?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[POWERPC] mpic_u3msi: Failed allocation unnoticed
[linux-2.6/next.git] / net / netlink / af_netlink.c
blob9b97f8006c9c6edfd0dea0dde522b6a7e64c171f
1 /*
2 * NETLINK Kernel-user communication protocol.
3 *
4 * Authors: Alan Cox <alan@redhat.com>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
22 */
23
24 #include <linux/module.h>
25
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
35 #include <linux/un.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
40 #include <linux/fs.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
54 #include <linux/mm.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/mutex.h>
58
59 #include <net/net_namespace.h>
60 #include <net/sock.h>
61 #include <net/scm.h>
62 #include <net/netlink.h>
63
64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
65 #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
66
67 struct netlink_sock {
68 /* struct sock has to be the first member of netlink_sock */
69 struct sock sk;
70 u32 pid;
71 u32 dst_pid;
72 u32 dst_group;
73 u32 flags;
74 u32 subscriptions;
75 u32 ngroups;
76 unsigned long *groups;
77 unsigned long state;
78 wait_queue_head_t wait;
79 struct netlink_callback *cb;
80 struct mutex *cb_mutex;
81 struct mutex cb_def_mutex;
82 void (*netlink_rcv)(struct sk_buff *skb);
83 struct module *module;
84 };
85
86 #define NETLINK_KERNEL_SOCKET 0x1
87 #define NETLINK_RECV_PKTINFO 0x2
88
89 static inline struct netlink_sock *nlk_sk(struct sock *sk)
90 {
91 return container_of(sk, struct netlink_sock, sk);
92 }
93
94 static inline int netlink_is_kernel(struct sock *sk)
95 {
96 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
97 }
98
99 struct nl_pid_hash {
100 struct hlist_head *table;
101 unsigned long rehash_time;
102
103 unsigned int mask;
104 unsigned int shift;
105
106 unsigned int entries;
107 unsigned int max_shift;
108
109 u32 rnd;
110 };
111
112 struct netlink_table {
113 struct nl_pid_hash hash;
114 struct hlist_head mc_list;
115 unsigned long *listeners;
116 unsigned int nl_nonroot;
117 unsigned int groups;
118 struct mutex *cb_mutex;
119 struct module *module;
120 int registered;
121 };
122
123 static struct netlink_table *nl_table;
124
125 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
126
127 static int netlink_dump(struct sock *sk);
128 static void netlink_destroy_callback(struct netlink_callback *cb);
129
130 static DEFINE_RWLOCK(nl_table_lock);
131 static atomic_t nl_table_users = ATOMIC_INIT(0);
132
133 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
134
135 static u32 netlink_group_mask(u32 group)
136 {
137 return group ? 1 << (group - 1) : 0;
138 }
139
140 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
141 {
142 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
143 }
144
145 static void netlink_sock_destruct(struct sock *sk)
146 {
147 struct netlink_sock *nlk = nlk_sk(sk);
148
149 if (nlk->cb) {
150 if (nlk->cb->done)
151 nlk->cb->done(nlk->cb);
152 netlink_destroy_callback(nlk->cb);
153 }
154
155 skb_queue_purge(&sk->sk_receive_queue);
156
157 if (!sock_flag(sk, SOCK_DEAD)) {
158 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
159 return;
160 }
161 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
162 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
163 BUG_TRAP(!nlk_sk(sk)->groups);
164 }
165
166 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
167 * SMP. Look, when several writers sleep and reader wakes them up, all but one
168 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
169 * this, _but_ remember, it adds useless work on UP machines.
170 */
171
172 static void netlink_table_grab(void)
173 __acquires(nl_table_lock)
174 {
175 write_lock_irq(&nl_table_lock);
176
177 if (atomic_read(&nl_table_users)) {
178 DECLARE_WAITQUEUE(wait, current);
179
180 add_wait_queue_exclusive(&nl_table_wait, &wait);
181 for (;;) {
182 set_current_state(TASK_UNINTERRUPTIBLE);
183 if (atomic_read(&nl_table_users) == 0)
184 break;
185 write_unlock_irq(&nl_table_lock);
186 schedule();
187 write_lock_irq(&nl_table_lock);
188 }
189
190 __set_current_state(TASK_RUNNING);
191 remove_wait_queue(&nl_table_wait, &wait);
192 }
193 }
194
195 static void netlink_table_ungrab(void)
196 __releases(nl_table_lock)
197 {
198 write_unlock_irq(&nl_table_lock);
199 wake_up(&nl_table_wait);
200 }
201
202 static inline void
203 netlink_lock_table(void)
204 {
205 /* read_lock() synchronizes us to netlink_table_grab */
206
207 read_lock(&nl_table_lock);
208 atomic_inc(&nl_table_users);
209 read_unlock(&nl_table_lock);
210 }
211
212 static inline void
213 netlink_unlock_table(void)
214 {
215 if (atomic_dec_and_test(&nl_table_users))
216 wake_up(&nl_table_wait);
217 }
218
219 static inline struct sock *netlink_lookup(struct net *net, int protocol,
220 u32 pid)
221 {
222 struct nl_pid_hash *hash = &nl_table[protocol].hash;
223 struct hlist_head *head;
224 struct sock *sk;
225 struct hlist_node *node;
226
227 read_lock(&nl_table_lock);
228 head = nl_pid_hashfn(hash, pid);
229 sk_for_each(sk, node, head) {
230 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
231 sock_hold(sk);
232 goto found;
233 }
234 }
235 sk = NULL;
236 found:
237 read_unlock(&nl_table_lock);
238 return sk;
239 }
240
241 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
242 {
243 if (size <= PAGE_SIZE)
244 return kzalloc(size, GFP_ATOMIC);
245 else
246 return (struct hlist_head *)
247 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
248 get_order(size));
249 }
250
251 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
252 {
253 if (size <= PAGE_SIZE)
254 kfree(table);
255 else
256 free_pages((unsigned long)table, get_order(size));
257 }
258
259 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
260 {
261 unsigned int omask, mask, shift;
262 size_t osize, size;
263 struct hlist_head *otable, *table;
264 int i;
265
266 omask = mask = hash->mask;
267 osize = size = (mask + 1) * sizeof(*table);
268 shift = hash->shift;
269
270 if (grow) {
271 if (++shift > hash->max_shift)
272 return 0;
273 mask = mask * 2 + 1;
