?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
p54pci: move tx cleanup into tasklet
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / netlink / af_netlink.c
bloba4957bf2ca60d6963e2d94f9b2e7679df3a0b4b6
1 /*
2 * NETLINK Kernel-user communication protocol.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 *
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
22 */
23
24 #include <linux/module.h>
25
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
35 #include <linux/un.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
40 #include <linux/fs.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
54 #include <linux/mm.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/mutex.h>
58
59 #include <net/net_namespace.h>
60 #include <net/sock.h>
61 #include <net/scm.h>
62 #include <net/netlink.h>
63
64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
65 #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
66
67 struct netlink_sock {
68 /* struct sock has to be the first member of netlink_sock */
69 struct sock sk;
70 u32 pid;
71 u32 dst_pid;
72 u32 dst_group;
73 u32 flags;
74 u32 subscriptions;
75 u32 ngroups;
76 unsigned long *groups;
77 unsigned long state;
78 wait_queue_head_t wait;
79 struct netlink_callback *cb;
80 struct mutex *cb_mutex;
81 struct mutex cb_def_mutex;
82 void (*netlink_rcv)(struct sk_buff *skb);
83 struct module *module;
84 };
85
86 struct listeners_rcu_head {
87 struct rcu_head rcu_head;
88 void *ptr;
89 };
90
91 #define NETLINK_KERNEL_SOCKET 0x1
92 #define NETLINK_RECV_PKTINFO 0x2
93 #define NETLINK_BROADCAST_SEND_ERROR 0x4
94 #define NETLINK_RECV_NO_ENOBUFS 0x8
95
96 static inline struct netlink_sock *nlk_sk(struct sock *sk)
97 {
98 return container_of(sk, struct netlink_sock, sk);
99 }
100
101 static inline int netlink_is_kernel(struct sock *sk)
102 {
103 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
104 }
105
106 struct nl_pid_hash {
107 struct hlist_head *table;
108 unsigned long rehash_time;
109
110 unsigned int mask;
111 unsigned int shift;
112
113 unsigned int entries;
114 unsigned int max_shift;
115
116 u32 rnd;
117 };
118
119 struct netlink_table {
120 struct nl_pid_hash hash;
121 struct hlist_head mc_list;
122 unsigned long *listeners;
123 unsigned int nl_nonroot;
124 unsigned int groups;
125 struct mutex *cb_mutex;
126 struct module *module;
127 int registered;
128 };
129
130 static struct netlink_table *nl_table;
131
132 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
133
134 static int netlink_dump(struct sock *sk);
135 static void netlink_destroy_callback(struct netlink_callback *cb);
136
137 static DEFINE_RWLOCK(nl_table_lock);
138 static atomic_t nl_table_users = ATOMIC_INIT(0);
139
140 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
141
142 static u32 netlink_group_mask(u32 group)
143 {
144 return group ? 1 << (group - 1) : 0;
145 }
146
147 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
148 {
149 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
150 }
151
152 static void netlink_sock_destruct(struct sock *sk)
153 {
154 struct netlink_sock *nlk = nlk_sk(sk);
155
156 if (nlk->cb) {
157 if (nlk->cb->done)
158 nlk->cb->done(nlk->cb);
159 netlink_destroy_callback(nlk->cb);
160 }
161
162 skb_queue_purge(&sk->sk_receive_queue);
163
164 if (!sock_flag(sk, SOCK_DEAD)) {
165 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
166 return;
167 }
168
169 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
170 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
171 WARN_ON(nlk_sk(sk)->groups);
172 }
173
174 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
175 * SMP. Look, when several writers sleep and reader wakes them up, all but one
176 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
177 * this, _but_ remember, it adds useless work on UP machines.
178 */
179
180 void netlink_table_grab(void)
181 __acquires(nl_table_lock)
182 {
183 might_sleep();
184
185 write_lock_irq(&nl_table_lock);
186
187 if (atomic_read(&nl_table_users)) {
188 DECLARE_WAITQUEUE(wait, current);
189
190 add_wait_queue_exclusive(&nl_table_wait, &wait);
191 for (;;) {
192 set_current_state(TASK_UNINTERRUPTIBLE);
193 if (atomic_read(&nl_table_users) == 0)
194 break;
195 write_unlock_irq(&nl_table_lock);
196 schedule();
197 write_lock_irq(&nl_table_lock);
198 }
199
200 __set_current_state(TASK_RUNNING);
201 remove_wait_queue(&nl_table_wait, &wait);
202 }
203 }
204
205 void netlink_table_ungrab(void)
206 __releases(nl_table_lock)
207 {
208 write_unlock_irq(&nl_table_lock);
209 wake_up(&nl_table_wait);
210 }
211
212 static inline void
213 netlink_lock_table(void)
214 {
215 /* read_lock() synchronizes us to netlink_table_grab */
216
217 read_lock(&nl_table_lock);
218 atomic_inc(&nl_table_users);
219 read_unlock(&nl_table_lock);
220 }
221
222 static inline void
223 netlink_unlock_table(void)
224 {
225 if (atomic_dec_and_test(&nl_table_users))
226 wake_up(&nl_table_wait);
227 }
228
229 static inline struct sock *netlink_lookup(struct net *net, int protocol,
230 u32 pid)
231 {
232 struct nl_pid_hash *hash = &nl_table[protocol].hash;
233 struct hlist_head *head;
234 struct sock *sk;
235 struct hlist_node *node;
236
237 read_lock(&nl_table_lock);
238 head = nl_pid_hashfn(hash, pid);
239 sk_for_each(sk, node, head) {
240 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
241 sock_hold(sk);
242 goto found;
243 }
244 }
245 sk = NULL;
246 found:
247 read_unlock(&nl_table_lock);
248 return sk;
249 }
250
251 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
252 {
253 if (size <= PAGE_SIZE)
254 return kzalloc(size, GFP_ATOMIC);
255 else
256 return (struct hlist_head *)
257 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
258 get_order(size));
259 }
260
261 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
262 {
263 if (size <= PAGE_SIZE)
264 kfree(table);
265 else
266 free_pages((unsigned long)table, get_order(size));
267 }
268
269 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
270 {
271 unsigned int omask, mask, shift;
272 size_t osize, size;
273 struct hlist_head *otable, *table;
274 int i;
275
276 omask = mask = hash->mask;
277 osize = size = (mask + 1) * sizeof(*table);
278 shift = hash->shift;
279
280 if (grow) {
281 if (++shift > hash->max_shift)
282 return 0;
283 mask = mask * 2 + 1;
284 size *= 2;
285 }
286
287 table = nl_pid_hash_zalloc(size);
288 if (!table)
289 return 0;
290
291 otable = hash->table;
292 hash->table = table;
293 hash->mask = mask;
294 hash->shift = shift;
295 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
296
297 for (i = 0; i <= omask; i++) {
298 struct sock *sk;
299 struct hlist_node *node, *tmp;
300
301 sk_for_each_safe(sk, node, tmp, &otable[i])
302 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
303 }
304
305 nl_pid_hash_free(otable, osize);
306 hash->rehash_time = jiffies + 10 * 60 * HZ;
307 return 1;
308 }
309
310 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
311 {
312 int avg = hash->entries >> hash->shift;
313
314 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
315 return 1;
316
317 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
318 nl_pid_hash_rehash(hash, 0);
319 return 1;
320 }
321
322 return 0;
323 }
324
325 static const struct proto_ops netlink_ops;
326
327 static void
328 netlink_update_listeners(struct sock *sk)
329 {
330 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
331 struct hlist_node *node;
332 unsigned long mask;
333 unsigned int i;
334
335 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
336 mask = 0;
337 sk_for_each_bound(sk, node, &tbl->mc_list) {
338 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
