carl9170: PHY/RF and MAC routines
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / netlink / af_netlink.c
blob2cbf380377d5e009fa972d85af18ded9971a5248
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,
432 int kern)
434 struct module *module = NULL;
435 struct mutex *cb_mutex;
436 struct netlink_sock *nlk;
437 int err = 0;
439 sock->state = SS_UNCONNECTED;
441 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
442 return -ESOCKTNOSUPPORT;
444 if (protocol < 0 || protocol >= MAX_LINKS)
445 return -EPROTONOSUPPORT;
447 netlink_lock_table();
448 #ifdef CONFIG_MODULES
449 if (!nl_table[protocol].registered) {
450 netlink_unlock_table();
451 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
452 netlink_lock_table();
454 #endif
455 if (nl_table[protocol].registered &&
456 try_module_get(nl_table[protocol].module))
457 module = nl_table[protocol].module;
458 else
459 err = -EPROTONOSUPPORT;
460 cb_mutex = nl_table[protocol].cb_mutex;
461 netlink_unlock_table();
463 if (err < 0)
464 goto out;
466 err = __netlink_create(net, sock, cb_mutex, protocol);
467 if (err < 0)
468 goto out_module;
470 local_bh_disable();
471 sock_prot_inuse_add(net, &netlink_proto, 1);
472 local_bh_enable();
474 nlk = nlk_sk(sock->sk);
475 nlk->module = module;
476 out:
477 return err;
479 out_module:
480 module_put(module);
481 goto out;
484 static int netlink_release(struct socket *sock)
486 struct sock *sk = sock->sk;
487 struct netlink_sock *nlk;
489 if (!sk)
490 return 0;
492 netlink_remove(sk);
493 sock_orphan(sk);
494 nlk = nlk_sk(sk);
497 * OK. Socket is unlinked, any packets that arrive now
498 * will be purged.
501 sock->sk = NULL;
502 wake_up_interruptible_all(&nlk->wait);
504 skb_queue_purge(&sk->sk_write_queue);
506 if (nlk->pid) {
507 struct netlink_notify n = {
508 .net = sock_net(sk),
509 .protocol = sk->sk_protocol,
510 .pid = nlk->pid,
512 atomic_notifier_call_chain(&netlink_chain,
513 NETLINK_URELEASE, &n);
516 module_put(nlk->module);
518 netlink_table_grab();
519 if (netlink_is_kernel(sk)) {
520 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
521 if (--nl_table[sk->sk_protocol].registered == 0) {
522 kfree(nl_table[sk->sk_protocol].listeners);
523 nl_table[sk->sk_protocol].module = NULL;
524 nl_table[sk->sk_protocol].registered = 0;
526 } else if (nlk->subscriptions)
527 netlink_update_listeners(sk);
528 netlink_table_ungrab();
530 kfree(nlk->groups);
531 nlk->groups = NULL;
533 local_bh_disable();
534 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
535 local_bh_enable();
536 sock_put(sk);
537 return 0;
540 static int netlink_autobind(struct socket *sock)
542 struct sock *sk = sock->sk;
543 struct net *net = sock_net(sk);
544 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
545 struct hlist_head *head;
546 struct sock *osk;
547 struct hlist_node *node;
548 s32 pid = task_tgid_vnr(current);
549 int err;
550 static s32 rover = -4097;
552 retry:
553 cond_resched();
554 netlink_table_grab();
555 head = nl_pid_hashfn(hash, pid);
556 sk_for_each(osk, node, head) {
557 if (!net_eq(sock_net(osk), net))
558 continue;
559 if (nlk_sk(osk)->pid == pid) {
560 /* Bind collision, search negative pid values. */
561 pid = rover--;
562 if (rover > -4097)
563 rover = -4097;
564 netlink_table_ungrab();
565 goto retry;
568 netlink_table_ungrab();
570 err = netlink_insert(sk, net, pid);
571 if (err == -EADDRINUSE)
572 goto retry;
574 /* If 2 threads race to autobind, that is fine. */
575 if (err == -EBUSY)
576 err = 0;
578 return err;
581 static inline int netlink_capable(struct socket *sock, unsigned int flag)
583 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
584 capable(CAP_NET_ADMIN);
587 static void
588 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
590 struct netlink_sock *nlk = nlk_sk(sk);
592 if (nlk->subscriptions && !subscriptions)
593 __sk_del_bind_node(sk);
594 else if (!nlk->subscriptions && subscriptions)
595 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
596 nlk->subscriptions = subscriptions;
599 static int netlink_realloc_groups(struct sock *sk)
601 struct netlink_sock *nlk = nlk_sk(sk);
602 unsigned int groups;
603 unsigned long *new_groups;
604 int err = 0;
606 netlink_table_grab();
608 groups = nl_table[sk->sk_protocol].groups;
609 if (!nl_table[sk->sk_protocol].registered) {
610 err = -ENOENT;
611 goto out_unlock;
614 if (nlk->ngroups >= groups)
615 goto out_unlock;
617 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
618 if (new_groups == NULL) {
619 err = -ENOMEM;
620 goto out_unlock;
622 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
623 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
625 nlk->groups = new_groups;
626 nlk->ngroups = groups;
627 out_unlock:
628 netlink_table_ungrab();
629 return err;
632 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
633 int addr_len)
635 struct sock *sk = sock->sk;
636 struct net *net = sock_net(sk);
637 struct netlink_sock *nlk = nlk_sk(sk);
638 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
639 int err;
641 if (nladdr->nl_family != AF_NETLINK)
642 return -EINVAL;
644 /* Only superuser is allowed to listen multicasts */
645 if (nladdr->nl_groups) {
646 if (!netlink_capable(sock, NL_NONROOT_RECV))
647 return -EPERM;
648 err = netlink_realloc_groups(sk);
649 if (err)
650 return err;
653 if (nlk->pid) {
654 if (nladdr->nl_pid != nlk->pid)
655 return -EINVAL;
656 } else {
657 err = nladdr->nl_pid ?
