fed up with those stupid warnings
[mmotm.git] / net / netlink / af_netlink.c
blob0cd2d88293136c271f72df7f1a4478919f448a06
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 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;
86 struct listeners_rcu_head {
87 struct rcu_head rcu_head;
88 void *ptr;
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
96 static inline struct netlink_sock *nlk_sk(struct sock *sk)
98 return container_of(sk, struct netlink_sock, sk);
101 static inline int netlink_is_kernel(struct sock *sk)
103 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
106 struct nl_pid_hash {
107 struct hlist_head *table;
108 unsigned long rehash_time;
110 unsigned int mask;
111 unsigned int shift;
113 unsigned int entries;
114 unsigned int max_shift;
116 u32 rnd;
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;
130 static struct netlink_table *nl_table;
132 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
134 static int netlink_dump(struct sock *sk);
135 static void netlink_destroy_callback(struct netlink_callback *cb);
137 static DEFINE_RWLOCK(nl_table_lock);
138 static atomic_t nl_table_users = ATOMIC_INIT(0);
140 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
142 static u32 netlink_group_mask(u32 group)
144 return group ? 1 << (group - 1) : 0;
147 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
149 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
152 static void netlink_sock_destruct(struct sock *sk)
154 struct netlink_sock *nlk = nlk_sk(sk);
156 if (nlk->cb) {
157 if (nlk->cb->done)
158 nlk->cb->done(nlk->cb);
159 netlink_destroy_callback(nlk->cb);
162 skb_queue_purge(&sk->sk_receive_queue);
164 if (!sock_flag(sk, SOCK_DEAD)) {
165 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
166 return;
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);
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.
180 void netlink_table_grab(void)
181 __acquires(nl_table_lock)
183 might_sleep();
185 write_lock_irq(&nl_table_lock);
187 if (atomic_read(&nl_table_users)) {
188 DECLARE_WAITQUEUE(wait, current);
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);
200 __set_current_state(TASK_RUNNING);
201 remove_wait_queue(&nl_table_wait, &wait);
205 void netlink_table_ungrab(void)
206 __releases(nl_table_lock)
208 write_unlock_irq(&nl_table_lock);
209 wake_up(&nl_table_wait);
212 static inline void
213 netlink_lock_table(void)
215 /* read_lock() synchronizes us to netlink_table_grab */
217 read_lock(&nl_table_lock);
218 atomic_inc(&nl_table_users);
219 read_unlock(&nl_table_lock);
222 static inline void
223 netlink_unlock_table(void)
225 if (atomic_dec_and_test(&nl_table_users))
226 wake_up(&nl_table_wait);
229 static inline struct sock *netlink_lookup(struct net *net, int protocol,
230 u32 pid)
232 struct nl_pid_hash *hash = &nl_table[protocol].hash;
233 struct hlist_head *head;
234 struct sock *sk;
235 struct hlist_node *node;
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;
245 sk = NULL;
246 found:
247 read_unlock(&nl_table_lock);
248 return sk;
251 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
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));
261 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
263 if (size <= PAGE_SIZE)
264 kfree(table);
265 else
266 free_pages((unsigned long)table, get_order(size));
269 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
271 unsigned int omask, mask, shift;
272 size_t osize, size;
273 struct hlist_head *otable, *table;
274 int i;
276 omask = mask = hash->mask;
277 osize = size = (mask + 1) * sizeof(*table);
278 shift = hash->shift;
280 if (grow) {
281 if (++shift > hash->max_shift)
282 return 0;
283 mask = mask * 2 + 1;
284 size *= 2;
287 table = nl_pid_hash_zalloc(size);
288 if (!table)
289 return 0;
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));
297 for (i = 0; i <= omask; i++) {
298 struct sock *sk;
299 struct hlist_node *node, *tmp;
301 sk_for_each_safe(sk, node, tmp, &otable[i])
302 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
305 nl_pid_hash_free(otable, osize);
306 hash->rehash_time = jiffies + 10 * 60 * HZ;
307 return 1;
310 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
312 int avg = hash->entries >> hash->shift;
314 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
315 return 1;
317 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
318 nl_pid_hash_rehash(hash, 0);
319 return 1;
322 return 0;
325 static const struct proto_ops netlink_ops;
327 static void
328 netlink_update_listeners(struct sock *sk)
330 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
331 struct hlist_node *node;
332 unsigned long mask;
333 unsigned int i;
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];
341 tbl->listeners[i] = mask;
343 /* this function is only called with the netlink table "grabbed", which
344 * makes sure updates are visible before bind or setsockopt return. */
347 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
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;
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++;
364 if (node)
365 goto err;
367 err = -EBUSY;
368 if (nlk_sk(sk)->pid)
369 goto err;
371 err = -ENOMEM;
372 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
373 goto err;
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;
382 err:
383 netlink_table_ungrab();
384 return err;
387 static void netlink_remove(struct sock *sk)
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();
397 static struct proto netlink_proto = {
398 .name = "NETLINK",
399 .owner = THIS_MODULE,
400 .obj_size = sizeof(struct netlink_sock),
403 static int __netlink_create(struct net *net, struct socket *sock,
404 struct mutex *cb_mutex, int protocol)
406 struct sock *sk;
407 struct netlink_sock *nlk;
409 sock->ops = &netlink_ops;
411 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
412 if (!sk)
413 return -ENOMEM;
415 sock_init_data(sock, sk);
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);
424 init_waitqueue_head(&nlk->wait);
426 sk->sk_destruct = netlink_sock_destruct;
427 sk->sk_protocol = protocol;
428 return 0;
431 static int netlink_create(struct net *net, struct socket *sock, int protocol)
433 struct module *module = NULL;
434 struct mutex *cb_mutex;
435 struct netlink_sock *nlk;
436 int err = 0;
438 sock->state = SS_UNCONNECTED;
440 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
441 return -ESOCKTNOSUPPORT;
443 if (protocol < 0 || protocol >= MAX_LINKS)
444 return -EPROTONOSUPPORT;
446 netlink_lock_table();
447 #ifdef CONFIG_MODULES
448 if (!nl_table[protocol].registered) {
449 netlink_unlock_table();
450 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
451 netlink_lock_table();
453 #endif
454 if (nl_table[protocol].registered &&
455 try_module_get(nl_table[protocol].module))
456 module = nl_table[protocol].module;
457 cb_mutex = nl_table[protocol].cb_mutex;
458 netlink_unlock_table();
460 err = __netlink_create(net, sock, cb_mutex, protocol);
461 if (err < 0)
462 goto out_module;
464 local_bh_disable();
465 sock_prot_inuse_add(net, &netlink_proto, 1);
466 local_bh_enable();
468 nlk = nlk_sk(sock->sk);
469 nlk->module = module;
470 out:
471 return err;
473 out_module:
474 module_put(module);
475 goto out;
478 static int netlink_release(struct socket *sock)
480 struct sock *sk = sock->sk;
481 struct netlink_sock *nlk;
483 if (!sk)
484 return 0;
486 netlink_remove(sk);
487 sock_orphan(sk);
488 nlk = nlk_sk(sk);
491 * OK. Socket is unlinked, any packets that arrive now
492 * will be purged.