339 mask |= nlk_sk(sk)->groups[i];
340 }
341 tbl->listeners[i] = mask;
342 }
343 /* this function is only called with the netlink table "grabbed", which
344 * makes sure updates are visible before bind or setsockopt return. */
345 }
346
347 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
348 {
349 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
350 struct hlist_head *head;
351 int err = -EADDRINUSE;
352 struct sock *osk;
353 struct hlist_node *node;
354 int len;
355
356 netlink_table_grab();
357 head = nl_pid_hashfn(hash, pid);
358 len = 0;
359 sk_for_each(osk, node, head) {
360 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
361 break;
362 len++;
363 }
364 if (node)
365 goto err;
366
367 err = -EBUSY;
368 if (nlk_sk(sk)->pid)
369 goto err;
370
371 err = -ENOMEM;
372 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
373 goto err;
374
375 if (len && nl_pid_hash_dilute(hash, len))
376 head = nl_pid_hashfn(hash, pid);
377 hash->entries++;
378 nlk_sk(sk)->pid = pid;
379 sk_add_node(sk, head);
380 err = 0;
381
382 err:
383 netlink_table_ungrab();
384 return err;
385 }
386
387 static void netlink_remove(struct sock *sk)
388 {
389 netlink_table_grab();
390 if (sk_del_node_init(sk))
391 nl_table[sk->sk_protocol].hash.entries--;
392 if (nlk_sk(sk)->subscriptions)
393 __sk_del_bind_node(sk);
394 netlink_table_ungrab();
395 }
396
397 static struct proto netlink_proto = {
398 .name = "NETLINK",
399 .owner = THIS_MODULE,
400 .obj_size = sizeof(struct netlink_sock),
401 };
402
403 static int __netlink_create(struct net *net, struct socket *sock,
404 struct mutex *cb_mutex, int protocol)
405 {
406 struct sock *sk;
407 struct netlink_sock *nlk;
408
409 sock->ops = &netlink_ops;
410
411 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
412 if (!sk)
413 return -ENOMEM;
414
415 sock_init_data(sock, sk);
416
417 nlk = nlk_sk(sk);
418 if (cb_mutex)
419 nlk->cb_mutex = cb_mutex;
420 else {
421 nlk->cb_mutex = &nlk->cb_def_mutex;
422 mutex_init(nlk->cb_mutex);
423 }
424 init_waitqueue_head(&nlk->wait);
425
426 sk->sk_destruct = netlink_sock_destruct;
427 sk->sk_protocol = protocol;
428 return 0;
429 }
430
431 static int netlink_create(struct net *net, struct socket *sock, int protocol,
432 int kern)
433 {
434 struct module *module = NULL;
435 struct mutex *cb_mutex;
436 struct netlink_sock *nlk;
437 int err = 0;
438
439 sock->state = SS_UNCONNECTED;
440
441 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
442 return -ESOCKTNOSUPPORT;
443
444 if (protocol < 0 || protocol >= MAX_LINKS)
445 return -EPROTONOSUPPORT;
446
447 netlink_lock_table();
448 #ifdef CONFIG_MODULES
449 if (!nl_table[protocol].registered) {
450 netlink_unlock_table();
451 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
452 netlink_lock_table();
453 }
454 #endif
455 if (nl_table[protocol].registered &&
456 try_module_get(nl_table[protocol].module))
457 module = nl_table[protocol].module;
458 cb_mutex = nl_table[protocol].cb_mutex;
459 netlink_unlock_table();
460
461 err = __netlink_create(net, sock, cb_mutex, protocol);
462 if (err < 0)
463 goto out_module;
464
465 local_bh_disable();
466 sock_prot_inuse_add(net, &netlink_proto, 1);
467 local_bh_enable();
468
469 nlk = nlk_sk(sock->sk);
470 nlk->module = module;
471 out:
472 return err;
473
474 out_module:
475 module_put(module);
476 goto out;
477 }
478
479 static int netlink_release(struct socket *sock)
480 {
481 struct sock *sk = sock->sk;
482 struct netlink_sock *nlk;
483
484 if (!sk)
485 return 0;
486
487 netlink_remove(sk);
488 sock_orphan(sk);
489 nlk = nlk_sk(sk);
490
491 /*
492 * OK. Socket is unlinked, any packets that arrive now
493 * will be purged.
494 */
495
496 sock->sk = NULL;
497 wake_up_interruptible_all(&nlk->wait);
498
499 skb_queue_purge(&sk->sk_write_queue);
500
501 if (nlk->pid) {
502 struct netlink_notify n = {
503 .net = sock_net(sk),
504 .protocol = sk->sk_protocol,
505 .pid = nlk->pid,
506 };
507 atomic_notifier_call_chain(&netlink_chain,
508 NETLINK_URELEASE, &n);
509 }
510
511 module_put(nlk->module);
512
513 netlink_table_grab();
514 if (netlink_is_kernel(sk)) {
515 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
516 if (--nl_table[sk->sk_protocol].registered == 0) {
517 kfree(nl_table[sk->sk_protocol].listeners);
518 nl_table[sk->sk_protocol].module = NULL;
519 nl_table[sk->sk_protocol].registered = 0;
520 }
521 } else if (nlk->subscriptions)
522 netlink_update_listeners(sk);
523 netlink_table_ungrab();
524
525 kfree(nlk->groups);
526 nlk->groups = NULL;
527
528 local_bh_disable();
529 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
530 local_bh_enable();
531 sock_put(sk);
532 return 0;
533 }
534
535 static int netlink_autobind(struct socket *sock)
536 {
537 struct sock *sk = sock->sk;
538 struct net *net = sock_net(sk);
539 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
540 struct hlist_head *head;
541 struct sock *osk;
542 struct hlist_node *node;
543 s32 pid = current->tgid;
544 int err;
545 static s32 rover = -4097;
546
547 retry:
548 cond_resched();
549 netlink_table_grab();
550 head = nl_pid_hashfn(hash, pid);
551 sk_for_each(osk, node, head) {
552 if (!net_eq(sock_net(osk), net))
553 continue;
554 if (nlk_sk(osk)->pid == pid) {
555 /* Bind collision, search negative pid values. */
556 pid = rover--;
557 if (rover > -4097)
558 rover = -4097;
559 netlink_table_ungrab();
560 goto retry;
561 }
562 }
563 netlink_table_ungrab();
564
565 err = netlink_insert(sk, net, pid);
566 if (err == -EADDRINUSE)
567 goto retry;
568
569 /* If 2 threads race to autobind, that is fine. */
570 if (err == -EBUSY)
571 err = 0;
572
573 return err;
574 }
575
576 static inline int netlink_capable(struct socket *sock, unsigned int flag)
577 {
578 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
579 capable(CAP_NET_ADMIN);
580 }
581
582 static void
583 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
584 {
585 struct netlink_sock *nlk = nlk_sk(sk);
586
587 if (nlk->subscriptions && !subscriptions)
588 __sk_del_bind_node(sk);
589 else if (!nlk->subscriptions && subscriptions)
590 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
591 nlk->subscriptions = subscriptions;
592 }
593
594 static int netlink_realloc_groups(struct sock *sk)
595 {
596 struct netlink_sock *nlk = nlk_sk(sk);
597 unsigned int groups;
598 unsigned long *new_groups;
599 int err = 0;
600
601 netlink_table_grab();
602
603 groups = nl_table[sk->sk_protocol].groups;
604 if (!nl_table[sk->sk_protocol].registered) {
605 err = -ENOENT;
606 goto out_unlock;
607 }
608
609 if (nlk->ngroups >= groups)
610 goto out_unlock;
611
612 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
613 if (new_groups == NULL) {
614 err = -ENOMEM;
615 goto out_unlock;
616 }
617 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
618 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
619
620 nlk->groups = new_groups;
621 nlk->ngroups = groups;
622 out_unlock:
623 netlink_table_ungrab();
624 return err;
625 }
626
627 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
628 int addr_len)
629 {
630 struct sock *sk = sock->sk;
631 struct net *net = sock_net(sk);
632 struct netlink_sock *nlk = nlk_sk(sk);
633 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
634 int err;
635
636 if (nladdr->nl_family != AF_NETLINK)
637 return -EINVAL;
638
639 /* Only superuser is allowed to listen multicasts */
640 if (nladdr->nl_groups) {
641 if (!netlink_capable(sock, NL_NONROOT_RECV))
642 return -EPERM;
643 err = netlink_realloc_groups(sk);
644 if (err)
645 return err;
646 }
647
648 if (nlk->pid) {
649 if (nladdr->nl_pid != nlk->pid)
650 return -EINVAL;
651 } else {
652 err = nladdr->nl_pid ?