658 netlink_insert(sk, net, nladdr->nl_pid) :
659 netlink_autobind(sock);
660 if (err)
661 return err;
664 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
665 return 0;
667 netlink_table_grab();
668 netlink_update_subscriptions(sk, nlk->subscriptions +
669 hweight32(nladdr->nl_groups) -
670 hweight32(nlk->groups[0]));
671 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
672 netlink_update_listeners(sk);
673 netlink_table_ungrab();
675 return 0;
678 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
679 int alen, int flags)
681 int err = 0;
682 struct sock *sk = sock->sk;
683 struct netlink_sock *nlk = nlk_sk(sk);
684 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
686 if (alen < sizeof(addr->sa_family))
687 return -EINVAL;
689 if (addr->sa_family == AF_UNSPEC) {
690 sk->sk_state = NETLINK_UNCONNECTED;
691 nlk->dst_pid = 0;
692 nlk->dst_group = 0;
693 return 0;
695 if (addr->sa_family != AF_NETLINK)
696 return -EINVAL;
698 /* Only superuser is allowed to send multicasts */
699 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
700 return -EPERM;
702 if (!nlk->pid)
703 err = netlink_autobind(sock);
705 if (err == 0) {
706 sk->sk_state = NETLINK_CONNECTED;
707 nlk->dst_pid = nladdr->nl_pid;
708 nlk->dst_group = ffs(nladdr->nl_groups);
711 return err;
714 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
715 int *addr_len, int peer)
717 struct sock *sk = sock->sk;
718 struct netlink_sock *nlk = nlk_sk(sk);
719 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
721 nladdr->nl_family = AF_NETLINK;
722 nladdr->nl_pad = 0;
723 *addr_len = sizeof(*nladdr);
725 if (peer) {
726 nladdr->nl_pid = nlk->dst_pid;
727 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
728 } else {
729 nladdr->nl_pid = nlk->pid;
730 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
732 return 0;
735 static void netlink_overrun(struct sock *sk)
737 struct netlink_sock *nlk = nlk_sk(sk);
739 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
740 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
741 sk->sk_err = ENOBUFS;
742 sk->sk_error_report(sk);
745 atomic_inc(&sk->sk_drops);
748 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
750 struct sock *sock;
751 struct netlink_sock *nlk;
753 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
754 if (!sock)
755 return ERR_PTR(-ECONNREFUSED);
757 /* Don't bother queuing skb if kernel socket has no input function */
758 nlk = nlk_sk(sock);
759 if (sock->sk_state == NETLINK_CONNECTED &&
760 nlk->dst_pid != nlk_sk(ssk)->pid) {
761 sock_put(sock);
762 return ERR_PTR(-ECONNREFUSED);
764 return sock;
767 struct sock *netlink_getsockbyfilp(struct file *filp)
769 struct inode *inode = filp->f_path.dentry->d_inode;
770 struct sock *sock;
772 if (!S_ISSOCK(inode->i_mode))
773 return ERR_PTR(-ENOTSOCK);
775 sock = SOCKET_I(inode)->sk;
776 if (sock->sk_family != AF_NETLINK)
777 return ERR_PTR(-EINVAL);
779 sock_hold(sock);
780 return sock;
784 * Attach a skb to a netlink socket.
785 * The caller must hold a reference to the destination socket. On error, the
786 * reference is dropped. The skb is not send to the destination, just all
787 * all error checks are performed and memory in the queue is reserved.
788 * Return values:
789 * < 0: error. skb freed, reference to sock dropped.
790 * 0: continue
791 * 1: repeat lookup - reference dropped while waiting for socket memory.
793 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
794 long *timeo, struct sock *ssk)
796 struct netlink_sock *nlk;
798 nlk = nlk_sk(sk);
800 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
801 test_bit(0, &nlk->state)) {
802 DECLARE_WAITQUEUE(wait, current);
803 if (!*timeo) {
804 if (!ssk || netlink_is_kernel(ssk))
805 netlink_overrun(sk);
806 sock_put(sk);
807 kfree_skb(skb);
808 return -EAGAIN;
811 __set_current_state(TASK_INTERRUPTIBLE);
812 add_wait_queue(&nlk->wait, &wait);
814 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
815 test_bit(0, &nlk->state)) &&
816 !sock_flag(sk, SOCK_DEAD))
817 *timeo = schedule_timeout(*timeo);
819 __set_current_state(TASK_RUNNING);
820 remove_wait_queue(&nlk->wait, &wait);
821 sock_put(sk);
823 if (signal_pending(current)) {
824 kfree_skb(skb);
825 return sock_intr_errno(*timeo);
827 return 1;
829 skb_set_owner_r(skb, sk);
830 return 0;
833 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
835 int len = skb->len;
837 skb_queue_tail(&sk->sk_receive_queue, skb);
838 sk->sk_data_ready(sk, len);
839 sock_put(sk);
840 return len;
843 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
845 kfree_skb(skb);
846 sock_put(sk);
849 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
850 gfp_t allocation)
852 int delta;
854 skb_orphan(skb);
856 delta = skb->end - skb->tail;
857 if (delta * 2 < skb->truesize)
858 return skb;
860 if (skb_shared(skb)) {
861 struct sk_buff *nskb = skb_clone(skb, allocation);
862 if (!nskb)
863 return skb;
864 kfree_skb(skb);
865 skb = nskb;
868 if (!pskb_expand_head(skb, 0, -delta, allocation))
869 skb->truesize -= delta;
871 return skb;
874 static inline void netlink_rcv_wake(struct sock *sk)
876 struct netlink_sock *nlk = nlk_sk(sk);
878 if (skb_queue_empty(&sk->sk_receive_queue))
879 clear_bit(0, &nlk->state);
880 if (!test_bit(0, &nlk->state))
881 wake_up_interruptible(&nlk->wait);
884 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
886 int ret;
887 struct netlink_sock *nlk = nlk_sk(sk);
889 ret = -ECONNREFUSED;
890 if (nlk->netlink_rcv != NULL) {
891 ret = skb->len;
892 skb_set_owner_r(skb, sk);
893 nlk->netlink_rcv(skb);
895 kfree_skb(skb);
896 sock_put(sk);
897 return ret;
900 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
901 u32 pid, int nonblock)
903 struct sock *sk;
904 int err;
