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/config.h>
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/smp_lock.h>
50 #include <linux/notifier.h>
51 #include <linux/security.h>
52 #include <linux/jhash.h>
53 #include <linux/jiffies.h>
54 #include <linux/random.h>
55 #include <linux/bitops.h>
57 #include <linux/types.h>
58 #include <linux/audit.h>
62 #include <net/netlink.h>
65 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
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;
81 void (*data_ready)(struct sock *sk, int bytes);
82 struct module *module;
85 #define NETLINK_KERNEL_SOCKET 0x1
86 #define NETLINK_RECV_PKTINFO 0x2
88 static inline struct netlink_sock *nlk_sk(struct sock *sk)
90 return (struct netlink_sock *)sk;
94 struct hlist_head *table;
95 unsigned long rehash_time;
100 unsigned int entries;
101 unsigned int max_shift;
106 struct netlink_table {
107 struct nl_pid_hash hash;
108 struct hlist_head mc_list;
109 unsigned long *listeners;
110 unsigned int nl_nonroot;
112 struct module *module;
116 static struct netlink_table *nl_table;
118 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
120 static int netlink_dump(struct sock *sk);
121 static void netlink_destroy_callback(struct netlink_callback *cb);
123 static DEFINE_RWLOCK(nl_table_lock);
124 static atomic_t nl_table_users = ATOMIC_INIT(0);
126 static struct notifier_block *netlink_chain;
128 static u32 netlink_group_mask(u32 group)
130 return group ? 1 << (group - 1) : 0;
133 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
135 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
138 static void netlink_sock_destruct(struct sock *sk)
140 skb_queue_purge(&sk->sk_receive_queue);
142 if (!sock_flag(sk, SOCK_DEAD)) {
143 printk("Freeing alive netlink socket %p\n", sk);
146 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
147 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
148 BUG_TRAP(!nlk_sk(sk)->cb);
149 BUG_TRAP(!nlk_sk(sk)->groups);
152 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
153 * Look, when several writers sleep and reader wakes them up, all but one
154 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
155 * this, _but_ remember, it adds useless work on UP machines.
158 static void netlink_table_grab(void)
160 write_lock_bh(&nl_table_lock);
162 if (atomic_read(&nl_table_users)) {
163 DECLARE_WAITQUEUE(wait, current);
165 add_wait_queue_exclusive(&nl_table_wait, &wait);
167 set_current_state(TASK_UNINTERRUPTIBLE);
168 if (atomic_read(&nl_table_users) == 0)
170 write_unlock_bh(&nl_table_lock);
172 write_lock_bh(&nl_table_lock);
175 __set_current_state(TASK_RUNNING);
176 remove_wait_queue(&nl_table_wait, &wait);
180 static __inline__ void netlink_table_ungrab(void)
182 write_unlock_bh(&nl_table_lock);
183 wake_up(&nl_table_wait);
186 static __inline__ void
187 netlink_lock_table(void)
189 /* read_lock() synchronizes us to netlink_table_grab */
191 read_lock(&nl_table_lock);
192 atomic_inc(&nl_table_users);
193 read_unlock(&nl_table_lock);
196 static __inline__ void
197 netlink_unlock_table(void)
199 if (atomic_dec_and_test(&nl_table_users))
200 wake_up(&nl_table_wait);
203 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
205 struct nl_pid_hash *hash = &nl_table[protocol].hash;
206 struct hlist_head *head;
208 struct hlist_node *node;
210 read_lock(&nl_table_lock);
211 head = nl_pid_hashfn(hash, pid);
212 sk_for_each(sk, node, head) {
213 if (nlk_sk(sk)->pid == pid) {
220 read_unlock(&nl_table_lock);
224 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
226 if (size <= PAGE_SIZE)
227 return kmalloc(size, GFP_ATOMIC);
229 return (struct hlist_head *)
230 __get_free_pages(GFP_ATOMIC, get_order(size));
233 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
235 if (size <= PAGE_SIZE)
238 free_pages((unsigned long)table, get_order(size));
241 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
243 unsigned int omask, mask, shift;
245 struct hlist_head *otable, *table;
248 omask = mask = hash->mask;
249 osize = size = (mask + 1) * sizeof(*table);
253 if (++shift > hash->max_shift)
259 table = nl_pid_hash_alloc(size);
263 memset(table, 0, size);
264 otable = hash->table;
268 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
270 for (i = 0; i <= omask; i++) {
272 struct hlist_node *node, *tmp;