653 netlink_insert(sk, net, nladdr->nl_pid) :
654 netlink_autobind(sock);
655 if (err)
656 return err;
657 }
658
659 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
660 return 0;
661
662 netlink_table_grab();
663 netlink_update_subscriptions(sk, nlk->subscriptions +
664 hweight32(nladdr->nl_groups) -
665 hweight32(nlk->groups[0]));
666 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
667 netlink_update_listeners(sk);
668 netlink_table_ungrab();
669
670 return 0;
671 }
672
673 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
674 int alen, int flags)
675 {
676 int err = 0;
677 struct sock *sk = sock->sk;
678 struct netlink_sock *nlk = nlk_sk(sk);
679 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
680
681 if (addr->sa_family == AF_UNSPEC) {
682 sk->sk_state = NETLINK_UNCONNECTED;
683 nlk->dst_pid = 0;
684 nlk->dst_group = 0;
685 return 0;
686 }
687 if (addr->sa_family != AF_NETLINK)
688 return -EINVAL;
689
690 /* Only superuser is allowed to send multicasts */
691 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
692 return -EPERM;
693
694 if (!nlk->pid)
695 err = netlink_autobind(sock);
696
697 if (err == 0) {
698 sk->sk_state = NETLINK_CONNECTED;
699 nlk->dst_pid = nladdr->nl_pid;
700 nlk->dst_group = ffs(nladdr->nl_groups);
701 }
702
703 return err;
704 }
705
706 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
707 int *addr_len, int peer)
708 {
709 struct sock *sk = sock->sk;
710 struct netlink_sock *nlk = nlk_sk(sk);
711 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
712
713 nladdr->nl_family = AF_NETLINK;
714 nladdr->nl_pad = 0;
715 *addr_len = sizeof(*nladdr);
716
717 if (peer) {
718 nladdr->nl_pid = nlk->dst_pid;
719 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
720 } else {
721 nladdr->nl_pid = nlk->pid;
722 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
723 }
724 return 0;
725 }
726
727 static void netlink_overrun(struct sock *sk)
728 {
729 struct netlink_sock *nlk = nlk_sk(sk);
730
731 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
732 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
733 sk->sk_err = ENOBUFS;
734 sk->sk_error_report(sk);
735 }
736 }
737 atomic_inc(&sk->sk_drops);
738 }
739
740 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
741 {
742 struct sock *sock;
743 struct netlink_sock *nlk;
744
745 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
746 if (!sock)
747 return ERR_PTR(-ECONNREFUSED);
748
749 /* Don't bother queuing skb if kernel socket has no input function */
750 nlk = nlk_sk(sock);
751 if (sock->sk_state == NETLINK_CONNECTED &&
752 nlk->dst_pid != nlk_sk(ssk)->pid) {
753 sock_put(sock);
754 return ERR_PTR(-ECONNREFUSED);
755 }
756 return sock;
757 }
758
759 struct sock *netlink_getsockbyfilp(struct file *filp)
760 {
761 struct inode *inode = filp->f_path.dentry->d_inode;
762 struct sock *sock;
763
764 if (!S_ISSOCK(inode->i_mode))
765 return ERR_PTR(-ENOTSOCK);
766
767 sock = SOCKET_I(inode)->sk;
768 if (sock->sk_family != AF_NETLINK)
769 return ERR_PTR(-EINVAL);
770
771 sock_hold(sock);
772 return sock;
773 }
774
775 /*
776 * Attach a skb to a netlink socket.
777 * The caller must hold a reference to the destination socket. On error, the
778 * reference is dropped. The skb is not send to the destination, just all
779 * all error checks are performed and memory in the queue is reserved.
780 * Return values:
781 * < 0: error. skb freed, reference to sock dropped.
782 * 0: continue
783 * 1: repeat lookup - reference dropped while waiting for socket memory.