905 long timeo;
907 skb = netlink_trim(skb, gfp_any());
909 timeo = sock_sndtimeo(ssk, nonblock);
910 retry:
911 sk = netlink_getsockbypid(ssk, pid);
912 if (IS_ERR(sk)) {
913 kfree_skb(skb);
914 return PTR_ERR(sk);
916 if (netlink_is_kernel(sk))
917 return netlink_unicast_kernel(sk, skb);
919 if (sk_filter(sk, skb)) {
920 err = skb->len;
921 kfree_skb(skb);
922 sock_put(sk);
923 return err;
926 err = netlink_attachskb(sk, skb, &timeo, ssk);
927 if (err == 1)
928 goto retry;
929 if (err)
930 return err;
932 return netlink_sendskb(sk, skb);
934 EXPORT_SYMBOL(netlink_unicast);
936 int netlink_has_listeners(struct sock *sk, unsigned int group)
938 int res = 0;
939 unsigned long *listeners;
941 BUG_ON(!netlink_is_kernel(sk));
943 rcu_read_lock();
944 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
946 if (group - 1 < nl_table[sk->sk_protocol].groups)
947 res = test_bit(group - 1, listeners);
949 rcu_read_unlock();
951 return res;
953 EXPORT_SYMBOL_GPL(netlink_has_listeners);
955 static inline int netlink_broadcast_deliver(struct sock *sk,
956 struct sk_buff *skb)
958 struct netlink_sock *nlk = nlk_sk(sk);
960 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
961 !test_bit(0, &nlk->state)) {
962 skb_set_owner_r(skb, sk);
963 skb_queue_tail(&sk->sk_receive_queue, skb);
964 sk->sk_data_ready(sk, skb->len);
965 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
967 return -1;
970 struct netlink_broadcast_data {
971 struct sock *exclude_sk;
972 struct net *net;
973 u32 pid;
974 u32 group;
975 int failure;
976 int delivery_failure;
977 int congested;
978 int delivered;
979 gfp_t allocation;
980 struct sk_buff *skb, *skb2;
981 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
982 void *tx_data;
985 static inline int do_one_broadcast(struct sock *sk,
986 struct netlink_broadcast_data *p)
988 struct netlink_sock *nlk = nlk_sk(sk);
989 int val;
991 if (p->exclude_sk == sk)
992 goto out;
994 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
995 !test_bit(p->group - 1, nlk->groups))
996 goto out;
998 if (!net_eq(sock_net(sk), p->net))
999 goto out;
1001 if (p->failure) {
1002 netlink_overrun(sk);
1003 goto out;
1006 sock_hold(sk);
1007 if (p->skb2 == NULL) {
1008 if (skb_shared(p->skb)) {
1009 p->skb2 = skb_clone(p->skb, p->allocation);
1010 } else {
1011 p->skb2 = skb_get(p->skb);
1013 * skb ownership may have been set when
1014 * delivered to a previous socket.
1016 skb_orphan(p->skb2);
1019 if (p->skb2 == NULL) {
1020 netlink_overrun(sk);
1021 /* Clone failed. Notify ALL listeners. */
1022 p->failure = 1;
1023 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1024 p->delivery_failure = 1;
1025 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1026 kfree_skb(p->skb2);
1027 p->skb2 = NULL;
1028 } else if (sk_filter(sk, p->skb2)) {
1029 kfree_skb(p->skb2);
1030 p->skb2 = NULL;
1031 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1032 netlink_overrun(sk);
1033 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1034 p->delivery_failure = 1;
1035 } else {
1036 p->congested |= val;
1037 p->delivered = 1;
1038 p->skb2 = NULL;
1040 sock_put(sk);
1042 out:
1043 return 0;
1046 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid,
1047 u32 group, gfp_t allocation,
1048 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1049 void *filter_data)
1051 struct net *net = sock_net(ssk);
1052 struct netlink_broadcast_data info;
1053 struct hlist_node *node;
1054 struct sock *sk;
1056 skb = netlink_trim(skb, allocation);
1058 info.exclude_sk = ssk;
1059 info.net = net;
1060 info.pid = pid;
1061 info.group = group;
1062 info.failure = 0;
1063 info.delivery_failure = 0;
1064 info.congested = 0;
1065 info.delivered = 0;
1066 info.allocation = allocation;
1067 info.skb = skb;
1068 info.skb2 = NULL;
1069 info.tx_filter = filter;
1070 info.tx_data = filter_data;
1072 /* While we sleep in clone, do not allow to change socket list */
1074 netlink_lock_table();
1076 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1077 do_one_broadcast(sk, &info);
1079 consume_skb(skb);
1081 netlink_unlock_table();
1083 if (info.delivery_failure) {
1084 kfree_skb(info.skb2);
1085 return -ENOBUFS;
1086 } else
1087 consume_skb(info.skb2);
1089 if (info.delivered) {
1090 if (info.congested && (allocation & __GFP_WAIT))
1091 yield();
1092 return 0;
1094 return -ESRCH;
1096 EXPORT_SYMBOL(netlink_broadcast_filtered);
1098 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1099 u32 group, gfp_t allocation)
1101 return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1102 NULL, NULL);
1104 EXPORT_SYMBOL(netlink_broadcast);
1106 struct netlink_set_err_data {
1107 struct sock *exclude_sk;
1108 u32 pid;
1109 u32 group;
1110 int code;
1113 static inline int do_one_set_err(struct sock *sk,
1114 struct netlink_set_err_data *p)
1116 struct netlink_sock *nlk = nlk_sk(sk);
1117 int ret = 0;
1119 if (sk == p->exclude_sk)
1120 goto out;
1122 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1123 goto out;
1125 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1126 !test_bit(p->group - 1, nlk->groups))
1127 goto out;
1129 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1130 ret = 1;
1131 goto out;
1134 sk->sk_err = p->code;
1135 sk->sk_error_report(sk);
1136 out:
1137 return ret;
1141 * netlink_set_err - report error to broadcast listeners
1142 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1143 * @pid: the PID of a process that we want to skip (if any)
1144 * @groups: the broadcast group that will notice the error
1145 * @code: error code, must be negative (as usual in kernelspace)
1147 * This function returns the number of broadcast listeners that have set the
1148 * NETLINK_RECV_NO_ENOBUFS socket option.
