msg, size_t len,
1318 int flags)
1319 {
1320 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1321 struct scm_cookie scm;
1322 struct sock *sk = sock->sk;
1323 struct netlink_sock *nlk = nlk_sk(sk);
1324 int noblock = flags&MSG_DONTWAIT;
1325 size_t copied;
1326 struct sk_buff *skb;
1327 int err;
1328
1329 if (flags&MSG_OOB)
1330 return -EOPNOTSUPP;
1331
1332 copied = 0;
1333
1334 skb = skb_recv_datagram(sk, flags, noblock, &err);
1335 if (skb == NULL)
1336 goto out;
1337
1338 msg->msg_namelen = 0;
1339
1340 copied = skb->len;
1341 if (len < copied) {
1342 msg->msg_flags |= MSG_TRUNC;
1343 copied = len;
1344 }
1345
1346 skb_reset_transport_header(skb);
1347 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1348
1349 if (msg->msg_name) {
1350 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1351 addr->nl_family = AF_NETLINK;
1352 addr->nl_pad = 0;
1353 addr->nl_pid = NETLINK_CB(skb).pid;
1354 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1355 msg->msg_namelen = sizeof(*addr);
1356 }
1357
1358 if (nlk->flags & NETLINK_RECV_PKTINFO)
1359 netlink_cmsg_recv_pktinfo(msg, skb);
1360
1361 if (NULL == siocb->scm) {
1362 memset(&scm, 0, sizeof(scm));
1363 siocb->scm = &scm;
1364 }
1365 siocb->scm->creds = *NETLINK_CREDS(skb);
1366 if (flags & MSG_TRUNC)
1367 copied = skb->len;
1368 skb_free_datagram(sk, skb);
1369
1370 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1371 netlink_dump(sk);
1372
1373 scm_recv(sock, msg, siocb->scm, flags);
1374 out:
1375 netlink_rcv_wake(sk);
1376 return err ? : copied;
1377 }
1378
1379 static void netlink_data_ready(struct sock *sk, int len)
1380 {
1381 BUG();
1382 }
1383
1384 /*
1385 * We export these functions to other modules. They provide a
1386 * complete set of kernel non-blocking support for message
1387 * queueing.
1388 */
1389
1390 struct sock *
1391 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1392 void (*input)(struct sk_buff *skb),
1393 struct mutex *cb_mutex, struct module *module)
1394 {
1395 struct socket *sock;
1396 struct sock *sk;
1397 struct netlink_sock *nlk;
1398 unsigned long *listeners = NULL;
1399
1400 BUG_ON(!nl_table);
1401
1402 if (unit < 0 || unit >= MAX_LINKS)
1403 return NULL;
1404
1405 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1406 return NULL;
1407
1408 /*
1409 * We have to just have a reference on the net from sk, but don't
1410 * get_net it. Besides, we cannot get and then put the net here.
1411 * So we create one inside init_net and the move it to net.
1412 */
1413
1414 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1415 goto out_sock_release_nosk;
1416
1417 sk = sock->sk;
1418 sk_change_net(sk, net);
1419
1420 if (groups < 32)
1421 groups = 32;
1422
1423 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1424 if (!listeners)
1425 goto out_sock_release;
1426
1427 sk->sk_data_ready = netlink_data_ready;
1428 if (input)
1429 nlk_sk(sk)->netlink_rcv = input;
1430
1431 if (netlink_insert(sk, net, 0))
1432 goto out_sock_release;
1433
1434 nlk = nlk_sk(sk);
1435 nlk->flags |= NETLINK_KERNEL_SOCKET;
1436
1437 netlink_table_grab();
1438 if (!nl_table[unit].registered) {
1439 nl_table[unit].groups = groups;
1440 nl_table[unit].listeners = listeners;
1441 nl_table[unit].cb_mutex = cb_mutex;
1442 nl_table[unit].module = module;
1443 nl_table[unit].registered = 1;
1444 } else {
1445 kfree(listeners);
1446 nl_table[unit].registered++;
1447 }
1448 netlink_table_ungrab();
1449 return sk;
1450
1451 out_sock_release:
1452 kfree(listeners);
1453 netlink_kernel_release(sk);
1454 return NULL;
1455
1456 out_sock_release_nosk:
1457 sock_release(sock);
1458 return NULL;
1459 }
1460 EXPORT_SYMBOL(netlink_kernel_create);
1461
1462
1463 void
1464 netlink_kernel_release(struct sock *sk)
1465 {
1466 sk_release_kernel(sk);
1467 }
1468 EXPORT_SYMBOL(netlink_kernel_release);
1469
1470
1471 /**
1472 * netlink_change_ngroups - change number of multicast groups
1473 *
1474 * This changes the number of multicast groups that are available
1475 * on a certain netlink family. Note that it is not possible to
1476 * change the number of groups to below 32. Also note that it does
1477 * not implicitly call netlink_clear_multicast_users() when the
1478 * number of groups is reduced.
