Linux 3.9-rc4
[linux-2.6/cjktty.git] / net / netlink / af_netlink.c
blob1e3fd5bfcd86a995fcbfd4277452738a7a823727
1 /*
2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
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.
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
24 #include <linux/module.h>
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>
59 #include <net/net_namespace.h>
60 #include <net/sock.h>
61 #include <net/scm.h>
62 #include <net/netlink.h>
64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
65 #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
67 struct netlink_sock {
68 /* struct sock has to be the first member of netlink_sock */
69 struct sock sk;
70 u32 portid;
71 u32 dst_portid;
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 void (*netlink_bind)(int group);
84 struct module *module;
87 struct listeners {
88 struct rcu_head rcu;
89 unsigned long masks[0];
92 #define NETLINK_KERNEL_SOCKET 0x1
93 #define NETLINK_RECV_PKTINFO 0x2
94 #define NETLINK_BROADCAST_SEND_ERROR 0x4
95 #define NETLINK_RECV_NO_ENOBUFS 0x8
97 static inline struct netlink_sock *nlk_sk(struct sock *sk)
99 return container_of(sk, struct netlink_sock, sk);
102 static inline int netlink_is_kernel(struct sock *sk)
104 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
107 struct nl_portid_hash {
108 struct hlist_head *table;
109 unsigned long rehash_time;
111 unsigned int mask;
112 unsigned int shift;
114 unsigned int entries;
115 unsigned int max_shift;
117 u32 rnd;
120 struct netlink_table {
121 struct nl_portid_hash hash;
122 struct hlist_head mc_list;
123 struct listeners __rcu *listeners;
124 unsigned int flags;
125 unsigned int groups;
126 struct mutex *cb_mutex;
127 struct module *module;
128 void (*bind)(int group);
129 int registered;
132 static struct netlink_table *nl_table;
134 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
136 static int netlink_dump(struct sock *sk);
138 static DEFINE_RWLOCK(nl_table_lock);
139 static atomic_t nl_table_users = ATOMIC_INIT(0);
141 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
143 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
145 static inline u32 netlink_group_mask(u32 group)
147 return group ? 1 << (group - 1) : 0;
150 static inline struct hlist_head *nl_portid_hashfn(struct nl_portid_hash *hash, u32 portid)
152 return &hash->table[jhash_1word(portid, hash->rnd) & hash->mask];
155 static void netlink_destroy_callback(struct netlink_callback *cb)
157 kfree_skb(cb->skb);
158 kfree(cb);
161 static void netlink_consume_callback(struct netlink_callback *cb)
163 consume_skb(cb->skb);
164 kfree(cb);
167 static void netlink_sock_destruct(struct sock *sk)
169 struct netlink_sock *nlk = nlk_sk(sk);
171 if (nlk->cb) {
172 if (nlk->cb->done)
173 nlk->cb->done(nlk->cb);
175 module_put(nlk->cb->module);
176 netlink_destroy_callback(nlk->cb);
179 skb_queue_purge(&sk->sk_receive_queue);
181 if (!sock_flag(sk, SOCK_DEAD)) {
182 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
183 return;
186 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
187 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
188 WARN_ON(nlk_sk(sk)->groups);
191 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
192 * SMP. Look, when several writers sleep and reader wakes them up, all but one
193 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
194 * this, _but_ remember, it adds useless work on UP machines.
197 void netlink_table_grab(void)
198 __acquires(nl_table_lock)
200 might_sleep();
202 write_lock_irq(&nl_table_lock);
204 if (atomic_read(&nl_table_users)) {
205 DECLARE_WAITQUEUE(wait, current);
207 add_wait_queue_exclusive(&nl_table_wait, &wait);
208 for (;;) {
209 set_current_state(TASK_UNINTERRUPTIBLE);
210 if (atomic_read(&nl_table_users) == 0)
211 break;
212 write_unlock_irq(&nl_table_lock);
213 schedule();
214 write_lock_irq(&nl_table_lock);
217 __set_current_state(TASK_RUNNING);
218 remove_wait_queue(&nl_table_wait, &wait);
222 void netlink_table_ungrab(void)
223 __releases(nl_table_lock)
225 write_unlock_irq(&nl_table_lock);
226 wake_up(&nl_table_wait);
229 static inline void
230 netlink_lock_table(void)
232 /* read_lock() synchronizes us to netlink_table_grab */
234 read_lock(&nl_table_lock);
235 atomic_inc(&nl_table_users);
236 read_unlock(&nl_table_lock);
239 static inline void
240 netlink_unlock_table(void)
242 if (atomic_dec_and_test(&nl_table_users))
243 wake_up(&nl_table_wait);
246 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
248 struct nl_portid_hash *hash = &nl_table[protocol].hash;
249 struct hlist_head *head;
250 struct sock *sk;
252 read_lock(&nl_table_lock);
253 head = nl_portid_hashfn(hash, portid);
254 sk_for_each(sk, head) {
255 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->portid == portid)) {
256 sock_hold(sk);
257 goto found;
260 sk = NULL;
261 found:
262 read_unlock(&nl_table_lock);
263 return sk;
266 static struct hlist_head *nl_portid_hash_zalloc(size_t size)
268 if (size <= PAGE_SIZE)
269 return kzalloc(size, GFP_ATOMIC);
270 else
271 return (struct hlist_head *)
272 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
273 get_order(size));
276 static void nl_portid_hash_free(struct hlist_head *table, size_t size)
278 if (size <= PAGE_SIZE)
279 kfree(table);
280 else
281 free_pages((unsigned long)table, get_order(size));
284 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
286 unsigned int omask, mask, shift;
287 size_t osize, size;
288 struct hlist_head *otable, *table;
289 int i;
291 omask = mask = hash->mask;
292 osize = size = (mask + 1) * sizeof(*table);
293 shift = hash->shift;
295 if (grow) {
296 if (++shift > hash->max_shift)
297 return 0;
298 mask = mask * 2 + 1;
299 size *= 2;
302 table = nl_portid_hash_zalloc(size);
303 if (!table)
304 return 0;
306 otable = hash->table;
307 hash->table = table;
308 hash->mask = mask;
309 hash->shift = shift;
310 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
312 for (i = 0; i <= omask; i++) {
313 struct sock *sk;
314 struct hlist_node *tmp;
316 sk_for_each_safe(sk, tmp, &otable[i])
317 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
320 nl_portid_hash_free(otable, osize);
321 hash->rehash_time = jiffies + 10 * 60 * HZ;
322 return 1;
325 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
327 int avg = hash->entries >> hash->shift;
329 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
330 return 1;
332 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
333 nl_portid_hash_rehash(hash, 0);
334 return 1;
337 return 0;
340 static const struct proto_ops netlink_ops;
342 static void
343 netlink_update_listeners(struct sock *sk)
345 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
346 unsigned long mask;
347 unsigned int i;
348 struct listeners *listeners;
350 listeners = nl_deref_protected(tbl->listeners);
351 if (!listeners)
352 return;
354 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
355 mask = 0;
356 sk_for_each_bound(sk, &tbl->mc_list) {
357 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
358 mask |= nlk_sk(sk)->groups[i];
360 listeners->masks[i] = mask;
362 /* this function is only called with the netlink table "grabbed", which
363 * makes sure updates are visible before bind or setsockopt return. */
366 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
368 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash;
369 struct hlist_head *head;
370 int err = -EADDRINUSE;
371 struct sock *osk;
372 int len;
374 netlink_table_grab();
375 head = nl_portid_hashfn(hash, portid);
376 len = 0;
377 sk_for_each(osk, head) {
378 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->portid == portid))
379 break;
380 len++;
382 if (osk)
383 goto err;
385 err = -EBUSY;
386 if (nlk_sk(sk)->portid)
387 goto err;
389 err = -ENOMEM;
390 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
391 goto err;
393 if (len && nl_portid_hash_dilute(hash, len))
394 head = nl_portid_hashfn(hash, portid);
395 hash->entries++;
396 nlk_sk(sk)->portid = portid;
397 sk_add_node(sk, head);
398 err = 0;
400 err:
401 netlink_table_ungrab();
402 return err;
405 static void netlink_remove(struct sock *sk)
407 netlink_table_grab();
408 if (sk_del_node_init(sk))
409 nl_table[sk->sk_protocol].hash.entries--;
410 if (nlk_sk(sk)->subscriptions)
411 __sk_del_bind_node(sk);
412 netlink_table_ungrab();
415 static struct proto netlink_proto = {
416 .name = "NETLINK",
417 .owner = THIS_MODULE,
418 .obj_size = sizeof(struct netlink_sock),
421 static int __netlink_create(struct net *net, struct socket *sock,
422 struct mutex *cb_mutex, int protocol)
424 struct sock *sk;
425 struct netlink_sock *nlk;
427 sock->ops = &netlink_ops;
429 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
430 if (!sk)
431 return -ENOMEM;
433 sock_init_data(sock, sk);
435 nlk = nlk_sk(sk);
436 if (cb_mutex) {
437 nlk->cb_mutex = cb_mutex;
438 } else {
439 nlk->cb_mutex = &nlk->cb_def_mutex;
440 mutex_init(nlk->cb_mutex);
442 init_waitqueue_head(&nlk->wait);
444 sk->sk_destruct = netlink_sock_destruct;
445 sk->sk_protocol = protocol;
446 return 0;
449 static int netlink_create(struct net *net, struct socket *sock, int protocol,
450 int kern)
452 struct module *module = NULL;
453 struct mutex *cb_mutex;
454 struct netlink_sock *nlk;
455 void (*bind)(int group);
456 int err = 0;
458 sock->state = SS_UNCONNECTED;
460 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
461 return -ESOCKTNOSUPPORT;
463 if (protocol < 0 || protocol >= MAX_LINKS)
464 return -EPROTONOSUPPORT;
466 netlink_lock_table();
467 #ifdef CONFIG_MODULES
468 if (!nl_table[protocol].registered) {
469 netlink_unlock_table();
470 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
471 netlink_lock_table();
473 #endif
474 if (nl_table[protocol].registered &&
475 try_module_get(nl_table[protocol].module))
476 module = nl_table[protocol].module;
477 else
478 err = -EPROTONOSUPPORT;
479 cb_mutex = nl_table[protocol].cb_mutex;
480 bind = nl_table[protocol].bind;
481 netlink_unlock_table();
483 if (err < 0)
484 goto out;
486 err = __netlink_create(net, sock, cb_mutex, protocol);
487 if (err < 0)
488 goto out_module;
490 local_bh_disable();
491 sock_prot_inuse_add(net, &netlink_proto, 1);
492 local_bh_enable();
494 nlk = nlk_sk(sock->sk);
495 nlk->module = module;
496 nlk->netlink_bind = bind;
497 out:
498 return err;
500 out_module:
501 module_put(module);
502 goto out;
505 static int netlink_release(struct socket *sock)
507 struct sock *sk = sock->sk;
508 struct netlink_sock *nlk;
510 if (!sk)
511 return 0;
513 netlink_remove(sk);
514 sock_orphan(sk);
515 nlk = nlk_sk(sk);
518 * OK. Socket is unlinked, any packets that arrive now
519 * will be purged.
522 sock->sk = NULL;
523 wake_up_interruptible_all(&nlk->wait);
525 skb_queue_purge(&sk->sk_write_queue);
527 if (nlk->portid) {
528 struct netlink_notify n = {
529 .net = sock_net(sk),
530 .protocol = sk->sk_protocol,
531 .portid = nlk->portid,
533 atomic_notifier_call_chain(&netlink_chain,
534 NETLINK_URELEASE, &n);
537 module_put(nlk->module);
539 netlink_table_grab();
540 if (netlink_is_kernel(sk)) {
541 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
542 if (--nl_table[sk->sk_protocol].registered == 0) {
543 struct listeners *old;
545 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
546 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
547 kfree_rcu(old, rcu);
548 nl_table[sk->sk_protocol].module = NULL;
549 nl_table[sk->sk_protocol].bind = NULL;
550 nl_table[sk->sk_protocol].flags = 0;
551 nl_table[sk->sk_protocol].registered = 0;
553 } else if (nlk->subscriptions) {
554 netlink_update_listeners(sk);
556 netlink_table_ungrab();
558 kfree(nlk->groups);
559 nlk->groups = NULL;
561 local_bh_disable();
562 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
563 local_bh_enable();
564 sock_put(sk);
565 return 0;
568 static int netlink_autobind(struct socket *sock)
570 struct sock *sk = sock->sk;
571 struct net *net = sock_net(sk);
572 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash;
573 struct hlist_head *head;
574 struct sock *osk;
575 s32 portid = task_tgid_vnr(current);
576 int err;
577 static s32 rover = -4097;
579 retry:
580 cond_resched();
581 netlink_table_grab();
582 head = nl_portid_hashfn(hash, portid);
583 sk_for_each(osk, head) {
584 if (!net_eq(sock_net(osk), net))
585 continue;
586 if (nlk_sk(osk)->portid == portid) {
587 /* Bind collision, search negative portid values. */
588 portid = rover--;
589 if (rover > -4097)
590 rover = -4097;
591 netlink_table_ungrab();
592 goto retry;
595 netlink_table_ungrab();
597 err = netlink_insert(sk, net, portid);
598 if (err == -EADDRINUSE)
599 goto retry;
601 /* If 2 threads race to autobind, that is fine. */
602 if (err == -EBUSY)
603 err = 0;
605 return err;
608 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
610 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
611 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
614 static void
615 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
617 struct netlink_sock *nlk = nlk_sk(sk);
619 if (nlk->subscriptions && !subscriptions)
620 __sk_del_bind_node(sk);
621 else if (!nlk->subscriptions && subscriptions)
622 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
623 nlk->subscriptions = subscriptions;
626 static int netlink_realloc_groups(struct sock *sk)
628 struct netlink_sock *nlk = nlk_sk(sk);
629 unsigned int groups;
630 unsigned long *new_groups;
631 int err = 0;
633 netlink_table_grab();
635 groups = nl_table[sk->sk_protocol].groups;
636 if (!nl_table[sk->sk_protocol].registered) {
637 err = -ENOENT;
638 goto out_unlock;
641 if (nlk->ngroups >= groups)
642 goto out_unlock;
644 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
645 if (new_groups == NULL) {
646 err = -ENOMEM;
647 goto out_unlock;
649 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
650 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
652 nlk->groups = new_groups;
653 nlk->ngroups = groups;
654 out_unlock:
655 netlink_table_ungrab();
656 return err;
659 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
660 int addr_len)
662 struct sock *sk = sock->sk;
663 struct net *net = sock_net(sk);
664 struct netlink_sock *nlk = nlk_sk(sk);
665 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
666 int err;
668 if (addr_len < sizeof(struct sockaddr_nl))
669 return -EINVAL;
671 if (nladdr->nl_family != AF_NETLINK)
672 return -EINVAL;
674 /* Only superuser is allowed to listen multicasts */
675 if (nladdr->nl_groups) {
676 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
677 return -EPERM;
678 err = netlink_realloc_groups(sk);
679 if (err)
680 return err;
683 if (nlk->portid) {
684 if (nladdr->nl_pid != nlk->portid)
685 return -EINVAL;
686 } else {
687 err = nladdr->nl_pid ?
688 netlink_insert(sk, net, nladdr->nl_pid) :
689 netlink_autobind(sock);
690 if (err)
691 return err;
694 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
695 return 0;
697 netlink_table_grab();
698 netlink_update_subscriptions(sk, nlk->subscriptions +
699 hweight32(nladdr->nl_groups) -
700 hweight32(nlk->groups[0]));
701 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
702 netlink_update_listeners(sk);
703 netlink_table_ungrab();
705 if (nlk->netlink_bind && nlk->groups[0]) {
706 int i;
708 for (i=0; i<nlk->ngroups; i++) {
709 if (test_bit(i, nlk->groups))
710 nlk->netlink_bind(i);
714 return 0;
717 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
718 int alen, int flags)
720 int err = 0;
721 struct sock *sk = sock->sk;
722 struct netlink_sock *nlk = nlk_sk(sk);
723 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
725 if (alen < sizeof(addr->sa_family))
726 return -EINVAL;
728 if (addr->sa_family == AF_UNSPEC) {
729 sk->sk_state = NETLINK_UNCONNECTED;
730 nlk->dst_portid = 0;
731 nlk->dst_group = 0;
732 return 0;
734 if (addr->sa_family != AF_NETLINK)
735 return -EINVAL;
737 /* Only superuser is allowed to send multicasts */
738 if (nladdr->nl_groups && !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
739 return -EPERM;
741 if (!nlk->portid)
742 err = netlink_autobind(sock);
744 if (err == 0) {
745 sk->sk_state = NETLINK_CONNECTED;
746 nlk->dst_portid = nladdr->nl_pid;
747 nlk->dst_group = ffs(nladdr->nl_groups);
750 return err;
753 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
754 int *addr_len, int peer)
756 struct sock *sk = sock->sk;
757 struct netlink_sock *nlk = nlk_sk(sk);
758 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
760 nladdr->nl_family = AF_NETLINK;
761 nladdr->nl_pad = 0;
762 *addr_len = sizeof(*nladdr);
764 if (peer) {
765 nladdr->nl_pid = nlk->dst_portid;
766 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
767 } else {
768 nladdr->nl_pid = nlk->portid;
769 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
771 return 0;
774 static void netlink_overrun(struct sock *sk)
776 struct netlink_sock *nlk = nlk_sk(sk);
778 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
779 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
780 sk->sk_err = ENOBUFS;
781 sk->sk_error_report(sk);
784 atomic_inc(&sk->sk_drops);
787 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
789 struct sock *sock;
790 struct netlink_sock *nlk;
792 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
793 if (!sock)
794 return ERR_PTR(-ECONNREFUSED);
796 /* Don't bother queuing skb if kernel socket has no input function */
797 nlk = nlk_sk(sock);
798 if (sock->sk_state == NETLINK_CONNECTED &&
799 nlk->dst_portid != nlk_sk(ssk)->portid) {
800 sock_put(sock);
801 return ERR_PTR(-ECONNREFUSED);
803 return sock;
806 struct sock *netlink_getsockbyfilp(struct file *filp)
808 struct inode *inode = file_inode(filp);
809 struct sock *sock;
811 if (!S_ISSOCK(inode->i_mode))
812 return ERR_PTR(-ENOTSOCK);
814 sock = SOCKET_I(inode)->sk;
815 if (sock->sk_family != AF_NETLINK)
816 return ERR_PTR(-EINVAL);
818 sock_hold(sock);
819 return sock;
823 * Attach a skb to a netlink socket.
824 * The caller must hold a reference to the destination socket. On error, the
825 * reference is dropped. The skb is not send to the destination, just all
826 * all error checks are performed and memory in the queue is reserved.
827 * Return values:
828 * < 0: error. skb freed, reference to sock dropped.
829 * 0: continue
830 * 1: repeat lookup - reference dropped while waiting for socket memory.
832 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
833 long *timeo, struct sock *ssk)
835 struct netlink_sock *nlk;
837 nlk = nlk_sk(sk);
839 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
840 test_bit(0, &nlk->state)) {
841 DECLARE_WAITQUEUE(wait, current);
842 if (!*timeo) {
843 if (!ssk || netlink_is_kernel(ssk))
844 netlink_overrun(sk);
845 sock_put(sk);
846 kfree_skb(skb);
847 return -EAGAIN;
850 __set_current_state(TASK_INTERRUPTIBLE);
851 add_wait_queue(&nlk->wait, &wait);
853 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
854 test_bit(0, &nlk->state)) &&
855 !sock_flag(sk, SOCK_DEAD))
856 *timeo = schedule_timeout(*timeo);
858 __set_current_state(TASK_RUNNING);
859 remove_wait_queue(&nlk->wait, &wait);
860 sock_put(sk);
862 if (signal_pending(current)) {
863 kfree_skb(skb);
864 return sock_intr_errno(*timeo);
866 return 1;
868 skb_set_owner_r(skb, sk);
869 return 0;
872 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
874 int len = skb->len;
876 skb_queue_tail(&sk->sk_receive_queue, skb);
877 sk->sk_data_ready(sk, len);
878 return len;
881 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
883 int len = __netlink_sendskb(sk, skb);
885 sock_put(sk);
886 return len;
889 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
891 kfree_skb(skb);
892 sock_put(sk);
895 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
897 int delta;
899 skb_orphan(skb);
901 delta = skb->end - skb->tail;
902 if (delta * 2 < skb->truesize)
903 return skb;
905 if (skb_shared(skb)) {
906 struct sk_buff *nskb = skb_clone(skb, allocation);
907 if (!nskb)
908 return skb;
909 consume_skb(skb);
910 skb = nskb;
913 if (!pskb_expand_head(skb, 0, -delta, allocation))
914 skb->truesize -= delta;
916 return skb;
919 static void netlink_rcv_wake(struct sock *sk)
921 struct netlink_sock *nlk = nlk_sk(sk);
923 if (skb_queue_empty(&sk->sk_receive_queue))
924 clear_bit(0, &nlk->state);
925 if (!test_bit(0, &nlk->state))
926 wake_up_interruptible(&nlk->wait);
929 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
930 struct sock *ssk)
932 int ret;
933 struct netlink_sock *nlk = nlk_sk(sk);
935 ret = -ECONNREFUSED;
936 if (nlk->netlink_rcv != NULL) {
937 ret = skb->len;
938 skb_set_owner_r(skb, sk);
939 NETLINK_CB(skb).ssk = ssk;
940 nlk->netlink_rcv(skb);
941 consume_skb(skb);
942 } else {
943 kfree_skb(skb);
945 sock_put(sk);
946 return ret;
949 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
950 u32 portid, int nonblock)
952 struct sock *sk;
953 int err;
954 long timeo;
956 skb = netlink_trim(skb, gfp_any());
958 timeo = sock_sndtimeo(ssk, nonblock);
959 retry:
960 sk = netlink_getsockbyportid(ssk, portid);
961 if (IS_ERR(sk)) {
962 kfree_skb(skb);
963 return PTR_ERR(sk);
965 if (netlink_is_kernel(sk))
966 return netlink_unicast_kernel(sk, skb, ssk);
968 if (sk_filter(sk, skb)) {
969 err = skb->len;
970 kfree_skb(skb);
971 sock_put(sk);
972 return err;
975 err = netlink_attachskb(sk, skb, &timeo, ssk);
976 if (err == 1)
977 goto retry;
978 if (err)
979 return err;
981 return netlink_sendskb(sk, skb);
983 EXPORT_SYMBOL(netlink_unicast);
985 int netlink_has_listeners(struct sock *sk, unsigned int group)
987 int res = 0;
988 struct listeners *listeners;
990 BUG_ON(!netlink_is_kernel(sk));
992 rcu_read_lock();
993 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
995 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
996 res = test_bit(group - 1, listeners->masks);
998 rcu_read_unlock();
1000 return res;
1002 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1004 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1006 struct netlink_sock *nlk = nlk_sk(sk);
1008 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1009 !test_bit(0, &nlk->state)) {
1010 skb_set_owner_r(skb, sk);
1011 __netlink_sendskb(sk, skb);
1012 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1014 return -1;
1017 struct netlink_broadcast_data {
1018 struct sock *exclude_sk;
1019 struct net *net;
1020 u32 portid;
1021 u32 group;
1022 int failure;
1023 int delivery_failure;
1024 int congested;
1025 int delivered;
1026 gfp_t allocation;
1027 struct sk_buff *skb, *skb2;
1028 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1029 void *tx_data;
1032 static int do_one_broadcast(struct sock *sk,
1033 struct netlink_broadcast_data *p)
1035 struct netlink_sock *nlk = nlk_sk(sk);
1036 int val;
1038 if (p->exclude_sk == sk)
1039 goto out;
1041 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1042 !test_bit(p->group - 1, nlk->groups))
1043 goto out;
1045 if (!net_eq(sock_net(sk), p->net))
1046 goto out;
1048 if (p->failure) {
1049 netlink_overrun(sk);
1050 goto out;
1053 sock_hold(sk);
1054 if (p->skb2 == NULL) {
1055 if (skb_shared(p->skb)) {
1056 p->skb2 = skb_clone(p->skb, p->allocation);
1057 } else {
1058 p->skb2 = skb_get(p->skb);
1060 * skb ownership may have been set when
1061 * delivered to a previous socket.
1063 skb_orphan(p->skb2);
1066 if (p->skb2 == NULL) {
1067 netlink_overrun(sk);
1068 /* Clone failed. Notify ALL listeners. */
1069 p->failure = 1;
1070 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1071 p->delivery_failure = 1;
1072 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1073 kfree_skb(p->skb2);
1074 p->skb2 = NULL;
1075 } else if (sk_filter(sk, p->skb2)) {
1076 kfree_skb(p->skb2);
1077 p->skb2 = NULL;
1078 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1079 netlink_overrun(sk);
1080 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1081 p->delivery_failure = 1;
1082 } else {
1083 p->congested |= val;
1084 p->delivered = 1;
1085 p->skb2 = NULL;
1087 sock_put(sk);
1089 out:
1090 return 0;
1093 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1094 u32 group, gfp_t allocation,
1095 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1096 void *filter_data)
1098 struct net *net = sock_net(ssk);
1099 struct netlink_broadcast_data info;
1100 struct sock *sk;
1102 skb = netlink_trim(skb, allocation);
1104 info.exclude_sk = ssk;
1105 info.net = net;
1106 info.portid = portid;
1107 info.group = group;
1108 info.failure = 0;
1109 info.delivery_failure = 0;
1110 info.congested = 0;
1111 info.delivered = 0;
1112 info.allocation = allocation;
1113 info.skb = skb;
1114 info.skb2 = NULL;
1115 info.tx_filter = filter;
1116 info.tx_data = filter_data;
1118 /* While we sleep in clone, do not allow to change socket list */
1120 netlink_lock_table();
1122 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1123 do_one_broadcast(sk, &info);
1125 consume_skb(skb);
1127 netlink_unlock_table();
1129 if (info.delivery_failure) {
1130 kfree_skb(info.skb2);
1131 return -ENOBUFS;
1133 consume_skb(info.skb2);
1135 if (info.delivered) {
1136 if (info.congested && (allocation & __GFP_WAIT))
1137 yield();
1138 return 0;
1140 return -ESRCH;
1142 EXPORT_SYMBOL(netlink_broadcast_filtered);
1144 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1145 u32 group, gfp_t allocation)
1147 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1148 NULL, NULL);
1150 EXPORT_SYMBOL(netlink_broadcast);
1152 struct netlink_set_err_data {
1153 struct sock *exclude_sk;
1154 u32 portid;
1155 u32 group;
1156 int code;
1159 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1161 struct netlink_sock *nlk = nlk_sk(sk);
1162 int ret = 0;
1164 if (sk == p->exclude_sk)
1165 goto out;
1167 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1168 goto out;
1170 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1171 !test_bit(p->group - 1, nlk->groups))
1172 goto out;
1174 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1175 ret = 1;
1176 goto out;
1179 sk->sk_err = p->code;
1180 sk->sk_error_report(sk);
1181 out:
1182 return ret;
1186 * netlink_set_err - report error to broadcast listeners
1187 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1188 * @portid: the PORTID of a process that we want to skip (if any)
1189 * @groups: the broadcast group that will notice the error
1190 * @code: error code, must be negative (as usual in kernelspace)
1192 * This function returns the number of broadcast listeners that have set the
1193 * NETLINK_RECV_NO_ENOBUFS socket option.
1195 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1197 struct netlink_set_err_data info;
1198 struct sock *sk;
1199 int ret = 0;
1201 info.exclude_sk = ssk;
1202 info.portid = portid;
1203 info.group = group;
1204 /* sk->sk_err wants a positive error value */
1205 info.code = -code;
1207 read_lock(&nl_table_lock);
1209 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1210 ret += do_one_set_err(sk, &info);
1212 read_unlock(&nl_table_lock);
1213 return ret;
1215 EXPORT_SYMBOL(netlink_set_err);
1217 /* must be called with netlink table grabbed */
1218 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1219 unsigned int group,
1220 int is_new)
1222 int old, new = !!is_new, subscriptions;
1224 old = test_bit(group - 1, nlk->groups);
1225 subscriptions = nlk->subscriptions - old + new;
1226 if (new)
1227 __set_bit(group - 1, nlk->groups);
1228 else
1229 __clear_bit(group - 1, nlk->groups);
1230 netlink_update_subscriptions(&nlk->sk, subscriptions);
1231 netlink_update_listeners(&nlk->sk);
1234 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1235 char __user *optval, unsigned int optlen)
1237 struct sock *sk = sock->sk;
1238 struct netlink_sock *nlk = nlk_sk(sk);
1239 unsigned int val = 0;
1240 int err;
1242 if (level != SOL_NETLINK)
1243 return -ENOPROTOOPT;
1245 if (optlen >= sizeof(int) &&
1246 get_user(val, (unsigned int __user *)optval))
1247 return -EFAULT;
1249 switch (optname) {
1250 case NETLINK_PKTINFO:
1251 if (val)
1252 nlk->flags |= NETLINK_RECV_PKTINFO;
1253 else
1254 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1255 err = 0;
1256 break;
1257 case NETLINK_ADD_MEMBERSHIP:
1258 case NETLINK_DROP_MEMBERSHIP: {
1259 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1260 return -EPERM;
1261 err = netlink_realloc_groups(sk);
1262 if (err)
1263 return err;
1264 if (!val || val - 1 >= nlk->ngroups)
1265 return -EINVAL;
1266 netlink_table_grab();
1267 netlink_update_socket_mc(nlk, val,
1268 optname == NETLINK_ADD_MEMBERSHIP);
1269 netlink_table_ungrab();
1271 if (nlk->netlink_bind)
1272 nlk->netlink_bind(val);
1274 err = 0;
1275 break;
1277 case NETLINK_BROADCAST_ERROR:
1278 if (val)
1279 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1280 else
1281 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1282 err = 0;
1283 break;
1284 case NETLINK_NO_ENOBUFS:
1285 if (val) {
1286 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1287 clear_bit(0, &nlk->state);
1288 wake_up_interruptible(&nlk->wait);
1289 } else {
1290 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1292 err = 0;
1293 break;
1294 default:
1295 err = -ENOPROTOOPT;
1297 return err;
1300 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1301 char __user *optval, int __user *optlen)
1303 struct sock *sk = sock->sk;
1304 struct netlink_sock *nlk = nlk_sk(sk);
1305 int len, val, err;
1307 if (level != SOL_NETLINK)
1308 return -ENOPROTOOPT;
1310 if (get_user(len, optlen))
1311 return -EFAULT;
1312 if (len < 0)
1313 return -EINVAL;
1315 switch (optname) {
1316 case NETLINK_PKTINFO:
1317 if (len < sizeof(int))
1318 return -EINVAL;
1319 len = sizeof(int);
1320 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1321 if (put_user(len, optlen) ||
1322 put_user(val, optval))
1323 return -EFAULT;
1324 err = 0;
1325 break;
1326 case NETLINK_BROADCAST_ERROR:
1327 if (len < sizeof(int))
1328 return -EINVAL;
1329 len = sizeof(int);
1330 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1331 if (put_user(len, optlen) ||
1332 put_user(val, optval))
1333 return -EFAULT;
1334 err = 0;
1335 break;
1336 case NETLINK_NO_ENOBUFS:
1337 if (len < sizeof(int))
1338 return -EINVAL;
1339 len = sizeof(int);
1340 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1341 if (put_user(len, optlen) ||
1342 put_user(val, optval))
1343 return -EFAULT;
1344 err = 0;
1345 break;
1346 default:
1347 err = -ENOPROTOOPT;
1349 return err;
1352 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1354 struct nl_pktinfo info;
1356 info.group = NETLINK_CB(skb).dst_group;
1357 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1360 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1361 struct msghdr *msg, size_t len)
1363 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1364 struct sock *sk = sock->sk;
1365 struct netlink_sock *nlk = nlk_sk(sk);
1366 struct sockaddr_nl *addr = msg->msg_name;
1367 u32 dst_portid;
1368 u32 dst_group;
1369 struct sk_buff *skb;
1370 int err;
1371 struct scm_cookie scm;
1373 if (msg->msg_flags&MSG_OOB)
1374 return -EOPNOTSUPP;
1376 if (NULL == siocb->scm)
1377 siocb->scm = &scm;
1379 err = scm_send(sock, msg, siocb->scm, true);
1380 if (err < 0)
1381 return err;
1383 if (msg->msg_namelen) {
1384 err = -EINVAL;
1385 if (addr->nl_family != AF_NETLINK)
1386 goto out;
1387 dst_portid = addr->nl_pid;
1388 dst_group = ffs(addr->nl_groups);
1389 err = -EPERM;
1390 if ((dst_group || dst_portid) &&
1391 !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
1392 goto out;
1393 } else {
1394 dst_portid = nlk->dst_portid;
1395 dst_group = nlk->dst_group;
1398 if (!nlk->portid) {
1399 err = netlink_autobind(sock);
1400 if (err)
1401 goto out;
1404 err = -EMSGSIZE;
1405 if (len > sk->sk_sndbuf - 32)
1406 goto out;
1407 err = -ENOBUFS;
1408 skb = alloc_skb(len, GFP_KERNEL);
1409 if (skb == NULL)
1410 goto out;
1412 NETLINK_CB(skb).portid = nlk->portid;
1413 NETLINK_CB(skb).dst_group = dst_group;
1414 NETLINK_CB(skb).creds = siocb->scm->creds;
1416 err = -EFAULT;
1417 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1418 kfree_skb(skb);
1419 goto out;
1422 err = security_netlink_send(sk, skb);
1423 if (err) {
1424 kfree_skb(skb);
1425 goto out;
1428 if (dst_group) {
1429 atomic_inc(&skb->users);
1430 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1432 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1434 out:
1435 scm_destroy(siocb->scm);
1436 return err;
1439 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1440 struct msghdr *msg, size_t len,
1441 int flags)
1443 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1444 struct scm_cookie scm;
1445 struct sock *sk = sock->sk;
1446 struct netlink_sock *nlk = nlk_sk(sk);
1447 int noblock = flags&MSG_DONTWAIT;
1448 size_t copied;
1449 struct sk_buff *skb, *data_skb;
1450 int err, ret;
1452 if (flags&MSG_OOB)
1453 return -EOPNOTSUPP;
1455 copied = 0;
1457 skb = skb_recv_datagram(sk, flags, noblock, &err);
1458 if (skb == NULL)
1459 goto out;
1461 data_skb = skb;
1463 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1464 if (unlikely(skb_shinfo(skb)->frag_list)) {
1466 * If this skb has a frag_list, then here that means that we
1467 * will have to use the frag_list skb's data for compat tasks
1468 * and the regular skb's data for normal (non-compat) tasks.
1470 * If we need to send the compat skb, assign it to the
1471 * 'data_skb' variable so that it will be used below for data
1472 * copying. We keep 'skb' for everything else, including
1473 * freeing both later.
1475 if (flags & MSG_CMSG_COMPAT)
1476 data_skb = skb_shinfo(skb)->frag_list;
1478 #endif
1480 msg->msg_namelen = 0;
1482 copied = data_skb->len;
1483 if (len < copied) {
1484 msg->msg_flags |= MSG_TRUNC;
1485 copied = len;
1488 skb_reset_transport_header(data_skb);
1489 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1491 if (msg->msg_name) {
1492 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1493 addr->nl_family = AF_NETLINK;
1494 addr->nl_pad = 0;
1495 addr->nl_pid = NETLINK_CB(skb).portid;
1496 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1497 msg->msg_namelen = sizeof(*addr);
1500 if (nlk->flags & NETLINK_RECV_PKTINFO)
1501 netlink_cmsg_recv_pktinfo(msg, skb);
1503 if (NULL == siocb->scm) {
1504 memset(&scm, 0, sizeof(scm));
1505 siocb->scm = &scm;
1507 siocb->scm->creds = *NETLINK_CREDS(skb);
1508 if (flags & MSG_TRUNC)
1509 copied = data_skb->len;
1511 skb_free_datagram(sk, skb);
1513 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1514 ret = netlink_dump(sk);
1515 if (ret) {
1516 sk->sk_err = ret;
1517 sk->sk_error_report(sk);
1521 scm_recv(sock, msg, siocb->scm, flags);
1522 out:
1523 netlink_rcv_wake(sk);
1524 return err ? : copied;
1527 static void netlink_data_ready(struct sock *sk, int len)
1529 BUG();
1533 * We export these functions to other modules. They provide a
1534 * complete set of kernel non-blocking support for message
1535 * queueing.
1538 struct sock *
1539 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1540 struct netlink_kernel_cfg *cfg)
1542 struct socket *sock;
1543 struct sock *sk;
1544 struct netlink_sock *nlk;
1545 struct listeners *listeners = NULL;
1546 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1547 unsigned int groups;
1549 BUG_ON(!nl_table);
1551 if (unit < 0 || unit >= MAX_LINKS)
1552 return NULL;
1554 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1555 return NULL;
1558 * We have to just have a reference on the net from sk, but don't
1559 * get_net it. Besides, we cannot get and then put the net here.
1560 * So we create one inside init_net and the move it to net.
1563 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1564 goto out_sock_release_nosk;
1566 sk = sock->sk;
1567 sk_change_net(sk, net);
1569 if (!cfg || cfg->groups < 32)
1570 groups = 32;
1571 else
1572 groups = cfg->groups;
1574 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1575 if (!listeners)
1576 goto out_sock_release;
1578 sk->sk_data_ready = netlink_data_ready;
1579 if (cfg && cfg->input)
1580 nlk_sk(sk)->netlink_rcv = cfg->input;
1582 if (netlink_insert(sk, net, 0))
1583 goto out_sock_release;
1585 nlk = nlk_sk(sk);
1586 nlk->flags |= NETLINK_KERNEL_SOCKET;
1588 netlink_table_grab();
1589 if (!nl_table[unit].registered) {
1590 nl_table[unit].groups = groups;
1591 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1592 nl_table[unit].cb_mutex = cb_mutex;
1593 nl_table[unit].module = module;
1594 if (cfg) {
1595 nl_table[unit].bind = cfg->bind;
1596 nl_table[unit].flags = cfg->flags;
1598 nl_table[unit].registered = 1;
1599 } else {
1600 kfree(listeners);
1601 nl_table[unit].registered++;
1603 netlink_table_ungrab();
1604 return sk;
1606 out_sock_release:
1607 kfree(listeners);
1608 netlink_kernel_release(sk);
1609 return NULL;
1611 out_sock_release_nosk:
1612 sock_release(sock);
1613 return NULL;
1615 EXPORT_SYMBOL(__netlink_kernel_create);
1617 void
1618 netlink_kernel_release(struct sock *sk)
1620 sk_release_kernel(sk);
1622 EXPORT_SYMBOL(netlink_kernel_release);
1624 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1626 struct listeners *new, *old;
1627 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1629 if (groups < 32)
1630 groups = 32;
1632 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1633 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1634 if (!new)
1635 return -ENOMEM;
1636 old = nl_deref_protected(tbl->listeners);
1637 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1638 rcu_assign_pointer(tbl->listeners, new);
1640 kfree_rcu(old, rcu);
1642 tbl->groups = groups;
1644 return 0;
1648 * netlink_change_ngroups - change number of multicast groups
1650 * This changes the number of multicast groups that are available
1651 * on a certain netlink family. Note that it is not possible to
1652 * change the number of groups to below 32. Also note that it does
1653 * not implicitly call netlink_clear_multicast_users() when the
1654 * number of groups is reduced.
1656 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1657 * @groups: The new number of groups.
1659 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1661 int err;
1663 netlink_table_grab();
1664 err = __netlink_change_ngroups(sk, groups);
1665 netlink_table_ungrab();
1667 return err;
1670 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1672 struct sock *sk;
1673 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1675 sk_for_each_bound(sk, &tbl->mc_list)
1676 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1680 * netlink_clear_multicast_users - kick off multicast listeners
1682 * This function removes all listeners from the given group.
1683 * @ksk: The kernel netlink socket, as returned by
1684 * netlink_kernel_create().
1685 * @group: The multicast group to clear.
1687 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1689 netlink_table_grab();
1690 __netlink_clear_multicast_users(ksk, group);
1691 netlink_table_ungrab();
1694 struct nlmsghdr *
1695 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
1697 struct nlmsghdr *nlh;
1698 int size = NLMSG_LENGTH(len);
1700 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
1701 nlh->nlmsg_type = type;
1702 nlh->nlmsg_len = size;
1703 nlh->nlmsg_flags = flags;
1704 nlh->nlmsg_pid = portid;
1705 nlh->nlmsg_seq = seq;
1706 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
1707 memset(NLMSG_DATA(nlh) + len, 0, NLMSG_ALIGN(size) - size);
1708 return nlh;
1710 EXPORT_SYMBOL(__nlmsg_put);
1713 * It looks a bit ugly.
1714 * It would be better to create kernel thread.
1717 static int netlink_dump(struct sock *sk)
1719 struct netlink_sock *nlk = nlk_sk(sk);
1720 struct netlink_callback *cb;
1721 struct sk_buff *skb = NULL;
1722 struct nlmsghdr *nlh;
1723 int len, err = -ENOBUFS;
1724 int alloc_size;
1726 mutex_lock(nlk->cb_mutex);
1728 cb = nlk->cb;
1729 if (cb == NULL) {
1730 err = -EINVAL;
1731 goto errout_skb;
1734 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
1736 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
1737 if (!skb)
1738 goto errout_skb;
1740 len = cb->dump(skb, cb);
1742 if (len > 0) {
1743 mutex_unlock(nlk->cb_mutex);
1745 if (sk_filter(sk, skb))
1746 kfree_skb(skb);
1747 else
1748 __netlink_sendskb(sk, skb);
1749 return 0;
1752 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1753 if (!nlh)
1754 goto errout_skb;
1756 nl_dump_check_consistent(cb, nlh);
1758 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1760 if (sk_filter(sk, skb))
1761 kfree_skb(skb);
1762 else
1763 __netlink_sendskb(sk, skb);
1765 if (cb->done)
1766 cb->done(cb);
1767 nlk->cb = NULL;
1768 mutex_unlock(nlk->cb_mutex);
1770 module_put(cb->module);
1771 netlink_consume_callback(cb);
1772 return 0;
1774 errout_skb:
1775 mutex_unlock(nlk->cb_mutex);
1776 kfree_skb(skb);
1777 return err;
1780 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1781 const struct nlmsghdr *nlh,
1782 struct netlink_dump_control *control)
1784 struct netlink_callback *cb;
1785 struct sock *sk;
1786 struct netlink_sock *nlk;
1787 int ret;
1789 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1790 if (cb == NULL)
1791 return -ENOBUFS;
1793 cb->dump = control->dump;
1794 cb->done = control->done;
1795 cb->nlh = nlh;
1796 cb->data = control->data;
1797 cb->module = control->module;
1798 cb->min_dump_alloc = control->min_dump_alloc;
1799 atomic_inc(&skb->users);
1800 cb->skb = skb;
1802 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
1803 if (sk == NULL) {
1804 netlink_destroy_callback(cb);
1805 return -ECONNREFUSED;
1807 nlk = nlk_sk(sk);
1809 mutex_lock(nlk->cb_mutex);
1810 /* A dump is in progress... */
1811 if (nlk->cb) {
1812 mutex_unlock(nlk->cb_mutex);
1813 netlink_destroy_callback(cb);
1814 ret = -EBUSY;
1815 goto out;
1817 /* add reference of module which cb->dump belongs to */
1818 if (!try_module_get(cb->module)) {
1819 mutex_unlock(nlk->cb_mutex);
1820 netlink_destroy_callback(cb);
1821 ret = -EPROTONOSUPPORT;
1822 goto out;
1825 nlk->cb = cb;
1826 mutex_unlock(nlk->cb_mutex);
1828 ret = netlink_dump(sk);
1829 out:
1830 sock_put(sk);
1832 if (ret)
1833 return ret;
1835 /* We successfully started a dump, by returning -EINTR we
1836 * signal not to send ACK even if it was requested.
1838 return -EINTR;
1840 EXPORT_SYMBOL(__netlink_dump_start);
1842 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1844 struct sk_buff *skb;
1845 struct nlmsghdr *rep;
1846 struct nlmsgerr *errmsg;
1847 size_t payload = sizeof(*errmsg);
1849 /* error messages get the original request appened */
1850 if (err)
1851 payload += nlmsg_len(nlh);
1853 skb = nlmsg_new(payload, GFP_KERNEL);
1854 if (!skb) {
1855 struct sock *sk;
1857 sk = netlink_lookup(sock_net(in_skb->sk),
1858 in_skb->sk->sk_protocol,
1859 NETLINK_CB(in_skb).portid);
1860 if (sk) {
1861 sk->sk_err = ENOBUFS;
1862 sk->sk_error_report(sk);
1863 sock_put(sk);
1865 return;
1868 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
1869 NLMSG_ERROR, payload, 0);
1870 errmsg = nlmsg_data(rep);
1871 errmsg->error = err;
1872 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1873 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
1875 EXPORT_SYMBOL(netlink_ack);
1877 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1878 struct nlmsghdr *))
1880 struct nlmsghdr *nlh;
1881 int err;
1883 while (skb->len >= nlmsg_total_size(0)) {
1884 int msglen;
1886 nlh = nlmsg_hdr(skb);
1887 err = 0;
1889 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1890 return 0;
1892 /* Only requests are handled by the kernel */
1893 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1894 goto ack;
1896 /* Skip control messages */
1897 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1898 goto ack;
1900 err = cb(skb, nlh);
1901 if (err == -EINTR)
1902 goto skip;
1904 ack:
1905 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1906 netlink_ack(skb, nlh, err);
1908 skip:
1909 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1910 if (msglen > skb->len)
1911 msglen = skb->len;
1912 skb_pull(skb, msglen);
1915 return 0;
1917 EXPORT_SYMBOL(netlink_rcv_skb);
1920 * nlmsg_notify - send a notification netlink message
1921 * @sk: netlink socket to use
1922 * @skb: notification message
1923 * @portid: destination netlink portid for reports or 0
1924 * @group: destination multicast group or 0
1925 * @report: 1 to report back, 0 to disable
1926 * @flags: allocation flags
1928 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
1929 unsigned int group, int report, gfp_t flags)
1931 int err = 0;
1933 if (group) {
1934 int exclude_portid = 0;
1936 if (report) {
1937 atomic_inc(&skb->users);
1938 exclude_portid = portid;
1941 /* errors reported via destination sk->sk_err, but propagate
1942 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1943 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
1946 if (report) {
1947 int err2;
1949 err2 = nlmsg_unicast(sk, skb, portid);
1950 if (!err || err == -ESRCH)
1951 err = err2;
1954 return err;
1956 EXPORT_SYMBOL(nlmsg_notify);
1958 #ifdef CONFIG_PROC_FS
1959 struct nl_seq_iter {
1960 struct seq_net_private p;
1961 int link;
1962 int hash_idx;
1965 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1967 struct nl_seq_iter *iter = seq->private;
1968 int i, j;
1969 struct sock *s;
1970 loff_t off = 0;
1972 for (i = 0; i < MAX_LINKS; i++) {
1973 struct nl_portid_hash *hash = &nl_table[i].hash;
1975 for (j = 0; j <= hash->mask; j++) {
1976 sk_for_each(s, &hash->table[j]) {
1977 if (sock_net(s) != seq_file_net(seq))
1978 continue;
1979 if (off == pos) {
1980 iter->link = i;
1981 iter->hash_idx = j;
1982 return s;
1984 ++off;
1988 return NULL;
1991 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1992 __acquires(nl_table_lock)
1994 read_lock(&nl_table_lock);
1995 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1998 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2000 struct sock *s;
2001 struct nl_seq_iter *iter;
2002 int i, j;
2004 ++*pos;
2006 if (v == SEQ_START_TOKEN)
2007 return netlink_seq_socket_idx(seq, 0);
2009 iter = seq->private;
2010 s = v;
2011 do {
2012 s = sk_next(s);
2013 } while (s && sock_net(s) != seq_file_net(seq));
2014 if (s)
2015 return s;
2017 i = iter->link;
2018 j = iter->hash_idx + 1;
2020 do {
2021 struct nl_portid_hash *hash = &nl_table[i].hash;
2023 for (; j <= hash->mask; j++) {
2024 s = sk_head(&hash->table[j]);
2025 while (s && sock_net(s) != seq_file_net(seq))
2026 s = sk_next(s);
2027 if (s) {
2028 iter->link = i;
2029 iter->hash_idx = j;
2030 return s;
2034 j = 0;
2035 } while (++i < MAX_LINKS);
2037 return NULL;
2040 static void netlink_seq_stop(struct seq_file *seq, void *v)
2041 __releases(nl_table_lock)
2043 read_unlock(&nl_table_lock);
2047 static int netlink_seq_show(struct seq_file *seq, void *v)
2049 if (v == SEQ_START_TOKEN) {
2050 seq_puts(seq,
2051 "sk Eth Pid Groups "
2052 "Rmem Wmem Dump Locks Drops Inode\n");
2053 } else {
2054 struct sock *s = v;
2055 struct netlink_sock *nlk = nlk_sk(s);
2057 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2059 s->sk_protocol,
2060 nlk->portid,
2061 nlk->groups ? (u32)nlk->groups[0] : 0,
2062 sk_rmem_alloc_get(s),
2063 sk_wmem_alloc_get(s),
2064 nlk->cb,
2065 atomic_read(&s->sk_refcnt),
2066 atomic_read(&s->sk_drops),
2067 sock_i_ino(s)
2071 return 0;
2074 static const struct seq_operations netlink_seq_ops = {
2075 .start = netlink_seq_start,
2076 .next = netlink_seq_next,
2077 .stop = netlink_seq_stop,
2078 .show = netlink_seq_show,
2082 static int netlink_seq_open(struct inode *inode, struct file *file)
2084 return seq_open_net(inode, file, &netlink_seq_ops,
2085 sizeof(struct nl_seq_iter));
2088 static const struct file_operations netlink_seq_fops = {
2089 .owner = THIS_MODULE,
2090 .open = netlink_seq_open,
2091 .read = seq_read,
2092 .llseek = seq_lseek,
2093 .release = seq_release_net,
2096 #endif
2098 int netlink_register_notifier(struct notifier_block *nb)
2100 return atomic_notifier_chain_register(&netlink_chain, nb);
2102 EXPORT_SYMBOL(netlink_register_notifier);
2104 int netlink_unregister_notifier(struct notifier_block *nb)
2106 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2108 EXPORT_SYMBOL(netlink_unregister_notifier);
2110 static const struct proto_ops netlink_ops = {
2111 .family = PF_NETLINK,
2112 .owner = THIS_MODULE,
2113 .release = netlink_release,
2114 .bind = netlink_bind,
2115 .connect = netlink_connect,
2116 .socketpair = sock_no_socketpair,
2117 .accept = sock_no_accept,
2118 .getname = netlink_getname,
2119 .poll = datagram_poll,
2120 .ioctl = sock_no_ioctl,
2121 .listen = sock_no_listen,
2122 .shutdown = sock_no_shutdown,
2123 .setsockopt = netlink_setsockopt,
2124 .getsockopt = netlink_getsockopt,
2125 .sendmsg = netlink_sendmsg,
2126 .recvmsg = netlink_recvmsg,
2127 .mmap = sock_no_mmap,
2128 .sendpage = sock_no_sendpage,
2131 static const struct net_proto_family netlink_family_ops = {
2132 .family = PF_NETLINK,
2133 .create = netlink_create,
2134 .owner = THIS_MODULE, /* for consistency 8) */
2137 static int __net_init netlink_net_init(struct net *net)
2139 #ifdef CONFIG_PROC_FS
2140 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2141 return -ENOMEM;
2142 #endif
2143 return 0;
2146 static void __net_exit netlink_net_exit(struct net *net)
2148 #ifdef CONFIG_PROC_FS
2149 remove_proc_entry("netlink", net->proc_net);
2150 #endif
2153 static void __init netlink_add_usersock_entry(void)
2155 struct listeners *listeners;
2156 int groups = 32;
2158 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2159 if (!listeners)
2160 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2162 netlink_table_grab();
2164 nl_table[NETLINK_USERSOCK].groups = groups;
2165 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2166 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2167 nl_table[NETLINK_USERSOCK].registered = 1;
2168 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2170 netlink_table_ungrab();
2173 static struct pernet_operations __net_initdata netlink_net_ops = {
2174 .init = netlink_net_init,
2175 .exit = netlink_net_exit,
2178 static int __init netlink_proto_init(void)
2180 int i;
2181 unsigned long limit;
2182 unsigned int order;
2183 int err = proto_register(&netlink_proto, 0);
2185 if (err != 0)
2186 goto out;
2188 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2190 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2191 if (!nl_table)
2192 goto panic;
2194 if (totalram_pages >= (128 * 1024))
2195 limit = totalram_pages >> (21 - PAGE_SHIFT);
2196 else
2197 limit = totalram_pages >> (23 - PAGE_SHIFT);
2199 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2200 limit = (1UL << order) / sizeof(struct hlist_head);
2201 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2203 for (i = 0; i < MAX_LINKS; i++) {
2204 struct nl_portid_hash *hash = &nl_table[i].hash;
2206 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
2207 if (!hash->table) {
2208 while (i-- > 0)
2209 nl_portid_hash_free(nl_table[i].hash.table,
2210 1 * sizeof(*hash->table));
2211 kfree(nl_table);
2212 goto panic;
2214 hash->max_shift = order;
2215 hash->shift = 0;
2216 hash->mask = 0;
2217 hash->rehash_time = jiffies;
2220 netlink_add_usersock_entry();
2222 sock_register(&netlink_family_ops);
2223 register_pernet_subsys(&netlink_net_ops);
2224 /* The netlink device handler may be needed early. */
2225 rtnetlink_init();
2226 out:
2227 return err;
2228 panic:
2229 panic("netlink_init: Cannot allocate nl_table\n");
2232 core_initcall(netlink_proto_init);