274 sk_for_each_safe(sk, node, tmp, &otable[i])
275 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
278 nl_pid_hash_free(otable, osize);
279 hash->rehash_time = jiffies + 10 * 60 * HZ;
283 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
285 int avg = hash->entries >> hash->shift;
287 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
290 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
291 nl_pid_hash_rehash(hash, 0);
298 static const struct proto_ops netlink_ops;
301 netlink_update_listeners(struct sock *sk)
303 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
304 struct hlist_node *node;
308 for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) {
310 sk_for_each_bound(sk, node, &tbl->mc_list)
311 mask |= nlk_sk(sk)->groups[i];
312 tbl->listeners[i] = mask;
314 /* this function is only called with the netlink table "grabbed", which
315 * makes sure updates are visible before bind or setsockopt return. */
318 static int netlink_insert(struct sock *sk, u32 pid)
320 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
321 struct hlist_head *head;
322 int err = -EADDRINUSE;
324 struct hlist_node *node;
327 netlink_table_grab();
328 head = nl_pid_hashfn(hash, pid);
330 sk_for_each(osk, node, head) {
331 if (nlk_sk(osk)->pid == pid)
343 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
346 if (len && nl_pid_hash_dilute(hash, len))
347 head = nl_pid_hashfn(hash, pid);
349 nlk_sk(sk)->pid = pid;
350 sk_add_node(sk, head);
354 netlink_table_ungrab();
358 static void netlink_remove(struct sock *sk)
360 netlink_table_grab();
361 if (sk_del_node_init(sk))
362 nl_table[sk->sk_protocol].hash.entries--;
363 if (nlk_sk(sk)->subscriptions)
364 __sk_del_bind_node(sk);
365 netlink_table_ungrab();
368 static struct proto netlink_proto = {
370 .owner = THIS_MODULE,
371 .obj_size = sizeof(struct netlink_sock),
374 static int __netlink_create(struct socket *sock, int protocol)
377 struct netlink_sock *nlk;
379 sock->ops = &netlink_ops;
381 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
385 sock_init_data(sock, sk);
388 spin_lock_init(&nlk->cb_lock);
389 init_waitqueue_head(&nlk->wait);
391 sk->sk_destruct = netlink_sock_destruct;
392 sk->sk_protocol = protocol;
396 static int netlink_create(struct socket *sock, int protocol)
398 struct module *module = NULL;
399 struct netlink_sock *nlk;
403 sock->state = SS_UNCONNECTED;
405 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
406 return -ESOCKTNOSUPPORT;
408 if (protocol<0 || protocol >= MAX_LINKS)
409 return -EPROTONOSUPPORT;
411 netlink_lock_table();
413 if (!nl_table[protocol].registered) {
414 netlink_unlock_table();
415 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
416 netlink_lock_table();
419 if (nl_table[protocol].registered &&
420 try_module_get(nl_table[protocol].module))
421 module = nl_table[protocol].module;
422 groups = nl_table[protocol].groups;
423 netlink_unlock_table();
425 if ((err = __netlink_create(sock, protocol)) < 0)
428 nlk = nlk_sk(sock->sk);
429 nlk->module = module;
438 static int netlink_release(struct socket *sock)
440 struct sock *sk = sock->sk;
441 struct netlink_sock *nlk;
449 spin_lock(&nlk->cb_lock);
452 nlk->cb->done(nlk->cb);
453 netlink_destroy_callback(nlk->cb);
456 spin_unlock(&nlk->cb_lock);
458 /* OK. Socket is unlinked, and, therefore,
459 no new packets will arrive */
463 wake_up_interruptible_all(&nlk->wait);
465 skb_queue_purge(&sk->sk_write_queue);
467 if (nlk->pid && !nlk->subscriptions) {
468 struct netlink_notify n = {
469 .protocol = sk->sk_protocol,
472 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
476 module_put(nlk->module);
478 netlink_table_grab();
479 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
480 kfree(nl_table[sk->sk_protocol].listeners);
481 nl_table[sk->sk_protocol].module = NULL;
482 nl_table[sk->sk_protocol].registered = 0;
483 } else if (nlk->subscriptions)
484 netlink_update_listeners(sk);
485 netlink_table_ungrab();
494 static int netlink_autobind(struct socket *sock)
496 struct sock *sk = sock->sk;
497 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
498 struct hlist_head *head;
500 struct hlist_node *node;
501 s32 pid = current->tgid;
503 static s32 rover = -4097;
507 netlink_table_grab();
508 head = nl_pid_hashfn(hash, pid);
509 sk_for_each(osk, node, head) {
510 if (nlk_sk(osk)->pid == pid) {