1150 int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1152 struct netlink_set_err_data info;
1153 struct hlist_node *node;
1154 struct sock *sk;
1155 int ret = 0;
1157 info.exclude_sk = ssk;
1158 info.pid = pid;
1159 info.group = group;
1160 /* sk->sk_err wants a positive error value */
1161 info.code = -code;
1163 read_lock(&nl_table_lock);
1165 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1166 ret += do_one_set_err(sk, &info);
1168 read_unlock(&nl_table_lock);
1169 return ret;
1171 EXPORT_SYMBOL(netlink_set_err);
1173 /* must be called with netlink table grabbed */
1174 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1175 unsigned int group,
1176 int is_new)
1178 int old, new = !!is_new, subscriptions;
1180 old = test_bit(group - 1, nlk->groups);
1181 subscriptions = nlk->subscriptions - old + new;
1182 if (new)
1183 __set_bit(group - 1, nlk->groups);
1184 else
1185 __clear_bit(group - 1, nlk->groups);
1186 netlink_update_subscriptions(&nlk->sk, subscriptions);
1187 netlink_update_listeners(&nlk->sk);
1190 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1191 char __user *optval, unsigned int optlen)
1193 struct sock *sk = sock->sk;
1194 struct netlink_sock *nlk = nlk_sk(sk);
1195 unsigned int val = 0;
1196 int err;
1198 if (level != SOL_NETLINK)
1199 return -ENOPROTOOPT;
1201 if (optlen >= sizeof(int) &&
1202 get_user(val, (unsigned int __user *)optval))
1203 return -EFAULT;
1205 switch (optname) {
1206 case NETLINK_PKTINFO:
1207 if (val)
1208 nlk->flags |= NETLINK_RECV_PKTINFO;
1209 else
1210 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1211 err = 0;
1212 break;
1213 case NETLINK_ADD_MEMBERSHIP:
1214 case NETLINK_DROP_MEMBERSHIP: {
1215 if (!netlink_capable(sock, NL_NONROOT_RECV))
1216 return -EPERM;
1217 err = netlink_realloc_groups(sk);
1218 if (err)
1219 return err;
1220 if (!val || val - 1 >= nlk->ngroups)
1221 return -EINVAL;
1222 netlink_table_grab();
1223 netlink_update_socket_mc(nlk, val,
1224 optname == NETLINK_ADD_MEMBERSHIP);
1225 netlink_table_ungrab();
1226 err = 0;
1227 break;
1229 case NETLINK_BROADCAST_ERROR:
1230 if (val)
1231 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1232 else
1233 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1234 err = 0;
1235 break;
1236 case NETLINK_NO_ENOBUFS:
1237 if (val) {
1238 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1239 clear_bit(0, &nlk->state);
1240 wake_up_interruptible(&nlk->wait);
1241 } else
1242 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1243 err = 0;
1244 break;
1245 default:
1246 err = -ENOPROTOOPT;
1248 return err;
1251 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1252 char __user *optval, int __user *optlen)
1254 struct sock *sk = sock->sk;
1255 struct netlink_sock *nlk = nlk_sk(sk);
1256 int len, val, err;
1258 if (level != SOL_NETLINK)
1259 return -ENOPROTOOPT;
1261 if (get_user(len, optlen))
1262 return -EFAULT;
1263 if (len < 0)
1264 return -EINVAL;
1266 switch (optname) {
1267 case NETLINK_PKTINFO:
1268 if (len < sizeof(int))
1269 return -EINVAL;
1270 len = sizeof(int);
1271 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1272 if (put_user(len, optlen) ||
1273 put_user(val, optval))
1274 return -EFAULT;
1275 err = 0;
1276 break;
1277 case NETLINK_BROADCAST_ERROR:
1278 if (len < sizeof(int))
1279 return -EINVAL;
1280 len = sizeof(int);
1281 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1282 if (put_user(len, optlen) ||
1283 put_user(val, optval))
1284 return -EFAULT;
1285 err = 0;
1286 break;
1287 case NETLINK_NO_ENOBUFS:
1288 if (len < sizeof(int))
1289 return -EINVAL;
1290 len = sizeof(int);
1291 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1292 if (put_user(len, optlen) ||
1293 put_user(val, optval))
1294 return -EFAULT;
1295 err = 0;
1296 break;
1297 default:
1298 err = -ENOPROTOOPT;
1300 return err;
1303 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1305 struct nl_pktinfo info;
1307 info.group = NETLINK_CB(skb).dst_group;
1308 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1311 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1312 struct msghdr *msg, size_t len)
1314 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1315 struct sock *sk = sock->sk;
1316 struct netlink_sock *nlk = nlk_sk(sk);
1317 struct sockaddr_nl *addr = msg->msg_name;
1318 u32 dst_pid;
1319 u32 dst_group;
1320 struct sk_buff *skb;
1321 int err;
1322 struct scm_cookie scm;
1324 if (msg->msg_flags&MSG_OOB)
1325 return -EOPNOTSUPP;
1327 if (NULL == siocb->scm) {
1328 siocb->scm = &scm;
1329 memset(&scm, 0, sizeof(scm));
1331 err = scm_send(sock, msg, siocb->scm);
1332 if (err < 0)
1333 return err;
1335 if (msg->msg_namelen) {
1336 err = -EINVAL;
1337 if (addr->nl_family != AF_NETLINK)
1338 goto out;
1339 dst_pid = addr->nl_pid;
1340 dst_group = ffs(addr->nl_groups);
1341 err = -EPERM;
1342 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1343 goto out;
1344 } else {
