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/mutex.h>
59 #include <net/net_namespace.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))
68 /* struct sock has to be the first member of netlink_sock */
76 unsigned long *groups;
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 #define NETLINK_KERNEL_SOCKET 0x1
87 #define NETLINK_RECV_PKTINFO 0x2
89 static inline struct netlink_sock *nlk_sk(struct sock *sk)
91 return container_of(sk, struct netlink_sock, sk);
94 static inline int netlink_is_kernel(struct sock *sk)
96 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
100 struct hlist_head *table;
101 unsigned long rehash_time;
106 unsigned int entries;
107 unsigned int max_shift;
112 struct netlink_table {
113 struct nl_pid_hash hash;
114 struct hlist_head mc_list;
115 unsigned long *listeners;
116 unsigned int nl_nonroot;
118 struct mutex *cb_mutex;
119 struct module *module;
123 static struct netlink_table *nl_table;
125 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
127 static int netlink_dump(struct sock *sk);
128 static void netlink_destroy_callback(struct netlink_callback *cb);
130 static DEFINE_RWLOCK(nl_table_lock);
131 static atomic_t nl_table_users = ATOMIC_INIT(0);
133 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
135 static u32 netlink_group_mask(u32 group)
137 return group ? 1 << (group - 1) : 0;
140 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
142 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
145 static void netlink_sock_destruct(struct sock *sk)
147 struct netlink_sock *nlk = nlk_sk(sk);
151 nlk->cb->done(nlk->cb);
152 netlink_destroy_callback(nlk->cb);
155 skb_queue_purge(&sk->sk_receive_queue);
157 if (!sock_flag(sk, SOCK_DEAD)) {
158 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
161 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
162 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
163 BUG_TRAP(!nlk_sk(sk)->groups);
166 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
167 * SMP. Look, when several writers sleep and reader wakes them up, all but one
168 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
169 * this, _but_ remember, it adds useless work on UP machines.
172 static void netlink_table_grab(void)
173 __acquires(nl_table_lock)
175 write_lock_irq(&nl_table_lock);
177 if (atomic_read(&nl_table_users)) {
178 DECLARE_WAITQUEUE(wait, current);
180 add_wait_queue_exclusive(&nl_table_wait, &wait);
182 set_current_state(TASK_UNINTERRUPTIBLE);
183 if (atomic_read(&nl_table_users) == 0)
185 write_unlock_irq(&nl_table_lock);
187 write_lock_irq(&nl_table_lock);
190 __set_current_state(TASK_RUNNING);
191 remove_wait_queue(&nl_table_wait, &wait);
195 static void netlink_table_ungrab(void)
196 __releases(nl_table_lock)
198 write_unlock_irq(&nl_table_lock);
199 wake_up(&nl_table_wait);
203 netlink_lock_table(void)
205 /* read_lock() synchronizes us to netlink_table_grab */
207 read_lock(&nl_table_lock);
208 atomic_inc(&nl_table_users);
209 read_unlock(&nl_table_lock);
213 netlink_unlock_table(void)
215 if (atomic_dec_and_test(&nl_table_users))
216 wake_up(&nl_table_wait);
219 static inline struct sock *netlink_lookup(struct net *net, int protocol,
222 struct nl_pid_hash *hash = &nl_table[protocol].hash;
223 struct hlist_head *head;
225 struct hlist_node *node;
227 read_lock(&nl_table_lock);
228 head = nl_pid_hashfn(hash, pid);
229 sk_for_each(sk, node, head) {
230 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
237 read_unlock(&nl_table_lock);
241 static inline struct hlist_head *nl_pid_hash_zalloc(size_t size)
243 if (size <= PAGE_SIZE)
244 return kzalloc(size, GFP_ATOMIC);
246 return (struct hlist_head *)
247 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
251 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
253 if (size <= PAGE_SIZE)
256 free_pages((unsigned long)table, get_order(size));
259 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
261 unsigned int omask, mask, shift;
263 struct hlist_head *otable, *table;
266 omask = mask = hash->mask;
267 osize = size = (mask + 1) * sizeof(*table);
271 if (++shift > hash->max_shift)
277 table = nl_pid_hash_zalloc(size);
281 otable = hash->table;
285 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
287 for (i = 0; i <= omask; i++) {
289 struct hlist_node *node, *tmp;
291 sk_for_each_safe(sk, node, tmp, &otable[i])
292 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
295 nl_pid_hash_free(otable, osize);
296 hash->rehash_time = jiffies + 10 * 60 * HZ;
300 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
302 int avg = hash->entries >> hash->shift;
304 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
307 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
308 nl_pid_hash_rehash(hash, 0);
315 static const struct proto_ops netlink_ops;
318 netlink_update_listeners(struct sock *sk)
320 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
321 struct hlist_node *node;
325 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
327 sk_for_each_bound(sk, node, &tbl->mc_list) {
328 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
329 mask |= nlk_sk(sk)->groups[i];
331 tbl->listeners[i] = mask;
333 /* this function is only called with the netlink table "grabbed", which
334 * makes sure updates are visible before bind or setsockopt return. */
337 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
339 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
340 struct hlist_head *head;
341 int err = -EADDRINUSE;
343 struct hlist_node *node;
346 netlink_table_grab();
347 head = nl_pid_hashfn(hash, pid);
349 sk_for_each(osk, node, head) {
350 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
362 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
365 if (len && nl_pid_hash_dilute(hash, len))
366 head = nl_pid_hashfn(hash, pid);
368 nlk_sk(sk)->pid = pid;
369 sk_add_node(sk, head);
373 netlink_table_ungrab();
377 static void netlink_remove(struct sock *sk)
379 netlink_table_grab();
380 if (sk_del_node_init(sk))
381 nl_table[sk->sk_protocol].hash.entries--;
382 if (nlk_sk(sk)->subscriptions)
383 __sk_del_bind_node(sk);
384 netlink_table_ungrab();
387 static struct proto netlink_proto = {
389 .owner = THIS_MODULE,
390 .obj_size = sizeof(struct netlink_sock),
393 static int __netlink_create(struct net *net, struct socket *sock,
394 struct mutex *cb_mutex, int protocol)
397 struct netlink_sock *nlk;
399 sock->ops = &netlink_ops;
401 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
405 sock_init_data(sock, sk);
409 nlk->cb_mutex = cb_mutex;
411 nlk->cb_mutex = &nlk->cb_def_mutex;
412 mutex_init(nlk->cb_mutex);
414 init_waitqueue_head(&nlk->wait);
416 sk->sk_destruct = netlink_sock_destruct;
417 sk->sk_protocol = protocol;
421 static int netlink_create(struct net *net, struct socket *sock, int protocol)
423 struct module *module = NULL;
424 struct mutex *cb_mutex;
425 struct netlink_sock *nlk;
428 sock->state = SS_UNCONNECTED;
430 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
431 return -ESOCKTNOSUPPORT;
433 if (protocol < 0 || protocol >= MAX_LINKS)
434 return -EPROTONOSUPPORT;
436 netlink_lock_table();
438 if (!nl_table[protocol].registered) {
439 netlink_unlock_table();
440 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
441 netlink_lock_table();
444 if (nl_table[protocol].registered &&
445 try_module_get(nl_table[protocol].module))
446 module = nl_table[protocol].module;
447 cb_mutex = nl_table[protocol].cb_mutex;
448 netlink_unlock_table();
450 err = __netlink_create(net, sock, cb_mutex, protocol);
454 nlk = nlk_sk(sock->sk);
455 nlk->module = module;
464 static int netlink_release(struct socket *sock)
466 struct sock *sk = sock->sk;
467 struct netlink_sock *nlk;
477 * OK. Socket is unlinked, any packets that arrive now
482 wake_up_interruptible_all(&nlk->wait);
484 skb_queue_purge(&sk->sk_write_queue);
486 if (nlk->pid && !nlk->subscriptions) {
487 struct netlink_notify n = {
489 .protocol = sk->sk_protocol,
492 atomic_notifier_call_chain(&netlink_chain,
493 NETLINK_URELEASE, &n);
496 module_put(nlk->module);
498 netlink_table_grab();
499 if (netlink_is_kernel(sk)) {
500 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
501 if (--nl_table[sk->sk_protocol].registered == 0) {
502 kfree(nl_table[sk->sk_protocol].listeners);
503 nl_table[sk->sk_protocol].module = NULL;
504 nl_table[sk->sk_protocol].registered = 0;
506 } else if (nlk->subscriptions)
507 netlink_update_listeners(sk);
508 netlink_table_ungrab();
517 static int netlink_autobind(struct socket *sock)
519 struct sock *sk = sock->sk;
520 struct net *net = sock_net(sk);
521 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
522 struct hlist_head *head;
524 struct hlist_node *node;
525 s32 pid = current->tgid;
527 static s32 rover = -4097;
531 netlink_table_grab();
532 head = nl_pid_hashfn(hash, pid);
533 sk_for_each(osk, node, head) {
534 if (!net_eq(sock_net(osk), net))
536 if (nlk_sk(osk)->pid == pid) {
537 /* Bind collision, search negative pid values. */
541 netlink_table_ungrab();
545 netlink_table_ungrab();
547 err = netlink_insert(sk, net, pid);
548 if (err == -EADDRINUSE)
551 /* If 2 threads race to autobind, that is fine. */
558 static inline int netlink_capable(struct socket *sock, unsigned int flag)
560 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
561 capable(CAP_NET_ADMIN);
565 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
567 struct netlink_sock *nlk = nlk_sk(sk);
569 if (nlk->subscriptions && !subscriptions)
570 __sk_del_bind_node(sk);
571 else if (!nlk->subscriptions && subscriptions)
572 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
573 nlk->subscriptions = subscriptions;
576 static int netlink_realloc_groups(struct sock *sk)
578 struct netlink_sock *nlk = nlk_sk(sk);
580 unsigned long *new_groups;
583 netlink_table_grab();
585 groups = nl_table[sk->sk_protocol].groups;
586 if (!nl_table[sk->sk_protocol].registered) {
591 if (nlk->ngroups >= groups)
594 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
595 if (new_groups == NULL) {
599 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
600 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
602 nlk->groups = new_groups;
603 nlk->ngroups = groups;
605 netlink_table_ungrab();
609 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
612 struct sock *sk = sock->sk;
613 struct net *net = sock_net(sk);
614 struct netlink_sock *nlk = nlk_sk(sk);
615 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
618 if (nladdr->nl_family != AF_NETLINK)
621 /* Only superuser is allowed to listen multicasts */
622 if (nladdr->nl_groups) {
623 if (!netlink_capable(sock, NL_NONROOT_RECV))
625 err = netlink_realloc_groups(sk);
631 if (nladdr->nl_pid != nlk->pid)
634 err = nladdr->nl_pid ?
635 netlink_insert(sk, net, nladdr->nl_pid) :
636 netlink_autobind(sock);
641 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
644 netlink_table_grab();
645 netlink_update_subscriptions(sk, nlk->subscriptions +
646 hweight32(nladdr->nl_groups) -
647 hweight32(nlk->groups[0]));
648 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
649 netlink_update_listeners(sk);
650 netlink_table_ungrab();
655 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
659 struct sock *sk = sock->sk;
660 struct netlink_sock *nlk = nlk_sk(sk);
661 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
663 if (addr->sa_family == AF_UNSPEC) {
664 sk->sk_state = NETLINK_UNCONNECTED;
669 if (addr->sa_family != AF_NETLINK)
672 /* Only superuser is allowed to send multicasts */
673 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
677 err = netlink_autobind(sock);
680 sk->sk_state = NETLINK_CONNECTED;
681 nlk->dst_pid = nladdr->nl_pid;
682 nlk->dst_group = ffs(nladdr->nl_groups);
688 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
689 int *addr_len, int peer)
691 struct sock *sk = sock->sk;
692 struct netlink_sock *nlk = nlk_sk(sk);
693 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
695 nladdr->nl_family = AF_NETLINK;
697 *addr_len = sizeof(*nladdr);
700 nladdr->nl_pid = nlk->dst_pid;
701 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
703 nladdr->nl_pid = nlk->pid;
704 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
709 static void netlink_overrun(struct sock *sk)
711 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
712 sk->sk_err = ENOBUFS;
713 sk->sk_error_report(sk);
717 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
720 struct netlink_sock *nlk;
722 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
724 return ERR_PTR(-ECONNREFUSED);
726 /* Don't bother queuing skb if kernel socket has no input function */
728 if (sock->sk_state == NETLINK_CONNECTED &&
729 nlk->dst_pid != nlk_sk(ssk)->pid) {
731 return ERR_PTR(-ECONNREFUSED);
736 struct sock *netlink_getsockbyfilp(struct file *filp)
738 struct inode *inode = filp->f_path.dentry->d_inode;
741 if (!S_ISSOCK(inode->i_mode))
742 return ERR_PTR(-ENOTSOCK);
744 sock = SOCKET_I(inode)->sk;
745 if (sock->sk_family != AF_NETLINK)
746 return ERR_PTR(-EINVAL);
753 * Attach a skb to a netlink socket.
754 * The caller must hold a reference to the destination socket. On error, the
755 * reference is dropped. The skb is not send to the destination, just all
756 * all error checks are performed and memory in the queue is reserved.
758 * < 0: error. skb freed, reference to sock dropped.
760 * 1: repeat lookup - reference dropped while waiting for socket memory.
762 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
763 long *timeo, struct sock *ssk)
765 struct netlink_sock *nlk;
769 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
770 test_bit(0, &nlk->state)) {
771 DECLARE_WAITQUEUE(wait, current);
773 if (!ssk || netlink_is_kernel(ssk))
780 __set_current_state(TASK_INTERRUPTIBLE);
781 add_wait_queue(&nlk->wait, &wait);
783 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
784 test_bit(0, &nlk->state)) &&
785 !sock_flag(sk, SOCK_DEAD))
786 *timeo = schedule_timeout(*timeo);
788 __set_current_state(TASK_RUNNING);
789 remove_wait_queue(&nlk->wait, &wait);
792 if (signal_pending(current)) {
794 return sock_intr_errno(*timeo);
798 skb_set_owner_r(skb, sk);
802 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
806 skb_queue_tail(&sk->sk_receive_queue, skb);
807 sk->sk_data_ready(sk, len);
812 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
818 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
825 delta = skb->end - skb->tail;
826 if (delta * 2 < skb->truesize)
829 if (skb_shared(skb)) {
830 struct sk_buff *nskb = skb_clone(skb, allocation);
837 if (!pskb_expand_head(skb, 0, -delta, allocation))
838 skb->truesize -= delta;
843 static inline void netlink_rcv_wake(struct sock *sk)
845 struct netlink_sock *nlk = nlk_sk(sk);
847 if (skb_queue_empty(&sk->sk_receive_queue))
848 clear_bit(0, &nlk->state);
849 if (!test_bit(0, &nlk->state))
850 wake_up_interruptible(&nlk->wait);
853 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
856 struct netlink_sock *nlk = nlk_sk(sk);
859 if (nlk->netlink_rcv != NULL) {
861 skb_set_owner_r(skb, sk);
862 nlk->netlink_rcv(skb);
869 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
870 u32 pid, int nonblock)
876 skb = netlink_trim(skb, gfp_any());
878 timeo = sock_sndtimeo(ssk, nonblock);
880 sk = netlink_getsockbypid(ssk, pid);
885 if (netlink_is_kernel(sk))
886 return netlink_unicast_kernel(sk, skb);
888 if (sk_filter(sk, skb)) {
895 err = netlink_attachskb(sk, skb, nonblock, &timeo, ssk);
901 return netlink_sendskb(sk, skb);
903 EXPORT_SYMBOL(netlink_unicast);
905 int netlink_has_listeners(struct sock *sk, unsigned int group)
908 unsigned long *listeners;
910 BUG_ON(!netlink_is_kernel(sk));
913 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
915 if (group - 1 < nl_table[sk->sk_protocol].groups)
916 res = test_bit(group - 1, listeners);
922 EXPORT_SYMBOL_GPL(netlink_has_listeners);
924 static inline int netlink_broadcast_deliver(struct sock *sk,
927 struct netlink_sock *nlk = nlk_sk(sk);
929 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
930 !test_bit(0, &nlk->state)) {
931 skb_set_owner_r(skb, sk);
932 skb_queue_tail(&sk->sk_receive_queue, skb);
933 sk->sk_data_ready(sk, skb->len);
934 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
939 struct netlink_broadcast_data {
940 struct sock *exclude_sk;
948 struct sk_buff *skb, *skb2;
951 static inline int do_one_broadcast(struct sock *sk,
952 struct netlink_broadcast_data *p)
954 struct netlink_sock *nlk = nlk_sk(sk);
957 if (p->exclude_sk == sk)
960 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
961 !test_bit(p->group - 1, nlk->groups))
964 if (!net_eq(sock_net(sk), p->net))
973 if (p->skb2 == NULL) {
974 if (skb_shared(p->skb)) {
975 p->skb2 = skb_clone(p->skb, p->allocation);
977 p->skb2 = skb_get(p->skb);
979 * skb ownership may have been set when
980 * delivered to a previous socket.
985 if (p->skb2 == NULL) {
987 /* Clone failed. Notify ALL listeners. */
989 } else if (sk_filter(sk, p->skb2)) {
992 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1005 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1006 u32 group, gfp_t allocation)
1008 struct net *net = sock_net(ssk);
1009 struct netlink_broadcast_data info;
1010 struct hlist_node *node;
1013 skb = netlink_trim(skb, allocation);
1015 info.exclude_sk = ssk;
1022 info.allocation = allocation;
1026 /* While we sleep in clone, do not allow to change socket list */
1028 netlink_lock_table();
1030 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1031 do_one_broadcast(sk, &info);
1035 netlink_unlock_table();
1038 kfree_skb(info.skb2);
1040 if (info.delivered) {
1041 if (info.congested && (allocation & __GFP_WAIT))
1049 EXPORT_SYMBOL(netlink_broadcast);
1051 struct netlink_set_err_data {
1052 struct sock *exclude_sk;
1058 static inline int do_one_set_err(struct sock *sk,
1059 struct netlink_set_err_data *p)
1061 struct netlink_sock *nlk = nlk_sk(sk);
1063 if (sk == p->exclude_sk)
1066 if (sock_net(sk) != sock_net(p->exclude_sk))
1069 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1070 !test_bit(p->group - 1, nlk->groups))
1073 sk->sk_err = p->code;
1074 sk->sk_error_report(sk);
1079 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1081 struct netlink_set_err_data info;
1082 struct hlist_node *node;
1085 info.exclude_sk = ssk;
1090 read_lock(&nl_table_lock);
1092 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1093 do_one_set_err(sk, &info);
1095 read_unlock(&nl_table_lock);
1098 /* must be called with netlink table grabbed */
1099 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1103 int old, new = !!is_new, subscriptions;
1105 old = test_bit(group - 1, nlk->groups);
1106 subscriptions = nlk->subscriptions - old + new;
1108 __set_bit(group - 1, nlk->groups);
1110 __clear_bit(group - 1, nlk->groups);
1111 netlink_update_subscriptions(&nlk->sk, subscriptions);
1112 netlink_update_listeners(&nlk->sk);
1115 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1116 char __user *optval, int optlen)
1118 struct sock *sk = sock->sk;
1119 struct netlink_sock *nlk = nlk_sk(sk);
1120 unsigned int val = 0;
1123 if (level != SOL_NETLINK)
1124 return -ENOPROTOOPT;
1126 if (optlen >= sizeof(int) &&
1127 get_user(val, (unsigned int __user *)optval))
1131 case NETLINK_PKTINFO:
1133 nlk->flags |= NETLINK_RECV_PKTINFO;
1135 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1138 case NETLINK_ADD_MEMBERSHIP:
1139 case NETLINK_DROP_MEMBERSHIP: {
1140 if (!netlink_capable(sock, NL_NONROOT_RECV))
1142 err = netlink_realloc_groups(sk);
1145 if (!val || val - 1 >= nlk->ngroups)
1147 netlink_table_grab();
1148 netlink_update_socket_mc(nlk, val,
1149 optname == NETLINK_ADD_MEMBERSHIP);
1150 netlink_table_ungrab();
1160 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1161 char __user *optval, int __user *optlen)
1163 struct sock *sk = sock->sk;
1164 struct netlink_sock *nlk = nlk_sk(sk);
1167 if (level != SOL_NETLINK)
1168 return -ENOPROTOOPT;
1170 if (get_user(len, optlen))
1176 case NETLINK_PKTINFO:
1177 if (len < sizeof(int))
1180 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1181 if (put_user(len, optlen) ||
1182 put_user(val, optval))
1192 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1194 struct nl_pktinfo info;
1196 info.group = NETLINK_CB(skb).dst_group;
1197 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1200 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1201 struct msghdr *msg, size_t len)
1203 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1204 struct sock *sk = sock->sk;
1205 struct netlink_sock *nlk = nlk_sk(sk);
1206 struct sockaddr_nl *addr = msg->msg_name;
1209 struct sk_buff *skb;
1211 struct scm_cookie scm;
1213 if (msg->msg_flags&MSG_OOB)
1216 if (NULL == siocb->scm)
1218 err = scm_send(sock, msg, siocb->scm);
1222 if (msg->msg_namelen) {
1223 if (addr->nl_family != AF_NETLINK)
1225 dst_pid = addr->nl_pid;
1226 dst_group = ffs(addr->nl_groups);
1227 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1230 dst_pid = nlk->dst_pid;
1231 dst_group = nlk->dst_group;
1235 err = netlink_autobind(sock);
1241 if (len > sk->sk_sndbuf - 32)
1244 skb = alloc_skb(len, GFP_KERNEL);
1248 NETLINK_CB(skb).pid = nlk->pid;
1249 NETLINK_CB(skb).dst_group = dst_group;
1250 NETLINK_CB(skb).loginuid = audit_get_loginuid(current);
1251 NETLINK_CB(skb).sessionid = audit_get_sessionid(current);
1252 security_task_getsecid(current, &(NETLINK_CB(skb).sid));
1253 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1255 /* What can I do? Netlink is asynchronous, so that
1256 we will have to save current capabilities to
1257 check them, when this message will be delivered
1258 to corresponding kernel module. --ANK (980802)
1262 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1267 err = security_netlink_send(sk, skb);
1274 atomic_inc(&skb->users);
1275 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1277 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1283 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1284 struct msghdr *msg, size_t len,
1287 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1288 struct scm_cookie scm;
1289 struct sock *sk = sock->sk;
1290 struct netlink_sock *nlk = nlk_sk(sk);
1291 int noblock = flags&MSG_DONTWAIT;
1293 struct sk_buff *skb;
1301 skb = skb_recv_datagram(sk, flags, noblock, &err);
1305 msg->msg_namelen = 0;
1309 msg->msg_flags |= MSG_TRUNC;
1313 skb_reset_transport_header(skb);
1314 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1316 if (msg->msg_name) {
1317 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1318 addr->nl_family = AF_NETLINK;
1320 addr->nl_pid = NETLINK_CB(skb).pid;
1321 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1322 msg->msg_namelen = sizeof(*addr);
1325 if (nlk->flags & NETLINK_RECV_PKTINFO)
1326 netlink_cmsg_recv_pktinfo(msg, skb);
1328 if (NULL == siocb->scm) {
1329 memset(&scm, 0, sizeof(scm));
1332 siocb->scm->creds = *NETLINK_CREDS(skb);
1333 if (flags & MSG_TRUNC)
1335 skb_free_datagram(sk, skb);
1337 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1340 scm_recv(sock, msg, siocb->scm, flags);
1342 netlink_rcv_wake(sk);
1343 return err ? : copied;
1346 static void netlink_data_ready(struct sock *sk, int len)
1352 * We export these functions to other modules. They provide a
1353 * complete set of kernel non-blocking support for message
1358 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1359 void (*input)(struct sk_buff *skb),
1360 struct mutex *cb_mutex, struct module *module)
1362 struct socket *sock;
1364 struct netlink_sock *nlk;
1365 unsigned long *listeners = NULL;
1369 if (unit < 0 || unit >= MAX_LINKS)
1372 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1376 * We have to just have a reference on the net from sk, but don't
1377 * get_net it. Besides, we cannot get and then put the net here.
1378 * So we create one inside init_net and the move it to net.
1381 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1382 goto out_sock_release_nosk;
1385 sk_change_net(sk, net);
1390 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1392 goto out_sock_release;
1394 sk->sk_data_ready = netlink_data_ready;
1396 nlk_sk(sk)->netlink_rcv = input;
1398 if (netlink_insert(sk, net, 0))
1399 goto out_sock_release;
1402 nlk->flags |= NETLINK_KERNEL_SOCKET;
1404 netlink_table_grab();
1405 if (!nl_table[unit].registered) {
1406 nl_table[unit].groups = groups;
1407 nl_table[unit].listeners = listeners;
1408 nl_table[unit].cb_mutex = cb_mutex;
1409 nl_table[unit].module = module;
1410 nl_table[unit].registered = 1;
1413 nl_table[unit].registered++;
1415 netlink_table_ungrab();
1420 netlink_kernel_release(sk);
1423 out_sock_release_nosk:
1427 EXPORT_SYMBOL(netlink_kernel_create);
1431 netlink_kernel_release(struct sock *sk)
1433 sk_release_kernel(sk);
1435 EXPORT_SYMBOL(netlink_kernel_release);
1439 * netlink_change_ngroups - change number of multicast groups
1441 * This changes the number of multicast groups that are available
1442 * on a certain netlink family. Note that it is not possible to
1443 * change the number of groups to below 32. Also note that it does
1444 * not implicitly call netlink_clear_multicast_users() when the
1445 * number of groups is reduced.
1447 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1448 * @groups: The new number of groups.
1450 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1452 unsigned long *listeners, *old = NULL;
1453 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1459 netlink_table_grab();
1460 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1461 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1466 old = tbl->listeners;
1467 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1468 rcu_assign_pointer(tbl->listeners, listeners);
1470 tbl->groups = groups;
1473 netlink_table_ungrab();
1478 EXPORT_SYMBOL(netlink_change_ngroups);
1481 * netlink_clear_multicast_users - kick off multicast listeners
1483 * This function removes all listeners from the given group.
1484 * @ksk: The kernel netlink socket, as returned by
1485 * netlink_kernel_create().
1486 * @group: The multicast group to clear.
1488 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1491 struct hlist_node *node;
1492 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1494 netlink_table_grab();
1496 sk_for_each_bound(sk, node, &tbl->mc_list)
1497 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1499 netlink_table_ungrab();
1501 EXPORT_SYMBOL(netlink_clear_multicast_users);
1503 void netlink_set_nonroot(int protocol, unsigned int flags)
1505 if ((unsigned int)protocol < MAX_LINKS)
1506 nl_table[protocol].nl_nonroot = flags;
1508 EXPORT_SYMBOL(netlink_set_nonroot);
1510 static void netlink_destroy_callback(struct netlink_callback *cb)
1518 * It looks a bit ugly.
1519 * It would be better to create kernel thread.
1522 static int netlink_dump(struct sock *sk)
1524 struct netlink_sock *nlk = nlk_sk(sk);
1525 struct netlink_callback *cb;
1526 struct sk_buff *skb;
1527 struct nlmsghdr *nlh;
1528 int len, err = -ENOBUFS;
1530 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1534 mutex_lock(nlk->cb_mutex);
1542 len = cb->dump(skb, cb);
1545 mutex_unlock(nlk->cb_mutex);
1547 if (sk_filter(sk, skb))
1550 skb_queue_tail(&sk->sk_receive_queue, skb);
1551 sk->sk_data_ready(sk, skb->len);
1556 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1560 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1562 if (sk_filter(sk, skb))
1565 skb_queue_tail(&sk->sk_receive_queue, skb);
1566 sk->sk_data_ready(sk, skb->len);
1572 mutex_unlock(nlk->cb_mutex);
1574 netlink_destroy_callback(cb);
1578 mutex_unlock(nlk->cb_mutex);
1584 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1585 struct nlmsghdr *nlh,
1586 int (*dump)(struct sk_buff *skb,
1587 struct netlink_callback *),
1588 int (*done)(struct netlink_callback *))
1590 struct netlink_callback *cb;
1592 struct netlink_sock *nlk;
1594 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1601 atomic_inc(&skb->users);
1604 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1606 netlink_destroy_callback(cb);
1607 return -ECONNREFUSED;
1610 /* A dump is in progress... */
1611 mutex_lock(nlk->cb_mutex);
1613 mutex_unlock(nlk->cb_mutex);
1614 netlink_destroy_callback(cb);
1619 mutex_unlock(nlk->cb_mutex);
1624 /* We successfully started a dump, by returning -EINTR we
1625 * signal not to send ACK even if it was requested.
1629 EXPORT_SYMBOL(netlink_dump_start);
1631 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1633 struct sk_buff *skb;
1634 struct nlmsghdr *rep;
1635 struct nlmsgerr *errmsg;
1636 size_t payload = sizeof(*errmsg);
1638 /* error messages get the original request appened */
1640 payload += nlmsg_len(nlh);
1642 skb = nlmsg_new(payload, GFP_KERNEL);
1646 sk = netlink_lookup(sock_net(in_skb->sk),
1647 in_skb->sk->sk_protocol,
1648 NETLINK_CB(in_skb).pid);
1650 sk->sk_err = ENOBUFS;
1651 sk->sk_error_report(sk);
1657 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1658 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1659 errmsg = nlmsg_data(rep);
1660 errmsg->error = err;
1661 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1662 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1664 EXPORT_SYMBOL(netlink_ack);
1666 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1669 struct nlmsghdr *nlh;
1672 while (skb->len >= nlmsg_total_size(0)) {
1675 nlh = nlmsg_hdr(skb);
1678 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1681 /* Only requests are handled by the kernel */
1682 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1685 /* Skip control messages */
1686 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1694 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1695 netlink_ack(skb, nlh, err);
1698 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1699 if (msglen > skb->len)
1701 skb_pull(skb, msglen);
1706 EXPORT_SYMBOL(netlink_rcv_skb);
1709 * nlmsg_notify - send a notification netlink message
1710 * @sk: netlink socket to use
1711 * @skb: notification message
1712 * @pid: destination netlink pid for reports or 0
1713 * @group: destination multicast group or 0
1714 * @report: 1 to report back, 0 to disable
1715 * @flags: allocation flags
1717 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1718 unsigned int group, int report, gfp_t flags)
1723 int exclude_pid = 0;
1726 atomic_inc(&skb->users);
1730 /* errors reported via destination sk->sk_err */
1731 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1735 err = nlmsg_unicast(sk, skb, pid);
1739 EXPORT_SYMBOL(nlmsg_notify);
1741 #ifdef CONFIG_PROC_FS
1742 struct nl_seq_iter {
1743 struct seq_net_private p;
1748 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1750 struct nl_seq_iter *iter = seq->private;
1753 struct hlist_node *node;
1756 for (i = 0; i < MAX_LINKS; i++) {
1757 struct nl_pid_hash *hash = &nl_table[i].hash;
1759 for (j = 0; j <= hash->mask; j++) {
1760 sk_for_each(s, node, &hash->table[j]) {
1761 if (sock_net(s) != seq_file_net(seq))
1775 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1776 __acquires(nl_table_lock)
1778 read_lock(&nl_table_lock);
1779 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1782 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1785 struct nl_seq_iter *iter;
1790 if (v == SEQ_START_TOKEN)
1791 return netlink_seq_socket_idx(seq, 0);
1793 iter = seq->private;
1797 } while (s && sock_net(s) != seq_file_net(seq));
1802 j = iter->hash_idx + 1;
1805 struct nl_pid_hash *hash = &nl_table[i].hash;
1807 for (; j <= hash->mask; j++) {
1808 s = sk_head(&hash->table[j]);
1809 while (s && sock_net(s) != seq_file_net(seq))
1819 } while (++i < MAX_LINKS);
1824 static void netlink_seq_stop(struct seq_file *seq, void *v)
1825 __releases(nl_table_lock)
1827 read_unlock(&nl_table_lock);
1831 static int netlink_seq_show(struct seq_file *seq, void *v)
1833 if (v == SEQ_START_TOKEN)
1835 "sk Eth Pid Groups "
1836 "Rmem Wmem Dump Locks\n");
1839 struct netlink_sock *nlk = nlk_sk(s);
1841 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1845 nlk->groups ? (u32)nlk->groups[0] : 0,
1846 atomic_read(&s->sk_rmem_alloc),
1847 atomic_read(&s->sk_wmem_alloc),
1849 atomic_read(&s->sk_refcnt)
1856 static const struct seq_operations netlink_seq_ops = {
1857 .start = netlink_seq_start,
1858 .next = netlink_seq_next,
1859 .stop = netlink_seq_stop,
1860 .show = netlink_seq_show,
1864 static int netlink_seq_open(struct inode *inode, struct file *file)
1866 return seq_open_net(inode, file, &netlink_seq_ops,
1867 sizeof(struct nl_seq_iter));
1870 static const struct file_operations netlink_seq_fops = {
1871 .owner = THIS_MODULE,
1872 .open = netlink_seq_open,
1874 .llseek = seq_lseek,
1875 .release = seq_release_net,
1880 int netlink_register_notifier(struct notifier_block *nb)
1882 return atomic_notifier_chain_register(&netlink_chain, nb);
1884 EXPORT_SYMBOL(netlink_register_notifier);
1886 int netlink_unregister_notifier(struct notifier_block *nb)
1888 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1890 EXPORT_SYMBOL(netlink_unregister_notifier);
1892 static const struct proto_ops netlink_ops = {
1893 .family = PF_NETLINK,
1894 .owner = THIS_MODULE,
1895 .release = netlink_release,
1896 .bind = netlink_bind,
1897 .connect = netlink_connect,
1898 .socketpair = sock_no_socketpair,
1899 .accept = sock_no_accept,
1900 .getname = netlink_getname,
1901 .poll = datagram_poll,
1902 .ioctl = sock_no_ioctl,
1903 .listen = sock_no_listen,
1904 .shutdown = sock_no_shutdown,
1905 .setsockopt = netlink_setsockopt,
1906 .getsockopt = netlink_getsockopt,
1907 .sendmsg = netlink_sendmsg,
1908 .recvmsg = netlink_recvmsg,
1909 .mmap = sock_no_mmap,
1910 .sendpage = sock_no_sendpage,
1913 static struct net_proto_family netlink_family_ops = {
1914 .family = PF_NETLINK,
1915 .create = netlink_create,
1916 .owner = THIS_MODULE, /* for consistency 8) */
1919 static int __net_init netlink_net_init(struct net *net)
1921 #ifdef CONFIG_PROC_FS
1922 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1928 static void __net_exit netlink_net_exit(struct net *net)
1930 #ifdef CONFIG_PROC_FS
1931 proc_net_remove(net, "netlink");
1935 static struct pernet_operations __net_initdata netlink_net_ops = {
1936 .init = netlink_net_init,
1937 .exit = netlink_net_exit,
1940 static int __init netlink_proto_init(void)
1942 struct sk_buff *dummy_skb;
1944 unsigned long limit;
1946 int err = proto_register(&netlink_proto, 0);
1951 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1953 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1957 if (num_physpages >= (128 * 1024))
1958 limit = num_physpages >> (21 - PAGE_SHIFT);
1960 limit = num_physpages >> (23 - PAGE_SHIFT);
1962 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1963 limit = (1UL << order) / sizeof(struct hlist_head);
1964 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1966 for (i = 0; i < MAX_LINKS; i++) {
1967 struct nl_pid_hash *hash = &nl_table[i].hash;
1969 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
1972 nl_pid_hash_free(nl_table[i].hash.table,
1973 1 * sizeof(*hash->table));
1977 hash->max_shift = order;
1980 hash->rehash_time = jiffies;
1983 sock_register(&netlink_family_ops);
1984 register_pernet_subsys(&netlink_net_ops);
1985 /* The netlink device handler may be needed early. */
1990 panic("netlink_init: Cannot allocate nl_table\n");
1993 core_initcall(netlink_proto_init);