511 /* Bind collision, search negative pid values. */
515 netlink_table_ungrab();
519 netlink_table_ungrab();
521 err = netlink_insert(sk, pid);
522 if (err == -EADDRINUSE)
525 /* If 2 threads race to autobind, that is fine. */
532 static inline int netlink_capable(struct socket *sock, unsigned int flag)
534 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
535 capable(CAP_NET_ADMIN);
539 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
541 struct netlink_sock *nlk = nlk_sk(sk);
543 if (nlk->subscriptions && !subscriptions)
544 __sk_del_bind_node(sk);
545 else if (!nlk->subscriptions && subscriptions)
546 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
547 nlk->subscriptions = subscriptions;
550 static int netlink_alloc_groups(struct sock *sk)
552 struct netlink_sock *nlk = nlk_sk(sk);
556 netlink_lock_table();
557 groups = nl_table[sk->sk_protocol].groups;
558 if (!nl_table[sk->sk_protocol].registered)
560 netlink_unlock_table();
565 nlk->groups = kmalloc(NLGRPSZ(groups), GFP_KERNEL);
566 if (nlk->groups == NULL)
568 memset(nlk->groups, 0, NLGRPSZ(groups));
569 nlk->ngroups = groups;
573 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
575 struct sock *sk = sock->sk;
576 struct netlink_sock *nlk = nlk_sk(sk);
577 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
580 if (nladdr->nl_family != AF_NETLINK)
583 /* Only superuser is allowed to listen multicasts */
584 if (nladdr->nl_groups) {
585 if (!netlink_capable(sock, NL_NONROOT_RECV))
587 if (nlk->groups == NULL) {
588 err = netlink_alloc_groups(sk);
595 if (nladdr->nl_pid != nlk->pid)
598 err = nladdr->nl_pid ?
599 netlink_insert(sk, nladdr->nl_pid) :
600 netlink_autobind(sock);
605 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
608 netlink_table_grab();
609 netlink_update_subscriptions(sk, nlk->subscriptions +
610 hweight32(nladdr->nl_groups) -
611 hweight32(nlk->groups[0]));
612 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
613 netlink_update_listeners(sk);
614 netlink_table_ungrab();
619 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
623 struct sock *sk = sock->sk;
624 struct netlink_sock *nlk = nlk_sk(sk);
625 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
627 if (addr->sa_family == AF_UNSPEC) {
628 sk->sk_state = NETLINK_UNCONNECTED;
633 if (addr->sa_family != AF_NETLINK)
636 /* Only superuser is allowed to send multicasts */
637 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
641 err = netlink_autobind(sock);
644 sk->sk_state = NETLINK_CONNECTED;
645 nlk->dst_pid = nladdr->nl_pid;
646 nlk->dst_group = ffs(nladdr->nl_groups);
652 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
654 struct sock *sk = sock->sk;
655 struct netlink_sock *nlk = nlk_sk(sk);
656 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
658 nladdr->nl_family = AF_NETLINK;
660 *addr_len = sizeof(*nladdr);
663 nladdr->nl_pid = nlk->dst_pid;
664 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
666 nladdr->nl_pid = nlk->pid;
667 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
672 static void netlink_overrun(struct sock *sk)
674 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
675 sk->sk_err = ENOBUFS;
676 sk->sk_error_report(sk);
680 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
682 int protocol = ssk->sk_protocol;
684 struct netlink_sock *nlk;
686 sock = netlink_lookup(protocol, pid);
688 return ERR_PTR(-ECONNREFUSED);
690 /* Don't bother queuing skb if kernel socket has no input function */
692 if ((nlk->pid == 0 && !nlk->data_ready) ||
693 (sock->sk_state == NETLINK_CONNECTED &&
694 nlk->dst_pid != nlk_sk(ssk)->pid)) {
696 return ERR_PTR(-ECONNREFUSED);
701 struct sock *netlink_getsockbyfilp(struct file *filp)
703 struct inode *inode = filp->f_dentry->d_inode;
706 if (!S_ISSOCK(inode->i_mode))
707 return ERR_PTR(-ENOTSOCK);
709 sock = SOCKET_I(inode)->sk;
710 if (sock->sk_family != AF_NETLINK)
711 return ERR_PTR(-EINVAL);
718 * Attach a skb to a netlink socket.
719 * The caller must hold a reference to the destination socket. On error, the
720 * reference is dropped. The skb is not send to the destination, just all
721 * all error checks are performed and memory in the queue is reserved.
723 * < 0: error. skb freed, reference to sock dropped.
725 * 1: repeat lookup - reference dropped while waiting for socket memory.