1345 dst_pid = nlk->dst_pid;
1346 dst_group = nlk->dst_group;
1349 if (!nlk->pid) {
1350 err = netlink_autobind(sock);
1351 if (err)
1352 goto out;
1355 err = -EMSGSIZE;
1356 if (len > sk->sk_sndbuf - 32)
1357 goto out;
1358 err = -ENOBUFS;
1359 skb = alloc_skb(len, GFP_KERNEL);
1360 if (skb == NULL)
1361 goto out;
1363 NETLINK_CB(skb).pid = nlk->pid;
1364 NETLINK_CB(skb).dst_group = dst_group;
1365 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1366 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1367 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1368 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1370 /* What can I do? Netlink is asynchronous, so that
1371 we will have to save current capabilities to
1372 check them, when this message will be delivered
1373 to corresponding kernel module. --ANK (980802)
1376 err = -EFAULT;
1377 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1378 kfree_skb(skb);
1379 goto out;
1382 err = security_netlink_send(sk, skb);
1383 if (err) {
1384 kfree_skb(skb);
1385 goto out;
1388 if (dst_group) {
1389 atomic_inc(&skb->users);
1390 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1392 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1394 out:
1395 scm_destroy(siocb->scm);
1396 return err;
1399 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1400 struct msghdr *msg, size_t len,
1401 int flags)
1403 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1404 struct scm_cookie scm;
1405 struct sock *sk = sock->sk;
1406 struct netlink_sock *nlk = nlk_sk(sk);
1407 int noblock = flags&MSG_DONTWAIT;
1408 size_t copied;
1409 struct sk_buff *skb;
1410 int err;
1412 if (flags&MSG_OOB)
1413 return -EOPNOTSUPP;
1415 copied = 0;
1417 skb = skb_recv_datagram(sk, flags, noblock, &err);
1418 if (skb == NULL)
1419 goto out;
1421 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1422 if (unlikely(skb_shinfo(skb)->frag_list)) {
1423 bool need_compat = !!(flags & MSG_CMSG_COMPAT);
1426 * If this skb has a frag_list, then here that means that
1427 * we will have to use the frag_list skb for compat tasks
1428 * and the regular skb for non-compat tasks.
1430 * The skb might (and likely will) be cloned, so we can't
1431 * just reset frag_list and go on with things -- we need to
1432 * keep that. For the compat case that's easy -- simply get
1433 * a reference to the compat skb and free the regular one
1434 * including the frag. For the non-compat case, we need to
1435 * avoid sending the frag to the user -- so assign NULL but
1436 * restore it below before freeing the skb.
1438 if (need_compat) {
1439 struct sk_buff *compskb = skb_shinfo(skb)->frag_list;
1440 skb_get(compskb);
1441 kfree_skb(skb);
1442 skb = compskb;
1443 } else {
1445 * Before setting frag_list to NULL, we must get a
1446 * private copy of skb if shared (because of MSG_PEEK)
1448 if (skb_shared(skb)) {
1449 struct sk_buff *nskb;
1451 nskb = pskb_copy(skb, GFP_KERNEL);
1452 kfree_skb(skb);
1453 skb = nskb;
1454 err = -ENOMEM;
1455 if (!skb)
1456 goto out;
1458 kfree_skb(skb_shinfo(skb)->frag_list);
1459 skb_shinfo(skb)->frag_list = NULL;
1462 #endif
1464 msg->msg_namelen = 0;
1466 copied = skb->len;
1467 if (len < copied) {
1468 msg->msg_flags |= MSG_TRUNC;
1469 copied = len;
1472 skb_reset_transport_header(skb);
1473 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1475 if (msg->msg_name) {
1476 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1477 addr->nl_family = AF_NETLINK;
1478 addr->nl_pad = 0;
1479 addr->nl_pid = NETLINK_CB(skb).pid;
1480 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1481 msg->msg_namelen = sizeof(*addr);
1484 if (nlk->flags & NETLINK_RECV_PKTINFO)
1485 netlink_cmsg_recv_pktinfo(msg, skb);
1487 if (NULL == siocb->scm) {
1488 memset(&scm, 0, sizeof(scm));
1489 siocb->scm = &scm;
1491 siocb->scm->creds = *NETLINK_CREDS(skb);
1492 if (flags & MSG_TRUNC)
1493 copied = skb->len;
1495 skb_free_datagram(sk, skb);
1497 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1498 netlink_dump(sk);
1500 scm_recv(sock, msg, siocb->scm, flags);
1501 out:
1502 netlink_rcv_wake(sk);
1503 return err ? : copied;
1506 static void netlink_data_ready(struct sock *sk, int len)
1508 BUG();
1512 * We export these functions to other modules. They provide a
1513 * complete set of kernel non-blocking support for message
1514 * queueing.
1517 struct sock *
1518 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1519 void (*input)(struct sk_buff *skb),
1520 struct mutex *cb_mutex, struct module *module)
1522 struct socket *sock;
1523 struct sock *sk;
1524 struct netlink_sock *nlk;
1525 unsigned long *listeners = NULL;
1527 BUG_ON(!nl_table);
1529 if (unit < 0 || unit >= MAX_LINKS)
1530 return NULL;
1532 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1533 return NULL;
1536 * We have to just have a reference on the net from sk, but don't
1537 * get_net it. Besides, we cannot get and then put the net here.
1538 * So we create one inside init_net and the move it to net.
1541 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1542 goto out_sock_release_nosk;
1544 sk = sock->sk;
1545 sk_change_net(sk, net);
1547 if (groups < 32)
1548 groups = 32;
1550 listeners = kzalloc(NLGRPSZ(groups) + sizeof(struct listeners_rcu_head),
1551 GFP_KERNEL);
1552 if (!listeners)
1553 goto out_sock_release;
1555 sk->sk_data_ready = netlink_data_ready;
1556 if (input)
1557 nlk_sk(sk)->netlink_rcv = input;
1559 if (netlink_insert(sk, net, 0))
1560 goto out_sock_release;
1562 nlk = nlk_sk(sk);
1563 nlk->flags |= NETLINK_KERNEL_SOCKET;
1565 netlink_table_grab();
1566 if (!nl_table[unit].registered) {
1567 nl_table[unit].groups = groups;
1568 nl_table[unit].listeners = listeners;
1569 nl_table[unit].cb_mutex = cb_mutex;
1570 nl_table[unit].module = module;
1571 nl_table[unit].registered = 1;
1572 } else {
1573 kfree(listeners);
1574 nl_table[unit].registered++;
1576 netlink_table_ungrab();
1577 return sk;
1579 out_sock_release:
1580 kfree(listeners);
1581 netlink_kernel_release(sk);
1582 return NULL;
1584 out_sock_release_nosk:
1585 sock_release(sock);
1586 return NULL;
1588 EXPORT_SYMBOL(netlink_kernel_create);
1591 void
1592 netlink_kernel_release(struct sock *sk)
1594 sk_release_kernel(sk);
1596 EXPORT_SYMBOL(netlink_kernel_release);
1599 static void netlink_free_old_listeners(struct rcu_head *rcu_head)
1601 struct listeners_rcu_head *lrh;
1603 lrh = container_of(rcu_head, struct listeners_rcu_head, rcu_head);
1604 kfree(lrh->ptr);
1607 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1609 unsigned long *listeners, *old = NULL;
1610 struct listeners_rcu_head *old_rcu_head;
1611 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1613 if (groups < 32)
1614 groups = 32;
1616 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1617 listeners = kzalloc(NLGRPSZ(groups) +
1618 sizeof(struct listeners_rcu_head),
1619 GFP_ATOMIC);
1620 if (!listeners)
1621 return -ENOMEM;
1622 old = tbl->listeners;
1623 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1624 rcu_assign_pointer(tbl->listeners, listeners);
1626 * Free the old memory after an RCU grace period so we
1627 * don't leak it. We use call_rcu() here in order to be
1628 * able to call this function from atomic contexts. The
1629 * allocation of this memory will have reserved enough
1630 * space for struct listeners_rcu_head at the end.
1632 old_rcu_head = (void *)(tbl->listeners +
1633 NLGRPLONGS(tbl->groups));
1634 old_rcu_head->ptr = old;
1635 call_rcu(&old_rcu_head->rcu_head, netlink_free_old_listeners);
1637 tbl->groups = groups;
1639 return 0;
1643 * netlink_change_ngroups - change number of multicast groups
1645 * This changes the number of multicast groups that are available
1646 * on a certain netlink family. Note that it is not possible to
1647 * change the number of groups to below 32. Also note that it does
1648 * not implicitly call netlink_clear_multicast_users() when the
1649 * number of groups is reduced.
1651 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1652 * @groups: The new number of groups.
1654 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1656 int err;
1658 netlink_table_grab();
1659 err = __netlink_change_ngroups(sk, groups);
1660 netlink_table_ungrab();
1662 return err;
1665 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1667 struct sock *sk;
1668 struct hlist_node *node;
1669 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1671 sk_for_each_bound(sk, node, &tbl->mc_list)
1672 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1676 * netlink_clear_multicast_users - kick off multicast listeners
1678 * This function removes all listeners from the given group.
1679 * @ksk: The kernel netlink socket, as returned by
1680 * netlink_kernel_create().
1681 * @group: The multicast group to clear.
1683 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1685 netlink_table_grab();
1686 __netlink_clear_multicast_users(ksk, group);
1687 netlink_table_ungrab();
1690 void netlink_set_nonroot(int protocol, unsigned int flags)
1692 if ((unsigned int)protocol < MAX_LINKS)
1693 nl_table[protocol].nl_nonroot = flags;
1695 EXPORT_SYMBOL(netlink_set_nonroot);
1697 static void netlink_destroy_callback(struct netlink_callback *cb)
1699 kfree_skb(cb->skb);
1700 kfree(cb);
1704 * It looks a bit ugly.
1705 * It would be better to create kernel thread.
1708 static int netlink_dump(struct sock *sk)
1710 struct netlink_sock *nlk = nlk_sk(sk);
1711 struct netlink_callback *cb;
1712 struct sk_buff *skb;
1713 struct nlmsghdr *nlh;
1714 int len, err = -ENOBUFS;
1716 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1717 if (!skb)
1718 goto errout;
1720 mutex_lock(nlk->cb_mutex);
1722 cb = nlk->cb;
1723 if (cb == NULL) {
1724 err = -EINVAL;
1725 goto errout_skb;
1728 len = cb->dump(skb, cb);
1730 if (len > 0) {
1731 mutex_unlock(nlk->cb_mutex);
1733 if (sk_filter(sk, skb))
1734 kfree_skb(skb);
1735 else {
1736 skb_queue_tail(&sk->sk_receive_queue, skb);
1737 sk->sk_data_ready(sk, skb->len);
1739 return 0;
1742 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1743 if (!nlh)
1744 goto errout_skb;
1746 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1748 if (sk_filter(sk, skb))
1749 kfree_skb(skb);
1750 else {
1751 skb_queue_tail(&sk->sk_receive_queue, skb);
1752 sk->sk_data_ready(sk, skb->len);
1755 if (cb->done)
1756 cb->done(cb);
1757 nlk->cb = NULL;
1758 mutex_unlock(nlk->cb_mutex);
1760 netlink_destroy_callback(cb);
1761 return 0;
1763 errout_skb:
1764 mutex_unlock(nlk->cb_mutex);
1765 kfree_skb(skb);
1766 errout:
1767 return err;
1770 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1771 const struct nlmsghdr *nlh,
1772 int (*dump)(struct sk_buff *skb,
1773 struct netlink_callback *),
1774 int (*done)(struct netlink_callback *))
1776 struct netlink_callback *cb;
1777 struct sock *sk;
1778 struct netlink_sock *nlk;
1780 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1781 if (cb == NULL)
1782 return -ENOBUFS;
1784 cb->dump = dump;
1785 cb->done = done;
1786 cb->nlh = nlh;
1787 atomic_inc(&skb->users);
1788 cb->skb = skb;
1790 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1791 if (sk == NULL) {
1792 netlink_destroy_callback(cb);
1793 return -ECONNREFUSED;
1795 nlk = nlk_sk(sk);
1796 /* A dump is in progress... */
1797 mutex_lock(nlk->cb_mutex);
1798 if (nlk->cb) {
1799 mutex_unlock(nlk->cb_mutex);
1800 netlink_destroy_callback(cb);
1801 sock_put(sk);
1802 return -EBUSY;
1804 nlk->cb = cb;
1805 mutex_unlock(nlk->cb_mutex);
1807 netlink_dump(sk);
1808 sock_put(sk);
1810 /* We successfully started a dump, by returning -EINTR we
1811 * signal not to send ACK even if it was requested.
1813 return -EINTR;
1815 EXPORT_SYMBOL(netlink_dump_start);
1817 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1819 struct sk_buff *skb;
1820 struct nlmsghdr *rep;
1821 struct nlmsgerr *errmsg;
1822 size_t payload = sizeof(*errmsg);
1824 /* error messages get the original request appened */
1825 if (err)
1826 payload += nlmsg_len(nlh);
1828 skb = nlmsg_new(payload, GFP_KERNEL);
1829 if (!skb) {
1830 struct sock *sk;
1832 sk = netlink_lookup(sock_net(in_skb->sk),
1833 in_skb->sk->sk_protocol,
1834 NETLINK_CB(in_skb).pid);
1835 if (sk) {
1836 sk->sk_err = ENOBUFS;
1837 sk->sk_error_report(sk);
1838 sock_put(sk);
1840 return;
1843 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1844 NLMSG_ERROR, payload, 0);
1845 errmsg = nlmsg_data(rep);
1846 errmsg->error = err;
1847 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1848 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1850 EXPORT_SYMBOL(netlink_ack);
1852 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1853 struct nlmsghdr *))
1855 struct nlmsghdr *nlh;
1856 int err;
1858 while (skb->len >= nlmsg_total_size(0)) {
1859 int msglen;
1861 nlh = nlmsg_hdr(skb);
1862 err = 0;
1864 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1865 return 0;
1867 /* Only requests are handled by the kernel */
1868 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1869 goto ack;
1871 /* Skip control messages */
1872 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1873 goto ack;
1875 err = cb(skb, nlh);
1876 if (err == -EINTR)
1877 goto skip;
1879 ack:
1880 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1881 netlink_ack(skb, nlh, err);
1883 skip:
1884 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1885 if (msglen > skb->len)
1886 msglen = skb->len;
1887 skb_pull(skb, msglen);
1890 return 0;
1892 EXPORT_SYMBOL(netlink_rcv_skb);
1895 * nlmsg_notify - send a notification netlink message
1896 * @sk: netlink socket to use
1897 * @skb: notification message
1898 * @pid: destination netlink pid for reports or 0
1899 * @group: destination multicast group or 0
1900 * @report: 1 to report back, 0 to disable
1901 * @flags: allocation flags
1903 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1904 unsigned int group, int report, gfp_t flags)
1906 int err = 0;
1908 if (group) {
1909 int exclude_pid = 0;
1911 if (report) {
1912 atomic_inc(&skb->users);
1913 exclude_pid = pid;
1916 /* errors reported via destination sk->sk_err, but propagate
1917 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1918 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1921 if (report) {
1922 int err2;
1924 err2 = nlmsg_unicast(sk, skb, pid);
1925 if (!err || err == -ESRCH)
1926 err = err2;
1929 return err;
1931 EXPORT_SYMBOL(nlmsg_notify);
1933 #ifdef CONFIG_PROC_FS
1934 struct nl_seq_iter {
1935 struct seq_net_private p;
1936 int link;
1937 int hash_idx;
1940 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1942 struct nl_seq_iter *iter = seq->private;
1943 int i, j;
1944 struct sock *s;
1945 struct hlist_node *node;
1946 loff_t off = 0;
1948 for (i = 0; i < MAX_LINKS; i++) {
1949 struct nl_pid_hash *hash = &nl_table[i].hash;
1951 for (j = 0; j <= hash->mask; j++) {