784 */
785 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
786 long *timeo, struct sock *ssk)
787 {
788 struct netlink_sock *nlk;
789
790 nlk = nlk_sk(sk);
791
792 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
793 test_bit(0, &nlk->state)) {
794 DECLARE_WAITQUEUE(wait, current);
795 if (!*timeo) {
796 if (!ssk || netlink_is_kernel(ssk))
797 netlink_overrun(sk);
798 sock_put(sk);
799 kfree_skb(skb);
800 return -EAGAIN;
801 }
802
803 __set_current_state(TASK_INTERRUPTIBLE);
804 add_wait_queue(&nlk->wait, &wait);
805
806 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
807 test_bit(0, &nlk->state)) &&
808 !sock_flag(sk, SOCK_DEAD))
809 *timeo = schedule_timeout(*timeo);
810
811 __set_current_state(TASK_RUNNING);
812 remove_wait_queue(&nlk->wait, &wait);
813 sock_put(sk);
814
815 if (signal_pending(current)) {
816 kfree_skb(skb);
817 return sock_intr_errno(*timeo);
818 }
819 return 1;
820 }
821 skb_set_owner_r(skb, sk);
822 return 0;
823 }
824
825 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
826 {
827 int len = skb->len;
828
829 skb_queue_tail(&sk->sk_receive_queue, skb);
830 sk->sk_data_ready(sk, len);
831 sock_put(sk);
832 return len;
833 }
834
835 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
836 {
837 kfree_skb(skb);
838 sock_put(sk);
839 }
840
841 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
842 gfp_t allocation)
843 {
844 int delta;
845
846 skb_orphan(skb);
847
848 delta = skb->end - skb->tail;
849 if (delta * 2 < skb->truesize)
850 return skb;
851
852 if (skb_shared(skb)) {
853 struct sk_buff *nskb = skb_clone(skb, allocation);
854 if (!nskb)
855 return skb;
856 kfree_skb(skb);
857 skb = nskb;
858 }
859
860 if (!pskb_expand_head(skb, 0, -delta, allocation))
861 skb->truesize -= delta;
862
863 return skb;
864 }
865
866 static inline void netlink_rcv_wake(struct sock *sk)
867 {
868 struct netlink_sock *nlk = nlk_sk(sk);
869
870 if (skb_queue_empty(&sk->sk_receive_queue))
871 clear_bit(0, &nlk->state);
872 if (!test_bit(0, &nlk->state))
873 wake_up_interruptible(&nlk->wait);
874 }
875
876 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
877 {
878 int ret;
879 struct netlink_sock *nlk = nlk_sk(sk);
880
881 ret = -ECONNREFUSED;
882 if (nlk->netlink_rcv != NULL) {
883 ret = skb->len;
884 skb_set_owner_r(skb, sk);
885 nlk->netlink_rcv(skb);
886 }
887 kfree_skb(skb);
888 sock_put(sk);
889 return ret;
890 }
891
892 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
893 u32 pid, int nonblock)
894 {
895 struct sock *sk;
896 int err;
897 long timeo;
898
899 skb = netlink_trim(skb, gfp_any());
900
901 timeo = sock_sndtimeo(ssk, nonblock);
902 retry:
903 sk = netlink_getsockbypid(ssk, pid);
904 if (IS_ERR(sk)) {
905 kfree_skb(skb);
906 return PTR_ERR(sk);
907 }
908 if (netlink_is_kernel(sk))
909 return netlink_unicast_kernel(sk, skb);
910
911 if (sk_filter(sk, skb)) {
912 err = skb->len;
913 kfree_skb(skb);
914 sock_put(sk);
915 return err;
916 }
917
918 err = netlink_attachskb(sk, skb, &timeo, ssk);
919 if (err == 1)
920 goto retry;
921 if (err)
922 return err;
923
924 return netlink_sendskb(sk, skb);
925 }
926 EXPORT_SYMBOL(netlink_unicast);
927
928 int netlink_has_listeners(struct sock *sk, unsigned int group)
929 {
930 int res = 0;
931 unsigned long *listeners;
932
933 BUG_ON(!netlink_is_kernel(sk));
934
935 rcu_read_lock();
936 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
937
938 if (group - 1 < nl_table[sk->sk_protocol].groups)
939 res = test_bit(group - 1, listeners);
940
941 rcu_read_unlock();
942
943 return res;
944 }
945 EXPORT_SYMBOL_GPL(netlink_has_listeners);
946
947 static inline int netlink_broadcast_deliver(struct sock *sk,
948 struct sk_buff *skb)
949 {
950 struct netlink_sock *nlk = nlk_sk(sk);
951
952 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
953 !test_bit(0, &nlk->state)) {
954 skb_set_owner_r(skb, sk);
955 skb_queue_tail(&sk->sk_receive_queue, skb);
956 sk->sk_data_ready(sk, skb->len);
957 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
958 }
959 return -1;
960 }
961
962 struct netlink_broadcast_data {
963 struct sock *exclude_sk;
964 struct net *net;
965 u32 pid;
966 u32 group;
967 int failure;
968 int delivery_failure;
969 int congested;
970 int delivered;
971 gfp_t allocation;
972 struct sk_buff *skb, *skb2;
973 };
974
975 static inline int do_one_broadcast(struct sock *sk,
976 struct netlink_broadcast_data *p)
977 {
978 struct netlink_sock *nlk = nlk_sk(sk);
979 int val;
980
981 if (p->exclude_sk == sk)
982 goto out;
983
984 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
985 !test_bit(p->group - 1, nlk->groups))
986 goto out;
987
988 if (!net_eq(sock_net(sk), p->net))
989 goto out;
990
991 if (p->failure) {
992 netlink_overrun(sk);
993 goto out;
994 }
995
996 sock_hold(sk);
997 if (p->skb2 == NULL) {
998 if (skb_shared(p->skb)) {
999 p->skb2 = skb_clone(p->skb, p->allocation);
1000 } else {
1001 p->skb2 = skb_get(p->skb);
1002 /*
1003 * skb ownership may have been set when
1004 * delivered to a previous socket.
1005 */
1006 skb_orphan(p->skb2);
1007 }
1008 }
1009 if (p->skb2 == NULL) {
1010 netlink_overrun(sk);
1011 /* Clone failed. Notify ALL listeners. */
1012 p->failure = 1;
1013 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1014 p->delivery_failure = 1;
1015 } else if (sk_filter(sk, p->skb2)) {
1016 kfree_skb(p->skb2);
1017 p->skb2 = NULL;
1018 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1019 netlink_overrun(sk);
1020 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1021 p->delivery_failure = 1;
1022 } else {
1023 p->congested |= val;
1024 p->delivered = 1;
1025 p->skb2 = NULL;
1026 }
1027 sock_put(sk);
1028
1029 out:
1030 return 0;
1031 }
1032
1033 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1034 u32 group, gfp_t allocation)
1035 {
1036 struct net *net = sock_net(ssk);
1037 struct netlink_broadcast_data info;
1038 struct hlist_node *node;
1039 struct sock *sk;
1040
1041 skb = netlink_trim(skb, allocation);
1042
1043 info.exclude_sk = ssk;
1044 info.net = net;
1045 info.pid = pid;
1046 info.group = group;
1047 info.failure = 0;
1048 info.delivery_failure = 0;
1049 info.congested = 0;
1050 info.delivered = 0;
1051 info.allocation = allocation;
1052 info.skb = skb;
1053 info.skb2 = NULL;
1054
1055 /* While we sleep in clone, do not allow to change socket list */
1056
1057 netlink_lock_table();
1058
1059 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1060 do_one_broadcast(sk, &info);
1061
1062 kfree_skb(skb);
1063
1064 netlink_unlock_table();
1065
1066 kfree_skb(info.skb2);
1067
1068 if (info.delivery_failure)
1069 return -ENOBUFS;
1070
1071 if (info.delivered) {
1072 if (info.congested && (allocation & __GFP_WAIT))
1073 yield();
1074 return 0;
1075 }
1076 return -ESRCH;
1077 }
1078 EXPORT_SYMBOL(netlink_broadcast);
1079
1080 struct netlink_set_err_data {
1081 struct sock *exclude_sk;
1082 u32 pid;
1083 u32 group;
1084 int code;
1085 };
1086
1087 static inline int do_one_set_err(struct sock *sk,
1088 struct netlink_set_err_data *p)
1089 {
1090 struct netlink_sock *nlk = nlk_sk(sk);
1091
1092 if (sk == p->exclude_sk)
1093 goto out;
1094
1095 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1096 goto out;
1097
1098 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1099 !test_bit(p->group - 1, nlk->groups))
1100 goto out;
1101
1102 sk->sk_err = p->code;
1103 sk->sk_error_report(sk);
1104 out:
1105 return 0;
1106 }
1107
1108 /**
1109 * netlink_set_err - report error to broadcast listeners