274 size *= 2;
275 }
276
277 table = nl_pid_hash_zalloc(size);
278 if (!table)
279 return 0;
280
281 otable = hash->table;
282 hash->table = table;
283 hash->mask = mask;
284 hash->shift = shift;
285 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
286
287 for (i = 0; i <= omask; i++) {
288 struct sock *sk;
289 struct hlist_node *node, *tmp;
290
291 sk_for_each_safe(sk, node, tmp, &otable[i])
292 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
293 }
294
295 nl_pid_hash_free(otable, osize);
296 hash->rehash_time = jiffies + 10 * 60 * HZ;
297 return 1;
298 }
299
300 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
301 {
302 int avg = hash->entries >> hash->shift;
303
304 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
305 return 1;
306
307 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
308 nl_pid_hash_rehash(hash, 0);
309 return 1;
310 }
311
312 return 0;
313 }
314
315 static const struct proto_ops netlink_ops;
316
317 static void
318 netlink_update_listeners(struct sock *sk)
319 {
320 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
321 struct hlist_node *node;
322 unsigned long mask;
323 unsigned int i;
324
325 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
326 mask = 0;
327 sk_for_each_bound(sk, node, &tbl->mc_list) {
328 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
329 mask |= nlk_sk(sk)->groups[i];
330 }
331 tbl->listeners[i] = mask;
332 }
333 /* this function is only called with the netlink table "grabbed", which
334 * makes sure updates are visible before bind or setsockopt return. */
335 }
336
337 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
338 {
339 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
340 struct hlist_head *head;
341 int err = -EADDRINUSE;
342 struct sock *osk;
343 struct hlist_node *node;
344 int len;
345
346 netlink_table_grab();
347 head = nl_pid_hashfn(hash, pid);
348 len = 0;
349 sk_for_each(osk, node, head) {
350 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
351 break;
352 len++;
353 }
354 if (node)
355 goto err;
356
357 err = -EBUSY;
358 if (nlk_sk(sk)->pid)
359 goto err;
360
361 err = -ENOMEM;
362 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
363 goto err;
364
365 if (len && nl_pid_hash_dilute(hash, len))
366 head = nl_pid_hashfn(hash, pid);
367 hash->entries++;
368 nlk_sk(sk)->pid = pid;
369 sk_add_node(sk, head);
370 err = 0;
371
372 err:
373 netlink_table_ungrab();
374 return err;
375 }
376
377 static void netlink_remove(struct sock *sk)
378 {
379 netlink_table_grab();
380 if (sk_del_node_init(sk))
381 nl_table[sk->sk_protocol].hash.entries--;
382 if (nlk_sk(sk)->subscriptions)
383 __sk_del_bind_node(sk);
384 netlink_table_ungrab();
385 }
386
387 static struct proto netlink_proto = {
388 .name = "NETLINK",
389 .owner = THIS_MODULE,
390 .obj_size = sizeof(struct netlink_sock),
391 };
392
393 static int __netlink_create(struct net *net, struct socket *sock,
394 struct mutex *cb_mutex, int protocol)
395 {
396 struct sock *sk;
397 struct netlink_sock *nlk;
398
399 sock->ops = &netlink_ops;
400
401 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
402 if (!sk)
403 return -ENOMEM;
404
405 sock_init_data(sock, sk);
406
407 nlk = nlk_sk(sk);
408 if (cb_mutex)
409 nlk->cb_mutex = cb_mutex;
410 else {
411 nlk->cb_mutex = &nlk->cb_def_mutex;
412 mutex_init(nlk->cb_mutex);
413 }
414 init_waitqueue_head(&nlk->wait);
415
416 sk->sk_destruct = netlink_sock_destruct;
417 sk->sk_protocol = protocol;
418 return 0;
419 }
420
421 static int netlink_create(struct net *net, struct socket *sock, int protocol)
422 {
423 struct module *module = NULL;
424 struct mutex *cb_mutex;
425 struct netlink_sock *nlk;
426 int err = 0;
427
428 sock->state = SS_UNCONNECTED;
429
430 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
431 return -ESOCKTNOSUPPORT;
432
433 if (protocol < 0 || protocol >= MAX_LINKS)
434 return -EPROTONOSUPPORT;
435
436 netlink_lock_table();
437 #ifdef CONFIG_KMOD
438 if (!nl_table[protocol].registered) {
439 netlink_unlock_table();
440 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
441 netlink_lock_table();
442 }
443 #endif
444 if (nl_table[protocol].registered &&
445 try_module_get(nl_table[protocol].module))
446 module = nl_table[protocol].module;
447 cb_mutex = nl_table[protocol].cb_mutex;
448 netlink_unlock_table();
449
450 err = __netlink_create(net, sock, cb_mutex, protocol);
451 if (err < 0)
452 goto out_module;
453
454 nlk = nlk_sk(sock->sk);
455 nlk->module = module;
456 out:
457 return err;
458
459 out_module:
460 module_put(module);
461 goto out;
462 }
463
464 static int netlink_release(struct socket *sock)
465 {
466 struct sock *sk = sock->sk;
467 struct netlink_sock *nlk;
468
469 if (!sk)
470 return 0;
471
472 netlink_remove(sk);
473 sock_orphan(sk);
474 nlk = nlk_sk(sk);
475
476 /*
477 * OK. Socket is unlinked, any packets that arrive now
478 * will be purged.
479 */
480
481 sock->sk = NULL;
482 wake_up_interruptible_all(&nlk->wait);
483
484 skb_queue_purge(&sk->sk_write_queue);
485
486 if (nlk->pid && !nlk->subscriptions) {
487 struct netlink_notify n = {
488 .net = sock_net(sk),
489 .protocol = sk->sk_protocol,
490 .pid = nlk->pid,
491 };
492 atomic_notifier_call_chain(&netlink_chain,
493 NETLINK_URELEASE, &n);
494 }
495
496 module_put(nlk->module);
497
498 netlink_table_grab();
499 if (netlink_is_kernel(sk)) {
500 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
501 if (--nl_table[sk->sk_protocol].registered == 0) {
502 kfree(nl_table[sk->sk_protocol].listeners);
503 nl_table[sk->sk_protocol].module = NULL;
504 nl_table[sk->sk_protocol].registered = 0;
505 }
506 } else if (nlk->subscriptions)
507 netlink_update_listeners(sk);
508 netlink_table_ungrab();
509
510 kfree(nlk->groups);
511 nlk->groups = NULL;
512
513 sock_put(sk);
514 return 0;
515 }
516
517 static int netlink_autobind(struct socket *sock)
518 {
519 struct sock *sk = sock->sk;
520 struct net *net = sock_net(sk);
521 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
522 struct hlist_head *head;
523 struct sock *osk;
524 struct hlist_node *node;
525 s32 pid = current->tgid;
526 int err;
527 static s32 rover = -4097;
528
529 retry:
530 cond_resched();
531 netlink_table_grab();
532 head = nl_pid_hashfn(hash, pid);
533 sk_for_each(osk, node, head) {
534 if (!net_eq(sock_net(osk), net))
535 continue;
536 if (nlk_sk(osk)->pid == pid) {
537 /* Bind collision, search negative pid values. */
538 pid = rover--;
539 if (rover > -4097)
540 rover = -4097;
541 netlink_table_ungrab();
542 goto retry;
543 }
544 }
545 netlink_table_ungrab();
546
547 err = netlink_insert(sk, net, pid);
548 if (err == -EADDRINUSE)
549 goto retry;
550
551 /* If 2 threads race to autobind, that is fine. */
552 if (err == -EBUSY)
553 err = 0;
554
555 return err;
556 }
557
558 static inline int netlink_capable(struct socket *sock, unsigned int flag)
559 {
560 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
561 capable(CAP_NET_ADMIN);
562 }
563
564 static void
565 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
566 {
567 struct netlink_sock *nlk = nlk_sk(sk);
568
569 if (nlk->subscriptions && !subscriptions)
570 __sk_del_bind_node(sk);
571 else if (!nlk->subscriptions && subscriptions)
572 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
573 nlk->subscriptions = subscriptions;
574 }
575
576 static int netlink_realloc_groups(struct sock *sk)
577 {
578 struct netlink_sock *nlk = nlk_sk(sk);
579 unsigned int groups;
580 unsigned long *new_groups;
581 int err = 0;
582
583 netlink_table_grab();
584
585 groups = nl_table[sk->sk_protocol].groups;
586 if (!nl_table[sk->sk_protocol].registered) {
587 err = -ENOENT;
588 goto out_unlock;
589 }
590
591 if (nlk->ngroups >= groups)
592 goto out_unlock;
593
594 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
595 if (new_groups == NULL) {
596 err = -ENOMEM;
597 goto out_unlock;
598 }
599 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
600 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
601
602 nlk->groups = new_groups;
603 nlk->ngroups = groups;
604 out_unlock:
605 netlink_table_ungrab();
606 return err;
607 }
608
609 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
610 int addr_len)
611 {
612 struct sock *sk = sock->sk;
613 struct net *net = sock_net(sk);
614 struct netlink_sock *nlk = nlk_sk(sk);
615 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
616 int err;
617
618 if (nladdr->nl_family != AF_NETLINK)
619 return -EINVAL;
620
621 /* Only superuser is allowed to listen multicasts */
622 if (nladdr->nl_groups) {
623 if (!netlink_capable(sock, NL_NONROOT_RECV))
624 return -EPERM;
625 err = netlink_realloc_groups(sk);
626 if (err)
627 return err;
628 }
629
630 if (nlk->pid) {
631 if (nladdr->nl_pid != nlk->pid)
632 return -EINVAL;
633 } else {
634 err = nladdr->nl_pid ?
635 netlink_insert(sk, net, nladdr->nl_pid) :
636 netlink_autobind(sock);
637 if (err)
638 return err;
639 }
640
641 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
642 return 0;
643
644 netlink_table_grab();
645 netlink_update_subscriptions(sk, nlk->subscriptions +
646 hweight32(nladdr->nl_groups) -
647 hweight32(nlk->groups[0]));
648 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
649 netlink_update_listeners(sk);
650 netlink_table_ungrab();
651
652 return 0;
653 }
654
655 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
656 int alen, int flags)
657 {
658 int err = 0;
659 struct sock *sk = sock->sk;
660 struct netlink_sock *nlk = nlk_sk(sk);
661 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
662
663 if (addr->sa_family == AF_UNSPEC) {
664 sk->sk_state = NETLINK_UNCONNECTED;
665 nlk->dst_pid = 0;
666 nlk->dst_group = 0;
667 return 0;
668 }
669 if (addr->sa_family != AF_NETLINK)
670 return -EINVAL;
671
672 /* Only superuser is allowed to send multicasts */
673 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
674 return -EPERM;
675
676 if (!nlk->pid)
677 err = netlink_autobind(sock);
678
679 if (err == 0) {
680 sk->sk_state = NETLINK_CONNECTED;
681 nlk->dst_pid = nladdr->nl_pid;
682 nlk->dst_group = ffs(nladdr->nl_groups);
683 }
684
685 return err;
686 }
687
688 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
689 int *addr_len, int peer)
690 {
691 struct sock *sk = sock->sk;
692 struct netlink_sock *nlk = nlk_sk(sk);
693 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
694
695 nladdr->nl_family = AF_NETLINK;
696 nladdr->nl_pad = 0;
697 *addr_len = sizeof(*nladdr);
698
699 if (peer) {
700 nladdr->nl_pid = nlk->dst_pid;
701 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
702 } else {
703 nladdr->nl_pid = nlk->pid;
704 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
705 }
706 return 0;
707 }
708
709 static void netlink_overrun(struct sock *sk)
710 {
711 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
712 sk->sk_err = ENOBUFS;
713 sk->sk_error_report(sk);
714 }
715 }
716
717 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
718 {
719 struct sock *sock;
720 struct netlink_sock *nlk;
721
722 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
723 if (!sock)
724 return ERR_PTR(-ECONNREFUSED);
725
726 /* Don't bother queuing skb if kernel socket has no input function */
727 nlk = nlk_sk(sock);
728 if (sock->sk_state == NETLINK_CONNECTED &&
729 nlk->dst_pid != nlk_sk(ssk)->pid) {
730 sock_put(sock);
731 return ERR_PTR(-ECONNREFUSED);
732 }
733 return sock;
734 }
735
736 struct sock *netlink_getsockbyfilp(struct file *filp)
737 {
738 struct inode *inode = filp->f_path.dentry->d_inode;
739 struct sock *sock;
740
741 if (!S_ISSOCK(inode->i_mode))
742 return ERR_PTR(-ENOTSOCK);
743
744 sock = SOCKET_I(inode)->sk;
745 if (sock->sk_family != AF_NETLINK)
746 return ERR_PTR(-EINVAL);
747
748 sock_hold(sock);
749 return sock;
750 }
751
752 /*
753 * Attach a skb to a netlink socket.
754 * The caller must hold a reference to the destination socket. On error, the
755 * reference is dropped. The skb is not send to the destination, just all
756 * all error checks are performed and memory in the queue is reserved.
757 * Return values:
758 * < 0: error. skb freed, reference to sock dropped.
759 * 0: continue
760 * 1: repeat lookup - reference dropped while waiting for socket memory.
761 */
762 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
763 long *timeo, struct sock *ssk)
764 {
765 struct netlink_sock *nlk;
766
767 nlk = nlk_sk(sk);
768
769 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
770 test_bit(0, &nlk->state)) {
771 DECLARE_WAITQUEUE(wait, current);
772 if (!*timeo) {
773 if (!ssk || netlink_is_kernel(ssk))
774 netlink_overrun(sk);
775 sock_put(sk);
776 kfree_skb(skb);
777 return -EAGAIN;
778 }
779
780 __set_current_state(TASK_INTERRUPTIBLE);
781 add_wait_queue(&nlk->wait, &wait);
782
783 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
784 test_bit(0, &nlk->state)) &&
785 !sock_flag(sk, SOCK_DEAD))
786 *timeo = schedule_timeout(*timeo);
787
788 __set_current_state(TASK_RUNNING);
789 remove_wait_queue(&nlk->wait, &wait);
790 sock_put(sk);
791
792 if (signal_pending(current)) {
793 kfree_skb(skb);
794 return sock_intr_errno(*timeo);
795 }
796 return 1;
797 }
798 skb_set_owner_r(skb, sk);
799 return 0;
800 }
801
802 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
803 {
804 int len = skb->len;
805
806 skb_queue_tail(&sk->sk_receive_queue, skb);
807 sk->sk_data_ready(sk, len);
808 sock_put(sk);
809 return len;
810 }
811
812 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
813 {
814 kfree_skb(skb);
815 sock_put(sk);
816 }
817
818 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
819 gfp_t allocation)
820 {
821 int delta;
822
823 skb_orphan(skb);
824
825 delta = skb->end - skb->tail;
826 if (delta * 2 < skb->truesize)
827 return skb;
828
829 if (skb_shared(skb)) {
830 struct sk_buff *nskb = skb_clone(skb, allocation);
831 if (!nskb)
832 return skb;
833 kfree_skb(skb);
834 skb = nskb;
835 }
836
837 if (!pskb_expand_head(skb, 0, -delta, allocation))
838 skb->truesize -= delta;
839
840 return skb;
841 }
842
843 static inline void netlink_rcv_wake(struct sock *sk)
844 {
845 struct netlink_sock *nlk = nlk_sk(sk);
846
847 if (skb_queue_empty(&sk->sk_receive_queue))
848 clear_bit(0, &nlk->state);
849 if (!test_bit(0, &nlk->state))
850 wake_up_interruptible(&nlk->wait);
851 }
852
853 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
854 {
855 int ret;
856 struct netlink_sock *nlk = nlk_sk(sk);
857
858 ret = -ECONNREFUSED;
859 if (nlk->netlink_rcv != NULL) {
860 ret = skb->len;
861 skb_set_owner_r(skb, sk);
862 nlk->netlink_rcv(skb);
863 }
864 kfree_skb(skb);
865 sock_put(sk);
866 return ret;
867 }
868
869 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
870 u32 pid, int nonblock)
871 {
872 struct sock *sk;
873 int err;
874 long timeo;
875
876 skb = netlink_trim(skb, gfp_any());
877
878 timeo = sock_sndtimeo(ssk, nonblock);
879 retry:
880 sk = netlink_getsockbypid(ssk, pid);
881 if (IS_ERR(sk)) {
882 kfree_skb(skb);
883 return PTR_ERR(sk);
884 }
885 if (netlink_is_kernel(sk))
886 return netlink_unicast_kernel(sk, skb);
887
888 if (sk_filter(sk, skb)) {
889 int err = skb->len;
890 kfree_skb(skb);
891 sock_put(sk);
892 return err;
893 }
894
895 err = netlink_attachskb(sk, skb, nonblock, &timeo, ssk);
896 if (err == 1)
897 goto retry;
898 if (err)
899 return err;
900
901 return netlink_sendskb(sk, skb);
902 }
903 EXPORT_SYMBOL(netlink_unicast);
904
905 int netlink_has_listeners(struct sock *sk, unsigned int group)
906 {
907 int res = 0;
908 unsigned long *listeners;
909
910 BUG_ON(!netlink_is_kernel(sk));
911
912 rcu_read_lock();
913 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
914
915 if (group - 1 < nl_table[sk->sk_protocol].groups)
916 res = test_bit(group - 1, listeners);
917
918 rcu_read_unlock();
919
920 return res;
921 }
922 EXPORT_SYMBOL_GPL(netlink_has_listeners);
923
924 static inline int netlink_broadcast_deliver(struct sock *sk,
925 struct sk_buff *skb)
926 {
927 struct netlink_sock *nlk = nlk_sk(sk);
928
929 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
930 !test_bit(0, &nlk->state)) {
931 skb_set_owner_r(skb, sk);
932 skb_queue_tail(&sk->sk_receive_queue, skb);
933 sk->sk_data_ready(sk, skb->len);
934 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
935 }
936 return -1;
937 }
938
939 struct netlink_broadcast_data {
940 struct sock *exclude_sk;
941 struct net *net;
942 u32 pid;
943 u32 group;
944 int failure;
945 int congested;
946 int delivered;
947 gfp_t allocation;
948 struct sk_buff *skb, *skb2;
949 };
950
951 static inline int do_one_broadcast(struct sock *sk,
952 struct netlink_broadcast_data *p)
953 {
954 struct netlink_sock *nlk = nlk_sk(sk);
955 int val;
956
957 if (p->exclude_sk == sk)
958 goto out;
959
960 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
961 !test_bit(p->group - 1, nlk->groups))
962 goto out;
963
964 if (!net_eq(sock_net(sk), p->net))
965 goto out;
966
967 if (p->failure) {
968 netlink_overrun(sk);
969 goto out;
970 }
971
972 sock_hold(sk);
973 if (p->skb2 == NULL) {
974 if (skb_shared(p->skb)) {
975 p->skb2 = skb_clone(p->skb, p->allocation);
976 } else {
977 p->skb2 = skb_get(p->skb);
978 /*
979 * skb ownership may have been set when
980 * delivered to a previous socket.