1479 *
1480 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1481 * @groups: The new number of groups.
1482 */
1483 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1484 {
1485 unsigned long *listeners, *old = NULL;
1486 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1487 int err = 0;
1488
1489 if (groups < 32)
1490 groups = 32;
1491
1492 netlink_table_grab();
1493 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1494 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1495 if (!listeners) {
1496 err = -ENOMEM;
1497 goto out_ungrab;
1498 }
1499 old = tbl->listeners;
1500 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1501 rcu_assign_pointer(tbl->listeners, listeners);
1502 }
1503 tbl->groups = groups;
1504
1505 out_ungrab:
1506 netlink_table_ungrab();
1507 synchronize_rcu();
1508 kfree(old);
1509 return err;
1510 }
1511 EXPORT_SYMBOL(netlink_change_ngroups);
1512
1513 /**
1514 * netlink_clear_multicast_users - kick off multicast listeners
1515 *
1516 * This function removes all listeners from the given group.
1517 * @ksk: The kernel netlink socket, as returned by
1518 * netlink_kernel_create().
1519 * @group: The multicast group to clear.
1520 */
1521 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1522 {
1523 struct sock *sk;
1524 struct hlist_node *node;
1525 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1526
1527 netlink_table_grab();
1528
1529 sk_for_each_bound(sk, node, &tbl->mc_list)
1530 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1531
1532 netlink_table_ungrab();
1533 }
1534 EXPORT_SYMBOL(netlink_clear_multicast_users);
1535
1536 void netlink_set_nonroot(int protocol, unsigned int flags)
1537 {
1538 if ((unsigned int)protocol < MAX_LINKS)
1539 nl_table[protocol].nl_nonroot = flags;
1540 }
1541 EXPORT_SYMBOL(netlink_set_nonroot);
1542
1543 static void netlink_destroy_callback(struct netlink_callback *cb)
1544 {
1545 if (cb->skb)
1546 kfree_skb(cb->skb);
1547 kfree(cb);
1548 }
1549
1550 /*
1551 * It looks a bit ugly.
1552 * It would be better to create kernel thread.
1553 */
1554
1555 static int netlink_dump(struct sock *sk)
1556 {
1557 struct netlink_sock *nlk = nlk_sk(sk);
1558 struct netlink_callback *cb;
1559 struct sk_buff *skb;
1560 struct nlmsghdr *nlh;
1561 int len, err = -ENOBUFS;
1562
1563 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1564 if (!skb)
1565 goto errout;
1566
1567 mutex_lock(nlk->cb_mutex);
1568
1569 cb = nlk->cb;
1570 if (cb == NULL) {
1571 err = -EINVAL;
1572 goto errout_skb;
1573 }
1574
1575 len = cb->dump(skb, cb);
1576
1577 if (len > 0) {
1578 mutex_unlock(nlk->cb_mutex);
1579
1580 if (sk_filter(sk, skb))
1581 kfree_skb(skb);
1582 else {
1583 skb_queue_tail(&sk->sk_receive_queue, skb);
1584 sk->sk_data_ready(sk, skb->len);
1585 }
1586 return 0;
1587 }
1588
1589 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1590 if (!nlh)
1591 goto errout_skb;
1592
1593 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1594
1595 if (sk_filter(sk, skb))
1596 kfree_skb(skb);
1597 else {
1598 skb_queue_tail(&sk->sk_receive_queue, skb);
1599 sk->sk_data_ready(sk, skb->len);
1600 }
1601
1602 if (cb->done)
1603 cb->done(cb);
1604 nlk->cb = NULL;
1605 mutex_unlock(nlk->cb_mutex);
1606
1607 netlink_destroy_callback(cb);
1608 return 0;
1609
1610 errout_skb:
1611 mutex_unlock(nlk->cb_mutex);
1612 kfree_skb(skb);
1613 errout:
1614 return err;
1615 }
1616
1617 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1618 struct nlmsghdr *nlh,