1110 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1111 * @pid: the PID of a process that we want to skip (if any)
1112 * @groups: the broadcast group that will notice the error
1113 * @code: error code, must be negative (as usual in kernelspace)
1114 */
1115 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1116 {
1117 struct netlink_set_err_data info;
1118 struct hlist_node *node;
1119 struct sock *sk;
1120
1121 info.exclude_sk = ssk;
1122 info.pid = pid;
1123 info.group = group;
1124 /* sk->sk_err wants a positive error value */
1125 info.code = -code;
1126
1127 read_lock(&nl_table_lock);
1128
1129 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1130 do_one_set_err(sk, &info);
1131
1132 read_unlock(&nl_table_lock);
1133 }
1134 EXPORT_SYMBOL(netlink_set_err);
1135
1136 /* must be called with netlink table grabbed */
1137 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1138 unsigned int group,
1139 int is_new)
1140 {
1141 int old, new = !!is_new, subscriptions;
1142
1143 old = test_bit(group - 1, nlk->groups);
1144 subscriptions = nlk->subscriptions - old + new;
1145 if (new)
1146 __set_bit(group - 1, nlk->groups);
1147 else
1148 __clear_bit(group - 1, nlk->groups);
1149 netlink_update_subscriptions(&nlk->sk, subscriptions);
1150 netlink_update_listeners(&nlk->sk);
1151 }
1152
1153 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1154 char __user *optval, unsigned int optlen)
1155 {
1156 struct sock *sk = sock->sk;
1157 struct netlink_sock *nlk = nlk_sk(sk);
1158 unsigned int val = 0;
1159 int err;
1160
1161 if (level != SOL_NETLINK)
1162 return -ENOPROTOOPT;
1163
1164 if (optlen >= sizeof(int) &&
1165 get_user(val, (unsigned int __user *)optval))
1166 return -EFAULT;
1167
1168 switch (optname) {
1169 case NETLINK_PKTINFO:
1170 if (val)
1171 nlk->flags |= NETLINK_RECV_PKTINFO;
1172 else
1173 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1174 err = 0;
1175 break;
1176 case NETLINK_ADD_MEMBERSHIP:
1177 case NETLINK_DROP_MEMBERSHIP: {
1178 if (!netlink_capable(sock, NL_NONROOT_RECV))
1179 return -EPERM;
1180 err = netlink_realloc_groups(sk);
1181 if (err)
1182 return err;
1183 if (!val || val - 1 >= nlk->ngroups)
1184 return -EINVAL;
1185 netlink_table_grab();
1186 netlink_update_socket_mc(nlk, val,
1187 optname == NETLINK_ADD_MEMBERSHIP);
1188 netlink_table_ungrab();
1189 err = 0;
1190 break;
1191 }
1192 case NETLINK_BROADCAST_ERROR:
1193 if (val)
1194 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1195 else
1196 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1197 err = 0;
1198 break;
1199 case NETLINK_NO_ENOBUFS:
1200 if (val) {
1201 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1202 clear_bit(0, &nlk->state);
1203 wake_up_interruptible(&nlk->wait);
1204 } else
1205 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1206 err = 0;
1207 break;
1208 default:
1209 err = -ENOPROTOOPT;
1210 }
1211 return err;
1212 }
1213
1214 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1215 char __user *optval, int __user *optlen)
1216 {
1217 struct sock *sk = sock->sk;
1218 struct netlink_sock *nlk = nlk_sk(sk);
1219 int len, val, err;
1220
1221 if (level != SOL_NETLINK)
1222 return -ENOPROTOOPT;
1223
1224 if (get_user(len, optlen))
1225 return -EFAULT;
1226 if (len < 0)
1227 return -EINVAL;
1228
1229 switch (optname) {
1230 case NETLINK_PKTINFO:
1231 if (len < sizeof(int))
1232 return -EINVAL;
1233 len = sizeof(int);
1234 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1235 if (put_user(len, optlen) ||
1236 put_user(val, optval))
1237 return -EFAULT;
1238 err = 0;
1239 break;
1240 case NETLINK_BROADCAST_ERROR:
1241 if (len < sizeof(int))
1242 return -EINVAL;
1243 len = sizeof(int);
1244 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1245 if (put_user(len, optlen) ||
1246 put_user(val, optval))
1247 return -EFAULT;
1248 err = 0;
1249 break;
1250 case NETLINK_NO_ENOBUFS:
1251 if (len < sizeof(int))
1252 return -EINVAL;
1253 len = sizeof(int);
1254 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1255 if (put_user(len, optlen) ||
1256 put_user(val, optval))
1257 return -EFAULT;
1258 err = 0;
1259 break;
1260 default:
1261 err = -ENOPROTOOPT;
1262 }
1263 return err;
1264 }
1265
1266 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1267 {
1268 struct nl_pktinfo info;
1269
1270 info.group = NETLINK_CB(skb).dst_group;
1271 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1272 }
1273
1274 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1275 struct msghdr *msg, size_t len)
1276 {
1277 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1278 struct sock *sk = sock->sk;
1279 struct netlink_sock *nlk = nlk_sk(sk);
1280 struct sockaddr_nl *addr = msg->msg_name;
1281 u32 dst_pid;
1282 u32 dst_group;
1283 struct sk_buff *skb;
1284 int err;
1285 struct scm_cookie scm;
1286
1287 if (msg->msg_flags&MSG_OOB)
1288 return -EOPNOTSUPP;
1289
1290 if (NULL == siocb->scm)
1291 siocb->scm = &scm;
1292 err = scm_send(sock, msg, siocb->scm);
1293 if (err < 0)
1294 return err;
1295
1296 if (msg->msg_namelen) {
1297 if (addr->nl_family != AF_NETLINK)
1298 return -EINVAL;
1299 dst_pid = addr->nl_pid;
1300 dst_group = ffs(addr->nl_groups);
1301 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1302 return -EPERM;
1303 } else {
1304 dst_pid = nlk->dst_pid;
1305 dst_group = nlk->dst_group;
1306 }
1307
1308 if (!nlk->pid) {
1309 err = netlink_autobind(sock);
1310 if (err)
1311 goto out;
1312 }
1313
1314 err = -EMSGSIZE;
1315 if (len > sk->sk_sndbuf - 32)
1316 goto out;
1317 err = -ENOBUFS;
1318 skb = alloc_skb(len, GFP_KERNEL);
1319 if (skb == NULL)
1320 goto out;
1321
1322 NETLINK_CB(skb).pid = nlk->pid;
1323 NETLINK_CB(skb).dst_group = dst_group;
1324 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1325 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1326 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1327 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1328
1329 /* What can I do? Netlink is asynchronous, so that
1330 we will have to save current capabilities to
1331 check them, when this message will be delivered
1332 to corresponding kernel module. --ANK (980802)
1333 */
1334
1335 err = -EFAULT;
1336 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1337 kfree_skb(skb);
1338 goto out;
1339 }
1340
1341 err = security_netlink_send(sk, skb);
1342 if (err) {
1343 kfree_skb(skb);
1344 goto out;
1345 }
1346
1347 if (dst_group) {
1348 atomic_inc(&skb->users);
1349 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1350 }
1351 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1352
1353 out:
1354 return err;
1355 }
1356
1357 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1358 struct msghdr *msg, size_t len,
1359 int flags)
1360 {
1361 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1362 struct scm_cookie scm;
1363 struct sock *sk = sock->sk;
1364 struct netlink_sock *nlk = nlk_sk(sk);
1365 int noblock = flags&MSG_DONTWAIT;
1366 size_t copied;
1367 struct sk_buff *skb, *frag __maybe_unused = NULL;
1368 int err;
1369
1370 if (flags&MSG_OOB)
1371 return -EOPNOTSUPP;
1372
1373 copied = 0;
1374
1375 skb = skb_recv_datagram(sk, flags, noblock, &err);
1376 if (skb == NULL)
1377 goto out;
1378
1379 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1380 if (unlikely(skb_shinfo(skb)->frag_list)) {
1381 bool need_compat = !!(flags & MSG_CMSG_COMPAT);
1382
1383 /*
1384 * If this skb has a frag_list, then here that means that
1385 * we will have to use the frag_list skb for compat tasks
1386 * and the regular skb for non-compat tasks.