495 sock->sk = NULL;
496 wake_up_interruptible_all(&nlk->wait);
498 skb_queue_purge(&sk->sk_write_queue);
500 if (nlk->pid && !nlk->subscriptions) {
501 struct netlink_notify n = {
502 .net = sock_net(sk),
503 .protocol = sk->sk_protocol,
504 .pid = nlk->pid,
506 atomic_notifier_call_chain(&netlink_chain,
507 NETLINK_URELEASE, &n);
510 module_put(nlk->module);
512 netlink_table_grab();
513 if (netlink_is_kernel(sk)) {
514 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
515 if (--nl_table[sk->sk_protocol].registered == 0) {
516 kfree(nl_table[sk->sk_protocol].listeners);
517 nl_table[sk->sk_protocol].module = NULL;
518 nl_table[sk->sk_protocol].registered = 0;
520 } else if (nlk->subscriptions)
521 netlink_update_listeners(sk);
522 netlink_table_ungrab();
524 kfree(nlk->groups);
525 nlk->groups = NULL;
527 local_bh_disable();
528 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
529 local_bh_enable();
530 sock_put(sk);
531 return 0;
534 static int netlink_autobind(struct socket *sock)
536 struct sock *sk = sock->sk;
537 struct net *net = sock_net(sk);
538 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
539 struct hlist_head *head;
540 struct sock *osk;
541 struct hlist_node *node;
542 s32 pid = current->tgid;
543 int err;
544 static s32 rover = -4097;
546 retry:
547 cond_resched();
548 netlink_table_grab();
549 head = nl_pid_hashfn(hash, pid);
550 sk_for_each(osk, node, head) {
551 if (!net_eq(sock_net(osk), net))
552 continue;
553 if (nlk_sk(osk)->pid == pid) {
554 /* Bind collision, search negative pid values. */
555 pid = rover--;
556 if (rover > -4097)
557 rover = -4097;
558 netlink_table_ungrab();
559 goto retry;
562 netlink_table_ungrab();
564 err = netlink_insert(sk, net, pid);
565 if (err == -EADDRINUSE)
566 goto retry;
568 /* If 2 threads race to autobind, that is fine. */
569 if (err == -EBUSY)
570 err = 0;
572 return err;
575 static inline int netlink_capable(struct socket *sock, unsigned int flag)
577 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
578 capable(CAP_NET_ADMIN);
581 static void
582 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
584 struct netlink_sock *nlk = nlk_sk(sk);
586 if (nlk->subscriptions && !subscriptions)
587 __sk_del_bind_node(sk);
588 else if (!nlk->subscriptions && subscriptions)
589 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
590 nlk->subscriptions = subscriptions;
593 static int netlink_realloc_groups(struct sock *sk)
595 struct netlink_sock *nlk = nlk_sk(sk);
596 unsigned int groups;
597 unsigned long *new_groups;
598 int err = 0;
600 netlink_table_grab();
602 groups = nl_table[sk->sk_protocol].groups;
603 if (!nl_table[sk->sk_protocol].registered) {
604 err = -ENOENT;
605 goto out_unlock;
608 if (nlk->ngroups >= groups)
609 goto out_unlock;
611 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
612 if (new_groups == NULL) {
613 err = -ENOMEM;
614 goto out_unlock;
616 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
617 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
619 nlk->groups = new_groups;
620 nlk->ngroups = groups;
621 out_unlock:
622 netlink_table_ungrab();
623 return err;
626 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
627 int addr_len)
629 struct sock *sk = sock->sk;
630 struct net *net = sock_net(sk);
631 struct netlink_sock *nlk = nlk_sk(sk);
632 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
633 int err;
635 if (nladdr->nl_family != AF_NETLINK)
636 return -EINVAL;
638 /* Only superuser is allowed to listen multicasts */
639 if (nladdr->nl_groups) {
640 if (!netlink_capable(sock, NL_NONROOT_RECV))
641 return -EPERM;
642 err = netlink_realloc_groups(sk);
643 if (err)
644 return err;
647 if (nlk->pid) {
648 if (nladdr->nl_pid != nlk->pid)
649 return -EINVAL;
650 } else {
651 err = nladdr->nl_pid ?