981 */
982 skb_orphan(p->skb2);
983 }
984 }
985 if (p->skb2 == NULL) {
986 netlink_overrun(sk);
987 /* Clone failed. Notify ALL listeners. */
988 p->failure = 1;
989 } else if (sk_filter(sk, p->skb2)) {
990 kfree_skb(p->skb2);
991 p->skb2 = NULL;
992 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
993 netlink_overrun(sk);
994 } else {
995 p->congested |= val;
996 p->delivered = 1;
997 p->skb2 = NULL;
998 }
999 sock_put(sk);
1000
1001 out:
1002 return 0;
1003 }
1004
1005 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1006 u32 group, gfp_t allocation)
1007 {
1008 struct net *net = sock_net(ssk);
1009 struct netlink_broadcast_data info;
1010 struct hlist_node *node;
1011 struct sock *sk;
1012
1013 skb = netlink_trim(skb, allocation);
1014
1015 info.exclude_sk = ssk;
1016 info.net = net;
1017 info.pid = pid;
1018 info.group = group;
1019 info.failure = 0;
1020 info.congested = 0;
1021 info.delivered = 0;
1022 info.allocation = allocation;
1023 info.skb = skb;
1024 info.skb2 = NULL;
1025
1026 /* While we sleep in clone, do not allow to change socket list */
1027
1028 netlink_lock_table();
1029
1030 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1031 do_one_broadcast(sk, &info);
1032
1033 kfree_skb(skb);
1034
1035 netlink_unlock_table();
1036
1037 if (info.skb2)
1038 kfree_skb(info.skb2);
1039
1040 if (info.delivered) {
1041 if (info.congested && (allocation & __GFP_WAIT))
1042 yield();
1043 return 0;
1044 }
1045 if (info.failure)
1046 return -ENOBUFS;
1047 return -ESRCH;
1048 }
1049 EXPORT_SYMBOL(netlink_broadcast);
1050
1051 struct netlink_set_err_data {
1052 struct sock *exclude_sk;
1053 u32 pid;
1054 u32 group;
1055 int code;
1056 };
1057
1058 static inline int do_one_set_err(struct sock *sk,
1059 struct netlink_set_err_data *p)
1060 {
1061 struct netlink_sock *nlk = nlk_sk(sk);
1062
1063 if (sk == p->exclude_sk)
1064 goto out;
1065
1066 if (sock_net(sk) != sock_net(p->exclude_sk))
1067 goto out;
1068
1069 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1070 !test_bit(p->group - 1, nlk->groups))
1071 goto out;
1072
1073 sk->sk_err = p->code;
1074 sk->sk_error_report(sk);
1075 out:
1076 return 0;
1077 }
1078
1079 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1080 {
1081 struct netlink_set_err_data info;
1082 struct hlist_node *node;
1083 struct sock *sk;
1084
1085 info.exclude_sk = ssk;
1086 info.pid = pid;
1087 info.group = group;
1088 info.code = code;
1089
1090 read_lock(&nl_table_lock);
1091
1092 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1093 do_one_set_err(sk, &info);
1094
1095 read_unlock(&nl_table_lock);
1096 }
1097
1098 /* must be called with netlink table grabbed */
1099 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1100 unsigned int group,
1101 int is_new)
1102 {
1103 int old, new = !!is_new, subscriptions;
1104
1105 old = test_bit(group - 1, nlk->groups);
1106 subscriptions = nlk->subscriptions - old + new;
1107 if (new)
1108 __set_bit(group - 1, nlk->groups);
1109 else
1110 __clear_bit(group - 1, nlk->groups);
1111 netlink_update_subscriptions(&nlk->sk, subscriptions);
1112 netlink_update_listeners(&nlk->sk);
1113 }
1114
1115 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1116 char __user *optval, int optlen)
1117 {
1118 struct sock *sk = sock->sk;
1119 struct netlink_sock *nlk = nlk_sk(sk);
1120 unsigned int val = 0;
1121 int err;
1122
1123 if (level != SOL_NETLINK)
1124 return -ENOPROTOOPT;
1125
1126 if (optlen >= sizeof(int) &&
1127 get_user(val, (unsigned int __user *)optval))
1128 return -EFAULT;
1129
1130 switch (optname) {
1131 case NETLINK_PKTINFO:
1132 if (val)
1133 nlk->flags |= NETLINK_RECV_PKTINFO;
1134 else
1135 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1136 err = 0;
1137 break;
1138 case NETLINK_ADD_MEMBERSHIP:
1139 case NETLINK_DROP_MEMBERSHIP: {
1140 if (!netlink_capable(sock, NL_NONROOT_RECV))
1141 return -EPERM;
1142 err = netlink_realloc_groups(sk);
1143 if (err)
1144 return err;
1145 if (!val || val - 1 >= nlk->ngroups)
1146 return -EINVAL;
1147 netlink_table_grab();
1148 netlink_update_socket_mc(nlk, val,
1149 optname == NETLINK_ADD_MEMBERSHIP);
1150 netlink_table_ungrab();
1151 err = 0;
1152 break;
1153 }
1154 default:
1155 err = -ENOPROTOOPT;
1156 }
1157 return err;
1158 }
1159
1160 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1161 char __user *optval, int __user *optlen)
1162 {
1163 struct sock *sk = sock->sk;
1164 struct netlink_sock *nlk = nlk_sk(sk);
1165 int len, val, err;
1166
1167 if (level != SOL_NETLINK)
1168 return -ENOPROTOOPT;
1169
1170 if (get_user(len, optlen))
1171 return -EFAULT;
1172 if (len < 0)
1173 return -EINVAL;
1174
1175 switch (optname) {
1176 case NETLINK_PKTINFO:
1177 if (len < sizeof(int))
1178 return -EINVAL;
1179 len = sizeof(int);
1180 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1181 if (put_user(len, optlen) ||
1182 put_user(val, optval))
1183 return -EFAULT;
1184 err = 0;
1185 break;
1186 default:
1187 err = -ENOPROTOOPT;
1188 }
1189 return err;
1190 }
1191
1192 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1193 {
1194 struct nl_pktinfo info;
1195
1196 info.group = NETLINK_CB(skb).dst_group;
1197 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1198 }
1199
1200 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1201 struct msghdr *msg, size_t len)
1202 {
1203 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1204 struct sock *sk = sock->sk;
1205 struct netlink_sock *nlk = nlk_sk(sk);
1206 struct sockaddr_nl *addr = msg->msg_name;
1207 u32 dst_pid;
1208 u32 dst_group;
1209 struct sk_buff *skb;
1210 int err;
1211 struct scm_cookie scm;
1212
1213 if (msg->msg_flags&MSG_OOB)
1214 return -EOPNOTSUPP;
1215
1216 if (NULL == siocb->scm)
1217 siocb->scm = &scm;
1218 err = scm_send(sock, msg, siocb->scm);
1219 if (err < 0)
1220 return err;
1221
1222 if (msg->msg_namelen) {
1223 if (addr->nl_family != AF_NETLINK)
1224 return -EINVAL;
1225 dst_pid = addr->nl_pid;
1226 dst_group = ffs(addr->nl_groups);
1227 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1228 return -EPERM;
1229 } else {
1230 dst_pid = nlk->dst_pid;
1231 dst_group = nlk->dst_group;
1232 }
1233
1234 if (!nlk->pid) {
1235 err = netlink_autobind(sock);
1236 if (err)
1237 goto out;
1238 }
1239
1240 err = -EMSGSIZE;
1241 if (len > sk->sk_sndbuf - 32)
1242 goto out;
1243 err = -ENOBUFS;
1244 skb = alloc_skb(len, GFP_KERNEL);
1245 if (skb == NULL)
1246 goto out;
1247
1248 NETLINK_CB(skb).pid = nlk->pid;
1249 NETLINK_CB(skb).dst_group = dst_group;
1250 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1251 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1252 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1253 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1254
1255 /* What can I do? Netlink is asynchronous, so that
1256 we will have to save current capabilities to
1257 check them, when this message will be delivered
1258 to corresponding kernel module. --ANK (980802)
1259 */
1260
1261 err = -EFAULT;
1262 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1263 kfree_skb(skb);
1264 goto out;
1265 }
1266
1267 err = security_netlink_send(sk, skb);
1268 if (err) {
1269 kfree_skb(skb);
1270 goto out;
1271 }
1272
1273 if (dst_group) {
1274 atomic_inc(&skb->users);
1275 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1276 }
1277 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1278
1279 out:
1280 return err;
1281 }
1282
1283 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1284 struct msghdr *msg, size_t len,
1285 int flags)
1286 {
1287 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1288 struct scm_cookie scm;
1289 struct sock *sk = sock->sk;
1290 struct netlink_sock *nlk = nlk_sk(sk);
1291 int noblock = flags&MSG_DONTWAIT;
1292 size_t copied;
1293 struct sk_buff *skb;
1294 int err;
1295
1296 if (flags&MSG_OOB)
1297 return -EOPNOTSUPP;
1298
1299 copied = 0;
1300
1301 skb = skb_recv_datagram(sk, flags, noblock, &err);
1302 if (skb == NULL)
1303 goto out;
1304
1305 msg->msg_namelen = 0;
1306
1307 copied = skb->len;
1308 if (len < copied) {
1309 msg->msg_flags |= MSG_TRUNC;
1310 copied = len;
1311 }
1312
1313 skb_reset_transport_header(skb);
1314 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1315
1316 if (msg->msg_name) {
1317 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1318 addr->nl_family = AF_NETLINK;
1319 addr->nl_pad = 0;
1320 addr->nl_pid = NETLINK_CB(skb).pid;
1321 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1322 msg->msg_namelen = sizeof(*addr);
1323 }
1324
1325 if (nlk->flags & NETLINK_RECV_PKTINFO)
1326 netlink_cmsg_recv_pktinfo(msg, skb);
1327
1328 if (NULL == siocb->scm) {
1329 memset(&scm, 0, sizeof(scm));
1330 siocb->scm = &scm;
1331 }
1332 siocb->scm->creds = *NETLINK_CREDS(skb);
1333 if (flags & MSG_TRUNC)
1334 copied = skb->len;
1335 skb_free_datagram(sk, skb);
1336
1337 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1338 netlink_dump(sk);
1339
1340 scm_recv(sock, msg, siocb->scm, flags);
1341 out:
1342 netlink_rcv_wake(sk);
1343 return err ? : copied;
1344 }
1345
1346 static void netlink_data_ready(struct sock *sk, int len)
1347 {
1348 BUG();
1349 }
1350
1351 /*
1352 * We export these functions to other modules. They provide a
1353 * complete set of kernel non-blocking support for message
1354 * queueing.
1355 */
1356
1357 struct sock *
1358 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1359 void (*input)(struct sk_buff *skb),
1360 struct mutex *cb_mutex, struct module *module)
1361 {
1362 struct socket *sock;
1363 struct sock *sk;
1364 struct netlink_sock *nlk;
1365 unsigned long *listeners = NULL;
1366
1367 BUG_ON(!nl_table);
1368
1369 if (unit < 0 || unit >= MAX_LINKS)
1370 return NULL;
1371
1372 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1373 return NULL;
1374
1375 /*
1376 * We have to just have a reference on the net from sk, but don't
1377 * get_net it. Besides, we cannot get and then put the net here.
1378 * So we create one inside init_net and the move it to net.
1379 */
1380
1381 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1382 goto out_sock_release_nosk;
1383
1384 sk = sock->sk;
1385 sk_change_net(sk, net);
1386
1387 if (groups < 32)
1388 groups = 32;
1389
1390 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1391 if (!listeners)
1392 goto out_sock_release;
1393
1394 sk->sk_data_ready = netlink_data_ready;
1395 if (input)
1396 nlk_sk(sk)->netlink_rcv = input;
1397
1398 if (netlink_insert(sk, net, 0))
1399 goto out_sock_release;
1400
1401 nlk = nlk_sk(sk);
1402 nlk->flags |= NETLINK_KERNEL_SOCKET;
1403
1404 netlink_table_grab();
1405 if (!nl_table[unit].registered) {
1406 nl_table[unit].groups = groups;
1407 nl_table[unit].listeners = listeners;
1408 nl_table[unit].cb_mutex = cb_mutex;
1409 nl_table[unit].module = module;
1410 nl_table[unit].registered = 1;
1411 } else {
1412 kfree(listeners);
1413 nl_table[unit].registered++;
1414 }
1415 netlink_table_ungrab();
1416 return sk;
1417
1418 out_sock_release:
1419 kfree(listeners);
1420 netlink_kernel_release(sk);
1421 return NULL;
1422
1423 out_sock_release_nosk:
1424 sock_release(sock);
1425 return NULL;
1426 }
1427 EXPORT_SYMBOL(netlink_kernel_create);
1428
1429
1430 void
1431 netlink_kernel_release(struct sock *sk)
1432 {
1433 sk_release_kernel(sk);
1434 }
1435 EXPORT_SYMBOL(netlink_kernel_release);
1436
1437
1438 /**
1439 * netlink_change_ngroups - change number of multicast groups
1440 *
1441 * This changes the number of multicast groups that are available
1442 * on a certain netlink family. Note that it is not possible to
1443 * change the number of groups to below 32. Also note that it does
1444 * not implicitly call netlink_clear_multicast_users() when the
1445 * number of groups is reduced.
1446 *
1447 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1448 * @groups: The new number of groups.
1449 */
1450 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1451 {
1452 unsigned long *listeners, *old = NULL;
1453 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1454 int err = 0;
1455
1456 if (groups < 32)
1457 groups = 32;
1458
1459 netlink_table_grab();
1460 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1461 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1462 if (!listeners) {
1463 err = -ENOMEM;
1464 goto out_ungrab;
1465 }
1466 old = tbl->listeners;
1467 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1468 rcu_assign_pointer(tbl->listeners, listeners);
1469 }
1470 tbl->groups = groups;
1471
1472 out_ungrab:
1473 netlink_table_ungrab();
1474 synchronize_rcu();
1475 kfree(old);
1476 return err;
1477 }
1478 EXPORT_SYMBOL(netlink_change_ngroups);
1479
1480 /**
1481 * netlink_clear_multicast_users - kick off multicast listeners
1482 *
1483 * This function removes all listeners from the given group.
1484 * @ksk: The kernel netlink socket, as returned by
1485 * netlink_kernel_create().
1486 * @group: The multicast group to clear.
1487 */
1488 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1489 {
1490 struct sock *sk;
1491 struct hlist_node *node;
1492 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1493
1494 netlink_table_grab();
1495
1496 sk_for_each_bound(sk, node, &tbl->mc_list)
1497 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1498
1499 netlink_table_ungrab();
1500 }
1501 EXPORT_SYMBOL(netlink_clear_multicast_users);
1502
1503 void netlink_set_nonroot(int protocol, unsigned int flags)
1504 {
1505 if ((unsigned int)protocol < MAX_LINKS)
1506 nl_table[protocol].nl_nonroot = flags;
1507 }
1508 EXPORT_SYMBOL(netlink_set_nonroot);
1509
1510 static void netlink_destroy_callback(struct netlink_callback *cb)
1511 {
1512 if (cb->skb)
1513 kfree_skb(cb->skb);
1514 kfree(cb);
1515 }
1516
1517 /*
1518 * It looks a bit ugly.
1519 * It would be better to create kernel thread.
1520 */
1521
1522 static int netlink_dump(struct sock *sk)
1523 {
1524 struct netlink_sock *nlk = nlk_sk(sk);
1525 struct netlink_callback *cb;
1526 struct sk_buff *skb;
1527 struct nlmsghdr *nlh;
1528 int len, err = -ENOBUFS;
1529
1530 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1531 if (!skb)
1532 goto errout;
1533
1534 mutex_lock(nlk->cb_mutex);
1535
1536 cb = nlk->cb;
1537 if (cb == NULL) {
1538 err = -EINVAL;
1539 goto errout_skb;
1540 }
1541
1542 len = cb->dump(skb, cb);
1543
1544 if (len > 0) {
1545 mutex_unlock(nlk->cb_mutex);
1546
1547 if (sk_filter(sk, skb))
1548 kfree_skb(skb);
1549 else {
1550 skb_queue_tail(&sk->sk_receive_queue, skb);
1551 sk->sk_data_ready(sk, skb->len);
1552 }
1553 return 0;
1554 }
1555
1556 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1557 if (!nlh)
1558 goto errout_skb;
1559
1560 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1561
1562 if (sk_filter(sk, skb))
1563 kfree_skb(skb);
1564 else {
1565 skb_queue_tail(&sk->sk_receive_queue, skb);
1566 sk->sk_data_ready(sk, skb->len);
1567 }
1568
1569 if (cb->done)
1570 cb->done(cb);
1571 nlk->cb = NULL;
1572 mutex_unlock(nlk->cb_mutex);
1573
1574 netlink_destroy_callback(cb);
1575 return 0;
1576
1577 errout_skb:
1578 mutex_unlock(nlk->cb_mutex);
1579 kfree_skb(skb);
1580 errout:
1581 return err;
1582 }
1583
1584 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1585 struct nlmsghdr *nlh,
1586 int (*dump)(struct sk_buff *skb,
1587 struct netlink_callback *),
1588 int (*done)(struct netlink_callback *))
1589 {
1590 struct netlink_callback *cb;
1591 struct sock *sk;
1592 struct netlink_sock *nlk;
1593
1594 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1595 if (cb == NULL)
1596 return -ENOBUFS;
1597
1598 cb->dump = dump;
1599 cb->done = done;
1600 cb->nlh = nlh;
1601 atomic_inc(&skb->users);
1602 cb->skb = skb;
1603
1604 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1605 if (sk == NULL) {
1606 netlink_destroy_callback(cb);
1607 return -ECONNREFUSED;
1608 }
1609 nlk = nlk_sk(sk);
1610 /* A dump is in progress... */
1611 mutex_lock(nlk->cb_mutex);
1612 if (nlk->cb) {
1613 mutex_unlock(nlk->cb_mutex);
1614 netlink_destroy_callback(cb);
1615 sock_put(sk);
1616 return -EBUSY;
1617 }
1618 nlk->cb = cb;
1619 mutex_unlock(nlk->cb_mutex);
1620
1621 netlink_dump(sk);
1622 sock_put(sk);
1623
1624 /* We successfully started a dump, by returning -EINTR we
1625 * signal not to send ACK even if it was requested.
1626 */
1627 return -EINTR;
1628 }
1629 EXPORT_SYMBOL(netlink_dump_start);
1630
1631 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1632 {
1633 struct sk_buff *skb;
1634 struct nlmsghdr *rep;
1635 struct nlmsgerr *errmsg;
1636 size_t payload = sizeof(*errmsg);
1637
1638 /* error messages get the original request appened */
1639 if (err)
1640 payload += nlmsg_len(nlh);
1641
1642 skb = nlmsg_new(payload, GFP_KERNEL);
1643 if (!skb) {
1644 struct sock *sk;
1645
1646 sk = netlink_lookup(sock_net(in_skb->sk),
1647 in_skb->sk->sk_protocol,
1648 NETLINK_CB(in_skb).pid);
1649 if (sk) {
1650 sk->sk_err = ENOBUFS;
1651 sk->sk_error_report(sk);
1652 sock_put(sk);
1653 }
1654 return;
1655 }
1656
1657 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1658 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1659 errmsg = nlmsg_data(rep);
1660 errmsg->error = err;
1661 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1662 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1663 }
1664 EXPORT_SYMBOL(netlink_ack);
1665
1666 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1667 struct nlmsghdr *))
1668 {
1669 struct nlmsghdr *nlh;
1670 int err;
1671
1672 while (skb->len >= nlmsg_total_size(0)) {
1673 int msglen;
1674
1675 nlh = nlmsg_hdr(skb);
1676 err = 0;
1677
1678 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1679 return 0;
1680
1681 /* Only requests are handled by the kernel */
1682 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1683 goto ack;
1684
1685 /* Skip control messages */
1686 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1687 goto ack;
1688
1689 err = cb(skb, nlh);
1690 if (err == -EINTR)
1691 goto skip;
1692
1693 ack:
1694 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1695 netlink_ack(skb, nlh, err);
1696
1697 skip:
1698 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1699 if (msglen > skb->len)
1700 msglen = skb->len;
1701 skb_pull(skb, msglen);
1702 }
1703
1704 return 0;
1705 }
1706 EXPORT_SYMBOL(netlink_rcv_skb);
1707
1708 /**
1709 * nlmsg_notify - send a notification netlink message
1710 * @sk: netlink socket to use
1711 * @skb: notification message
1712 * @pid: destination netlink pid for reports or 0
1713 * @group: destination multicast group or 0
1714 * @report: 1 to report back, 0 to disable
1715 * @flags: allocation flags
1716 */
1717 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1718 unsigned int group, int report, gfp_t flags)
1719 {
1720 int err = 0;
1721
1722 if (group) {
1723 int exclude_pid = 0;
1724
1725 if (report) {
1726 atomic_inc(&skb->users);
1727 exclude_pid = pid;
1728 }
1729
1730 /* errors reported via destination sk->sk_err */
1731 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1732 }
1733
1734 if (report)
1735 err = nlmsg_unicast(sk, skb, pid);
1736
1737 return err;
1738 }
1739 EXPORT_SYMBOL(nlmsg_notify);
1740
1741 #ifdef CONFIG_PROC_FS
1742 struct nl_seq_iter {
1743 struct seq_net_private p;
1744 int link;
1745 int hash_idx;
1746 };
1747
1748 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1749 {
1750 struct nl_seq_iter *iter = seq->private;
1751 int i, j;
1752 struct sock *s;
1753 struct hlist_node *node;
1754 loff_t off = 0;
1755
1756 for (i = 0; i < MAX_LINKS; i++) {
1757 struct nl_pid_hash *hash = &nl_table[i].hash;
1758
1759 for (j = 0; j <= hash->mask; j++) {
1760 sk_for_each(s, node, &hash->table[j]) {
1761 if (sock_net(s) != seq_file_net(seq))
1762 continue;
1763 if (off == pos) {
1764 iter->link = i;
1765 iter->hash_idx = j;
1766 return s;
1767 }
1768 ++off;
1769 }
1770 }
1771 }
1772 return NULL;
1773 }
1774
1775 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1776 __acquires(nl_table_lock)
1777 {
1778 read_lock(&nl_table_lock);
1779 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1780 }
1781
1782 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1783 {
1784 struct sock *s;
1785 struct nl_seq_iter *iter;
1786 int i, j;
1787
1788 ++*pos;
1789
1790 if (v == SEQ_START_TOKEN)
1791 return netlink_seq_socket_idx(seq, 0);
1792
1793 iter = seq->private;
1794 s = v;
1795 do {
1796 s = sk_next(s);
1797 } while (s && sock_net(s) != seq_file_net(seq));
1798 if (s)
1799 return s;
1800
1801 i = iter->link;
1802 j = iter->hash_idx + 1;
1803
1804 do {
1805 struct nl_pid_hash *hash = &nl_table[i].hash;
1806
1807 for (; j <= hash->mask; j++) {
1808 s = sk_head(&hash->table[j]);
1809 while (s && sock_net(s) != seq_file_net(seq))
1810 s = sk_next(s);
1811 if (s) {
1812 iter->link = i;
1813 iter->hash_idx = j;
1814 return s;
1815 }
1816 }
1817
1818 j = 0;
1819 } while (++i < MAX_LINKS);
1820
1821 return NULL;
1822 }
1823
1824 static void netlink_seq_stop(struct seq_file *seq, void *v)
1825 __releases(nl_table_lock)
1826 {
1827 read_unlock(&nl_table_lock);
1828 }
1829
1830
1831 static int netlink_seq_show(struct seq_file *seq, void *v)
1832 {
1833 if (v == SEQ_START_TOKEN)
1834 seq_puts(seq,
1835 "sk Eth Pid Groups "
1836 "Rmem Wmem Dump Locks\n");
1837 else {
1838 struct sock *s = v;
1839 struct netlink_sock *nlk = nlk_sk(s);
1840
1841 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1842 s,
1843 s->sk_protocol,
1844 nlk->pid,
1845 nlk->groups ? (u32)nlk->groups[0] : 0,
1846 atomic_read(&s->sk_rmem_alloc),
1847 atomic_read(&s->sk_wmem_alloc),
1848 nlk->cb,
1849 atomic_read(&s->sk_refcnt)
1850 );
1851
1852 }
1853 return 0;
1854 }
1855
1856 static const struct seq_operations netlink_seq_ops = {
1857 .start = netlink_seq_start,
1858 .next = netlink_seq_next,
1859 .stop = netlink_seq_stop,
1860 .show = netlink_seq_show,
1861 };
1862
1863
1864 static int netlink_seq_open(struct inode *inode, struct file *file)
1865 {
1866 return seq_open_net(inode, file, &netlink_seq_ops,
1867 sizeof(struct nl_seq_iter));
1868 }
1869
1870 static const struct file_operations netlink_seq_fops = {
1871 .owner = THIS_MODULE,
1872 .open = netlink_seq_open,
1873 .read = seq_read,
1874 .llseek = seq_lseek,
1875 .release = seq_release_net,
1876 };
1877
1878 #endif
1879
1880 int netlink_register_notifier(struct notifier_block *nb)
1881 {
1882 return atomic_notifier_chain_register(&netlink_chain, nb);
1883 }
1884 EXPORT_SYMBOL(netlink_register_notifier);
1885
1886 int netlink_unregister_notifier(struct notifier_block *nb)
1887 {
1888 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1889 }
1890 EXPORT_SYMBOL(netlink_unregister_notifier);
1891
1892 static const struct proto_ops netlink_ops = {
1893 .family = PF_NETLINK,
1894 .owner = THIS_MODULE,
1895 .release = netlink_release,
1896 .bind = netlink_bind,
1897 .connect = netlink_connect,
1898 .socketpair = sock_no_socketpair,
1899 .accept = sock_no_accept,
1900 .getname = netlink_getname,
1901 .poll = datagram_poll,
1902 .ioctl = sock_no_ioctl,
1903 .listen = sock_no_listen,
1904 .shutdown = sock_no_shutdown,
1905 .setsockopt = netlink_setsockopt,
1906 .getsockopt = netlink_getsockopt,
1907 .sendmsg = netlink_sendmsg,
1908 .recvmsg = netlink_recvmsg,
1909 .mmap = sock_no_mmap,
1910 .sendpage = sock_no_sendpage,
1911 };
1912
1913 static struct net_proto_family netlink_family_ops = {
1914 .family = PF_NETLINK,
1915 .create = netlink_create,
1916 .owner = THIS_MODULE, /* for consistency 8) */
1917 };
1918
1919 static int __net_init netlink_net_init(struct net *net)
1920 {
1921 #ifdef CONFIG_PROC_FS
1922 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1923 return -ENOMEM;
1924 #endif
1925 return 0;
1926 }
1927
1928 static void __net_exit netlink_net_exit(struct net *net)
1929 {
1930 #ifdef CONFIG_PROC_FS
1931 proc_net_remove(net, "netlink");
1932 #endif
1933 }
1934
1935 static struct pernet_operations __net_initdata netlink_net_ops = {
1936 .init = netlink_net_init,
1937 .exit = netlink_net_exit,
1938 };
1939
1940 static int __init netlink_proto_init(void)
1941 {
1942 struct sk_buff *dummy_skb;
1943 int i;
1944 unsigned long limit;
1945 unsigned int order;
1946 int err = proto_register(&netlink_proto, 0);
1947
1948 if (err != 0)
1949 goto out;
1950
1951 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1952
1953 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1954 if (!nl_table)
1955 goto panic;
1956
1957 if (num_physpages >= (128 * 1024))
1958 limit = num_physpages >> (21 - PAGE_SHIFT);
1959 else
1960 limit = num_physpages >> (23 - PAGE_SHIFT);
1961
1962 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1963 limit = (1UL << order) / sizeof(struct hlist_head);
1964 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1965
1966 for (i = 0; i < MAX_LINKS; i++) {
1967 struct nl_pid_hash *hash = &nl_table[i].hash;
1968
1969 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1970 if (!hash->table) {
1971 while (i-- > 0)
1972 nl_pid_hash_free(nl_table[i].hash.table,
1973 1 * sizeof(*hash->table));
1974 kfree(nl_table);
1975 goto panic;
1976 }
1977 hash->max_shift = order;
1978 hash->shift = 0;
1979 hash->mask = 0;
1980 hash->rehash_time = jiffies;
1981 }
1982
1983 sock_register(&netlink_family_ops);
1984 register_pernet_subsys(&netlink_net_ops);
1985 /* The netlink device handler may be needed early. */
1986 rtnetlink_init();
1987 out:
1988 return err;
1989 panic:
1990 panic("netlink_init: Cannot allocate nl_table\n");
1991 }
1992
1993 core_initcall(netlink_proto_init);
linux-next