1387 *
1388 * The skb might (and likely will) be cloned, so we can't
1389 * just reset frag_list and go on with things -- we need to
1390 * keep that. For the compat case that's easy -- simply get
1391 * a reference to the compat skb and free the regular one
1392 * including the frag. For the non-compat case, we need to
1393 * avoid sending the frag to the user -- so assign NULL but
1394 * restore it below before freeing the skb.
1395 */
1396 if (need_compat) {
1397 struct sk_buff *compskb = skb_shinfo(skb)->frag_list;
1398 skb_get(compskb);
1399 kfree_skb(skb);
1400 skb = compskb;
1401 } else {
1402 frag = skb_shinfo(skb)->frag_list;
1403 skb_shinfo(skb)->frag_list = NULL;
1404 }
1405 }
1406 #endif
1407
1408 msg->msg_namelen = 0;
1409
1410 copied = skb->len;
1411 if (len < copied) {
1412 msg->msg_flags |= MSG_TRUNC;
1413 copied = len;
1414 }
1415
1416 skb_reset_transport_header(skb);
1417 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1418
1419 if (msg->msg_name) {
1420 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1421 addr->nl_family = AF_NETLINK;
1422 addr->nl_pad = 0;
1423 addr->nl_pid = NETLINK_CB(skb).pid;
1424 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1425 msg->msg_namelen = sizeof(*addr);
1426 }
1427
1428 if (nlk->flags & NETLINK_RECV_PKTINFO)
1429 netlink_cmsg_recv_pktinfo(msg, skb);
1430
1431 if (NULL == siocb->scm) {
1432 memset(&scm, 0, sizeof(scm));
1433 siocb->scm = &scm;
1434 }
1435 siocb->scm->creds = *NETLINK_CREDS(skb);
1436 if (flags & MSG_TRUNC)
1437 copied = skb->len;
1438
1439 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1440 skb_shinfo(skb)->frag_list = frag;
1441 #endif
1442
1443 skb_free_datagram(sk, skb);
1444
1445 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1446 netlink_dump(sk);
1447
1448 scm_recv(sock, msg, siocb->scm, flags);
1449 out:
1450 netlink_rcv_wake(sk);
1451 return err ? : copied;
1452 }
1453
1454 static void netlink_data_ready(struct sock *sk, int len)
1455 {
1456 BUG();
1457 }
1458
1459 /*
1460 * We export these functions to other modules. They provide a
1461 * complete set of kernel non-blocking support for message
1462 * queueing.
1463 */
1464
1465 struct sock *
1466 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1467 void (*input)(struct sk_buff *skb),
1468 struct mutex *cb_mutex, struct module *module)
1469 {
1470 struct socket *sock;
1471 struct sock *sk;
1472 struct netlink_sock *nlk;
1473 unsigned long *listeners = NULL;
1474
1475 BUG_ON(!nl_table);
1476
1477 if (unit < 0 || unit >= MAX_LINKS)
1478 return NULL;
1479
1480 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1481 return NULL;
1482
1483 /*
1484 * We have to just have a reference on the net from sk, but don't
1485 * get_net it. Besides, we cannot get and then put the net here.
1486 * So we create one inside init_net and the move it to net.
1487 */
1488
1489 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1490 goto out_sock_release_nosk;
1491
1492 sk = sock->sk;
1493 sk_change_net(sk, net);
1494
1495 if (groups < 32)
1496 groups = 32;
1497
1498 listeners = kzalloc(NLGRPSZ(groups) + sizeof(struct listeners_rcu_head),
1499 GFP_KERNEL);
1500 if (!listeners)
1501 goto out_sock_release;
1502
1503 sk->sk_data_ready = netlink_data_ready;
1504 if (input)
1505 nlk_sk(sk)->netlink_rcv = input;
1506
1507 if (netlink_insert(sk, net, 0))
1508 goto out_sock_release;
1509
1510 nlk = nlk_sk(sk);
1511 nlk->flags |= NETLINK_KERNEL_SOCKET;
1512
1513 netlink_table_grab();
1514 if (!nl_table[unit].registered) {
1515 nl_table[unit].groups = groups;
1516 nl_table[unit].listeners = listeners;
1517 nl_table[unit].cb_mutex = cb_mutex;
1518 nl_table[unit].module = module;
1519 nl_table[unit].registered = 1;
1520 } else {
1521 kfree(listeners);
1522 nl_table[unit].registered++;
1523 }
1524 netlink_table_ungrab();
1525 return sk;
1526
1527 out_sock_release:
1528 kfree(listeners);
1529 netlink_kernel_release(sk);
1530 return NULL;
1531
1532 out_sock_release_nosk:
1533 sock_release(sock);
1534 return NULL;
1535 }
1536 EXPORT_SYMBOL(netlink_kernel_create);
1537
1538
1539 void
1540 netlink_kernel_release(struct sock *sk)
1541 {
1542 sk_release_kernel(sk);
1543 }
1544 EXPORT_SYMBOL(netlink_kernel_release);
1545
1546
1547 static void netlink_free_old_listeners(struct rcu_head *rcu_head)
1548 {
1549 struct listeners_rcu_head *lrh;
1550
1551 lrh = container_of(rcu_head, struct listeners_rcu_head, rcu_head);
1552 kfree(lrh->ptr);
1553 }
1554
1555 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1556 {
1557 unsigned long *listeners, *old = NULL;
1558 struct listeners_rcu_head *old_rcu_head;
1559 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1560
1561 if (groups < 32)
1562 groups = 32;
1563
1564 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1565 listeners = kzalloc(NLGRPSZ(groups) +
1566 sizeof(struct listeners_rcu_head),
1567 GFP_ATOMIC);
1568 if (!listeners)
1569 return -ENOMEM;
1570 old = tbl->listeners;
1571 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1572 rcu_assign_pointer(tbl->listeners, listeners);
1573 /*
1574 * Free the old memory after an RCU grace period so we
1575 * don't leak it. We use call_rcu() here in order to be
1576 * able to call this function from atomic contexts. The
1577 * allocation of this memory will have reserved enough
1578 * space for struct listeners_rcu_head at the end.
1579 */
1580 old_rcu_head = (void *)(tbl->listeners +
1581 NLGRPLONGS(tbl->groups));
1582 old_rcu_head->ptr = old;
1583 call_rcu(&old_rcu_head->rcu_head, netlink_free_old_listeners);
1584 }
1585 tbl->groups = groups;
1586
1587 return 0;
1588 }
1589
1590 /**
1591 * netlink_change_ngroups - change number of multicast groups
1592 *
1593 * This changes the number of multicast groups that are available
1594 * on a certain netlink family. Note that it is not possible to
1595 * change the number of groups to below 32. Also note that it does
1596 * not implicitly call netlink_clear_multicast_users() when the
1597 * number of groups is reduced.
1598 *
1599 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1600 * @groups: The new number of groups.
1601 */
1602 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1603 {
1604 int err;
1605
1606 netlink_table_grab();
1607 err = __netlink_change_ngroups(sk, groups);
1608 netlink_table_ungrab();
1609
1610 return err;
1611 }
1612
1613 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1614 {
1615 struct sock *sk;
1616 struct hlist_node *node;
1617 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1618
1619 sk_for_each_bound(sk, node, &tbl->mc_list)
1620 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1621 }
1622
1623 /**
1624 * netlink_clear_multicast_users - kick off multicast listeners
1625 *
1626 * This function removes all listeners from the given group.
1627 * @ksk: The kernel netlink socket, as returned by
1628 * netlink_kernel_create().