652 netlink_insert(sk, net, nladdr->nl_pid) :
653 netlink_autobind(sock);
654 if (err)
655 return err;
658 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
659 return 0;
661 netlink_table_grab();
662 netlink_update_subscriptions(sk, nlk->subscriptions +
663 hweight32(nladdr->nl_groups) -
664 hweight32(nlk->groups[0]));
665 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
666 netlink_update_listeners(sk);
667 netlink_table_ungrab();
669 return 0;
672 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
673 int alen, int flags)
675 int err = 0;
676 struct sock *sk = sock->sk;
677 struct netlink_sock *nlk = nlk_sk(sk);
678 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
680 if (addr->sa_family == AF_UNSPEC) {
681 sk->sk_state = NETLINK_UNCONNECTED;
682 nlk->dst_pid = 0;
683 nlk->dst_group = 0;
684 return 0;
686 if (addr->sa_family != AF_NETLINK)
687 return -EINVAL;
689 /* Only superuser is allowed to send multicasts */
690 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
691 return -EPERM;
693 if (!nlk->pid)
694 err = netlink_autobind(sock);
696 if (err == 0) {
697 sk->sk_state = NETLINK_CONNECTED;
698 nlk->dst_pid = nladdr->nl_pid;
699 nlk->dst_group = ffs(nladdr->nl_groups);
702 return err;
705 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
706 int *addr_len, int peer)
708 struct sock *sk = sock->sk;
709 struct netlink_sock *nlk = nlk_sk(sk);
710 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
712 nladdr->nl_family = AF_NETLINK;
713 nladdr->nl_pad = 0;
714 *addr_len = sizeof(*nladdr);
716 if (peer) {
717 nladdr->nl_pid = nlk->dst_pid;
718 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
719 } else {
720 nladdr->nl_pid = nlk->pid;
721 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
723 return 0;
726 static void netlink_overrun(struct sock *sk)
728 struct netlink_sock *nlk = nlk_sk(sk);
730 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
731 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
732 sk->sk_err = ENOBUFS;
733 sk->sk_error_report(sk);
736 atomic_inc(&sk->sk_drops);
739 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
741 struct sock *sock;
742 struct netlink_sock *nlk;
744 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
745 if (!sock)
746 return ERR_PTR(-ECONNREFUSED);
748 /* Don't bother queuing skb if kernel socket has no input function */
749 nlk = nlk_sk(sock);
750 if (sock->sk_state == NETLINK_CONNECTED &&
751 nlk->dst_pid != nlk_sk(ssk)->pid) {
752 sock_put(sock);
753 return ERR_PTR(-ECONNREFUSED);
755 return sock;
758 struct sock *netlink_getsockbyfilp(struct file *filp)
760 struct inode *inode = filp->f_path.dentry->d_inode;
761 struct sock *sock;
763 if (!S_ISSOCK(inode->i_mode))
764 return ERR_PTR(-ENOTSOCK);
766 sock = SOCKET_I(inode)->sk;
767 if (sock->sk_family != AF_NETLINK)
768 return ERR_PTR(-EINVAL);
770 sock_hold(sock);
771 return sock;
775 * Attach a skb to a netlink socket.
776 * The caller must hold a reference to the destination socket. On error, the
777 * reference is dropped. The skb is not send to the destination, just all
778 * all error checks are performed and memory in the queue is reserved.
779 * Return values:
780 * < 0: error. skb freed, reference to sock dropped.
781 * 0: continue
782 * 1: repeat lookup - reference dropped while waiting for socket memory.
784 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
785 long *timeo, struct sock *ssk)
787 struct netlink_sock *nlk;
789 nlk = nlk_sk(sk);
791 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
792 test_bit(0, &nlk->state)) {
793 DECLARE_WAITQUEUE(wait, current);
794 if (!*timeo) {
795 if (!ssk || netlink_is_kernel(ssk))
796 netlink_overrun(sk);
797 sock_put(sk);
798 kfree_skb(skb);
799 return -EAGAIN;
802 __set_current_state(TASK_INTERRUPTIBLE);
803 add_wait_queue(&nlk->wait, &wait);
805 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
806 test_bit(0, &nlk->state)) &&
807 !sock_flag(sk, SOCK_DEAD))
808 *timeo = schedule_timeout(*timeo);
810 __set_current_state(TASK_RUNNING);
811 remove_wait_queue(&nlk->wait, &wait);
812 sock_put(sk);
814 if (signal_pending(current)) {
815 kfree_skb(skb);
816 return sock_intr_errno(*timeo);
818 return 1;
820 skb_set_owner_r(skb, sk);
821 return 0;
824 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
826 int len = skb->len;
828 skb_queue_tail(&sk->sk_receive_queue, skb);
829 sk->sk_data_ready(sk, len);
830 sock_put(sk);
831 return len;
834 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
836 kfree_skb(skb);
837 sock_put(sk);
840 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
841 gfp_t allocation)
843 int delta;
845 skb_orphan(skb);
847 delta = skb->end - skb->tail;
848 if (delta * 2 < skb->truesize)
849 return skb;
851 if (skb_shared(skb)) {
852 struct sk_buff *nskb = skb_clone(skb, allocation);
853 if (!nskb)
854 return skb;
855 kfree_skb(skb);
856 skb = nskb;
859 if (!pskb_expand_head(skb, 0, -delta, allocation))
860 skb->truesize -= delta;
862 return skb;
865 static inline void netlink_rcv_wake(struct sock *sk)
867 struct netlink_sock *nlk = nlk_sk(sk);
869 if (skb_queue_empty(&sk->sk_receive_queue))
870 clear_bit(0, &nlk->state);
871 if (!test_bit(0, &nlk->state))
872 wake_up_interruptible(&nlk->wait);
875 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
877 int ret;
878 struct netlink_sock *nlk = nlk_sk(sk);
880 ret = -ECONNREFUSED;
881 if (nlk->netlink_rcv != NULL) {
882 ret = skb->len;
883 skb_set_owner_r(skb, sk);
884 nlk->netlink_rcv(skb);
886 kfree_skb(skb);
887 sock_put(sk);
888 return ret;
891 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
892 u32 pid, int nonblock)
894 struct sock *sk;
895 int err;
896 long timeo;
898 skb = netlink_trim(skb, gfp_any());
900 timeo = sock_sndtimeo(ssk, nonblock);
901 retry:
902 sk = netlink_getsockbypid(ssk, pid);
903 if (IS_ERR(sk)) {
904 kfree_skb(skb);
905 return PTR_ERR(sk);
907 if (netlink_is_kernel(sk))
908 return netlink_unicast_kernel(sk, skb);
910 if (sk_filter(sk, skb)) {
911 err = skb->len;
912 kfree_skb(skb);
913 sock_put(sk);
914 return err;
917 err = netlink_attachskb(sk, skb, &timeo, ssk);
918 if (err == 1)
919 goto retry;
920 if (err)
921 return err;
923 return netlink_sendskb(sk, skb);
925 EXPORT_SYMBOL(netlink_unicast);
927 int netlink_has_listeners(struct sock *sk, unsigned int group)
929 int res = 0;
930 unsigned long *listeners;
932 BUG_ON(!netlink_is_kernel(sk));
934 rcu_read_lock();
935 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
937 if (group - 1 < nl_table[sk->sk_protocol].groups)
938 res = test_bit(group - 1, listeners);
940 rcu_read_unlock();
942 return res;
944 EXPORT_SYMBOL_GPL(netlink_has_listeners);
946 static inline int netlink_broadcast_deliver(struct sock *sk,
947 struct sk_buff *skb)
949 struct netlink_sock *nlk = nlk_sk(sk);
951 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
952 !test_bit(0, &nlk->state)) {
953 skb_set_owner_r(skb, sk);
954 skb_queue_tail(&sk->sk_receive_queue, skb);
955 sk->sk_data_ready(sk, skb->len);
956 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
958 return -1;
961 struct netlink_broadcast_data {
962 struct sock *exclude_sk;
963 struct net *net;
964 u32 pid;
965 u32 group;
966 int failure;
967 int delivery_failure;
968 int congested;
969 int delivered;
970 gfp_t allocation;
971 struct sk_buff *skb, *skb2;
974 static inline int do_one_broadcast(struct sock *sk,
975 struct netlink_broadcast_data *p)
977 struct netlink_sock *nlk = nlk_sk(sk);
978 int val;
980 if (p->exclude_sk == sk)
981 goto out;
983 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
984 !test_bit(p->group - 1, nlk->groups))
985 goto out;
987 if (!net_eq(sock_net(sk), p->net))
988 goto out;
990 if (p->failure) {
991 netlink_overrun(sk);
992 goto out;
995 sock_hold(sk);
996 if (p->skb2 == NULL) {
997 if (skb_shared(p->skb)) {
998 p->skb2 = skb_clone(p->skb, p->allocation);
999 } else {
1000 p->skb2 = skb_get(p->skb);
1002 * skb ownership may have been set when
1003 * delivered to a previous socket.
1005 skb_orphan(p->skb2);
1008 if (p->skb2 == NULL) {
1009 netlink_overrun(sk);
1010 /* Clone failed. Notify ALL listeners. */
1011 p->failure = 1;
1012 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1013 p->delivery_failure = 1;
1014 } else if (sk_filter(sk, p->skb2)) {
1015 kfree_skb(p->skb2);
1016 p->skb2 = NULL;
1017 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1018 netlink_overrun(sk);
1019 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1020 p->delivery_failure = 1;
1021 } else {
1022 p->congested |= val;
1023 p->delivered = 1;
1024 p->skb2 = NULL;
1026 sock_put(sk);
1028 out:
1029 return 0;
1032 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1033 u32 group, gfp_t allocation)
1035 struct net *net = sock_net(ssk);
1036 struct netlink_broadcast_data info;
1037 struct hlist_node *node;
1038 struct sock *sk;
1040 skb = netlink_trim(skb, allocation);
1042 info.exclude_sk = ssk;
1043 info.net = net;
1044 info.pid = pid;
1045 info.group = group;
1046 info.failure = 0;
1047 info.delivery_failure = 0;
1048 info.congested = 0;
1049 info.delivered = 0;
1050 info.allocation = allocation;
1051 info.skb = skb;
1052 info.skb2 = NULL;
1054 /* While we sleep in clone, do not allow to change socket list */
1056 netlink_lock_table();
1058 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1059 do_one_broadcast(sk, &info);
1061 kfree_skb(skb);
1063 netlink_unlock_table();
1065 kfree_skb(info.skb2);
1067 if (info.delivery_failure)
1068 return -ENOBUFS;
1070 if (info.delivered) {
1071 if (info.congested && (allocation & __GFP_WAIT))
1072 yield();
1073 return 0;
1075 return -ESRCH;
1077 EXPORT_SYMBOL(netlink_broadcast);
1079 struct netlink_set_err_data {
1080 struct sock *exclude_sk;
1081 u32 pid;
1082 u32 group;
1083 int code;
1086 static inline int do_one_set_err(struct sock *sk,
1087 struct netlink_set_err_data *p)
1089 struct netlink_sock *nlk = nlk_sk(sk);
1091 if (sk == p->exclude_sk)
1092 goto out;
1094 if (sock_net(sk) != sock_net(p->exclude_sk))
1095 goto out;
1097 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1098 !test_bit(p->group - 1, nlk->groups))
1099 goto out;
1101 sk->sk_err = p->code;
1102 sk->sk_error_report(sk);
1103 out:
1104 return 0;
1108 * netlink_set_err - report error to broadcast listeners
1109 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1110 * @pid: the PID of a process that we want to skip (if any)
1111 * @groups: the broadcast group that will notice the error
1112 * @code: error code, must be negative (as usual in kernelspace)
1114 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1116 struct netlink_set_err_data info;
1117 struct hlist_node *node;
1118 struct sock *sk;
1120 info.exclude_sk = ssk;
1121 info.pid = pid;
1122 info.group = group;
1123 /* sk->sk_err wants a positive error value */
1124 info.code = -code;
1126 read_lock(&nl_table_lock);
1128 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1129 do_one_set_err(sk, &info);
1131 read_unlock(&nl_table_lock);
1133 EXPORT_SYMBOL(netlink_set_err);
1135 /* must be called with netlink table grabbed */
1136 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1137 unsigned int group,
1138 int is_new)
1140 int old, new = !!is_new, subscriptions;
1142 old = test_bit(group - 1, nlk->groups);
1143 subscriptions = nlk->subscriptions - old + new;
1144 if (new)
1145 __set_bit(group - 1, nlk->groups);
1146 else
1147 __clear_bit(group - 1, nlk->groups);
1148 netlink_update_subscriptions(&nlk->sk, subscriptions);
1149 netlink_update_listeners(&nlk->sk);
1152 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1153 char __user *optval, unsigned int optlen)
1155 struct sock *sk = sock->sk;
1156 struct netlink_sock *nlk = nlk_sk(sk);
1157 unsigned int val = 0;
1158 int err;
1160 if (level != SOL_NETLINK)
1161 return -ENOPROTOOPT;
1163 if (optlen >= sizeof(int) &&
1164 get_user(val, (unsigned int __user *)optval))
1165 return -EFAULT;
1167 switch (optname) {
1168 case NETLINK_PKTINFO:
1169 if (val)
1170 nlk->flags |= NETLINK_RECV_PKTINFO;
1171 else
1172 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1173 err = 0;
1174 break;
1175 case NETLINK_ADD_MEMBERSHIP:
1176 case NETLINK_DROP_MEMBERSHIP: {
1177 if (!netlink_capable(sock, NL_NONROOT_RECV))
1178 return -EPERM;
1179 err = netlink_realloc_groups(sk);
1180 if (err)
1181 return err;
1182 if (!val || val - 1 >= nlk->ngroups)
1183 return -EINVAL;
1184 netlink_table_grab();
1185 netlink_update_socket_mc(nlk, val,
1186 optname == NETLINK_ADD_MEMBERSHIP);
1187 netlink_table_ungrab();
1188 err = 0;
1189 break;
1191 case NETLINK_BROADCAST_ERROR:
1192 if (val)
1193 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1194 else
1195 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1196 err = 0;
1197 break;
1198 case NETLINK_NO_ENOBUFS:
1199 if (val) {
1200 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1201 clear_bit(0, &nlk->state);
1202 wake_up_interruptible(&nlk->wait);
1203 } else