1619 int (*dump)(struct sk_buff *skb,
1620 struct netlink_callback *),
1621 int (*done)(struct netlink_callback *))
1622 {
1623 struct netlink_callback *cb;
1624 struct sock *sk;
1625 struct netlink_sock *nlk;
1626
1627 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1628 if (cb == NULL)
1629 return -ENOBUFS;
1630
1631 cb->dump = dump;
1632 cb->done = done;
1633 cb->nlh = nlh;
1634 atomic_inc(&skb->users);
1635 cb->skb = skb;
1636
1637 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1638 if (sk == NULL) {
1639 netlink_destroy_callback(cb);
1640 return -ECONNREFUSED;
1641 }
1642 nlk = nlk_sk(sk);
1643 /* A dump is in progress... */
1644 mutex_lock(nlk->cb_mutex);
1645 if (nlk->cb) {
1646 mutex_unlock(nlk->cb_mutex);
1647 netlink_destroy_callback(cb);
1648 sock_put(sk);
1649 return -EBUSY;
1650 }
1651 nlk->cb = cb;
1652 mutex_unlock(nlk->cb_mutex);
1653
1654 netlink_dump(sk);
1655 sock_put(sk);
1656
1657 /* We successfully started a dump, by returning -EINTR we
1658 * signal not to send ACK even if it was requested.
1659 */
1660 return -EINTR;
1661 }
1662 EXPORT_SYMBOL(netlink_dump_start);
1663
1664 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1665 {
1666 struct sk_buff *skb;
1667 struct nlmsghdr *rep;
1668 struct nlmsgerr *errmsg;
1669 size_t payload = sizeof(*errmsg);
1670
1671 /* error messages get the original request appened */
1672 if (err)
1673 payload += nlmsg_len(nlh);
1674
1675 skb = nlmsg_new(payload, GFP_KERNEL);
1676 if (!skb) {
1677 struct sock *sk;
1678
1679 sk = netlink_lookup(sock_net(in_skb->sk),
1680 in_skb->sk->sk_protocol,
1681 NETLINK_CB(in_skb).pid);
1682 if (sk) {
1683 sk->sk_err = ENOBUFS;
1684 sk->sk_error_report(sk);
1685 sock_put(sk);
1686 }
1687 return;
1688 }
1689
1690 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1691 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1692 errmsg = nlmsg_data(rep);
1693 errmsg->error = err;
1694 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1695 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1696 }
1697 EXPORT_SYMBOL(netlink_ack);
1698
1699 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1700 struct nlmsghdr *))
1701 {
1702 struct nlmsghdr *nlh;
1703 int err;
1704
1705 while (skb->len >= nlmsg_total_size(0)) {
1706 int msglen;
1707
1708 nlh = nlmsg_hdr(skb);
1709 err = 0;
1710
1711 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1712 return 0;
1713
1714 /* Only requests are handled by the kernel */
1715 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1716 goto ack;
1717
1718 /* Skip control messages */
1719 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1720 goto ack;
1721
1722 err = cb(skb, nlh);
1723 if (err == -EINTR)
1724 goto skip;
1725
1726 ack:
1727 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1728 netlink_ack(skb, nlh, err);
1729
1730 skip:
1731 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1732 if (msglen > skb->len)
1733 msglen = skb->len;
1734 skb_pull(skb, msglen);
1735 }
1736
1737 return 0;
1738 }
1739 EXPORT_SYMBOL(netlink_rcv_skb);
1740
1741 /**
1742 * nlmsg_notify - send a notification netlink message
1743 * @sk: netlink socket to use
1744 * @skb: notification message
1745 * @pid: destination netlink pid for reports or 0
1746 * @group: destination multicast group or 0
1747 * @report: 1 to report back, 0 to disable
1748 * @flags: allocation flags
1749 */