1629 * @group: The multicast group to clear.
1630 */
1631 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1632 {
1633 netlink_table_grab();
1634 __netlink_clear_multicast_users(ksk, group);
1635 netlink_table_ungrab();
1636 }
1637
1638 void netlink_set_nonroot(int protocol, unsigned int flags)
1639 {
1640 if ((unsigned int)protocol < MAX_LINKS)
1641 nl_table[protocol].nl_nonroot = flags;
1642 }
1643 EXPORT_SYMBOL(netlink_set_nonroot);
1644
1645 static void netlink_destroy_callback(struct netlink_callback *cb)
1646 {
1647 kfree_skb(cb->skb);
1648 kfree(cb);
1649 }
1650
1651 /*
1652 * It looks a bit ugly.
1653 * It would be better to create kernel thread.
1654 */
1655
1656 static int netlink_dump(struct sock *sk)
1657 {
1658 struct netlink_sock *nlk = nlk_sk(sk);
1659 struct netlink_callback *cb;
1660 struct sk_buff *skb;
1661 struct nlmsghdr *nlh;
1662 int len, err = -ENOBUFS;
1663
1664 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1665 if (!skb)
1666 goto errout;
1667
1668 mutex_lock(nlk->cb_mutex);
1669
1670 cb = nlk->cb;
1671 if (cb == NULL) {
1672 err = -EINVAL;
1673 goto errout_skb;
1674 }
1675
1676 len = cb->dump(skb, cb);
1677
1678 if (len > 0) {
1679 mutex_unlock(nlk->cb_mutex);
1680
1681 if (sk_filter(sk, skb))
1682 kfree_skb(skb);
1683 else {
1684 skb_queue_tail(&sk->sk_receive_queue, skb);
1685 sk->sk_data_ready(sk, skb->len);
1686 }
1687 return 0;
1688 }
1689
1690 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1691 if (!nlh)
1692 goto errout_skb;
1693
1694 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1695
1696 if (sk_filter(sk, skb))
1697 kfree_skb(skb);
1698 else {
1699 skb_queue_tail(&sk->sk_receive_queue, skb);
1700 sk->sk_data_ready(sk, skb->len);
1701 }
1702
1703 if (cb->done)
1704 cb->done(cb);
1705 nlk->cb = NULL;
1706 mutex_unlock(nlk->cb_mutex);
1707
1708 netlink_destroy_callback(cb);
1709 return 0;
1710
1711 errout_skb:
1712 mutex_unlock(nlk->cb_mutex);
1713 kfree_skb(skb);
1714 errout:
1715 return err;
1716 }
1717
1718 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1719 const struct nlmsghdr *nlh,
1720 int (*dump)(struct sk_buff *skb,
1721 struct netlink_callback *),
1722 int (*done)(struct netlink_callback *))
1723 {
1724 struct netlink_callback *cb;
1725 struct sock *sk;
1726 struct netlink_sock *nlk;
1727
1728 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1729 if (cb == NULL)
1730 return -ENOBUFS;
1731
1732 cb->dump = dump;
1733 cb->done = done;
1734 cb->nlh = nlh;
1735 atomic_inc(&skb->users);
1736 cb->skb = skb;
1737
1738 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1739 if (sk == NULL) {
1740 netlink_destroy_callback(cb);
1741 return -ECONNREFUSED;
1742 }
1743 nlk = nlk_sk(sk);
1744 /* A dump is in progress... */
1745 mutex_lock(nlk->cb_mutex);
1746 if (nlk->cb) {
1747 mutex_unlock(nlk->cb_mutex);
1748 netlink_destroy_callback(cb);
1749 sock_put(sk);
1750 return -EBUSY;
1751 }
1752 nlk->cb = cb;
1753 mutex_unlock(nlk->cb_mutex);
1754
1755 netlink_dump(sk);
1756 sock_put(sk);
1757
1758 /* We successfully started a dump, by returning -EINTR we
1759 * signal not to send ACK even if it was requested.
1760 */
1761 return -EINTR;
1762 }
1763 EXPORT_SYMBOL(netlink_dump_start);
1764
1765 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1766 {
1767 struct sk_buff *skb;
1768 struct nlmsghdr *rep;
1769 struct nlmsgerr *errmsg;
1770 size_t payload = sizeof(*errmsg);
1771
1772 /* error messages get the original request appened */
1773 if (err)
1774 payload += nlmsg_len(nlh);
1775
1776 skb = nlmsg_new(payload, GFP_KERNEL);
1777 if (!skb) {
1778 struct sock *sk;
1779
1780 sk = netlink_lookup(sock_net(in_skb->sk),
1781 in_skb->sk->sk_protocol,
1782 NETLINK_CB(in_skb).pid);
1783 if (sk) {
1784 sk->sk_err = ENOBUFS;
1785 sk->sk_error_report(sk);
1786 sock_put(sk);
1787 }
1788 return;
1789 }
1790
1791 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1792 NLMSG_ERROR, payload, 0);
1793 errmsg = nlmsg_data(rep);
1794 errmsg->error = err;
1795 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1796 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1797 }
1798 EXPORT_SYMBOL(netlink_ack);
1799
1800 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1801 struct nlmsghdr *))
1802 {
1803 struct nlmsghdr *nlh;
1804 int err;
1805
1806 while (skb->len >= nlmsg_total_size(0)) {
1807 int msglen;
1808
1809 nlh = nlmsg_hdr(skb);
1810 err = 0;
1811
1812 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1813 return 0;
1814
1815 /* Only requests are handled by the kernel */
1816 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1817 goto ack;
1818
1819 /* Skip control messages */
1820 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1821 goto ack;
1822
1823 err = cb(skb, nlh);
1824 if (err == -EINTR)
1825 goto skip;
1826
1827 ack:
1828 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1829 netlink_ack(skb, nlh, err);
1830
1831 skip:
1832 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1833 if (msglen > skb->len)
1834 msglen = skb->len;
1835 skb_pull(skb, msglen);
1836 }
1837
1838 return 0;
1839 }
1840 EXPORT_SYMBOL(netlink_rcv_skb);
1841
1842 /**
1843 * nlmsg_notify - send a notification netlink message
1844 * @sk: netlink socket to use
1845 * @skb: notification message
1846 * @pid: destination netlink pid for reports or 0
1847 * @group: destination multicast group or 0
1848 * @report: 1 to report back, 0 to disable
1849 * @flags: allocation flags
1850 */
1851 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1852 unsigned int group, int report, gfp_t flags)
1853 {
1854 int err = 0;
1855
1856 if (group) {
1857 int exclude_pid = 0;
1858
1859 if (report) {
1860 atomic_inc(&skb->users);
1861 exclude_pid = pid;
1862 }
1863
1864 /* errors reported via destination sk->sk_err, but propagate
1865 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1866 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1867 }
1868
1869 if (report) {
1870 int err2;
1871
1872 err2 = nlmsg_unicast(sk, skb, pid);
1873 if (!err || err == -ESRCH)
1874 err = err2;
1875 }
1876
1877 return err;
1878 }
1879 EXPORT_SYMBOL(nlmsg_notify);
1880
1881 #ifdef CONFIG_PROC_FS
1882 struct nl_seq_iter {
1883 struct seq_net_private p;
1884 int link;
1885 int hash_idx;
1886 };
1887
1888 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1889 {
1890 struct nl_seq_iter *iter = seq->private;
1891 int i, j;
1892 struct sock *s;
1893 struct hlist_node *node;
1894 loff_t off = 0;
1895
1896 for (i = 0; i < MAX_LINKS; i++) {
1897 struct nl_pid_hash *hash = &nl_table[i].hash;
1898
1899 for (j = 0; j <= hash->mask; j++) {
1900 sk_for_each(s, node, &hash->table[j]) {
1901 if (sock_net(s) != seq_file_net(seq))
1902 continue;
1903 if (off == pos) {
1904 iter->link = i;
1905 iter->hash_idx = j;
1906 return s;
1907 }
1908 ++off;
1909 }
1910 }
1911 }
1912 return NULL;
1913 }
1914
1915 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1916 __acquires(nl_table_lock)
1917 {
1918 read_lock(&nl_table_lock);
1919 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1920 }
1921
1922 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1923 {
1924 struct sock *s;
1925 struct nl_seq_iter *iter;
1926 int i, j;
1927
1928 ++*pos;
1929
1930 if (v == SEQ_START_TOKEN)
1931 return netlink_seq_socket_idx(seq, 0);
1932
1933 iter = seq->private;
1934 s = v;
1935 do {
1936 s = sk_next(s);
1937 } while (s && sock_net(s) != seq_file_net(seq));
1938 if (s)
1939 return s;
1940
1941 i = iter->link;
1942 j = iter->hash_idx + 1;
1943
1944 do {
1945 struct nl_pid_hash *hash = &nl_table[i].hash;
1946
1947 for (; j <= hash->mask; j++) {
1948 s = sk_head(&hash->table[j]);
1949 while (s && sock_net(s) != seq_file_net(seq))
1950 s = sk_next(s);
1951 if (s) {
1952 iter->link = i;
1953 iter->hash_idx = j;
1954 return s;
1955 }
1956 }
1957
1958 j = 0;
1959 } while (++i < MAX_LINKS);
1960
1961 return NULL;
1962 }
1963
1964 static void netlink_seq_stop(struct seq_file *seq, void *v)
1965 __releases(nl_table_lock)
1966 {
1967 read_unlock(&nl_table_lock);
1968 }
1969
1970
1971 static int netlink_seq_show(struct seq_file *seq, void *v)
1972 {
1973 if (v == SEQ_START_TOKEN)
1974 seq_puts(seq,
1975 "sk Eth Pid Groups "
1976 "Rmem Wmem Dump Locks Drops\n");
1977 else {
1978 struct sock *s = v;
1979 struct netlink_sock *nlk = nlk_sk(s);
1980
1981 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %-8d %-8d\n",
1982 s,
1983 s->sk_protocol,
1984 nlk->pid,
1985 nlk->groups ? (u32)nlk->groups[0] : 0,
1986 sk_rmem_alloc_get(s),
1987 sk_wmem_alloc_get(s),
1988 nlk->cb,
1989 atomic_read(&s->sk_refcnt),
1990 atomic_read(&s->sk_drops)
1991 );
1992
1993 }
1994 return 0;
1995 }
1996
1997 static const struct seq_operations netlink_seq_ops = {
1998 .start = netlink_seq_start,
1999 .next = netlink_seq_next,
2000 .stop = netlink_seq_stop,
2001 .show = netlink_seq_show,
2002 };
2003
2004
2005 static int netlink_seq_open(struct inode *inode, struct file *file)
2006 {
2007 return seq_open_net(inode, file, &netlink_seq_ops,
2008 sizeof(struct nl_seq_iter));
2009 }
2010
2011 static const struct file_operations netlink_seq_fops = {
2012 .owner = THIS_MODULE,
2013 .open = netlink_seq_open,
2014 .read = seq_read,
2015 .llseek = seq_lseek,
2016 .release = seq_release_net,
2017 };
2018
2019 #endif
2020
2021 int netlink_register_notifier(struct notifier_block *nb)
2022 {
2023 return atomic_notifier_chain_register(&netlink_chain, nb);
2024 }
2025 EXPORT_SYMBOL(netlink_register_notifier);
2026
2027 int netlink_unregister_notifier(struct notifier_block *nb)
2028 {
2029 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2030 }
2031 EXPORT_SYMBOL(netlink_unregister_notifier);
2032
2033 static const struct proto_ops netlink_ops = {
2034 .family = PF_NETLINK,
2035 .owner = THIS_MODULE,
2036 .release = netlink_release,
2037 .bind = netlink_bind,
2038 .connect = netlink_connect,
2039 .socketpair = sock_no_socketpair,
2040 .accept = sock_no_accept,
2041 .getname = netlink_getname,
2042 .poll = datagram_poll,
2043 .ioctl = sock_no_ioctl,
2044 .listen = sock_no_listen,
2045 .shutdown = sock_no_shutdown,
2046 .setsockopt = netlink_setsockopt,
2047 .getsockopt = netlink_getsockopt,
2048 .sendmsg = netlink_sendmsg,
2049 .recvmsg = netlink_recvmsg,
2050 .mmap = sock_no_mmap,
2051 .sendpage = sock_no_sendpage,
2052 };
2053
2054 static const struct net_proto_family netlink_family_ops = {
2055 .family = PF_NETLINK,
2056 .create = netlink_create,
2057 .owner = THIS_MODULE, /* for consistency 8) */
2058 };
2059
2060 static int __net_init netlink_net_init(struct net *net)
2061 {
2062 #ifdef CONFIG_PROC_FS
2063 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2064 return -ENOMEM;
2065 #endif
2066 return 0;
2067 }
2068
2069 static void __net_exit netlink_net_exit(struct net *net)
2070 {
2071 #ifdef CONFIG_PROC_FS
2072 proc_net_remove(net, "netlink");
2073 #endif
2074 }
2075
2076 static struct pernet_operations __net_initdata netlink_net_ops = {
2077 .init = netlink_net_init,
2078 .exit = netlink_net_exit,
2079 };
2080
2081 static int __init netlink_proto_init(void)
2082 {
2083 struct sk_buff *dummy_skb;
2084 int i;
2085 unsigned long limit;
2086 unsigned int order;
2087 int err = proto_register(&netlink_proto, 0);
2088
2089 if (err != 0)
2090 goto out;
2091
2092 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2093
2094 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2095 if (!nl_table)
2096 goto panic;
2097
2098 if (totalram_pages >= (128 * 1024))
2099 limit = totalram_pages >> (21 - PAGE_SHIFT);
2100 else
2101 limit = totalram_pages >> (23 - PAGE_SHIFT);
2102
2103 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2104 limit = (1UL << order) / sizeof(struct hlist_head);
2105 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2106
2107 for (i = 0; i < MAX_LINKS; i++) {
2108 struct nl_pid_hash *hash = &nl_table[i].hash;
2109
2110 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2111 if (!hash->table) {
2112 while (i-- > 0)
2113 nl_pid_hash_free(nl_table[i].hash.table,
2114 1 * sizeof(*hash->table));
2115 kfree(nl_table);
2116 goto panic;
2117 }
2118 hash->max_shift = order;
2119 hash->shift = 0;
2120 hash->mask = 0;
2121 hash->rehash_time = jiffies;
2122 }
2123
2124 sock_register(&netlink_family_ops);
2125 register_pernet_subsys(&netlink_net_ops);
2126 /* The netlink device handler may be needed early. */
2127 rtnetlink_init();
2128 out:
2129 return err;
2130 panic:
2131 panic("netlink_init: Cannot allocate nl_table\n");
2132 }
2133
2134 core_initcall(netlink_proto_init);
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree