2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@redhat.com>
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>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.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>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/selinux.h>
58 #include <linux/mutex.h>
60 #include <net/net_namespace.h>
63 #include <net/netlink.h>
65 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
66 #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
69 /* struct sock has to be the first member of netlink_sock */
77 unsigned long *groups;
79 wait_queue_head_t wait;
80 struct netlink_callback *cb;
81 struct mutex *cb_mutex;
82 struct mutex cb_def_mutex;
83 void (*netlink_rcv)(struct sk_buff *skb);
84 struct module *module;
87 #define NETLINK_KERNEL_SOCKET 0x1
88 #define NETLINK_RECV_PKTINFO 0x2
90 static inline struct netlink_sock *nlk_sk(struct sock *sk)
92 return container_of(sk, struct netlink_sock, sk);
95 static inline int netlink_is_kernel(struct sock *sk)
97 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
101 struct hlist_head *table;
102 unsigned long rehash_time;
107 unsigned int entries;
108 unsigned int max_shift;
113 struct netlink_table {
114 struct nl_pid_hash hash;
115 struct hlist_head mc_list;
116 unsigned long *listeners;
117 unsigned int nl_nonroot;
119 struct mutex *cb_mutex;
120 struct module *module;
124 static struct netlink_table *nl_table;
126 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
128 static int netlink_dump(struct sock *sk);
129 static void netlink_destroy_callback(struct netlink_callback *cb);
131 static DEFINE_RWLOCK(nl_table_lock);
132 static atomic_t nl_table_users = ATOMIC_INIT(0);
134 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
136 static u32 netlink_group_mask(u32 group)
138 return group ? 1 << (group - 1) : 0;
141 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
143 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
146 static void netlink_sock_destruct(struct sock *sk)
148 struct netlink_sock *nlk = nlk_sk(sk);
152 nlk->cb->done(nlk->cb);
153 netlink_destroy_callback(nlk->cb);
156 skb_queue_purge(&sk->sk_receive_queue);
158 if (!sock_flag(sk, SOCK_DEAD)) {
159 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
162 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
163 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
164 BUG_TRAP(!nlk_sk(sk)->groups);
167 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
168 * SMP. Look, when several writers sleep and reader wakes them up, all but one
169 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
170 * this, _but_ remember, it adds useless work on UP machines.
173 static void netlink_table_grab(void)
174 __acquires(nl_table_lock)
176 write_lock_irq(&nl_table_lock);
178 if (atomic_read(&nl_table_users)) {
179 DECLARE_WAITQUEUE(wait, current);
181 add_wait_queue_exclusive(&nl_table_wait, &wait);
183 set_current_state(TASK_UNINTERRUPTIBLE);
184 if (atomic_read(&nl_table_users) == 0)
186 write_unlock_irq(&nl_table_lock);
188 write_lock_irq(&nl_table_lock);
191 __set_current_state(TASK_RUNNING);
192 remove_wait_queue(&nl_table_wait, &wait);
196 static inline void netlink_table_ungrab(void)
197 __releases(nl_table_lock)
199 write_unlock_irq(&nl_table_lock);
200 wake_up(&nl_table_wait);
204 netlink_lock_table(void)
206 /* read_lock() synchronizes us to netlink_table_grab */
208 read_lock(&nl_table_lock);
209 atomic_inc(&nl_table_users);
210 read_unlock(&nl_table_lock);
214 netlink_unlock_table(void)
216 if (atomic_dec_and_test(&nl_table_users))
217 wake_up(&nl_table_wait);
220 static inline struct sock *netlink_lookup(struct net *net, int protocol,
223 struct nl_pid_hash *hash = &nl_table[protocol].hash;
224 struct hlist_head *head;
226 struct hlist_node *node;
228 read_lock(&nl_table_lock);
229 head = nl_pid_hashfn(hash, pid);
230 sk_for_each(sk, node, head) {
231 if ((sk->sk_net == net) && (nlk_sk(sk)->pid == pid)) {
238 read_unlock(&nl_table_lock);
242 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
244 if (size <= PAGE_SIZE)
245 return kzalloc(size, GFP_ATOMIC);
247 return (struct hlist_head *)
248 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
252 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
254 if (size <= PAGE_SIZE)
257 free_pages((unsigned long)table, get_order(size));
260 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
262 unsigned int omask, mask, shift;
264 struct hlist_head *otable, *table;
267 omask = mask = hash->mask;
268 osize = size = (mask + 1) * sizeof(*table);
272 if (++shift > hash->max_shift)
278 table = nl_pid_hash_zalloc(size);
282 otable = hash->table;
286 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
288 for (i = 0; i <= omask; i++) {
290 struct hlist_node *node, *tmp;
292 sk_for_each_safe(sk, node, tmp, &otable[i])
293 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
296 nl_pid_hash_free(otable, osize);
297 hash->rehash_time = jiffies + 10 * 60 * HZ;
301 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
303 int avg = hash->entries >> hash->shift;
305 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
308 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
309 nl_pid_hash_rehash(hash, 0);
316 static const struct proto_ops netlink_ops;
319 netlink_update_listeners(struct sock *sk)
321 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
322 struct hlist_node *node;
326 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
328 sk_for_each_bound(sk, node, &tbl->mc_list) {
329 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
330 mask |= nlk_sk(sk)->groups[i];
332 tbl->listeners[i] = mask;
334 /* this function is only called with the netlink table "grabbed", which
335 * makes sure updates are visible before bind or setsockopt return. */
338 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
340 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
341 struct hlist_head *head;
342 int err = -EADDRINUSE;
344 struct hlist_node *node;
347 netlink_table_grab();
348 head = nl_pid_hashfn(hash, pid);
350 sk_for_each(osk, node, head) {
351 if ((osk->sk_net == net) && (nlk_sk(osk)->pid == pid))
363 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
366 if (len && nl_pid_hash_dilute(hash, len))
367 head = nl_pid_hashfn(hash, pid);
369 nlk_sk(sk)->pid = pid;
370 sk_add_node(sk, head);
374 netlink_table_ungrab();
378 static void netlink_remove(struct sock *sk)
380 netlink_table_grab();
381 if (sk_del_node_init(sk))
382 nl_table[sk->sk_protocol].hash.entries--;
383 if (nlk_sk(sk)->subscriptions)
384 __sk_del_bind_node(sk);
385 netlink_table_ungrab();
388 static struct proto netlink_proto = {
390 .owner = THIS_MODULE,
391 .obj_size = sizeof(struct netlink_sock),
394 static int __netlink_create(struct net *net, struct socket *sock,
395 struct mutex *cb_mutex, int protocol)
398 struct netlink_sock *nlk;
400 sock->ops = &netlink_ops;
402 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
406 sock_init_data(sock, sk);
410 nlk->cb_mutex = cb_mutex;
412 nlk->cb_mutex = &nlk->cb_def_mutex;
413 mutex_init(nlk->cb_mutex);
415 init_waitqueue_head(&nlk->wait);
417 sk->sk_destruct = netlink_sock_destruct;
418 sk->sk_protocol = protocol;
422 static int netlink_create(struct net *net, struct socket *sock, int protocol)
424 struct module *module = NULL;
425 struct mutex *cb_mutex;
426 struct netlink_sock *nlk;
429 sock->state = SS_UNCONNECTED;
431 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
432 return -ESOCKTNOSUPPORT;
434 if (protocol < 0 || protocol >= MAX_LINKS)
435 return -EPROTONOSUPPORT;
437 netlink_lock_table();
439 if (!nl_table[protocol].registered) {
440 netlink_unlock_table();
441 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
442 netlink_lock_table();
445 if (nl_table[protocol].registered &&
446 try_module_get(nl_table[protocol].module))
447 module = nl_table[protocol].module;
448 cb_mutex = nl_table[protocol].cb_mutex;
449 netlink_unlock_table();
451 err = __netlink_create(net, sock, cb_mutex, protocol);
455 nlk = nlk_sk(sock->sk);
456 nlk->module = module;
465 static int netlink_release(struct socket *sock)
467 struct sock *sk = sock->sk;
468 struct netlink_sock *nlk;
478 * OK. Socket is unlinked, any packets that arrive now
483 wake_up_interruptible_all(&nlk->wait);
485 skb_queue_purge(&sk->sk_write_queue);
487 if (nlk->pid && !nlk->subscriptions) {
488 struct netlink_notify n = {
490 .protocol = sk->sk_protocol,
493 atomic_notifier_call_chain(&netlink_chain,
494 NETLINK_URELEASE, &n);
497 module_put(nlk->module);
499 netlink_table_grab();
500 if (netlink_is_kernel(sk)) {
501 kfree(nl_table[sk->sk_protocol].listeners);
502 nl_table[sk->sk_protocol].module = NULL;
503 nl_table[sk->sk_protocol].registered = 0;
504 } else if (nlk->subscriptions)
505 netlink_update_listeners(sk);
506 netlink_table_ungrab();
515 static int netlink_autobind(struct socket *sock)
517 struct sock *sk = sock->sk;
518 struct net *net = sk->sk_net;
519 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
520 struct hlist_head *head;
522 struct hlist_node *node;
523 s32 pid = current->tgid;
525 static s32 rover = -4097;
529 netlink_table_grab();
530 head = nl_pid_hashfn(hash, pid);
531 sk_for_each(osk, node, head) {
532 if ((osk->sk_net != net))
534 if (nlk_sk(osk)->pid == pid) {
535 /* Bind collision, search negative pid values. */
539 netlink_table_ungrab();
543 netlink_table_ungrab();
545 err = netlink_insert(sk, net, pid);
546 if (err == -EADDRINUSE)
549 /* If 2 threads race to autobind, that is fine. */
556 static inline int netlink_capable(struct socket *sock, unsigned int flag)
558 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
559 capable(CAP_NET_ADMIN);
563 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
565 struct netlink_sock *nlk = nlk_sk(sk);
567 if (nlk->subscriptions && !subscriptions)
568 __sk_del_bind_node(sk);
569 else if (!nlk->subscriptions && subscriptions)
570 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
571 nlk->subscriptions = subscriptions;
574 static int netlink_realloc_groups(struct sock *sk)
576 struct netlink_sock *nlk = nlk_sk(sk);
578 unsigned long *new_groups;
581 netlink_table_grab();
583 groups = nl_table[sk->sk_protocol].groups;
584 if (!nl_table[sk->sk_protocol].registered) {
589 if (nlk->ngroups >= groups)
592 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
593 if (new_groups == NULL) {
597 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
598 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
600 nlk->groups = new_groups;
601 nlk->ngroups = groups;
603 netlink_table_ungrab();
607 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
610 struct sock *sk = sock->sk;
611 struct net *net = sk->sk_net;
612 struct netlink_sock *nlk = nlk_sk(sk);
613 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
616 if (nladdr->nl_family != AF_NETLINK)
619 /* Only superuser is allowed to listen multicasts */
620 if (nladdr->nl_groups) {
621 if (!netlink_capable(sock, NL_NONROOT_RECV))
623 err = netlink_realloc_groups(sk);
629 if (nladdr->nl_pid != nlk->pid)
632 err = nladdr->nl_pid ?
633 netlink_insert(sk, net, nladdr->nl_pid) :
634 netlink_autobind(sock);
639 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
642 netlink_table_grab();
643 netlink_update_subscriptions(sk, nlk->subscriptions +
644 hweight32(nladdr->nl_groups) -
645 hweight32(nlk->groups[0]));
646 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
647 netlink_update_listeners(sk);
648 netlink_table_ungrab();
653 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
657 struct sock *sk = sock->sk;
658 struct netlink_sock *nlk = nlk_sk(sk);
659 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
661 if (addr->sa_family == AF_UNSPEC) {
662 sk->sk_state = NETLINK_UNCONNECTED;
667 if (addr->sa_family != AF_NETLINK)
670 /* Only superuser is allowed to send multicasts */
671 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
675 err = netlink_autobind(sock);
678 sk->sk_state = NETLINK_CONNECTED;
679 nlk->dst_pid = nladdr->nl_pid;
680 nlk->dst_group = ffs(nladdr->nl_groups);
686 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
687 int *addr_len, int peer)
689 struct sock *sk = sock->sk;
690 struct netlink_sock *nlk = nlk_sk(sk);
691 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
693 nladdr->nl_family = AF_NETLINK;
695 *addr_len = sizeof(*nladdr);
698 nladdr->nl_pid = nlk->dst_pid;
699 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
701 nladdr->nl_pid = nlk->pid;
702 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
707 static void netlink_overrun(struct sock *sk)
709 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
710 sk->sk_err = ENOBUFS;
711 sk->sk_error_report(sk);
715 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
718 struct netlink_sock *nlk;
720 sock = netlink_lookup(ssk->sk_net, ssk->sk_protocol, pid);
722 return ERR_PTR(-ECONNREFUSED);
724 /* Don't bother queuing skb if kernel socket has no input function */
726 if (sock->sk_state == NETLINK_CONNECTED &&
727 nlk->dst_pid != nlk_sk(ssk)->pid) {
729 return ERR_PTR(-ECONNREFUSED);
734 struct sock *netlink_getsockbyfilp(struct file *filp)
736 struct inode *inode = filp->f_path.dentry->d_inode;
739 if (!S_ISSOCK(inode->i_mode))
740 return ERR_PTR(-ENOTSOCK);
742 sock = SOCKET_I(inode)->sk;
743 if (sock->sk_family != AF_NETLINK)
744 return ERR_PTR(-EINVAL);
751 * Attach a skb to a netlink socket.
752 * The caller must hold a reference to the destination socket. On error, the
753 * reference is dropped. The skb is not send to the destination, just all
754 * all error checks are performed and memory in the queue is reserved.
756 * < 0: error. skb freed, reference to sock dropped.
758 * 1: repeat lookup - reference dropped while waiting for socket memory.
760 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
761 long *timeo, struct sock *ssk)
763 struct netlink_sock *nlk;
767 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
768 test_bit(0, &nlk->state)) {
769 DECLARE_WAITQUEUE(wait, current);
771 if (!ssk || netlink_is_kernel(ssk))
778 __set_current_state(TASK_INTERRUPTIBLE);
779 add_wait_queue(&nlk->wait, &wait);
781 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
782 test_bit(0, &nlk->state)) &&
783 !sock_flag(sk, SOCK_DEAD))
784 *timeo = schedule_timeout(*timeo);
786 __set_current_state(TASK_RUNNING);
787 remove_wait_queue(&nlk->wait, &wait);
790 if (signal_pending(current)) {
792 return sock_intr_errno(*timeo);
796 skb_set_owner_r(skb, sk);
800 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
804 skb_queue_tail(&sk->sk_receive_queue, skb);
805 sk->sk_data_ready(sk, len);
810 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
816 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
823 delta = skb->end - skb->tail;
824 if (delta * 2 < skb->truesize)
827 if (skb_shared(skb)) {
828 struct sk_buff *nskb = skb_clone(skb, allocation);
835 if (!pskb_expand_head(skb, 0, -delta, allocation))
836 skb->truesize -= delta;
841 static inline void netlink_rcv_wake(struct sock *sk)
843 struct netlink_sock *nlk = nlk_sk(sk);
845 if (skb_queue_empty(&sk->sk_receive_queue))
846 clear_bit(0, &nlk->state);
847 if (!test_bit(0, &nlk->state))
848 wake_up_interruptible(&nlk->wait);
851 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
854 struct netlink_sock *nlk = nlk_sk(sk);
857 if (nlk->netlink_rcv != NULL) {
859 skb_set_owner_r(skb, sk);
860 nlk->netlink_rcv(skb);
867 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
868 u32 pid, int nonblock)
874 skb = netlink_trim(skb, gfp_any());
876 timeo = sock_sndtimeo(ssk, nonblock);
878 sk = netlink_getsockbypid(ssk, pid);
883 if (netlink_is_kernel(sk))
884 return netlink_unicast_kernel(sk, skb);
886 err = netlink_attachskb(sk, skb, nonblock, &timeo, ssk);
892 return netlink_sendskb(sk, skb);
894 EXPORT_SYMBOL(netlink_unicast);
896 int netlink_has_listeners(struct sock *sk, unsigned int group)
899 unsigned long *listeners;
901 BUG_ON(!netlink_is_kernel(sk));
904 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
906 if (group - 1 < nl_table[sk->sk_protocol].groups)
907 res = test_bit(group - 1, listeners);
913 EXPORT_SYMBOL_GPL(netlink_has_listeners);
915 static inline int netlink_broadcast_deliver(struct sock *sk,
918 struct netlink_sock *nlk = nlk_sk(sk);
920 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
921 !test_bit(0, &nlk->state)) {
922 skb_set_owner_r(skb, sk);
923 skb_queue_tail(&sk->sk_receive_queue, skb);
924 sk->sk_data_ready(sk, skb->len);
925 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
930 struct netlink_broadcast_data {
931 struct sock *exclude_sk;
939 struct sk_buff *skb, *skb2;
942 static inline int do_one_broadcast(struct sock *sk,
943 struct netlink_broadcast_data *p)
945 struct netlink_sock *nlk = nlk_sk(sk);
948 if (p->exclude_sk == sk)
951 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
952 !test_bit(p->group - 1, nlk->groups))
955 if ((sk->sk_net != p->net))
964 if (p->skb2 == NULL) {
965 if (skb_shared(p->skb)) {
966 p->skb2 = skb_clone(p->skb, p->allocation);
968 p->skb2 = skb_get(p->skb);
970 * skb ownership may have been set when
971 * delivered to a previous socket.
976 if (p->skb2 == NULL) {
978 /* Clone failed. Notify ALL listeners. */
980 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
993 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
994 u32 group, gfp_t allocation)
996 struct net *net = ssk->sk_net;
997 struct netlink_broadcast_data info;
998 struct hlist_node *node;
1001 skb = netlink_trim(skb, allocation);
1003 info.exclude_sk = ssk;
1010 info.allocation = allocation;
1014 /* While we sleep in clone, do not allow to change socket list */
1016 netlink_lock_table();
1018 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1019 do_one_broadcast(sk, &info);
1023 netlink_unlock_table();
1026 kfree_skb(info.skb2);
1028 if (info.delivered) {
1029 if (info.congested && (allocation & __GFP_WAIT))
1037 EXPORT_SYMBOL(netlink_broadcast);
1039 struct netlink_set_err_data {
1040 struct sock *exclude_sk;
1046 static inline int do_one_set_err(struct sock *sk,
1047 struct netlink_set_err_data *p)
1049 struct netlink_sock *nlk = nlk_sk(sk);
1051 if (sk == p->exclude_sk)
1054 if (sk->sk_net != p->exclude_sk->sk_net)
1057 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1058 !test_bit(p->group - 1, nlk->groups))
1061 sk->sk_err = p->code;
1062 sk->sk_error_report(sk);
1067 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1069 struct netlink_set_err_data info;
1070 struct hlist_node *node;
1073 info.exclude_sk = ssk;
1078 read_lock(&nl_table_lock);
1080 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1081 do_one_set_err(sk, &info);
1083 read_unlock(&nl_table_lock);
1086 /* must be called with netlink table grabbed */
1087 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1091 int old, new = !!is_new, subscriptions;
1093 old = test_bit(group - 1, nlk->groups);
1094 subscriptions = nlk->subscriptions - old + new;
1096 __set_bit(group - 1, nlk->groups);
1098 __clear_bit(group - 1, nlk->groups);
1099 netlink_update_subscriptions(&nlk->sk, subscriptions);
1100 netlink_update_listeners(&nlk->sk);
1103 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1104 char __user *optval, int optlen)
1106 struct sock *sk = sock->sk;
1107 struct netlink_sock *nlk = nlk_sk(sk);
1108 unsigned int val = 0;
1111 if (level != SOL_NETLINK)
1112 return -ENOPROTOOPT;
1114 if (optlen >= sizeof(int) &&
1115 get_user(val, (unsigned int __user *)optval))
1119 case NETLINK_PKTINFO:
1121 nlk->flags |= NETLINK_RECV_PKTINFO;
1123 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1126 case NETLINK_ADD_MEMBERSHIP:
1127 case NETLINK_DROP_MEMBERSHIP: {
1128 if (!netlink_capable(sock, NL_NONROOT_RECV))
1130 err = netlink_realloc_groups(sk);
1133 if (!val || val - 1 >= nlk->ngroups)
1135 netlink_table_grab();
1136 netlink_update_socket_mc(nlk, val,
1137 optname == NETLINK_ADD_MEMBERSHIP);
1138 netlink_table_ungrab();
1148 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1149 char __user *optval, int __user *optlen)
1151 struct sock *sk = sock->sk;
1152 struct netlink_sock *nlk = nlk_sk(sk);
1155 if (level != SOL_NETLINK)
1156 return -ENOPROTOOPT;
1158 if (get_user(len, optlen))
1164 case NETLINK_PKTINFO:
1165 if (len < sizeof(int))
1168 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1169 if (put_user(len, optlen) ||
1170 put_user(val, optval))
1180 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1182 struct nl_pktinfo info;
1184 info.group = NETLINK_CB(skb).dst_group;
1185 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1188 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1189 struct msghdr *msg, size_t len)
1191 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1192 struct sock *sk = sock->sk;
1193 struct netlink_sock *nlk = nlk_sk(sk);
1194 struct sockaddr_nl *addr = msg->msg_name;
1197 struct sk_buff *skb;
1199 struct scm_cookie scm;
1201 if (msg->msg_flags&MSG_OOB)
1204 if (NULL == siocb->scm)
1206 err = scm_send(sock, msg, siocb->scm);
1210 if (msg->msg_namelen) {
1211 if (addr->nl_family != AF_NETLINK)
1213 dst_pid = addr->nl_pid;
1214 dst_group = ffs(addr->nl_groups);
1215 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1218 dst_pid = nlk->dst_pid;
1219 dst_group = nlk->dst_group;
1223 err = netlink_autobind(sock);
1229 if (len > sk->sk_sndbuf - 32)
1232 skb = alloc_skb(len, GFP_KERNEL);
1236 NETLINK_CB(skb).pid = nlk->pid;
1237 NETLINK_CB(skb).dst_group = dst_group;
1238 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1239 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1240 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1242 /* What can I do? Netlink is asynchronous, so that
1243 we will have to save current capabilities to
1244 check them, when this message will be delivered
1245 to corresponding kernel module. --ANK (980802)
1249 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1254 err = security_netlink_send(sk, skb);
1261 atomic_inc(&skb->users);
1262 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1264 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1270 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1271 struct msghdr *msg, size_t len,
1274 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1275 struct scm_cookie scm;
1276 struct sock *sk = sock->sk;
1277 struct netlink_sock *nlk = nlk_sk(sk);
1278 int noblock = flags&MSG_DONTWAIT;
1280 struct sk_buff *skb;
1288 skb = skb_recv_datagram(sk, flags, noblock, &err);
1292 msg->msg_namelen = 0;
1296 msg->msg_flags |= MSG_TRUNC;
1300 skb_reset_transport_header(skb);
1301 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1303 if (msg->msg_name) {
1304 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1305 addr->nl_family = AF_NETLINK;
1307 addr->nl_pid = NETLINK_CB(skb).pid;
1308 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1309 msg->msg_namelen = sizeof(*addr);
1312 if (nlk->flags & NETLINK_RECV_PKTINFO)
1313 netlink_cmsg_recv_pktinfo(msg, skb);
1315 if (NULL == siocb->scm) {
1316 memset(&scm, 0, sizeof(scm));
1319 siocb->scm->creds = *NETLINK_CREDS(skb);
1320 if (flags & MSG_TRUNC)
1322 skb_free_datagram(sk, skb);
1324 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1327 scm_recv(sock, msg, siocb->scm, flags);
1329 netlink_rcv_wake(sk);
1330 return err ? : copied;
1333 static void netlink_data_ready(struct sock *sk, int len)
1339 * We export these functions to other modules. They provide a
1340 * complete set of kernel non-blocking support for message
1345 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1346 void (*input)(struct sk_buff *skb),
1347 struct mutex *cb_mutex, struct module *module)
1349 struct socket *sock;
1351 struct netlink_sock *nlk;
1352 unsigned long *listeners = NULL;
1356 if (unit < 0 || unit >= MAX_LINKS)
1359 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1362 if (__netlink_create(net, sock, cb_mutex, unit) < 0)
1363 goto out_sock_release;
1368 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1370 goto out_sock_release;
1373 sk->sk_data_ready = netlink_data_ready;
1375 nlk_sk(sk)->netlink_rcv = input;
1377 if (netlink_insert(sk, net, 0))
1378 goto out_sock_release;
1381 nlk->flags |= NETLINK_KERNEL_SOCKET;
1383 netlink_table_grab();
1384 if (!nl_table[unit].registered) {
1385 nl_table[unit].groups = groups;
1386 nl_table[unit].listeners = listeners;
1387 nl_table[unit].cb_mutex = cb_mutex;
1388 nl_table[unit].module = module;
1389 nl_table[unit].registered = 1;
1393 netlink_table_ungrab();
1402 EXPORT_SYMBOL(netlink_kernel_create);
1405 * netlink_change_ngroups - change number of multicast groups
1407 * This changes the number of multicast groups that are available
1408 * on a certain netlink family. Note that it is not possible to
1409 * change the number of groups to below 32. Also note that it does
1410 * not implicitly call netlink_clear_multicast_users() when the
1411 * number of groups is reduced.
1413 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1414 * @groups: The new number of groups.
1416 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1418 unsigned long *listeners, *old = NULL;
1419 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1425 netlink_table_grab();
1426 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1427 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1432 old = tbl->listeners;
1433 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1434 rcu_assign_pointer(tbl->listeners, listeners);
1436 tbl->groups = groups;
1439 netlink_table_ungrab();
1444 EXPORT_SYMBOL(netlink_change_ngroups);
1447 * netlink_clear_multicast_users - kick off multicast listeners
1449 * This function removes all listeners from the given group.
1450 * @ksk: The kernel netlink socket, as returned by
1451 * netlink_kernel_create().
1452 * @group: The multicast group to clear.
1454 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1457 struct hlist_node *node;
1458 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1460 netlink_table_grab();
1462 sk_for_each_bound(sk, node, &tbl->mc_list)
1463 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1465 netlink_table_ungrab();
1467 EXPORT_SYMBOL(netlink_clear_multicast_users);
1469 void netlink_set_nonroot(int protocol, unsigned int flags)
1471 if ((unsigned int)protocol < MAX_LINKS)
1472 nl_table[protocol].nl_nonroot = flags;
1474 EXPORT_SYMBOL(netlink_set_nonroot);
1476 static void netlink_destroy_callback(struct netlink_callback *cb)
1484 * It looks a bit ugly.
1485 * It would be better to create kernel thread.
1488 static int netlink_dump(struct sock *sk)
1490 struct netlink_sock *nlk = nlk_sk(sk);
1491 struct netlink_callback *cb;
1492 struct sk_buff *skb;
1493 struct nlmsghdr *nlh;
1494 int len, err = -ENOBUFS;
1496 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1500 mutex_lock(nlk->cb_mutex);
1508 len = cb->dump(skb, cb);
1511 mutex_unlock(nlk->cb_mutex);
1512 skb_queue_tail(&sk->sk_receive_queue, skb);
1513 sk->sk_data_ready(sk, len);
1517 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1521 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1523 skb_queue_tail(&sk->sk_receive_queue, skb);
1524 sk->sk_data_ready(sk, skb->len);
1529 mutex_unlock(nlk->cb_mutex);
1531 netlink_destroy_callback(cb);
1535 mutex_unlock(nlk->cb_mutex);
1541 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1542 struct nlmsghdr *nlh,
1543 int (*dump)(struct sk_buff *skb,
1544 struct netlink_callback *),
1545 int (*done)(struct netlink_callback *))
1547 struct netlink_callback *cb;
1549 struct netlink_sock *nlk;
1551 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1558 atomic_inc(&skb->users);
1561 sk = netlink_lookup(ssk->sk_net, ssk->sk_protocol, NETLINK_CB(skb).pid);
1563 netlink_destroy_callback(cb);
1564 return -ECONNREFUSED;
1567 /* A dump is in progress... */
1568 mutex_lock(nlk->cb_mutex);
1570 mutex_unlock(nlk->cb_mutex);
1571 netlink_destroy_callback(cb);
1576 mutex_unlock(nlk->cb_mutex);
1581 /* We successfully started a dump, by returning -EINTR we
1582 * signal not to send ACK even if it was requested.
1586 EXPORT_SYMBOL(netlink_dump_start);
1588 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1590 struct sk_buff *skb;
1591 struct nlmsghdr *rep;
1592 struct nlmsgerr *errmsg;
1593 size_t payload = sizeof(*errmsg);
1595 /* error messages get the original request appened */
1597 payload += nlmsg_len(nlh);
1599 skb = nlmsg_new(payload, GFP_KERNEL);
1603 sk = netlink_lookup(in_skb->sk->sk_net,
1604 in_skb->sk->sk_protocol,
1605 NETLINK_CB(in_skb).pid);
1607 sk->sk_err = ENOBUFS;
1608 sk->sk_error_report(sk);
1614 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1615 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1616 errmsg = nlmsg_data(rep);
1617 errmsg->error = err;
1618 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1619 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1621 EXPORT_SYMBOL(netlink_ack);
1623 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1626 struct nlmsghdr *nlh;
1629 while (skb->len >= nlmsg_total_size(0)) {
1632 nlh = nlmsg_hdr(skb);
1635 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1638 /* Only requests are handled by the kernel */
1639 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1642 /* Skip control messages */
1643 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1651 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1652 netlink_ack(skb, nlh, err);
1655 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1656 if (msglen > skb->len)
1658 skb_pull(skb, msglen);
1663 EXPORT_SYMBOL(netlink_rcv_skb);
1666 * nlmsg_notify - send a notification netlink message
1667 * @sk: netlink socket to use
1668 * @skb: notification message
1669 * @pid: destination netlink pid for reports or 0
1670 * @group: destination multicast group or 0
1671 * @report: 1 to report back, 0 to disable
1672 * @flags: allocation flags
1674 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1675 unsigned int group, int report, gfp_t flags)
1680 int exclude_pid = 0;
1683 atomic_inc(&skb->users);
1687 /* errors reported via destination sk->sk_err */
1688 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1692 err = nlmsg_unicast(sk, skb, pid);
1696 EXPORT_SYMBOL(nlmsg_notify);
1698 #ifdef CONFIG_PROC_FS
1699 struct nl_seq_iter {
1700 struct seq_net_private p;
1705 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1707 struct nl_seq_iter *iter = seq->private;
1710 struct hlist_node *node;
1713 for (i = 0; i < MAX_LINKS; i++) {
1714 struct nl_pid_hash *hash = &nl_table[i].hash;
1716 for (j = 0; j <= hash->mask; j++) {
1717 sk_for_each(s, node, &hash->table[j]) {
1718 if (iter->p.net != s->sk_net)
1732 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1733 __acquires(nl_table_lock)
1735 read_lock(&nl_table_lock);
1736 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1739 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1742 struct nl_seq_iter *iter;
1747 if (v == SEQ_START_TOKEN)
1748 return netlink_seq_socket_idx(seq, 0);
1750 iter = seq->private;
1754 } while (s && (iter->p.net != s->sk_net));
1759 j = iter->hash_idx + 1;
1762 struct nl_pid_hash *hash = &nl_table[i].hash;
1764 for (; j <= hash->mask; j++) {
1765 s = sk_head(&hash->table[j]);
1766 while (s && (iter->p.net != s->sk_net))
1776 } while (++i < MAX_LINKS);
1781 static void netlink_seq_stop(struct seq_file *seq, void *v)
1782 __releases(nl_table_lock)
1784 read_unlock(&nl_table_lock);
1788 static int netlink_seq_show(struct seq_file *seq, void *v)
1790 if (v == SEQ_START_TOKEN)
1792 "sk Eth Pid Groups "
1793 "Rmem Wmem Dump Locks\n");
1796 struct netlink_sock *nlk = nlk_sk(s);
1798 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1802 nlk->groups ? (u32)nlk->groups[0] : 0,
1803 atomic_read(&s->sk_rmem_alloc),
1804 atomic_read(&s->sk_wmem_alloc),
1806 atomic_read(&s->sk_refcnt)
1813 static const struct seq_operations netlink_seq_ops = {
1814 .start = netlink_seq_start,
1815 .next = netlink_seq_next,
1816 .stop = netlink_seq_stop,
1817 .show = netlink_seq_show,
1821 static int netlink_seq_open(struct inode *inode, struct file *file)
1823 return seq_open_net(inode, file, &netlink_seq_ops,
1824 sizeof(struct nl_seq_iter));
1827 static const struct file_operations netlink_seq_fops = {
1828 .owner = THIS_MODULE,
1829 .open = netlink_seq_open,
1831 .llseek = seq_lseek,
1832 .release = seq_release_net,
1837 int netlink_register_notifier(struct notifier_block *nb)
1839 return atomic_notifier_chain_register(&netlink_chain, nb);
1841 EXPORT_SYMBOL(netlink_register_notifier);
1843 int netlink_unregister_notifier(struct notifier_block *nb)
1845 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1847 EXPORT_SYMBOL(netlink_unregister_notifier);
1849 static const struct proto_ops netlink_ops = {
1850 .family = PF_NETLINK,
1851 .owner = THIS_MODULE,
1852 .release = netlink_release,
1853 .bind = netlink_bind,
1854 .connect = netlink_connect,
1855 .socketpair = sock_no_socketpair,
1856 .accept = sock_no_accept,
1857 .getname = netlink_getname,
1858 .poll = datagram_poll,
1859 .ioctl = sock_no_ioctl,
1860 .listen = sock_no_listen,
1861 .shutdown = sock_no_shutdown,
1862 .setsockopt = netlink_setsockopt,
1863 .getsockopt = netlink_getsockopt,
1864 .sendmsg = netlink_sendmsg,
1865 .recvmsg = netlink_recvmsg,
1866 .mmap = sock_no_mmap,
1867 .sendpage = sock_no_sendpage,
1870 static struct net_proto_family netlink_family_ops = {
1871 .family = PF_NETLINK,
1872 .create = netlink_create,
1873 .owner = THIS_MODULE, /* for consistency 8) */
1876 static int __net_init netlink_net_init(struct net *net)
1878 #ifdef CONFIG_PROC_FS
1879 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1885 static void __net_exit netlink_net_exit(struct net *net)
1887 #ifdef CONFIG_PROC_FS
1888 proc_net_remove(net, "netlink");
1892 static struct pernet_operations __net_initdata netlink_net_ops = {
1893 .init = netlink_net_init,
1894 .exit = netlink_net_exit,
1897 static int __init netlink_proto_init(void)
1899 struct sk_buff *dummy_skb;
1901 unsigned long limit;
1903 int err = proto_register(&netlink_proto, 0);
1908 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1910 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1914 if (num_physpages >= (128 * 1024))
1915 limit = num_physpages >> (21 - PAGE_SHIFT);
1917 limit = num_physpages >> (23 - PAGE_SHIFT);
1919 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1920 limit = (1UL << order) / sizeof(struct hlist_head);
1921 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1923 for (i = 0; i < MAX_LINKS; i++) {
1924 struct nl_pid_hash *hash = &nl_table[i].hash;
1926 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1929 nl_pid_hash_free(nl_table[i].hash.table,
1930 1 * sizeof(*hash->table));
1934 hash->max_shift = order;
1937 hash->rehash_time = jiffies;
1940 sock_register(&netlink_family_ops);
1941 register_pernet_subsys(&netlink_net_ops);
1942 /* The netlink device handler may be needed early. */
1947 panic("netlink_init: Cannot allocate nl_table\n");
1950 core_initcall(netlink_proto_init);