1204 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1205 err = 0;
1206 break;
1207 default:
1208 err = -ENOPROTOOPT;
1210 return err;
1213 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1214 char __user *optval, int __user *optlen)
1216 struct sock *sk = sock->sk;
1217 struct netlink_sock *nlk = nlk_sk(sk);
1218 int len, val, err;
1220 if (level != SOL_NETLINK)
1221 return -ENOPROTOOPT;
1223 if (get_user(len, optlen))
1224 return -EFAULT;
1225 if (len < 0)
1226 return -EINVAL;
1228 switch (optname) {
1229 case NETLINK_PKTINFO:
1230 if (len < sizeof(int))
1231 return -EINVAL;
1232 len = sizeof(int);
1233 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1234 if (put_user(len, optlen) ||
1235 put_user(val, optval))
1236 return -EFAULT;
1237 err = 0;
1238 break;
1239 case NETLINK_BROADCAST_ERROR:
1240 if (len < sizeof(int))
1241 return -EINVAL;
1242 len = sizeof(int);
1243 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1244 if (put_user(len, optlen) ||
1245 put_user(val, optval))
1246 return -EFAULT;
1247 err = 0;
1248 break;
1249 case NETLINK_NO_ENOBUFS:
1250 if (len < sizeof(int))
1251 return -EINVAL;
1252 len = sizeof(int);
1253 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1254 if (put_user(len, optlen) ||
1255 put_user(val, optval))
1256 return -EFAULT;
1257 err = 0;
1258 break;
1259 default:
1260 err = -ENOPROTOOPT;
1262 return err;
1265 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1267 struct nl_pktinfo info;
1269 info.group = NETLINK_CB(skb).dst_group;
1270 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1273 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1274 struct msghdr *msg, size_t len)
1276 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1277 struct sock *sk = sock->sk;
1278 struct netlink_sock *nlk = nlk_sk(sk);
1279 struct sockaddr_nl *addr = msg->msg_name;
1280 u32 dst_pid;
1281 u32 dst_group;
1282 struct sk_buff *skb;
1283 int err;
1284 struct scm_cookie scm;
1286 if (msg->msg_flags&MSG_OOB)
1287 return -EOPNOTSUPP;
1289 if (NULL == siocb->scm)
1290 siocb->scm = &scm;
1291 err = scm_send(sock, msg, siocb->scm);
1292 if (err < 0)
1293 return err;
1295 if (msg->msg_namelen) {
1296 if (addr->nl_family != AF_NETLINK)
1297 return -EINVAL;
1298 dst_pid = addr->nl_pid;
1299 dst_group = ffs(addr->nl_groups);
1300 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1301 return -EPERM;
1302 } else {
1303 dst_pid = nlk->dst_pid;
1304 dst_group = nlk->dst_group;
1307 if (!nlk->pid) {
1308 err = netlink_autobind(sock);
1309 if (err)
1310 goto out;
1313 err = -EMSGSIZE;
1314 if (len > sk->sk_sndbuf - 32)
1315 goto out;
1316 err = -ENOBUFS;
1317 skb = alloc_skb(len, GFP_KERNEL);
1318 if (skb == NULL)
1319 goto out;
1321 NETLINK_CB(skb).pid = nlk->pid;
1322 NETLINK_CB(skb).dst_group = dst_group;
1323 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1324 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1325 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1326 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1328 /* What can I do? Netlink is asynchronous, so that
1329 we will have to save current capabilities to
1330 check them, when this message will be delivered
1331 to corresponding kernel module. --ANK (980802)
1334 err = -EFAULT;
1335 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1336 kfree_skb(skb);
1337 goto out;
1340 err = security_netlink_send(sk, skb);
1341 if (err) {
1342 kfree_skb(skb);
1343 goto out;
1346 if (dst_group) {
1347 atomic_inc(&skb->users);
1348 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1350 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1352 out:
1353 return err;
1356 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1357 struct msghdr *msg, size_t len,
1358 int flags)
1360 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1361 struct scm_cookie scm;
1362 struct sock *sk = sock->sk;
1363 struct netlink_sock *nlk = nlk_sk(sk);
1364 int noblock = flags&MSG_DONTWAIT;
1365 size_t copied;
1366 struct sk_buff *skb, *frag __maybe_unused = NULL;
1367 int err;
1369 if (flags&MSG_OOB)
1370 return -EOPNOTSUPP;
1372 copied = 0;
1374 skb = skb_recv_datagram(sk, flags, noblock, &err);
1375 if (skb == NULL)
1376 goto out;
1378 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1379 if (unlikely(skb_shinfo(skb)->frag_list)) {
1380 bool need_compat = !!(flags & MSG_CMSG_COMPAT);
1383 * If this skb has a frag_list, then here that means that
1384 * we will have to use the frag_list skb for compat tasks
1385 * and the regular skb for non-compat tasks.
1387 * The skb might (and likely will) be cloned, so we can't
1388 * just reset frag_list and go on with things -- we need to
1389 * keep that. For the compat case that's easy -- simply get
1390 * a reference to the compat skb and free the regular one
1391 * including the frag. For the non-compat case, we need to
1392 * avoid sending the frag to the user -- so assign NULL but
1393 * restore it below before freeing the skb.
1395 if (need_compat) {
1396 struct sk_buff *compskb = skb_shinfo(skb)->frag_list;
1397 skb_get(compskb);
1398 kfree_skb(skb);
1399 skb = compskb;
1400 } else {
1401 frag = skb_shinfo(skb)->frag_list;
1402 skb_shinfo(skb)->frag_list = NULL;
1405 #endif
1407 msg->msg_namelen = 0;
1409 copied = skb->len;
1410 if (len < copied) {
1411 msg->msg_flags |= MSG_TRUNC;
1412 copied = len;
1415 skb_reset_transport_header(skb);
1416 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1418 if (msg->msg_name) {
1419 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1420 addr->nl_family = AF_NETLINK;
1421 addr->nl_pad = 0;
1422 addr->nl_pid = NETLINK_CB(skb).pid;
1423 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1424 msg->msg_namelen = sizeof(*addr);
1427 if (nlk->flags & NETLINK_RECV_PKTINFO)
1428 netlink_cmsg_recv_pktinfo(msg, skb);
1430 if (NULL == siocb->scm) {
1431 memset(&scm, 0, sizeof(scm));
1432 siocb->scm = &scm;
1434 siocb->scm->creds = *NETLINK_CREDS(skb);
1435 if (flags & MSG_TRUNC)
1436 copied = skb->len;
1438 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1439 skb_shinfo(skb)->frag_list = frag;
1440 #endif
1442 skb_free_datagram(sk, skb);
1444 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1445 netlink_dump(sk);
1447 scm_recv(sock, msg, siocb->scm, flags);
1448 out:
1449 netlink_rcv_wake(sk);
1450 return err ? : copied;
1453 static void netlink_data_ready(struct sock *sk, int len)
1455 BUG();
1459 * We export these functions to other modules. They provide a
1460 * complete set of kernel non-blocking support for message
1461 * queueing.
1464 struct sock *
1465 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1466 void (*input)(struct sk_buff *skb),
1467 struct mutex *cb_mutex, struct module *module)
1469 struct socket *sock;
1470 struct sock *sk;
1471 struct netlink_sock *nlk;
1472 unsigned long *listeners = NULL;
1474 BUG_ON(!nl_table);
1476 if (unit < 0 || unit >= MAX_LINKS)
1477 return NULL;
1479 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1480 return NULL;
1483 * We have to just have a reference on the net from sk, but don't
1484 * get_net it. Besides, we cannot get and then put the net here.
1485 * So we create one inside init_net and the move it to net.
1488 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1489 goto out_sock_release_nosk;
1491 sk = sock->sk;
1492 sk_change_net(sk, net);
1494 if (groups < 32)
1495 groups = 32;
1497 listeners = kzalloc(NLGRPSZ(groups) + sizeof(struct listeners_rcu_head),
1498 GFP_KERNEL);
1499 if (!listeners)
1500 goto out_sock_release;
1502 sk->sk_data_ready = netlink_data_ready;
1503 if (input)
1504 nlk_sk(sk)->netlink_rcv = input;
1506 if (netlink_insert(sk, net, 0))
1507 goto out_sock_release;
1509 nlk = nlk_sk(sk);
1510 nlk->flags |= NETLINK_KERNEL_SOCKET;
1512 netlink_table_grab();
1513 if (!nl_table[unit].registered) {
1514 nl_table[unit].groups = groups;
1515 nl_table[unit].listeners = listeners;
1516 nl_table[unit].cb_mutex = cb_mutex;
1517 nl_table[unit].module = module;
1518 nl_table[unit].registered = 1;
1519 } else {
1520 kfree(listeners);
1521 nl_table[unit].registered++;
1523 netlink_table_ungrab();
1524 return sk;
1526 out_sock_release:
1527 kfree(listeners);
1528 netlink_kernel_release(sk);
1529 return NULL;
1531 out_sock_release_nosk:
1532 sock_release(sock);
1533 return NULL;
1535 EXPORT_SYMBOL(netlink_kernel_create);
1538 void
1539 netlink_kernel_release(struct sock *sk)
1541 sk_release_kernel(sk);
1543 EXPORT_SYMBOL(netlink_kernel_release);
1546 static void netlink_free_old_listeners(struct rcu_head *rcu_head)
1548 struct listeners_rcu_head *lrh;
1550 lrh = container_of(rcu_head, struct listeners_rcu_head, rcu_head);
1551 kfree(lrh->ptr);
1554 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1556 unsigned long *listeners, *old = NULL;
1557 struct listeners_rcu_head *old_rcu_head;
1558 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1560 if (groups < 32)
1561 groups = 32;
1563 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1564 listeners = kzalloc(NLGRPSZ(groups) +
1565 sizeof(struct listeners_rcu_head),
1566 GFP_ATOMIC);
1567 if (!listeners)
1568 return -ENOMEM;
1569 old = tbl->listeners;
1570 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1571 rcu_assign_pointer(tbl->listeners, listeners);
1573 * Free the old memory after an RCU grace period so we
1574 * don't leak it. We use call_rcu() here in order to be
1575 * able to call this function from atomic contexts. The
1576 * allocation of this memory will have reserved enough
1577 * space for struct listeners_rcu_head at the end.
1579 old_rcu_head = (void *)(tbl->listeners +
1580 NLGRPLONGS(tbl->groups));
1581 old_rcu_head->ptr = old;
1582 call_rcu(&old_rcu_head->rcu_head, netlink_free_old_listeners);
1584 tbl->groups = groups;
1586 return 0;
1590 * netlink_change_ngroups - change number of multicast groups
1592 * This changes the number of multicast groups that are available
1593 * on a certain netlink family. Note that it is not possible to
1594 * change the number of groups to below 32. Also note that it does
1595 * not implicitly call netlink_clear_multicast_users() when the
1596 * number of groups is reduced.
1598 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1599 * @groups: The new number of groups.
1601 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1603 int err;
1605 netlink_table_grab();
1606 err = __netlink_change_ngroups(sk, groups);
1607 netlink_table_ungrab();
1609 return err;
1612 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1614 struct sock *sk;
1615 struct hlist_node *node;
1616 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1618 sk_for_each_bound(sk, node, &tbl->mc_list)
1619 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1623 * netlink_clear_multicast_users - kick off multicast listeners
1625 * This function removes all listeners from the given group.
1626 * @ksk: The kernel netlink socket, as returned by
1627 * netlink_kernel_create().
1628 * @group: The multicast group to clear.
1630 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1632 netlink_table_grab();
1633 __netlink_clear_multicast_users(ksk, group);
1634 netlink_table_ungrab();
1637 void netlink_set_nonroot(int protocol, unsigned int flags)
1639 if ((unsigned int)protocol < MAX_LINKS)
1640 nl_table[protocol].nl_nonroot = flags;
1642 EXPORT_SYMBOL(netlink_set_nonroot);
1644 static void netlink_destroy_callback(struct netlink_callback *cb)
1646 kfree_skb(cb->skb);
1647 kfree(cb);
1651 * It looks a bit ugly.
1652 * It would be better to create kernel thread.
1655 static int netlink_dump(struct sock *sk)
1657 struct netlink_sock *nlk = nlk_sk(sk);
1658 struct netlink_callback *cb;
1659 struct sk_buff *skb;
1660 struct nlmsghdr *nlh;
1661 int len, err = -ENOBUFS;
1663 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1664 if (!skb)
1665 goto errout;
1667 mutex_lock(nlk->cb_mutex);
1669 cb = nlk->cb;
1670 if (cb == NULL) {
1671 err = -EINVAL;
1672 goto errout_skb;
1675 len = cb->dump(skb, cb);
1677 if (len > 0) {
1678 mutex_unlock(nlk->cb_mutex);
1680 if (sk_filter(sk, skb))
1681 kfree_skb(skb);
1682 else {
1683 skb_queue_tail(&sk->sk_receive_queue, skb);
1684 sk->sk_data_ready(sk, skb->len);
1686 return 0;
1689 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1690 if (!nlh)
1691 goto errout_skb;
1693 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1695 if (sk_filter(sk, skb))
1696 kfree_skb(skb);
1697 else {
1698 skb_queue_tail(&sk->sk_receive_queue, skb);
1699 sk->sk_data_ready(sk, skb->len);
1702 if (cb->done)
1703 cb->done(cb);
1704 nlk->cb = NULL;
1705 mutex_unlock(nlk->cb_mutex);
1707 netlink_destroy_callback(cb);
1708 return 0;
1710 errout_skb:
1711 mutex_unlock(nlk->cb_mutex);
1712 kfree_skb(skb);
1713 errout:
1714 return err;
1717 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1718 const struct nlmsghdr *nlh,
1719 int (*dump)(struct sk_buff *skb,
1720 struct netlink_callback *),
1721 int (*done)(struct netlink_callback *))
1723 struct netlink_callback *cb;
1724 struct sock *sk;
1725 struct netlink_sock *nlk;
1727 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1728 if (cb == NULL)
1729 return -ENOBUFS;
1731 cb->dump = dump;
1732 cb->done = done;
1733 cb->nlh = nlh;
1734 atomic_inc(&skb->users);
1735 cb->skb = skb;
1737 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1738 if (sk == NULL) {
1739 netlink_destroy_callback(cb);
1740 return -ECONNREFUSED;
1742 nlk = nlk_sk(sk);
1743 /* A dump is in progress... */
1744 mutex_lock(nlk->cb_mutex);
1745 if (nlk->cb) {
1746 mutex_unlock(nlk->cb_mutex);
1747 netlink_destroy_callback(cb);
1748 sock_put(sk);
1749 return -EBUSY;
1751 nlk->cb = cb;
1752 mutex_unlock(nlk->cb_mutex);
1754 netlink_dump(sk);
1755 sock_put(sk);
1757 /* We successfully started a dump, by returning -EINTR we
1758 * signal not to send ACK even if it was requested.
1760 return -EINTR;
1762 EXPORT_SYMBOL(netlink_dump_start);
1764 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1766 struct sk_buff *skb;
1767 struct nlmsghdr *rep;
1768 struct nlmsgerr *errmsg;
1769 size_t payload = sizeof(*errmsg);
1771 /* error messages get the original request appened */
1772 if (err)
1773 payload += nlmsg_len(nlh);
1775 skb = nlmsg_new(payload, GFP_KERNEL);
1776 if (!skb) {
1777 struct sock *sk;
1779 sk = netlink_lookup(sock_net(in_skb->sk),
1780 in_skb->sk->sk_protocol,
1781 NETLINK_CB(in_skb).pid);
1782 if (sk) {
1783 sk->sk_err = ENOBUFS;
1784 sk->sk_error_report(sk);
1785 sock_put(sk);
1787 return;
1790 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1791 NLMSG_ERROR, payload, 0);
1792 errmsg = nlmsg_data(rep);
1793 errmsg->error = err;
1794 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1795 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1797 EXPORT_SYMBOL(netlink_ack);
1799 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1800 struct nlmsghdr *))
1802 struct nlmsghdr *nlh;
1803 int err;
1805 while (skb->len >= nlmsg_total_size(0)) {
1806 int msglen;
1808 nlh = nlmsg_hdr(skb);
1809 err = 0;
1811 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1812 return 0;
1814 /* Only requests are handled by the kernel */
1815 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1816 goto ack;
1818 /* Skip control messages */
1819 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1820 goto ack;
1822 err = cb(skb, nlh);
1823 if (err == -EINTR)
1824 goto skip;
1826 ack:
1827 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1828 netlink_ack(skb, nlh, err);
1830 skip:
1831 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1832 if (msglen > skb->len)
1833 msglen = skb->len;
1834 skb_pull(skb, msglen);
1837 return 0;
1839 EXPORT_SYMBOL(netlink_rcv_skb);
1842 * nlmsg_notify - send a notification netlink message
1843 * @sk: netlink socket to use
1844 * @skb: notification message
1845 * @pid: destination netlink pid for reports or 0
1846 * @group: destination multicast group or 0
1847 * @report: 1 to report back, 0 to disable
1848 * @flags: allocation flags
1850 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1851 unsigned int group, int report, gfp_t flags)
1853 int err = 0;
1855 if (group) {
1856 int exclude_pid = 0;
1858 if (report) {
1859 atomic_inc(&skb->users);
1860 exclude_pid = pid;
1863 /* errors reported via destination sk->sk_err, but propagate
1864 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1865 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1868 if (report) {
1869 int err2;
1871 err2 = nlmsg_unicast(sk, skb, pid);
1872 if (!err || err == -ESRCH)
1873 err = err2;
1876 return err;
1878 EXPORT_SYMBOL(nlmsg_notify);
1880 #ifdef CONFIG_PROC_FS
1881 struct nl_seq_iter {
1882 struct seq_net_private p;
1883 int link;
1884 int hash_idx;
1887 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1889 struct nl_seq_iter *iter = seq->private;
1890 int i, j;
1891 struct sock *s;
1892 struct hlist_node *node;
1893 loff_t off = 0;
1895 for (i = 0; i < MAX_LINKS; i++) {
1896 struct nl_pid_hash *hash = &nl_table[i].hash;
1898 for (j = 0; j <= hash->mask; j++) {
1899 sk_for_each(s, node, &hash->table[j]) {
1900 if (sock_net(s) != seq_file_net(seq))
1901 continue;
1902 if (off == pos) {
1903 iter->link = i;
1904 iter->hash_idx = j;
1905 return s;
1907 ++off;
1911 return NULL;
1914 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1915 __acquires(nl_table_lock)
1917 read_lock(&nl_table_lock);
1918 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1921 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1923 struct sock *s;
1924 struct nl_seq_iter *iter;
1925 int i, j;
1927 ++*pos;
1929 if (v == SEQ_START_TOKEN)
1930 return netlink_seq_socket_idx(seq, 0);
1932 iter = seq->private;
1933 s = v;
1934 do {
1935 s = sk_next(s);
1936 } while (s && sock_net(s) != seq_file_net(seq));
1937 if (s)
1938 return s;
1940 i = iter->link;
1941 j = iter->hash_idx + 1;
1943 do {
1944 struct nl_pid_hash *hash = &nl_table[i].hash;
1946 for (; j <= hash->mask; j++) {
1947 s = sk_head(&hash->table[j]);
1948 while (s && sock_net(s) != seq_file_net(seq))
1949 s = sk_next(s);
1950 if (s) {
1951 iter->link = i;
1952 iter->hash_idx = j;
1953 return s;
1957 j = 0;
1958 } while (++i < MAX_LINKS);
1960 return NULL;
1963 static void netlink_seq_stop(struct seq_file *seq, void *v)
1964 __releases(nl_table_lock)
1966 read_unlock(&nl_table_lock);
1970 static int netlink_seq_show(struct seq_file *seq, void *v)
1972 if (v == SEQ_START_TOKEN)
1973 seq_puts(seq,
1974 "sk Eth Pid Groups "
1975 "Rmem Wmem Dump Locks Drops\n");
1976 else {
1977 struct sock *s = v;
1978 struct netlink_sock *nlk = nlk_sk(s);
1980 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %-8d %-8d\n",
1982 s->sk_protocol,
1983 nlk->pid,
1984 nlk->groups ? (u32)nlk->groups[0] : 0,
1985 sk_rmem_alloc_get(s),
1986 sk_wmem_alloc_get(s),
1987 nlk->cb,
1988 atomic_read(&s->sk_refcnt),
1989 atomic_read(&s->sk_drops)
1993 return 0;
1996 static const struct seq_operations netlink_seq_ops = {
1997 .start = netlink_seq_start,
1998 .next = netlink_seq_next,
1999 .stop = netlink_seq_stop,
2000 .show = netlink_seq_show,
2004 static int netlink_seq_open(struct inode *inode, struct file *file)
2006 return seq_open_net(inode, file, &netlink_seq_ops,
2007 sizeof(struct nl_seq_iter));
2010 static const struct file_operations netlink_seq_fops = {
2011 .owner = THIS_MODULE,
2012 .open = netlink_seq_open,
2013 .read = seq_read,
2014 .llseek = seq_lseek,
2015 .release = seq_release_net,
2018 #endif
2020 int netlink_register_notifier(struct notifier_block *nb)
2022 return atomic_notifier_chain_register(&netlink_chain, nb);
2024 EXPORT_SYMBOL(netlink_register_notifier);
2026 int netlink_unregister_notifier(struct notifier_block *nb)
2028 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2030 EXPORT_SYMBOL(netlink_unregister_notifier);
2032 static const struct proto_ops netlink_ops = {
2033 .family = PF_NETLINK,
2034 .owner = THIS_MODULE,
2035 .release = netlink_release,
2036 .bind = netlink_bind,
2037 .connect = netlink_connect,
2038 .socketpair = sock_no_socketpair,
2039 .accept = sock_no_accept,
2040 .getname = netlink_getname,
2041 .poll = datagram_poll,
2042 .ioctl = sock_no_ioctl,
2043 .listen = sock_no_listen,
2044 .shutdown = sock_no_shutdown,
2045 .setsockopt = netlink_setsockopt,
2046 .getsockopt = netlink_getsockopt,
2047 .sendmsg = netlink_sendmsg,
2048 .recvmsg = netlink_recvmsg,
2049 .mmap = sock_no_mmap,
2050 .sendpage = sock_no_sendpage,
2053 static const struct net_proto_family netlink_family_ops = {
2054 .family = PF_NETLINK,
2055 .create = netlink_create,
2056 .owner = THIS_MODULE, /* for consistency 8) */
2059 static int __net_init netlink_net_init(struct net *net)
2061 #ifdef CONFIG_PROC_FS
2062 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2063 return -ENOMEM;
2064 #endif
2065 return 0;
2068 static void __net_exit netlink_net_exit(struct net *net)
2070 #ifdef CONFIG_PROC_FS
2071 proc_net_remove(net, "netlink");
2072 #endif
2075 static struct pernet_operations __net_initdata netlink_net_ops = {
2076 .init = netlink_net_init,
2077 .exit = netlink_net_exit,
2080 static int __init netlink_proto_init(void)
2082 struct sk_buff *dummy_skb;
2083 int i;
2084 unsigned long limit;
2085 unsigned int order;
2086 int err = proto_register(&netlink_proto, 0);
2088 if (err != 0)
2089 goto out;
2091 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2093 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2094 if (!nl_table)
2095 goto panic;
2097 if (totalram_pages >= (128 * 1024))
2098 limit = totalram_pages >> (21 - PAGE_SHIFT);
2099 else
2100 limit = totalram_pages >> (23 - PAGE_SHIFT);
2102 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2103 limit = (1UL << order) / sizeof(struct hlist_head);
2104 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2106 for (i = 0; i < MAX_LINKS; i++) {
2107 struct nl_pid_hash *hash = &nl_table[i].hash;
2109 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2110 if (!hash->table) {
2111 while (i-- > 0)
2112 nl_pid_hash_free(nl_table[i].hash.table,
2113 1 * sizeof(*hash->table));
2114 kfree(nl_table);
2115 goto panic;
2117 hash->max_shift = order;
2118 hash->shift = 0;
2119 hash->mask = 0;
2120 hash->rehash_time = jiffies;
2123 sock_register(&netlink_family_ops);
2124 register_pernet_subsys(&netlink_net_ops);
2125 /* The netlink device handler may be needed early. */
2126 rtnetlink_init();
2127 out:
2128 return err;
2129 panic:
2130 panic("netlink_init: Cannot allocate nl_table\n");
2133 core_initcall(netlink_proto_init);