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("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 SMP.
168 * 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);
201 static __inline__ void
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);
211 static __inline__ void
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, u32 pid)
220 struct nl_pid_hash *hash = &nl_table[protocol].hash;
221 struct hlist_head *head;
223 struct hlist_node *node;
225 read_lock(&nl_table_lock);
226 head = nl_pid_hashfn(hash, pid);
227 sk_for_each(sk, node, head) {
228 if ((sk->sk_net == net) && (nlk_sk(sk)->pid == pid)) {
235 read_unlock(&nl_table_lock);
239 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
241 if (size <= PAGE_SIZE)
242 return kmalloc(size, GFP_ATOMIC);
244 return (struct hlist_head *)
245 __get_free_pages(GFP_ATOMIC, get_order(size));
248 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
250 if (size <= PAGE_SIZE)
253 free_pages((unsigned long)table, get_order(size));
256 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
258 unsigned int omask, mask, shift;
260 struct hlist_head *otable, *table;
263 omask = mask = hash->mask;
264 osize = size = (mask + 1) * sizeof(*table);
268 if (++shift > hash->max_shift)
274 table = nl_pid_hash_alloc(size);
278 memset(table, 0, size);
279 otable = hash->table;
283 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
285 for (i = 0; i <= omask; i++) {
287 struct hlist_node *node, *tmp;
289 sk_for_each_safe(sk, node, tmp, &otable[i])
290 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
293 nl_pid_hash_free(otable, osize);
294 hash->rehash_time = jiffies + 10 * 60 * HZ;
298 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
300 int avg = hash->entries >> hash->shift;
302 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
305 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
306 nl_pid_hash_rehash(hash, 0);
313 static const struct proto_ops netlink_ops;
316 netlink_update_listeners(struct sock *sk)
318 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
319 struct hlist_node *node;
323 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
325 sk_for_each_bound(sk, node, &tbl->mc_list) {
326 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
327 mask |= nlk_sk(sk)->groups[i];
329 tbl->listeners[i] = mask;
331 /* this function is only called with the netlink table "grabbed", which
332 * makes sure updates are visible before bind or setsockopt return. */
335 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
337 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
338 struct hlist_head *head;
339 int err = -EADDRINUSE;
341 struct hlist_node *node;
344 netlink_table_grab();
345 head = nl_pid_hashfn(hash, pid);
347 sk_for_each(osk, node, head) {
348 if ((osk->sk_net == net) && (nlk_sk(osk)->pid == pid))
360 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
363 if (len && nl_pid_hash_dilute(hash, len))
364 head = nl_pid_hashfn(hash, pid);
366 nlk_sk(sk)->pid = pid;
367 sk_add_node(sk, head);
371 netlink_table_ungrab();
375 static void netlink_remove(struct sock *sk)
377 netlink_table_grab();
378 if (sk_del_node_init(sk))
379 nl_table[sk->sk_protocol].hash.entries--;
380 if (nlk_sk(sk)->subscriptions)
381 __sk_del_bind_node(sk);
382 netlink_table_ungrab();
385 static struct proto netlink_proto = {
387 .owner = THIS_MODULE,
388 .obj_size = sizeof(struct netlink_sock),
391 static int __netlink_create(struct net *net, struct socket *sock,
392 struct mutex *cb_mutex, int protocol)
395 struct netlink_sock *nlk;
397 sock->ops = &netlink_ops;
399 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
403 sock_init_data(sock, sk);
407 nlk->cb_mutex = cb_mutex;
409 nlk->cb_mutex = &nlk->cb_def_mutex;
410 mutex_init(nlk->cb_mutex);
412 init_waitqueue_head(&nlk->wait);
414 sk->sk_destruct = netlink_sock_destruct;
415 sk->sk_protocol = protocol;
419 static int netlink_create(struct net *net, struct socket *sock, int protocol)
421 struct module *module = NULL;
422 struct mutex *cb_mutex;
423 struct netlink_sock *nlk;
426 sock->state = SS_UNCONNECTED;
428 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
429 return -ESOCKTNOSUPPORT;
431 if (protocol<0 || protocol >= MAX_LINKS)
432 return -EPROTONOSUPPORT;
434 netlink_lock_table();
436 if (!nl_table[protocol].registered) {
437 netlink_unlock_table();
438 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
439 netlink_lock_table();
442 if (nl_table[protocol].registered &&
443 try_module_get(nl_table[protocol].module))
444 module = nl_table[protocol].module;
445 cb_mutex = nl_table[protocol].cb_mutex;
446 netlink_unlock_table();
448 if ((err = __netlink_create(net, sock, cb_mutex, protocol)) < 0)
451 nlk = nlk_sk(sock->sk);
452 nlk->module = module;
461 static int netlink_release(struct socket *sock)
463 struct sock *sk = sock->sk;
464 struct netlink_sock *nlk;
474 * OK. Socket is unlinked, any packets that arrive now
479 wake_up_interruptible_all(&nlk->wait);
481 skb_queue_purge(&sk->sk_write_queue);
483 if (nlk->pid && !nlk->subscriptions) {
484 struct netlink_notify n = {
486 .protocol = sk->sk_protocol,
489 atomic_notifier_call_chain(&netlink_chain,
490 NETLINK_URELEASE, &n);
493 module_put(nlk->module);
495 netlink_table_grab();
496 if (netlink_is_kernel(sk)) {
497 kfree(nl_table[sk->sk_protocol].listeners);
498 nl_table[sk->sk_protocol].module = NULL;
499 nl_table[sk->sk_protocol].registered = 0;
500 } else if (nlk->subscriptions)
501 netlink_update_listeners(sk);
502 netlink_table_ungrab();
511 static int netlink_autobind(struct socket *sock)
513 struct sock *sk = sock->sk;
514 struct net *net = sk->sk_net;
515 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
516 struct hlist_head *head;
518 struct hlist_node *node;
519 s32 pid = current->tgid;
521 static s32 rover = -4097;
525 netlink_table_grab();
526 head = nl_pid_hashfn(hash, pid);
527 sk_for_each(osk, node, head) {
528 if ((osk->sk_net != net))
530 if (nlk_sk(osk)->pid == pid) {
531 /* Bind collision, search negative pid values. */
535 netlink_table_ungrab();
539 netlink_table_ungrab();
541 err = netlink_insert(sk, net, pid);
542 if (err == -EADDRINUSE)
545 /* If 2 threads race to autobind, that is fine. */
552 static inline int netlink_capable(struct socket *sock, unsigned int flag)
554 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
555 capable(CAP_NET_ADMIN);
559 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
561 struct netlink_sock *nlk = nlk_sk(sk);
563 if (nlk->subscriptions && !subscriptions)
564 __sk_del_bind_node(sk);
565 else if (!nlk->subscriptions && subscriptions)
566 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
567 nlk->subscriptions = subscriptions;
570 static int netlink_realloc_groups(struct sock *sk)
572 struct netlink_sock *nlk = nlk_sk(sk);
574 unsigned long *new_groups;
577 netlink_table_grab();
579 groups = nl_table[sk->sk_protocol].groups;
580 if (!nl_table[sk->sk_protocol].registered) {
585 if (nlk->ngroups >= groups)
588 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
589 if (new_groups == NULL) {
593 memset((char*)new_groups + NLGRPSZ(nlk->ngroups), 0,
594 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
596 nlk->groups = new_groups;
597 nlk->ngroups = groups;
599 netlink_table_ungrab();
603 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
605 struct sock *sk = sock->sk;
606 struct net *net = sk->sk_net;
607 struct netlink_sock *nlk = nlk_sk(sk);
608 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
611 if (nladdr->nl_family != AF_NETLINK)
614 /* Only superuser is allowed to listen multicasts */
615 if (nladdr->nl_groups) {
616 if (!netlink_capable(sock, NL_NONROOT_RECV))
618 err = netlink_realloc_groups(sk);
624 if (nladdr->nl_pid != nlk->pid)
627 err = nladdr->nl_pid ?
628 netlink_insert(sk, net, nladdr->nl_pid) :
629 netlink_autobind(sock);
634 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
637 netlink_table_grab();
638 netlink_update_subscriptions(sk, nlk->subscriptions +
639 hweight32(nladdr->nl_groups) -
640 hweight32(nlk->groups[0]));
641 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
642 netlink_update_listeners(sk);
643 netlink_table_ungrab();
648 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
652 struct sock *sk = sock->sk;
653 struct netlink_sock *nlk = nlk_sk(sk);
654 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
656 if (addr->sa_family == AF_UNSPEC) {
657 sk->sk_state = NETLINK_UNCONNECTED;
662 if (addr->sa_family != AF_NETLINK)
665 /* Only superuser is allowed to send multicasts */
666 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
670 err = netlink_autobind(sock);
673 sk->sk_state = NETLINK_CONNECTED;
674 nlk->dst_pid = nladdr->nl_pid;
675 nlk->dst_group = ffs(nladdr->nl_groups);
681 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
683 struct sock *sk = sock->sk;
684 struct netlink_sock *nlk = nlk_sk(sk);
685 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
687 nladdr->nl_family = AF_NETLINK;
689 *addr_len = sizeof(*nladdr);
692 nladdr->nl_pid = nlk->dst_pid;
693 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
695 nladdr->nl_pid = nlk->pid;
696 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
701 static void netlink_overrun(struct sock *sk)
703 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
704 sk->sk_err = ENOBUFS;
705 sk->sk_error_report(sk);
709 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
712 struct netlink_sock *nlk;
714 sock = netlink_lookup(ssk->sk_net, ssk->sk_protocol, pid);
716 return ERR_PTR(-ECONNREFUSED);
718 /* Don't bother queuing skb if kernel socket has no input function */
720 if (sock->sk_state == NETLINK_CONNECTED &&
721 nlk->dst_pid != nlk_sk(ssk)->pid) {
723 return ERR_PTR(-ECONNREFUSED);
728 struct sock *netlink_getsockbyfilp(struct file *filp)
730 struct inode *inode = filp->f_path.dentry->d_inode;
733 if (!S_ISSOCK(inode->i_mode))
734 return ERR_PTR(-ENOTSOCK);
736 sock = SOCKET_I(inode)->sk;
737 if (sock->sk_family != AF_NETLINK)
738 return ERR_PTR(-EINVAL);
745 * Attach a skb to a netlink socket.
746 * The caller must hold a reference to the destination socket. On error, the
747 * reference is dropped. The skb is not send to the destination, just all
748 * all error checks are performed and memory in the queue is reserved.
750 * < 0: error. skb freed, reference to sock dropped.
752 * 1: repeat lookup - reference dropped while waiting for socket memory.
754 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
755 long timeo, struct sock *ssk)
757 struct netlink_sock *nlk;
761 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
762 test_bit(0, &nlk->state)) {
763 DECLARE_WAITQUEUE(wait, current);
765 if (!ssk || netlink_is_kernel(ssk))
772 __set_current_state(TASK_INTERRUPTIBLE);
773 add_wait_queue(&nlk->wait, &wait);
775 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
776 test_bit(0, &nlk->state)) &&
777 !sock_flag(sk, SOCK_DEAD))
778 timeo = schedule_timeout(timeo);
780 __set_current_state(TASK_RUNNING);
781 remove_wait_queue(&nlk->wait, &wait);
784 if (signal_pending(current)) {
786 return sock_intr_errno(timeo);
790 skb_set_owner_r(skb, sk);
794 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
798 skb_queue_tail(&sk->sk_receive_queue, skb);
799 sk->sk_data_ready(sk, len);
804 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
810 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
817 delta = skb->end - skb->tail;
818 if (delta * 2 < skb->truesize)
821 if (skb_shared(skb)) {
822 struct sk_buff *nskb = skb_clone(skb, allocation);
829 if (!pskb_expand_head(skb, 0, -delta, allocation))
830 skb->truesize -= delta;
835 static inline void netlink_rcv_wake(struct sock *sk)
837 struct netlink_sock *nlk = nlk_sk(sk);
839 if (skb_queue_empty(&sk->sk_receive_queue))
840 clear_bit(0, &nlk->state);
841 if (!test_bit(0, &nlk->state))
842 wake_up_interruptible(&nlk->wait);
845 static inline int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
848 struct netlink_sock *nlk = nlk_sk(sk);
851 if (nlk->netlink_rcv != NULL) {
853 skb_set_owner_r(skb, sk);
854 nlk->netlink_rcv(skb);
861 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
862 u32 pid, int nonblock)
868 skb = netlink_trim(skb, gfp_any());
870 timeo = sock_sndtimeo(ssk, nonblock);
872 sk = netlink_getsockbypid(ssk, pid);
877 if (netlink_is_kernel(sk))
878 return netlink_unicast_kernel(sk, skb);
880 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
886 return netlink_sendskb(sk, skb);
889 int netlink_has_listeners(struct sock *sk, unsigned int group)
892 unsigned long *listeners;
894 BUG_ON(!netlink_is_kernel(sk));
897 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
899 if (group - 1 < nl_table[sk->sk_protocol].groups)
900 res = test_bit(group - 1, listeners);
906 EXPORT_SYMBOL_GPL(netlink_has_listeners);
908 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
910 struct netlink_sock *nlk = nlk_sk(sk);
912 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
913 !test_bit(0, &nlk->state)) {
914 skb_set_owner_r(skb, sk);
915 skb_queue_tail(&sk->sk_receive_queue, skb);
916 sk->sk_data_ready(sk, skb->len);
917 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
922 struct netlink_broadcast_data {
923 struct sock *exclude_sk;
931 struct sk_buff *skb, *skb2;
934 static inline int do_one_broadcast(struct sock *sk,
935 struct netlink_broadcast_data *p)
937 struct netlink_sock *nlk = nlk_sk(sk);
940 if (p->exclude_sk == sk)
943 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
944 !test_bit(p->group - 1, nlk->groups))
947 if ((sk->sk_net != p->net))
956 if (p->skb2 == NULL) {
957 if (skb_shared(p->skb)) {
958 p->skb2 = skb_clone(p->skb, p->allocation);
960 p->skb2 = skb_get(p->skb);
962 * skb ownership may have been set when
963 * delivered to a previous socket.
968 if (p->skb2 == NULL) {
970 /* Clone failed. Notify ALL listeners. */
972 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
985 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
986 u32 group, gfp_t allocation)
988 struct net *net = ssk->sk_net;
989 struct netlink_broadcast_data info;
990 struct hlist_node *node;
993 skb = netlink_trim(skb, allocation);
995 info.exclude_sk = ssk;
1002 info.allocation = allocation;
1006 /* While we sleep in clone, do not allow to change socket list */
1008 netlink_lock_table();
1010 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1011 do_one_broadcast(sk, &info);
1015 netlink_unlock_table();
1018 kfree_skb(info.skb2);
1020 if (info.delivered) {
1021 if (info.congested && (allocation & __GFP_WAIT))
1030 struct netlink_set_err_data {
1031 struct sock *exclude_sk;
1037 static inline int do_one_set_err(struct sock *sk,
1038 struct netlink_set_err_data *p)
1040 struct netlink_sock *nlk = nlk_sk(sk);
1042 if (sk == p->exclude_sk)
1045 if (sk->sk_net != p->exclude_sk->sk_net)
1048 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1049 !test_bit(p->group - 1, nlk->groups))
1052 sk->sk_err = p->code;
1053 sk->sk_error_report(sk);
1058 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1060 struct netlink_set_err_data info;
1061 struct hlist_node *node;
1064 info.exclude_sk = ssk;
1069 read_lock(&nl_table_lock);
1071 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1072 do_one_set_err(sk, &info);
1074 read_unlock(&nl_table_lock);
1077 /* must be called with netlink table grabbed */
1078 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1082 int old, new = !!is_new, subscriptions;
1084 old = test_bit(group - 1, nlk->groups);
1085 subscriptions = nlk->subscriptions - old + new;
1087 __set_bit(group - 1, nlk->groups);
1089 __clear_bit(group - 1, nlk->groups);
1090 netlink_update_subscriptions(&nlk->sk, subscriptions);
1091 netlink_update_listeners(&nlk->sk);
1094 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1095 char __user *optval, int optlen)
1097 struct sock *sk = sock->sk;
1098 struct netlink_sock *nlk = nlk_sk(sk);
1099 unsigned int val = 0;
1102 if (level != SOL_NETLINK)
1103 return -ENOPROTOOPT;
1105 if (optlen >= sizeof(int) &&
1106 get_user(val, (unsigned int __user *)optval))
1110 case NETLINK_PKTINFO:
1112 nlk->flags |= NETLINK_RECV_PKTINFO;
1114 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1117 case NETLINK_ADD_MEMBERSHIP:
1118 case NETLINK_DROP_MEMBERSHIP: {
1119 if (!netlink_capable(sock, NL_NONROOT_RECV))
1121 err = netlink_realloc_groups(sk);
1124 if (!val || val - 1 >= nlk->ngroups)
1126 netlink_table_grab();
1127 netlink_update_socket_mc(nlk, val,
1128 optname == NETLINK_ADD_MEMBERSHIP);
1129 netlink_table_ungrab();
1139 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1140 char __user *optval, int __user *optlen)
1142 struct sock *sk = sock->sk;
1143 struct netlink_sock *nlk = nlk_sk(sk);
1146 if (level != SOL_NETLINK)
1147 return -ENOPROTOOPT;
1149 if (get_user(len, optlen))
1155 case NETLINK_PKTINFO:
1156 if (len < sizeof(int))
1159 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1160 if (put_user(len, optlen) ||
1161 put_user(val, optval))
1171 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1173 struct nl_pktinfo info;
1175 info.group = NETLINK_CB(skb).dst_group;
1176 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1179 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1180 struct msghdr *msg, size_t len)
1182 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1183 struct sock *sk = sock->sk;
1184 struct netlink_sock *nlk = nlk_sk(sk);
1185 struct sockaddr_nl *addr=msg->msg_name;
1188 struct sk_buff *skb;
1190 struct scm_cookie scm;
1192 if (msg->msg_flags&MSG_OOB)
1195 if (NULL == siocb->scm)
1197 err = scm_send(sock, msg, siocb->scm);
1201 if (msg->msg_namelen) {
1202 if (addr->nl_family != AF_NETLINK)
1204 dst_pid = addr->nl_pid;
1205 dst_group = ffs(addr->nl_groups);
1206 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1209 dst_pid = nlk->dst_pid;
1210 dst_group = nlk->dst_group;
1214 err = netlink_autobind(sock);
1220 if (len > sk->sk_sndbuf - 32)
1223 skb = alloc_skb(len, GFP_KERNEL);
1227 NETLINK_CB(skb).pid = nlk->pid;
1228 NETLINK_CB(skb).dst_group = dst_group;
1229 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1230 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1231 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1233 /* What can I do? Netlink is asynchronous, so that
1234 we will have to save current capabilities to
1235 check them, when this message will be delivered
1236 to corresponding kernel module. --ANK (980802)
1240 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1245 err = security_netlink_send(sk, skb);
1252 atomic_inc(&skb->users);
1253 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1255 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1261 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1262 struct msghdr *msg, size_t len,
1265 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1266 struct scm_cookie scm;
1267 struct sock *sk = sock->sk;
1268 struct netlink_sock *nlk = nlk_sk(sk);
1269 int noblock = flags&MSG_DONTWAIT;
1271 struct sk_buff *skb;
1279 skb = skb_recv_datagram(sk,flags,noblock,&err);
1283 msg->msg_namelen = 0;
1287 msg->msg_flags |= MSG_TRUNC;
1291 skb_reset_transport_header(skb);
1292 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1294 if (msg->msg_name) {
1295 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1296 addr->nl_family = AF_NETLINK;
1298 addr->nl_pid = NETLINK_CB(skb).pid;
1299 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1300 msg->msg_namelen = sizeof(*addr);
1303 if (nlk->flags & NETLINK_RECV_PKTINFO)
1304 netlink_cmsg_recv_pktinfo(msg, skb);
1306 if (NULL == siocb->scm) {
1307 memset(&scm, 0, sizeof(scm));
1310 siocb->scm->creds = *NETLINK_CREDS(skb);
1311 if (flags & MSG_TRUNC)
1313 skb_free_datagram(sk, skb);
1315 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1318 scm_recv(sock, msg, siocb->scm, flags);
1320 netlink_rcv_wake(sk);
1321 return err ? : copied;
1324 static void netlink_data_ready(struct sock *sk, int len)
1330 * We export these functions to other modules. They provide a
1331 * complete set of kernel non-blocking support for message
1336 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1337 void (*input)(struct sk_buff *skb),
1338 struct mutex *cb_mutex, struct module *module)
1340 struct socket *sock;
1342 struct netlink_sock *nlk;
1343 unsigned long *listeners = NULL;
1347 if (unit<0 || unit>=MAX_LINKS)
1350 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1353 if (__netlink_create(net, sock, cb_mutex, unit) < 0)
1354 goto out_sock_release;
1359 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1361 goto out_sock_release;
1364 sk->sk_data_ready = netlink_data_ready;
1366 nlk_sk(sk)->netlink_rcv = input;
1368 if (netlink_insert(sk, net, 0))
1369 goto out_sock_release;
1372 nlk->flags |= NETLINK_KERNEL_SOCKET;
1374 netlink_table_grab();
1375 if (!nl_table[unit].registered) {
1376 nl_table[unit].groups = groups;
1377 nl_table[unit].listeners = listeners;
1378 nl_table[unit].cb_mutex = cb_mutex;
1379 nl_table[unit].module = module;
1380 nl_table[unit].registered = 1;
1384 netlink_table_ungrab();
1395 * netlink_change_ngroups - change number of multicast groups
1397 * This changes the number of multicast groups that are available
1398 * on a certain netlink family. Note that it is not possible to
1399 * change the number of groups to below 32. Also note that it does
1400 * not implicitly call netlink_clear_multicast_users() when the
1401 * number of groups is reduced.
1403 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1404 * @groups: The new number of groups.
1406 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1408 unsigned long *listeners, *old = NULL;
1409 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1415 netlink_table_grab();
1416 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1417 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1422 old = tbl->listeners;
1423 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1424 rcu_assign_pointer(tbl->listeners, listeners);
1426 tbl->groups = groups;
1429 netlink_table_ungrab();
1434 EXPORT_SYMBOL(netlink_change_ngroups);
1437 * netlink_clear_multicast_users - kick off multicast listeners
1439 * This function removes all listeners from the given group.
1440 * @ksk: The kernel netlink socket, as returned by
1441 * netlink_kernel_create().
1442 * @group: The multicast group to clear.
1444 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1447 struct hlist_node *node;
1448 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1450 netlink_table_grab();
1452 sk_for_each_bound(sk, node, &tbl->mc_list)
1453 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1455 netlink_table_ungrab();
1457 EXPORT_SYMBOL(netlink_clear_multicast_users);
1459 void netlink_set_nonroot(int protocol, unsigned int flags)
1461 if ((unsigned int)protocol < MAX_LINKS)
1462 nl_table[protocol].nl_nonroot = flags;
1465 static void netlink_destroy_callback(struct netlink_callback *cb)
1473 * It looks a bit ugly.
1474 * It would be better to create kernel thread.
1477 static int netlink_dump(struct sock *sk)
1479 struct netlink_sock *nlk = nlk_sk(sk);
1480 struct netlink_callback *cb;
1481 struct sk_buff *skb;
1482 struct nlmsghdr *nlh;
1483 int len, err = -ENOBUFS;
1485 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1489 mutex_lock(nlk->cb_mutex);
1497 len = cb->dump(skb, cb);
1500 mutex_unlock(nlk->cb_mutex);
1501 skb_queue_tail(&sk->sk_receive_queue, skb);
1502 sk->sk_data_ready(sk, len);
1506 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1510 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1512 skb_queue_tail(&sk->sk_receive_queue, skb);
1513 sk->sk_data_ready(sk, skb->len);
1518 mutex_unlock(nlk->cb_mutex);
1520 netlink_destroy_callback(cb);
1524 mutex_unlock(nlk->cb_mutex);
1530 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1531 struct nlmsghdr *nlh,
1532 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1533 int (*done)(struct netlink_callback*))
1535 struct netlink_callback *cb;
1537 struct netlink_sock *nlk;
1539 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1546 atomic_inc(&skb->users);
1549 sk = netlink_lookup(ssk->sk_net, ssk->sk_protocol, NETLINK_CB(skb).pid);
1551 netlink_destroy_callback(cb);
1552 return -ECONNREFUSED;
1555 /* A dump is in progress... */
1556 mutex_lock(nlk->cb_mutex);
1558 mutex_unlock(nlk->cb_mutex);
1559 netlink_destroy_callback(cb);
1564 mutex_unlock(nlk->cb_mutex);
1569 /* We successfully started a dump, by returning -EINTR we
1570 * signal not to send ACK even if it was requested.
1575 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1577 struct sk_buff *skb;
1578 struct nlmsghdr *rep;
1579 struct nlmsgerr *errmsg;
1580 size_t payload = sizeof(*errmsg);
1582 /* error messages get the original request appened */
1584 payload += nlmsg_len(nlh);
1586 skb = nlmsg_new(payload, GFP_KERNEL);
1590 sk = netlink_lookup(in_skb->sk->sk_net,
1591 in_skb->sk->sk_protocol,
1592 NETLINK_CB(in_skb).pid);
1594 sk->sk_err = ENOBUFS;
1595 sk->sk_error_report(sk);
1601 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1602 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1603 errmsg = nlmsg_data(rep);
1604 errmsg->error = err;
1605 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1606 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1609 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1612 struct nlmsghdr *nlh;
1615 while (skb->len >= nlmsg_total_size(0)) {
1618 nlh = nlmsg_hdr(skb);
1621 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1624 /* Only requests are handled by the kernel */
1625 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1628 /* Skip control messages */
1629 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1637 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1638 netlink_ack(skb, nlh, err);
1641 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1642 if (msglen > skb->len)
1644 skb_pull(skb, msglen);
1651 * nlmsg_notify - send a notification netlink message
1652 * @sk: netlink socket to use
1653 * @skb: notification message
1654 * @pid: destination netlink pid for reports or 0
1655 * @group: destination multicast group or 0
1656 * @report: 1 to report back, 0 to disable
1657 * @flags: allocation flags
1659 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1660 unsigned int group, int report, gfp_t flags)
1665 int exclude_pid = 0;
1668 atomic_inc(&skb->users);
1672 /* errors reported via destination sk->sk_err */
1673 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1677 err = nlmsg_unicast(sk, skb, pid);
1682 #ifdef CONFIG_PROC_FS
1683 struct nl_seq_iter {
1689 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1691 struct nl_seq_iter *iter = seq->private;
1694 struct hlist_node *node;
1697 for (i=0; i<MAX_LINKS; i++) {
1698 struct nl_pid_hash *hash = &nl_table[i].hash;
1700 for (j = 0; j <= hash->mask; j++) {
1701 sk_for_each(s, node, &hash->table[j]) {
1702 if (iter->net != s->sk_net)
1716 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1718 read_lock(&nl_table_lock);
1719 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1722 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1725 struct nl_seq_iter *iter;
1730 if (v == SEQ_START_TOKEN)
1731 return netlink_seq_socket_idx(seq, 0);
1733 iter = seq->private;
1737 } while (s && (iter->net != s->sk_net));
1742 j = iter->hash_idx + 1;
1745 struct nl_pid_hash *hash = &nl_table[i].hash;
1747 for (; j <= hash->mask; j++) {
1748 s = sk_head(&hash->table[j]);
1749 while (s && (iter->net != s->sk_net))
1759 } while (++i < MAX_LINKS);
1764 static void netlink_seq_stop(struct seq_file *seq, void *v)
1766 read_unlock(&nl_table_lock);
1770 static int netlink_seq_show(struct seq_file *seq, void *v)
1772 if (v == SEQ_START_TOKEN)
1774 "sk Eth Pid Groups "
1775 "Rmem Wmem Dump Locks\n");
1778 struct netlink_sock *nlk = nlk_sk(s);
1780 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1784 nlk->groups ? (u32)nlk->groups[0] : 0,
1785 atomic_read(&s->sk_rmem_alloc),
1786 atomic_read(&s->sk_wmem_alloc),
1788 atomic_read(&s->sk_refcnt)
1795 static const struct seq_operations netlink_seq_ops = {
1796 .start = netlink_seq_start,
1797 .next = netlink_seq_next,
1798 .stop = netlink_seq_stop,
1799 .show = netlink_seq_show,
1803 static int netlink_seq_open(struct inode *inode, struct file *file)
1805 struct nl_seq_iter *iter;
1807 iter = __seq_open_private(file, &netlink_seq_ops, sizeof(*iter));
1811 iter->net = get_proc_net(inode);
1813 seq_release_private(inode, file);
1820 static int netlink_seq_release(struct inode *inode, struct file *file)
1822 struct seq_file *seq = file->private_data;
1823 struct nl_seq_iter *iter = seq->private;
1825 return seq_release_private(inode, file);
1828 static const struct file_operations netlink_seq_fops = {
1829 .owner = THIS_MODULE,
1830 .open = netlink_seq_open,
1832 .llseek = seq_lseek,
1833 .release = netlink_seq_release,
1838 int netlink_register_notifier(struct notifier_block *nb)
1840 return atomic_notifier_chain_register(&netlink_chain, nb);
1843 int netlink_unregister_notifier(struct notifier_block *nb)
1845 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1848 static const struct proto_ops netlink_ops = {
1849 .family = PF_NETLINK,
1850 .owner = THIS_MODULE,
1851 .release = netlink_release,
1852 .bind = netlink_bind,
1853 .connect = netlink_connect,
1854 .socketpair = sock_no_socketpair,
1855 .accept = sock_no_accept,
1856 .getname = netlink_getname,
1857 .poll = datagram_poll,
1858 .ioctl = sock_no_ioctl,
1859 .listen = sock_no_listen,
1860 .shutdown = sock_no_shutdown,
1861 .setsockopt = netlink_setsockopt,
1862 .getsockopt = netlink_getsockopt,
1863 .sendmsg = netlink_sendmsg,
1864 .recvmsg = netlink_recvmsg,
1865 .mmap = sock_no_mmap,
1866 .sendpage = sock_no_sendpage,
1869 static struct net_proto_family netlink_family_ops = {
1870 .family = PF_NETLINK,
1871 .create = netlink_create,
1872 .owner = THIS_MODULE, /* for consistency 8) */
1875 static int __net_init netlink_net_init(struct net *net)
1877 #ifdef CONFIG_PROC_FS
1878 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1884 static void __net_exit netlink_net_exit(struct net *net)
1886 #ifdef CONFIG_PROC_FS
1887 proc_net_remove(net, "netlink");
1891 static struct pernet_operations netlink_net_ops = {
1892 .init = netlink_net_init,
1893 .exit = netlink_net_exit,
1896 static int __init netlink_proto_init(void)
1898 struct sk_buff *dummy_skb;
1900 unsigned long limit;
1902 int err = proto_register(&netlink_proto, 0);
1907 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1909 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1913 if (num_physpages >= (128 * 1024))
1914 limit = num_physpages >> (21 - PAGE_SHIFT);
1916 limit = num_physpages >> (23 - PAGE_SHIFT);
1918 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
1919 limit = (1UL << order) / sizeof(struct hlist_head);
1920 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
1922 for (i = 0; i < MAX_LINKS; i++) {
1923 struct nl_pid_hash *hash = &nl_table[i].hash;
1925 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1928 nl_pid_hash_free(nl_table[i].hash.table,
1929 1 * sizeof(*hash->table));
1933 memset(hash->table, 0, 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);
1952 EXPORT_SYMBOL(netlink_ack);
1953 EXPORT_SYMBOL(netlink_rcv_skb);
1954 EXPORT_SYMBOL(netlink_broadcast);
1955 EXPORT_SYMBOL(netlink_dump_start);
1956 EXPORT_SYMBOL(netlink_kernel_create);
1957 EXPORT_SYMBOL(netlink_register_notifier);
1958 EXPORT_SYMBOL(netlink_set_nonroot);
1959 EXPORT_SYMBOL(netlink_unicast);
1960 EXPORT_SYMBOL(netlink_unregister_notifier);
1961 EXPORT_SYMBOL(nlmsg_notify);