727 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
728 long timeo, struct sock *ssk)
730 struct netlink_sock *nlk;
734 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
735 test_bit(0, &nlk->state)) {
736 DECLARE_WAITQUEUE(wait, current);
738 if (!ssk || nlk_sk(ssk)->pid == 0)
745 __set_current_state(TASK_INTERRUPTIBLE);
746 add_wait_queue(&nlk->wait, &wait);
748 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
749 test_bit(0, &nlk->state)) &&
750 !sock_flag(sk, SOCK_DEAD))
751 timeo = schedule_timeout(timeo);
753 __set_current_state(TASK_RUNNING);
754 remove_wait_queue(&nlk->wait, &wait);
757 if (signal_pending(current)) {
759 return sock_intr_errno(timeo);
763 skb_set_owner_r(skb, sk);
767 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
771 skb_queue_tail(&sk->sk_receive_queue, skb);
772 sk->sk_data_ready(sk, len);
777 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
783 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
790 delta = skb->end - skb->tail;
791 if (delta * 2 < skb->truesize)
794 if (skb_shared(skb)) {
795 struct sk_buff *nskb = skb_clone(skb, allocation);
802 if (!pskb_expand_head(skb, 0, -delta, allocation))
803 skb->truesize -= delta;
808 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
814 skb = netlink_trim(skb, gfp_any());
816 timeo = sock_sndtimeo(ssk, nonblock);
818 sk = netlink_getsockbypid(ssk, pid);
823 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
829 return netlink_sendskb(sk, skb, ssk->sk_protocol);
832 int netlink_has_listeners(struct sock *sk, unsigned int group)
836 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
837 if (group - 1 < nl_table[sk->sk_protocol].groups)
838 res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners);
841 EXPORT_SYMBOL_GPL(netlink_has_listeners);
843 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
845 struct netlink_sock *nlk = nlk_sk(sk);
847 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
848 !test_bit(0, &nlk->state)) {
849 skb_set_owner_r(skb, sk);
850 skb_queue_tail(&sk->sk_receive_queue, skb);
851 sk->sk_data_ready(sk, skb->len);
852 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
857 struct netlink_broadcast_data {
858 struct sock *exclude_sk;
865 struct sk_buff *skb, *skb2;
868 static inline int do_one_broadcast(struct sock *sk,
869 struct netlink_broadcast_data *p)
871 struct netlink_sock *nlk = nlk_sk(sk);
874 if (p->exclude_sk == sk)
877 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
878 !test_bit(p->group - 1, nlk->groups))
887 if (p->skb2 == NULL) {
888 if (skb_shared(p->skb)) {
889 p->skb2 = skb_clone(p->skb, p->allocation);
891 p->skb2 = skb_get(p->skb);
893 * skb ownership may have been set when
894 * delivered to a previous socket.
899 if (p->skb2 == NULL) {
901 /* Clone failed. Notify ALL listeners. */
903 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
916 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
917 u32 group, gfp_t allocation)
919 struct netlink_broadcast_data info;
920 struct hlist_node *node;
923 skb = netlink_trim(skb, allocation);
925 info.exclude_sk = ssk;
931 info.allocation = allocation;
935 /* While we sleep in clone, do not allow to change socket list */
937 netlink_lock_table();
939 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
940 do_one_broadcast(sk, &info);
944 netlink_unlock_table();
947 kfree_skb(info.skb2);
949 if (info.delivered) {
950 if (info.congested && (allocation & __GFP_WAIT))
959 struct netlink_set_err_data {
960 struct sock *exclude_sk;
966 static inline int do_one_set_err(struct sock *sk,
967 struct netlink_set_err_data *p)
969 struct netlink_sock *nlk = nlk_sk(sk);
971 if (sk == p->exclude_sk)
974 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
975 !test_bit(p->group - 1, nlk->groups))
978 sk->sk_err = p->code;
979 sk->sk_error_report(sk);
984 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
986 struct netlink_set_err_data info;
987 struct hlist_node *node;
990 info.exclude_sk = ssk;
995 read_lock(&nl_table_lock);
997 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
998 do_one_set_err(sk, &info);
1000 read_unlock(&nl_table_lock);
1003 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1004 char __user *optval, int optlen)
1006 struct sock *sk = sock->sk;
1007 struct netlink_sock *nlk = nlk_sk(sk);
1010 if (level != SOL_NETLINK)