1750 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1751 unsigned int group, int report, gfp_t flags)
1752 {
1753 int err = 0;
1754
1755 if (group) {
1756 int exclude_pid = 0;
1757
1758 if (report) {
1759 atomic_inc(&skb->users);
1760 exclude_pid = pid;
1761 }
1762
1763 /* errors reported via destination sk->sk_err */
1764 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1765 }
1766
1767 if (report)
1768 err = nlmsg_unicast(sk, skb, pid);
1769
1770 return err;
1771 }
1772 EXPORT_SYMBOL(nlmsg_notify);
1773
1774 #ifdef CONFIG_PROC_FS
1775 struct nl_seq_iter {
1776 struct seq_net_private p;
1777 int link;
1778 int hash_idx;
1779 };
1780
1781 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1782 {
1783 struct nl_seq_iter *iter = seq->private;
1784 int i, j;
1785 struct sock *s;
1786 struct hlist_node *node;
1787 loff_t off = 0;
1788
1789 for (i = 0; i < MAX_LINKS; i++) {
1790 struct nl_pid_hash *hash = &nl_table[i].hash;
1791
1792 for (j = 0; j <= hash->mask; j++) {
1793 sk_for_each(s, node, &hash->table[j]) {
1794 if (sock_net(s) != seq_file_net(seq))
1795 continue;
1796 if (off == pos) {
1797 iter->link = i;
1798 iter->hash_idx = j;
1799 return s;
1800 }
1801 ++off;
1802 }
1803 }
1804 }
1805 return NULL;
1806 }
1807
1808 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1809 __acquires(nl_table_lock)
1810 {
1811 read_lock(&nl_table_lock);
1812 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1813 }
1814
1815 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1816 {
1817 struct sock *s;
1818 struct nl_seq_iter *iter;
1819 int i, j;
1820
1821 ++*pos;
1822
1823 if (v == SEQ_START_TOKEN)
1824 return netlink_seq_socket_idx(seq, 0);
1825
1826 iter = seq->private;
1827 s = v;
1828 do {
1829 s = sk_next(s);
1830 } while (s && sock_net(s) != seq_file_net(seq));
1831 if (s)
1832 return s;
1833
1834 i = iter->link;
1835 j = iter->hash_idx + 1;
1836
1837 do {
1838 struct nl_pid_hash *hash = &nl_table[i].hash;
1839
1840 for (; j <= hash->mask; j++) {
1841 s = sk_head(&hash->table[j]);
1842 while (s && sock_net(s) != seq_file_net(seq))
1843 s = sk_next(s);
1844 if (s) {
1845 iter->link = i;
1846 iter->hash_idx = j;
1847 return s;
1848 }
1849 }
1850
1851 j = 0;
1852 } while (++i < MAX_LINKS);
1853
1854 return NULL;
1855 }
1856
1857 static void netlink_seq_stop(struct seq_file *seq, void *v)
1858 __releases(nl_table_lock)
1859 {
1860 read_unlock(&nl_table_lock);
1861 }
1862
1863
1864 static int netlink_seq_show(struct seq_file *seq, void *v)
1865 {
1866 if (v == SEQ_START_TOKEN)
1867 seq_puts(seq,
1868 "sk Eth Pid Groups "
1869 "Rmem Wmem Dump Locks\n");
1870 else {
1871 struct sock *s = v;
1872 struct netlink_sock *nlk = nlk_sk(s);
1873
1874 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1875 s,
1876 s->sk_protocol,
1877 nlk->pid,
1878 nlk->groups ? (u32)nlk->groups[0] : 0,
1879 atomic_read(&s->sk_rmem_alloc),
1880 atomic_read(&s->sk_wmem_alloc),
1881 nlk->cb,
1882 atomic_read(&s->sk_refcnt)
1883 );
1884
1885 }
1886 return 0;
1887 }
1888
1889 static const struct seq_operations netlink_seq_ops = {
1890 .start = netlink_seq_start,
1891 .next = netlink_seq_next,
1892 .stop = netlink_seq_stop,
1893 .show = netlink_seq_show,
1894 };
1895
1896
1897 static int netlink_seq_open(struct inode *inode, struct file *file)
1898 {
1899 return seq_open_net(inode, file, &netlink_seq_ops,
1900 sizeof(struct nl_seq_iter));
1901 }
1902
1903 static const struct file_operations netlink_seq_fops = {
1904 .owner = THIS_MODULE,
1905 .open = netlink_seq_open,
1906 .read = seq_read,
1907 .llseek = seq_lseek,
1908 .release = seq_release_net,
1909 };
1910
1911 #endif
1912
1913 int netlink_register_notifier(struct notifier_block *nb)
1914 {
1915 return atomic_notifier_chain_register(&netlink_chain, nb);
1916 }
1917 EXPORT_SYMBOL(netlink_register_notifier);
1918
1919 int netlink_unregister_notifier(struct notifier_block *nb)
1920 {
1921 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1922 }
1923 EXPORT_SYMBOL(netlink_unregister_notifier);
1924
1925 static const struct proto_ops netlink_ops = {
1926 .family = PF_NETLINK,
1927 .owner = THIS_MODULE,
1928 .release = netlink_release,
1929 .bind = netlink_bind,
1930 .connect = netlink_connect,
1931 .socketpair = sock_no_socketpair,
1932 .accept = sock_no_accept,
1933 .getname = netlink_getname,
1934 .poll = datagram_poll,
1935 .ioctl = sock_no_ioctl,
1936 .listen = sock_no_listen,
1937 .shutdown = sock_no_shutdown,
1938 .setsockopt = netlink_setsockopt,
1939 .getsockopt = netlink_getsockopt,
1940 .sendmsg = netlink_sendmsg,
1941 .recvmsg = netlink_recvmsg,
1942 .mmap = sock_no_mmap,
1943 .sendpage = sock_no_sendpage,
1944 };
1945
1946 static struct net_proto_family netlink_family_ops = {
1947 .family = PF_NETLINK,
1948 .create = netlink_create,
1949 .owner = THIS_MODULE, /* for consistency 8) */
1950 };
1951
1952 static int __net_init netlink_net_init(struct net *net)
1953 {
1954 #ifdef CONFIG_PROC_FS
1955 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1956 return -ENOMEM;
1957 #endif
1958 return 0;
1959 }
1960
1961 static void __net_exit netlink_net_exit(struct net *net)
1962 {
1963 #ifdef CONFIG_PROC_FS
1964 proc_net_remove(net, "netlink");
1965 #endif
1966 }
1967
1968 static struct pernet_operations __net_initdata netlink_net_ops = {
1969 .init = netlink_net_init,
1970 .exit = netlink_net_exit,
1971 };
1972
1973 static int __init netlink_proto_init(void)
1974 {
1975 struct sk_buff *dummy_skb;
1976 int i;
1977 unsigned long limit;
1978 unsigned int order;
1979 int err = proto_register(&netlink_proto, 0);
1980
1981 if (err != 0)
1982 goto out;
1983
1984 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1985
1986 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1987 if (!nl_table)
1988 goto panic;
1989
1990 if (num_physpages >= (128 * 1024))
1991 limit = num_physpages >> (21 - PAGE_SHIFT);
1992 else
1993 limit = num_physpages >> (23 - PAGE_SHIFT);
1994
1995 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1996 limit = (1UL << order) / sizeof(struct hlist_head);
1997 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1998
1999 for (i = 0; i < MAX_LINKS; i++) {
2000 struct nl_pid_hash *hash = &nl_table[i].hash;
2001
2002 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2003 if (!hash->table) {
2004 while (i-- > 0)
2005 nl_pid_hash_free(nl_table[i].hash.table,
2006 1 * sizeof(*hash->table));
2007 kfree(nl_table);
2008 goto panic;
2009 }
2010 hash->max_shift = order;
2011 hash->shift = 0;
2012 hash->mask = 0;
2013 hash->rehash_time = jiffies;
2014 }
2015
2016 sock_register(&netlink_family_ops);
2017 register_pernet_subsys(&netlink_net_ops);
2018 /* The netlink device handler may be needed early. */
2019 rtnetlink_init();
2020 out:
2021 return err;
2022 panic:
2023 panic("netlink_init: Cannot allocate nl_table\n");
2024 }
2025
2026 core_initcall(netlink_proto_init);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree