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)
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 inline void netlink_table_ungrab(void)
197 write_unlock_irq(&nl_table_lock);
198 wake_up(&nl_table_wait);
202 netlink_lock_table(void)
204 /* read_lock() synchronizes us to netlink_table_grab */
206 read_lock(&nl_table_lock);
207 atomic_inc(&nl_table_users);
208 read_unlock(&nl_table_lock);
212 netlink_unlock_table(void)
214 if (atomic_dec_and_test(&nl_table_users))
215 wake_up(&nl_table_wait);
218 static inline struct sock *netlink_lookup(struct net *net, int protocol,
221 struct nl_pid_hash *hash = &nl_table[protocol].hash;
222 struct hlist_head *head;
224 struct hlist_node *node;
226 read_lock(&nl_table_lock);
227 head = nl_pid_hashfn(hash, pid);
228 sk_for_each(sk, node, head) {
229 if ((sk->sk_net == net) && (nlk_sk(sk)->pid == pid)) {
236 read_unlock(&nl_table_lock);
240 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
242 if (size <= PAGE_SIZE)
243 return kmalloc(size, GFP_ATOMIC);
245 return (struct hlist_head *)
246 __get_free_pages(GFP_ATOMIC, get_order(size));
249 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
251 if (size <= PAGE_SIZE)
254 free_pages((unsigned long)table, get_order(size));
257 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
259 unsigned int omask, mask, shift;
261 struct hlist_head *otable, *table;
264 omask = mask = hash->mask;
265 osize = size = (mask + 1) * sizeof(*table);
269 if (++shift > hash->max_shift)
275 table = nl_pid_hash_alloc(size);
279 memset(table, 0, size);
280 otable = hash->table;
284 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
286 for (i = 0; i <= omask; i++) {
288 struct hlist_node *node, *tmp;
290 sk_for_each_safe(sk, node, tmp, &otable[i])
291 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
294 nl_pid_hash_free(otable, osize);
295 hash->rehash_time = jiffies + 10 * 60 * HZ;
299 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
301 int avg = hash->entries >> hash->shift;
303 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
306 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
307 nl_pid_hash_rehash(hash, 0);
314 static const struct proto_ops netlink_ops;
317 netlink_update_listeners(struct sock *sk)
319 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
320 struct hlist_node *node;
324 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
326 sk_for_each_bound(sk, node, &tbl->mc_list) {
327 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
328 mask |= nlk_sk(sk)->groups[i];
330 tbl->listeners[i] = mask;
332 /* this function is only called with the netlink table "grabbed", which
333 * makes sure updates are visible before bind or setsockopt return. */
336 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
338 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
339 struct hlist_head *head;
340 int err = -EADDRINUSE;
342 struct hlist_node *node;
345 netlink_table_grab();
346 head = nl_pid_hashfn(hash, pid);
348 sk_for_each(osk, node, head) {
349 if ((osk->sk_net == net) && (nlk_sk(osk)->pid == pid))
361 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
364 if (len && nl_pid_hash_dilute(hash, len))
365 head = nl_pid_hashfn(hash, pid);
367 nlk_sk(sk)->pid = pid;
368 sk_add_node(sk, head);
372 netlink_table_ungrab();
376 static void netlink_remove(struct sock *sk)
378 netlink_table_grab();
379 if (sk_del_node_init(sk))
380 nl_table[sk->sk_protocol].hash.entries--;
381 if (nlk_sk(sk)->subscriptions)
382 __sk_del_bind_node(sk);
383 netlink_table_ungrab();
386 static struct proto netlink_proto = {
388 .owner = THIS_MODULE,
389 .obj_size = sizeof(struct netlink_sock),
392 static int __netlink_create(struct net *net, struct socket *sock,
393 struct mutex *cb_mutex, int protocol)
396 struct netlink_sock *nlk;
398 sock->ops = &netlink_ops;
400 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
404 sock_init_data(sock, sk);
408 nlk->cb_mutex = cb_mutex;
410 nlk->cb_mutex = &nlk->cb_def_mutex;
411 mutex_init(nlk->cb_mutex);
413 init_waitqueue_head(&nlk->wait);
415 sk->sk_destruct = netlink_sock_destruct;
416 sk->sk_protocol = protocol;
420 static int netlink_create(struct net *net, struct socket *sock, int protocol)
422 struct module *module = NULL;
423 struct mutex *cb_mutex;
424 struct netlink_sock *nlk;
427 sock->state = SS_UNCONNECTED;
429 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
430 return -ESOCKTNOSUPPORT;
432 if (protocol < 0 || protocol >= MAX_LINKS)
433 return -EPROTONOSUPPORT;
435 netlink_lock_table();
437 if (!nl_table[protocol].registered) {
438 netlink_unlock_table();
439 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
440 netlink_lock_table();
443 if (nl_table[protocol].registered &&
444 try_module_get(nl_table[protocol].module))
445 module = nl_table[protocol].module;
446 cb_mutex = nl_table[protocol].cb_mutex;
447 netlink_unlock_table();
449 err = __netlink_create(net, sock, cb_mutex, protocol);
453 nlk = nlk_sk(sock->sk);
454 nlk->module = module;
463 static int netlink_release(struct socket *sock)
465 struct sock *sk = sock->sk;
466 struct netlink_sock *nlk;
476 * OK. Socket is unlinked, any packets that arrive now
481 wake_up_interruptible_all(&nlk->wait);
483 skb_queue_purge(&sk->sk_write_queue);
485 if (nlk->pid && !nlk->subscriptions) {
486 struct netlink_notify n = {
488 .protocol = sk->sk_protocol,
491 atomic_notifier_call_chain(&netlink_chain,
492 NETLINK_URELEASE, &n);
495 module_put(nlk->module);
497 netlink_table_grab();
498 if (netlink_is_kernel(sk)) {
499 kfree(nl_table[sk->sk_protocol].listeners);
500 nl_table[sk->sk_protocol].module = NULL;
501 nl_table[sk->sk_protocol].registered = 0;
502 } else if (nlk->subscriptions)
503 netlink_update_listeners(sk);
504 netlink_table_ungrab();
513 static int netlink_autobind(struct socket *sock)
515 struct sock *sk = sock->sk;
516 struct net *net = sk->sk_net;
517 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
518 struct hlist_head *head;
520 struct hlist_node *node;
521 s32 pid = current->tgid;
523 static s32 rover = -4097;
527 netlink_table_grab();
528 head = nl_pid_hashfn(hash, pid);
529 sk_for_each(osk, node, head) {
530 if ((osk->sk_net != net))
532 if (nlk_sk(osk)->pid == pid) {
533 /* Bind collision, search negative pid values. */
537 netlink_table_ungrab();
541 netlink_table_ungrab();
543 err = netlink_insert(sk, net, pid);
544 if (err == -EADDRINUSE)
547 /* If 2 threads race to autobind, that is fine. */
554 static inline int netlink_capable(struct socket *sock, unsigned int flag)
556 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
557 capable(CAP_NET_ADMIN);
561 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
563 struct netlink_sock *nlk = nlk_sk(sk);
565 if (nlk->subscriptions && !subscriptions)
566 __sk_del_bind_node(sk);
567 else if (!nlk->subscriptions && subscriptions)
568 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
569 nlk->subscriptions = subscriptions;
572 static int netlink_realloc_groups(struct sock *sk)
574 struct netlink_sock *nlk = nlk_sk(sk);
576 unsigned long *new_groups;
579 netlink_table_grab();
581 groups = nl_table[sk->sk_protocol].groups;
582 if (!nl_table[sk->sk_protocol].registered) {
587 if (nlk->ngroups >= groups)
590 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
591 if (new_groups == NULL) {
595 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
596 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
598 nlk->groups = new_groups;
599 nlk->ngroups = groups;
601 netlink_table_ungrab();
605 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
608 struct sock *sk = sock->sk;
609 struct net *net = sk->sk_net;
610 struct netlink_sock *nlk = nlk_sk(sk);
611 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
614 if (nladdr->nl_family != AF_NETLINK)
617 /* Only superuser is allowed to listen multicasts */
618 if (nladdr->nl_groups) {
619 if (!netlink_capable(sock, NL_NONROOT_RECV))
621 err = netlink_realloc_groups(sk);
627 if (nladdr->nl_pid != nlk->pid)
630 err = nladdr->nl_pid ?
631 netlink_insert(sk, net, nladdr->nl_pid) :
632 netlink_autobind(sock);
637 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
640 netlink_table_grab();
641 netlink_update_subscriptions(sk, nlk->subscriptions +
642 hweight32(nladdr->nl_groups) -
643 hweight32(nlk->groups[0]));
644 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
645 netlink_update_listeners(sk);
646 netlink_table_ungrab();
651 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
655 struct sock *sk = sock->sk;
656 struct netlink_sock *nlk = nlk_sk(sk);
657 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
659 if (addr->sa_family == AF_UNSPEC) {
660 sk->sk_state = NETLINK_UNCONNECTED;
665 if (addr->sa_family != AF_NETLINK)
668 /* Only superuser is allowed to send multicasts */
669 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
673 err = netlink_autobind(sock);
676 sk->sk_state = NETLINK_CONNECTED;
677 nlk->dst_pid = nladdr->nl_pid;
678 nlk->dst_group = ffs(nladdr->nl_groups);
684 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
685 int *addr_len, int peer)
687 struct sock *sk = sock->sk;
688 struct netlink_sock *nlk = nlk_sk(sk);
689 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
691 nladdr->nl_family = AF_NETLINK;
693 *addr_len = sizeof(*nladdr);
696 nladdr->nl_pid = nlk->dst_pid;
697 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
699 nladdr->nl_pid = nlk->pid;
700 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
705 static void netlink_overrun(struct sock *sk)
707 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
708 sk->sk_err = ENOBUFS;
709 sk->sk_error_report(sk);
713 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
716 struct netlink_sock *nlk;
718 sock = netlink_lookup(ssk->sk_net, ssk->sk_protocol, pid);
720 return ERR_PTR(-ECONNREFUSED);
722 /* Don't bother queuing skb if kernel socket has no input function */
724 if (sock->sk_state == NETLINK_CONNECTED &&
725 nlk->dst_pid != nlk_sk(ssk)->pid) {
727 return ERR_PTR(-ECONNREFUSED);
732 struct sock *netlink_getsockbyfilp(struct file *filp)
734 struct inode *inode = filp->f_path.dentry->d_inode;
737 if (!S_ISSOCK(inode->i_mode))
738 return ERR_PTR(-ENOTSOCK);
740 sock = SOCKET_I(inode)->sk;
741 if (sock->sk_family != AF_NETLINK)
742 return ERR_PTR(-EINVAL);
749 * Attach a skb to a netlink socket.
750 * The caller must hold a reference to the destination socket. On error, the
751 * reference is dropped. The skb is not send to the destination, just all
752 * all error checks are performed and memory in the queue is reserved.
754 * < 0: error. skb freed, reference to sock dropped.
756 * 1: repeat lookup - reference dropped while waiting for socket memory.
758 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
759 long *timeo, struct sock *ssk)
761 struct netlink_sock *nlk;
765 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
766 test_bit(0, &nlk->state)) {
767 DECLARE_WAITQUEUE(wait, current);
769 if (!ssk || netlink_is_kernel(ssk))
776 __set_current_state(TASK_INTERRUPTIBLE);
777 add_wait_queue(&nlk->wait, &wait);
779 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
780 test_bit(0, &nlk->state)) &&
781 !sock_flag(sk, SOCK_DEAD))
782 *timeo = schedule_timeout(*timeo);
784 __set_current_state(TASK_RUNNING);
785 remove_wait_queue(&nlk->wait, &wait);
788 if (signal_pending(current)) {
790 return sock_intr_errno(*timeo);
794 skb_set_owner_r(skb, sk);
798 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
802 skb_queue_tail(&sk->sk_receive_queue, skb);
803 sk->sk_data_ready(sk, len);
808 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
814 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
821 delta = skb->end - skb->tail;
822 if (delta * 2 < skb->truesize)
825 if (skb_shared(skb)) {
826 struct sk_buff *nskb = skb_clone(skb, allocation);
833 if (!pskb_expand_head(skb, 0, -delta, allocation))
834 skb->truesize -= delta;
839 static inline void netlink_rcv_wake(struct sock *sk)
841 struct netlink_sock *nlk = nlk_sk(sk);
843 if (skb_queue_empty(&sk->sk_receive_queue))
844 clear_bit(0, &nlk->state);
845 if (!test_bit(0, &nlk->state))
846 wake_up_interruptible(&nlk->wait);
849 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
852 struct netlink_sock *nlk = nlk_sk(sk);
855 if (nlk->netlink_rcv != NULL) {
857 skb_set_owner_r(skb, sk);
858 nlk->netlink_rcv(skb);
865 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
866 u32 pid, int nonblock)
872 skb = netlink_trim(skb, gfp_any());
874 timeo = sock_sndtimeo(ssk, nonblock);
876 sk = netlink_getsockbypid(ssk, pid);
881 if (netlink_is_kernel(sk))
882 return netlink_unicast_kernel(sk, skb);
884 err = netlink_attachskb(sk, skb, nonblock, &timeo, ssk);
890 return netlink_sendskb(sk, skb);
892 EXPORT_SYMBOL(netlink_unicast);
894 int netlink_has_listeners(struct sock *sk, unsigned int group)
897 unsigned long *listeners;
899 BUG_ON(!netlink_is_kernel(sk));
902 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
904 if (group - 1 < nl_table[sk->sk_protocol].groups)
905 res = test_bit(group - 1, listeners);
911 EXPORT_SYMBOL_GPL(netlink_has_listeners);
913 static inline int netlink_broadcast_deliver(struct sock *sk,
916 struct netlink_sock *nlk = nlk_sk(sk);
918 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
919 !test_bit(0, &nlk->state)) {
920 skb_set_owner_r(skb, sk);
921 skb_queue_tail(&sk->sk_receive_queue, skb);
922 sk->sk_data_ready(sk, skb->len);
923 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
928 struct netlink_broadcast_data {
929 struct sock *exclude_sk;
937 struct sk_buff *skb, *skb2;
940 static inline int do_one_broadcast(struct sock *sk,
941 struct netlink_broadcast_data *p)
943 struct netlink_sock *nlk = nlk_sk(sk);
946 if (p->exclude_sk == sk)
949 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
950 !test_bit(p->group - 1, nlk->groups))
953 if ((sk->sk_net != p->net))
962 if (p->skb2 == NULL) {
963 if (skb_shared(p->skb)) {
964 p->skb2 = skb_clone(p->skb, p->allocation);
966 p->skb2 = skb_get(p->skb);
968 * skb ownership may have been set when
969 * delivered to a previous socket.
974 if (p->skb2 == NULL) {
976 /* Clone failed. Notify ALL listeners. */
978 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
991 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
992 u32 group, gfp_t allocation)
994 struct net *net = ssk->sk_net;
995 struct netlink_broadcast_data info;
996 struct hlist_node *node;
999 skb = netlink_trim(skb, allocation);
1001 info.exclude_sk = ssk;
1008 info.allocation = allocation;
1012 /* While we sleep in clone, do not allow to change socket list */
1014 netlink_lock_table();
1016 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1017 do_one_broadcast(sk, &info);
1021 netlink_unlock_table();
1024 kfree_skb(info.skb2);
1026 if (info.delivered) {
1027 if (info.congested && (allocation & __GFP_WAIT))
1035 EXPORT_SYMBOL(netlink_broadcast);
1037 struct netlink_set_err_data {
1038 struct sock *exclude_sk;
1044 static inline int do_one_set_err(struct sock *sk,
1045 struct netlink_set_err_data *p)
1047 struct netlink_sock *nlk = nlk_sk(sk);
1049 if (sk == p->exclude_sk)
1052 if (sk->sk_net != p->exclude_sk->sk_net)
1055 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1056 !test_bit(p->group - 1, nlk->groups))
1059 sk->sk_err = p->code;
1060 sk->sk_error_report(sk);
1065 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1067 struct netlink_set_err_data info;
1068 struct hlist_node *node;
1071 info.exclude_sk = ssk;
1076 read_lock(&nl_table_lock);
1078 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1079 do_one_set_err(sk, &info);
1081 read_unlock(&nl_table_lock);
1084 /* must be called with netlink table grabbed */
1085 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1089 int old, new = !!is_new, subscriptions;
1091 old = test_bit(group - 1, nlk->groups);
1092 subscriptions = nlk->subscriptions - old + new;
1094 __set_bit(group - 1, nlk->groups);
1096 __clear_bit(group - 1, nlk->groups);
1097 netlink_update_subscriptions(&nlk->sk, subscriptions);
1098 netlink_update_listeners(&nlk->sk);
1101 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1102 char __user *optval, int optlen)
1104 struct sock *sk = sock->sk;
1105 struct netlink_sock *nlk = nlk_sk(sk);
1106 unsigned int val = 0;
1109 if (level != SOL_NETLINK)
1110 return -ENOPROTOOPT;
1112 if (optlen >= sizeof(int) &&
1113 get_user(val, (unsigned int __user *)optval))
1117 case NETLINK_PKTINFO:
1119 nlk->flags |= NETLINK_RECV_PKTINFO;
1121 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1124 case NETLINK_ADD_MEMBERSHIP:
1125 case NETLINK_DROP_MEMBERSHIP: {
1126 if (!netlink_capable(sock, NL_NONROOT_RECV))
1128 err = netlink_realloc_groups(sk);
1131 if (!val || val - 1 >= nlk->ngroups)
1133 netlink_table_grab();
1134 netlink_update_socket_mc(nlk, val,
1135 optname == NETLINK_ADD_MEMBERSHIP);
1136 netlink_table_ungrab();
1146 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1147 char __user *optval, int __user *optlen)
1149 struct sock *sk = sock->sk;
1150 struct netlink_sock *nlk = nlk_sk(sk);
1153 if (level != SOL_NETLINK)
1154 return -ENOPROTOOPT;
1156 if (get_user(len, optlen))
1162 case NETLINK_PKTINFO:
1163 if (len < sizeof(int))
1166 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1167 if (put_user(len, optlen) ||
1168 put_user(val, optval))
1178 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1180 struct nl_pktinfo info;
1182 info.group = NETLINK_CB(skb).dst_group;
1183 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1186 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1187 struct msghdr *msg, size_t len)
1189 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1190 struct sock *sk = sock->sk;
1191 struct netlink_sock *nlk = nlk_sk(sk);
1192 struct sockaddr_nl *addr = msg->msg_name;
1195 struct sk_buff *skb;
1197 struct scm_cookie scm;
1199 if (msg->msg_flags&MSG_OOB)
1202 if (NULL == siocb->scm)
1204 err = scm_send(sock, msg, siocb->scm);
1208 if (msg->msg_namelen) {
1209 if (addr->nl_family != AF_NETLINK)
1211 dst_pid = addr->nl_pid;
1212 dst_group = ffs(addr->nl_groups);
1213 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1216 dst_pid = nlk->dst_pid;
1217 dst_group = nlk->dst_group;
1221 err = netlink_autobind(sock);
1227 if (len > sk->sk_sndbuf - 32)
1230 skb = alloc_skb(len, GFP_KERNEL);
1234 NETLINK_CB(skb).pid = nlk->pid;
1235 NETLINK_CB(skb).dst_group = dst_group;
1236 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1237 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1238 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1240 /* What can I do? Netlink is asynchronous, so that
1241 we will have to save current capabilities to
1242 check them, when this message will be delivered
1243 to corresponding kernel module. --ANK (980802)
1247 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1252 err = security_netlink_send(sk, skb);
1259 atomic_inc(&skb->users);
1260 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1262 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1268 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1269 struct msghdr *msg, size_t len,
1272 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1273 struct scm_cookie scm;
1274 struct sock *sk = sock->sk;
1275 struct netlink_sock *nlk = nlk_sk(sk);
1276 int noblock = flags&MSG_DONTWAIT;
1278 struct sk_buff *skb;
1286 skb = skb_recv_datagram(sk, flags, noblock, &err);
1290 msg->msg_namelen = 0;
1294 msg->msg_flags |= MSG_TRUNC;
1298 skb_reset_transport_header(skb);
1299 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1301 if (msg->msg_name) {
1302 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1303 addr->nl_family = AF_NETLINK;
1305 addr->nl_pid = NETLINK_CB(skb).pid;
1306 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1307 msg->msg_namelen = sizeof(*addr);
1310 if (nlk->flags & NETLINK_RECV_PKTINFO)
1311 netlink_cmsg_recv_pktinfo(msg, skb);
1313 if (NULL == siocb->scm) {
1314 memset(&scm, 0, sizeof(scm));
1317 siocb->scm->creds = *NETLINK_CREDS(skb);
1318 if (flags & MSG_TRUNC)
1320 skb_free_datagram(sk, skb);
1322 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1325 scm_recv(sock, msg, siocb->scm, flags);
1327 netlink_rcv_wake(sk);
1328 return err ? : copied;
1331 static void netlink_data_ready(struct sock *sk, int len)
1337 * We export these functions to other modules. They provide a
1338 * complete set of kernel non-blocking support for message
1343 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1344 void (*input)(struct sk_buff *skb),
1345 struct mutex *cb_mutex, struct module *module)
1347 struct socket *sock;
1349 struct netlink_sock *nlk;
1350 unsigned long *listeners = NULL;
1354 if (unit < 0 || unit >= MAX_LINKS)
1357 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1360 if (__netlink_create(net, sock, cb_mutex, unit) < 0)
1361 goto out_sock_release;
1366 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1368 goto out_sock_release;
1371 sk->sk_data_ready = netlink_data_ready;
1373 nlk_sk(sk)->netlink_rcv = input;
1375 if (netlink_insert(sk, net, 0))
1376 goto out_sock_release;
1379 nlk->flags |= NETLINK_KERNEL_SOCKET;
1381 netlink_table_grab();
1382 if (!nl_table[unit].registered) {
1383 nl_table[unit].groups = groups;
1384 nl_table[unit].listeners = listeners;
1385 nl_table[unit].cb_mutex = cb_mutex;
1386 nl_table[unit].module = module;
1387 nl_table[unit].registered = 1;
1391 netlink_table_ungrab();
1400 EXPORT_SYMBOL(netlink_kernel_create);
1403 * netlink_change_ngroups - change number of multicast groups
1405 * This changes the number of multicast groups that are available
1406 * on a certain netlink family. Note that it is not possible to
1407 * change the number of groups to below 32. Also note that it does
1408 * not implicitly call netlink_clear_multicast_users() when the
1409 * number of groups is reduced.
1411 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1412 * @groups: The new number of groups.
1414 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1416 unsigned long *listeners, *old = NULL;
1417 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1423 netlink_table_grab();
1424 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1425 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1430 old = tbl->listeners;
1431 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1432 rcu_assign_pointer(tbl->listeners, listeners);
1434 tbl->groups = groups;
1437 netlink_table_ungrab();
1442 EXPORT_SYMBOL(netlink_change_ngroups);
1445 * netlink_clear_multicast_users - kick off multicast listeners
1447 * This function removes all listeners from the given group.
1448 * @ksk: The kernel netlink socket, as returned by
1449 * netlink_kernel_create().
1450 * @group: The multicast group to clear.
1452 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1455 struct hlist_node *node;
1456 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1458 netlink_table_grab();
1460 sk_for_each_bound(sk, node, &tbl->mc_list)
1461 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1463 netlink_table_ungrab();
1465 EXPORT_SYMBOL(netlink_clear_multicast_users);
1467 void netlink_set_nonroot(int protocol, unsigned int flags)
1469 if ((unsigned int)protocol < MAX_LINKS)
1470 nl_table[protocol].nl_nonroot = flags;
1472 EXPORT_SYMBOL(netlink_set_nonroot);
1474 static void netlink_destroy_callback(struct netlink_callback *cb)
1482 * It looks a bit ugly.
1483 * It would be better to create kernel thread.
1486 static int netlink_dump(struct sock *sk)
1488 struct netlink_sock *nlk = nlk_sk(sk);
1489 struct netlink_callback *cb;
1490 struct sk_buff *skb;
1491 struct nlmsghdr *nlh;
1492 int len, err = -ENOBUFS;
1494 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1498 mutex_lock(nlk->cb_mutex);
1506 len = cb->dump(skb, cb);
1509 mutex_unlock(nlk->cb_mutex);
1510 skb_queue_tail(&sk->sk_receive_queue, skb);
1511 sk->sk_data_ready(sk, len);
1515 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1519 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1521 skb_queue_tail(&sk->sk_receive_queue, skb);
1522 sk->sk_data_ready(sk, skb->len);
1527 mutex_unlock(nlk->cb_mutex);
1529 netlink_destroy_callback(cb);
1533 mutex_unlock(nlk->cb_mutex);
1539 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1540 struct nlmsghdr *nlh,
1541 int (*dump)(struct sk_buff *skb,
1542 struct netlink_callback *),
1543 int (*done)(struct netlink_callback *))
1545 struct netlink_callback *cb;
1547 struct netlink_sock *nlk;
1549 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1556 atomic_inc(&skb->users);
1559 sk = netlink_lookup(ssk->sk_net, ssk->sk_protocol, NETLINK_CB(skb).pid);
1561 netlink_destroy_callback(cb);
1562 return -ECONNREFUSED;
1565 /* A dump is in progress... */
1566 mutex_lock(nlk->cb_mutex);
1568 mutex_unlock(nlk->cb_mutex);
1569 netlink_destroy_callback(cb);
1574 mutex_unlock(nlk->cb_mutex);
1579 /* We successfully started a dump, by returning -EINTR we
1580 * signal not to send ACK even if it was requested.
1584 EXPORT_SYMBOL(netlink_dump_start);
1586 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1588 struct sk_buff *skb;
1589 struct nlmsghdr *rep;
1590 struct nlmsgerr *errmsg;
1591 size_t payload = sizeof(*errmsg);
1593 /* error messages get the original request appened */
1595 payload += nlmsg_len(nlh);
1597 skb = nlmsg_new(payload, GFP_KERNEL);
1601 sk = netlink_lookup(in_skb->sk->sk_net,
1602 in_skb->sk->sk_protocol,
1603 NETLINK_CB(in_skb).pid);
1605 sk->sk_err = ENOBUFS;
1606 sk->sk_error_report(sk);
1612 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1613 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1614 errmsg = nlmsg_data(rep);
1615 errmsg->error = err;
1616 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1617 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1619 EXPORT_SYMBOL(netlink_ack);
1621 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1624 struct nlmsghdr *nlh;
1627 while (skb->len >= nlmsg_total_size(0)) {
1630 nlh = nlmsg_hdr(skb);
1633 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1636 /* Only requests are handled by the kernel */
1637 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1640 /* Skip control messages */
1641 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1649 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1650 netlink_ack(skb, nlh, err);
1653 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1654 if (msglen > skb->len)
1656 skb_pull(skb, msglen);
1661 EXPORT_SYMBOL(netlink_rcv_skb);
1664 * nlmsg_notify - send a notification netlink message
1665 * @sk: netlink socket to use
1666 * @skb: notification message
1667 * @pid: destination netlink pid for reports or 0
1668 * @group: destination multicast group or 0
1669 * @report: 1 to report back, 0 to disable
1670 * @flags: allocation flags
1672 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1673 unsigned int group, int report, gfp_t flags)
1678 int exclude_pid = 0;
1681 atomic_inc(&skb->users);
1685 /* errors reported via destination sk->sk_err */
1686 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1690 err = nlmsg_unicast(sk, skb, pid);
1694 EXPORT_SYMBOL(nlmsg_notify);
1696 #ifdef CONFIG_PROC_FS
1697 struct nl_seq_iter {
1698 struct seq_net_private p;
1703 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1705 struct nl_seq_iter *iter = seq->private;
1708 struct hlist_node *node;
1711 for (i = 0; i < MAX_LINKS; i++) {
1712 struct nl_pid_hash *hash = &nl_table[i].hash;
1714 for (j = 0; j <= hash->mask; j++) {
1715 sk_for_each(s, node, &hash->table[j]) {
1716 if (iter->p.net != s->sk_net)
1730 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1732 read_lock(&nl_table_lock);
1733 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1736 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1739 struct nl_seq_iter *iter;
1744 if (v == SEQ_START_TOKEN)
1745 return netlink_seq_socket_idx(seq, 0);
1747 iter = seq->private;
1751 } while (s && (iter->p.net != s->sk_net));
1756 j = iter->hash_idx + 1;
1759 struct nl_pid_hash *hash = &nl_table[i].hash;
1761 for (; j <= hash->mask; j++) {
1762 s = sk_head(&hash->table[j]);
1763 while (s && (iter->p.net != s->sk_net))
1773 } while (++i < MAX_LINKS);
1778 static void netlink_seq_stop(struct seq_file *seq, void *v)
1780 read_unlock(&nl_table_lock);
1784 static int netlink_seq_show(struct seq_file *seq, void *v)
1786 if (v == SEQ_START_TOKEN)
1788 "sk Eth Pid Groups "
1789 "Rmem Wmem Dump Locks\n");
1792 struct netlink_sock *nlk = nlk_sk(s);
1794 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1798 nlk->groups ? (u32)nlk->groups[0] : 0,
1799 atomic_read(&s->sk_rmem_alloc),
1800 atomic_read(&s->sk_wmem_alloc),
1802 atomic_read(&s->sk_refcnt)
1809 static const struct seq_operations netlink_seq_ops = {
1810 .start = netlink_seq_start,
1811 .next = netlink_seq_next,
1812 .stop = netlink_seq_stop,
1813 .show = netlink_seq_show,
1817 static int netlink_seq_open(struct inode *inode, struct file *file)
1819 return seq_open_net(inode, file, &netlink_seq_ops,
1820 sizeof(struct nl_seq_iter));
1823 static const struct file_operations netlink_seq_fops = {
1824 .owner = THIS_MODULE,
1825 .open = netlink_seq_open,
1827 .llseek = seq_lseek,
1828 .release = seq_release_net,
1833 int netlink_register_notifier(struct notifier_block *nb)
1835 return atomic_notifier_chain_register(&netlink_chain, nb);
1837 EXPORT_SYMBOL(netlink_register_notifier);
1839 int netlink_unregister_notifier(struct notifier_block *nb)
1841 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1843 EXPORT_SYMBOL(netlink_unregister_notifier);
1845 static const struct proto_ops netlink_ops = {
1846 .family = PF_NETLINK,
1847 .owner = THIS_MODULE,
1848 .release = netlink_release,
1849 .bind = netlink_bind,
1850 .connect = netlink_connect,
1851 .socketpair = sock_no_socketpair,
1852 .accept = sock_no_accept,
1853 .getname = netlink_getname,
1854 .poll = datagram_poll,
1855 .ioctl = sock_no_ioctl,
1856 .listen = sock_no_listen,
1857 .shutdown = sock_no_shutdown,
1858 .setsockopt = netlink_setsockopt,
1859 .getsockopt = netlink_getsockopt,
1860 .sendmsg = netlink_sendmsg,
1861 .recvmsg = netlink_recvmsg,
1862 .mmap = sock_no_mmap,
1863 .sendpage = sock_no_sendpage,
1866 static struct net_proto_family netlink_family_ops = {
1867 .family = PF_NETLINK,
1868 .create = netlink_create,
1869 .owner = THIS_MODULE, /* for consistency 8) */
1872 static int __net_init netlink_net_init(struct net *net)
1874 #ifdef CONFIG_PROC_FS
1875 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1881 static void __net_exit netlink_net_exit(struct net *net)
1883 #ifdef CONFIG_PROC_FS
1884 proc_net_remove(net, "netlink");
1888 static struct pernet_operations __net_initdata netlink_net_ops = {
1889 .init = netlink_net_init,
1890 .exit = netlink_net_exit,
1893 static int __init netlink_proto_init(void)
1895 struct sk_buff *dummy_skb;
1897 unsigned long limit;
1899 int err = proto_register(&netlink_proto, 0);
1904 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1906 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1910 if (num_physpages >= (128 * 1024))
1911 limit = num_physpages >> (21 - PAGE_SHIFT);
1913 limit = num_physpages >> (23 - PAGE_SHIFT);
1915 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1916 limit = (1UL << order) / sizeof(struct hlist_head);
1917 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1919 for (i = 0; i < MAX_LINKS; i++) {
1920 struct nl_pid_hash *hash = &nl_table[i].hash;
1922 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1925 nl_pid_hash_free(nl_table[i].hash.table,
1926 1 * sizeof(*hash->table));
1930 memset(hash->table, 0, 1 * sizeof(*hash->table));
1931 hash->max_shift = order;
1934 hash->rehash_time = jiffies;
1937 sock_register(&netlink_family_ops);
1938 register_pernet_subsys(&netlink_net_ops);
1939 /* The netlink device handler may be needed early. */
1944 panic("netlink_init: Cannot allocate nl_table\n");
1947 core_initcall(netlink_proto_init);