1952 sk_for_each(s, node, &hash->table[j]) {
1953 if (sock_net(s) != seq_file_net(seq))
1954 continue;
1955 if (off == pos) {
1956 iter->link = i;
1957 iter->hash_idx = j;
1958 return s;
1960 ++off;
1964 return NULL;
1967 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1968 __acquires(nl_table_lock)
1970 read_lock(&nl_table_lock);
1971 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1974 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1976 struct sock *s;
1977 struct nl_seq_iter *iter;
1978 int i, j;
1980 ++*pos;
1982 if (v == SEQ_START_TOKEN)
1983 return netlink_seq_socket_idx(seq, 0);
1985 iter = seq->private;
1986 s = v;
1987 do {
1988 s = sk_next(s);
1989 } while (s && sock_net(s) != seq_file_net(seq));
1990 if (s)
1991 return s;
1993 i = iter->link;
1994 j = iter->hash_idx + 1;
1996 do {
1997 struct nl_pid_hash *hash = &nl_table[i].hash;
1999 for (; j <= hash->mask; j++) {
2000 s = sk_head(&hash->table[j]);
2001 while (s && sock_net(s) != seq_file_net(seq))
2002 s = sk_next(s);
2003 if (s) {
2004 iter->link = i;
2005 iter->hash_idx = j;
2006 return s;
2010 j = 0;
2011 } while (++i < MAX_LINKS);
2013 return NULL;
2016 static void netlink_seq_stop(struct seq_file *seq, void *v)
2017 __releases(nl_table_lock)
2019 read_unlock(&nl_table_lock);
2023 static int netlink_seq_show(struct seq_file *seq, void *v)
2025 if (v == SEQ_START_TOKEN)
2026 seq_puts(seq,
2027 "sk Eth Pid Groups "
2028 "Rmem Wmem Dump Locks Drops Inode\n");
2029 else {
2030 struct sock *s = v;
2031 struct netlink_sock *nlk = nlk_sk(s);
2033 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %-8d %-8d %-8lu\n",
2035 s->sk_protocol,
2036 nlk->pid,
2037 nlk->groups ? (u32)nlk->groups[0] : 0,
2038 sk_rmem_alloc_get(s),
2039 sk_wmem_alloc_get(s),
2040 nlk->cb,
2041 atomic_read(&s->sk_refcnt),
2042 atomic_read(&s->sk_drops),
2043 sock_i_ino(s)
2047 return 0;
2050 static const struct seq_operations netlink_seq_ops = {
2051 .start = netlink_seq_start,
2052 .next = netlink_seq_next,
2053 .stop = netlink_seq_stop,
2054 .show = netlink_seq_show,
2058 static int netlink_seq_open(struct inode *inode, struct file *file)
2060 return seq_open_net(inode, file, &netlink_seq_ops,
2061 sizeof(struct nl_seq_iter));
2064 static const struct file_operations netlink_seq_fops = {
2065 .owner = THIS_MODULE,
2066 .open = netlink_seq_open,
2067 .read = seq_read,
2068 .llseek = seq_lseek,
2069 .release = seq_release_net,
2072 #endif
2074 int netlink_register_notifier(struct notifier_block *nb)
2076 return atomic_notifier_chain_register(&netlink_chain, nb);
2078 EXPORT_SYMBOL(netlink_register_notifier);
2080 int netlink_unregister_notifier(struct notifier_block *nb)
2082 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2084 EXPORT_SYMBOL(netlink_unregister_notifier);
2086 static const struct proto_ops netlink_ops = {
2087 .family = PF_NETLINK,
2088 .owner = THIS_MODULE,
2089 .release = netlink_release,
2090 .bind = netlink_bind,
2091 .connect = netlink_connect,
2092 .socketpair = sock_no_socketpair,
2093 .accept = sock_no_accept,
2094 .getname = netlink_getname,
2095 .poll = datagram_poll,
2096 .ioctl = sock_no_ioctl,
2097 .listen = sock_no_listen,
2098 .shutdown = sock_no_shutdown,
2099 .setsockopt = netlink_setsockopt,
2100 .getsockopt = netlink_getsockopt,
2101 .sendmsg = netlink_sendmsg,
2102 .recvmsg = netlink_recvmsg,
2103 .mmap = sock_no_mmap,
2104 .sendpage = sock_no_sendpage,
2107 static const struct net_proto_family netlink_family_ops = {
2108 .family = PF_NETLINK,
2109 .create = netlink_create,
2110 .owner = THIS_MODULE, /* for consistency 8) */
2113 static int __net_init netlink_net_init(struct net *net)
2115 #ifdef CONFIG_PROC_FS
2116 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2117 return -ENOMEM;
2118 #endif
2119 return 0;
2122 static void __net_exit netlink_net_exit(struct net *net)
2124 #ifdef CONFIG_PROC_FS
2125 proc_net_remove(net, "netlink");
2126 #endif
2129 static struct pernet_operations __net_initdata netlink_net_ops = {
2130 .init = netlink_net_init,
2131 .exit = netlink_net_exit,
2134 static int __init netlink_proto_init(void)
2136 struct sk_buff *dummy_skb;
2137 int i;
2138 unsigned long limit;
2139 unsigned int order;
2140 int err = proto_register(&netlink_proto, 0);
2142 if (err != 0)
2143 goto out;
2145 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2147 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2148 if (!nl_table)
2149 goto panic;
2151 if (totalram_pages >= (128 * 1024))
2152 limit = totalram_pages >> (21 - PAGE_SHIFT);
2153 else
2154 limit = totalram_pages >> (23 - PAGE_SHIFT);
2156 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2157 limit = (1UL << order) / sizeof(struct hlist_head);
2158 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2160 for (i = 0; i < MAX_LINKS; i++) {
2161 struct nl_pid_hash *hash = &nl_table[i].hash;
2163 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2164 if (!hash->table) {
2165 while (i-- > 0)
2166 nl_pid_hash_free(nl_table[i].hash.table,
2167 1 * sizeof(*hash->table));
2168 kfree(nl_table);
2169 goto panic;
2171 hash->max_shift = order;
2172 hash->shift = 0;
2173 hash->mask = 0;
2174 hash->rehash_time = jiffies;
2177 sock_register(&netlink_family_ops);
2178 register_pernet_subsys(&netlink_net_ops);
2179 /* The netlink device handler may be needed early. */
2180 rtnetlink_init();
2181 out:
2182 return err;
2183 panic:
2184 panic("netlink_init: Cannot allocate nl_table\n");
2187 core_initcall(netlink_proto_init);