1011 return -ENOPROTOOPT;
1013 if (optlen >= sizeof(int) &&
1014 get_user(val, (int __user *)optval))
1018 case NETLINK_PKTINFO:
1020 nlk->flags |= NETLINK_RECV_PKTINFO;
1022 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1025 case NETLINK_ADD_MEMBERSHIP:
1026 case NETLINK_DROP_MEMBERSHIP: {
1027 unsigned int subscriptions;
1028 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1030 if (!netlink_capable(sock, NL_NONROOT_RECV))
1032 if (nlk->groups == NULL) {
1033 err = netlink_alloc_groups(sk);
1037 if (!val || val - 1 >= nlk->ngroups)
1039 netlink_table_grab();
1040 old = test_bit(val - 1, nlk->groups);
1041 subscriptions = nlk->subscriptions - old + new;
1043 __set_bit(val - 1, nlk->groups);
1045 __clear_bit(val - 1, nlk->groups);
1046 netlink_update_subscriptions(sk, subscriptions);
1047 netlink_update_listeners(sk);
1048 netlink_table_ungrab();
1058 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1059 char __user *optval, int __user *optlen)
1061 struct sock *sk = sock->sk;
1062 struct netlink_sock *nlk = nlk_sk(sk);
1065 if (level != SOL_NETLINK)
1066 return -ENOPROTOOPT;
1068 if (get_user(len, optlen))
1074 case NETLINK_PKTINFO:
1075 if (len < sizeof(int))
1078 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1079 put_user(len, optlen);
1080 put_user(val, optval);
1089 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1091 struct nl_pktinfo info;
1093 info.group = NETLINK_CB(skb).dst_group;
1094 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1097 static inline void netlink_rcv_wake(struct sock *sk)
1099 struct netlink_sock *nlk = nlk_sk(sk);
1101 if (skb_queue_empty(&sk->sk_receive_queue))
1102 clear_bit(0, &nlk->state);
1103 if (!test_bit(0, &nlk->state))
1104 wake_up_interruptible(&nlk->wait);
1107 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1108 struct msghdr *msg, size_t len)
1110 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1111 struct sock *sk = sock->sk;
1112 struct netlink_sock *nlk = nlk_sk(sk);
1113 struct sockaddr_nl *addr=msg->msg_name;
1116 struct sk_buff *skb;
1118 struct scm_cookie scm;
1120 if (msg->msg_flags&MSG_OOB)
1123 if (NULL == siocb->scm)
1125 err = scm_send(sock, msg, siocb->scm);
1129 if (msg->msg_namelen) {
1130 if (addr->nl_family != AF_NETLINK)
1132 dst_pid = addr->nl_pid;
1133 dst_group = ffs(addr->nl_groups);
1134 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1137 dst_pid = nlk->dst_pid;
1138 dst_group = nlk->dst_group;
1142 err = netlink_autobind(sock);
1148 if (len > sk->sk_sndbuf - 32)
1151 skb = alloc_skb(len, GFP_KERNEL);
1155 NETLINK_CB(skb).pid = nlk->pid;
1156 NETLINK_CB(skb).dst_pid = dst_pid;
1157 NETLINK_CB(skb).dst_group = dst_group;
1158 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1159 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1161 /* What can I do? Netlink is asynchronous, so that
1162 we will have to save current capabilities to
1163 check them, when this message will be delivered
1164 to corresponding kernel module. --ANK (980802)
1168 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1173 err = security_netlink_send(sk, skb);
1180 atomic_inc(&skb->users);
1181 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1183 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1189 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1190 struct msghdr *msg, size_t len,
1193 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1194 struct scm_cookie scm;
1195 struct sock *sk = sock->sk;
1196 struct netlink_sock *nlk = nlk_sk(sk);
1197 int noblock = flags&MSG_DONTWAIT;
1199 struct sk_buff *skb;
1207 skb = skb_recv_datagram(sk,flags,noblock,&err);
1211 msg->msg_namelen = 0;
1215 msg->msg_flags |= MSG_TRUNC;
1219 skb->h.raw = skb->data;
1220 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1222 if (msg->msg_name) {
1223 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1224 addr->nl_family = AF_NETLINK;
1226 addr->nl_pid = NETLINK_CB(skb).pid;
1227 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1228 msg->msg_namelen = sizeof(*addr);
1231 if (nlk->flags & NETLINK_RECV_PKTINFO)
1232 netlink_cmsg_recv_pktinfo(msg, skb);
1234 if (NULL == siocb->scm) {
1235 memset(&scm, 0, sizeof(scm));
1238 siocb->scm->creds = *NETLINK_CREDS(skb);
1239 skb_free_datagram(sk, skb);
1241 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1244 scm_recv(sock, msg, siocb->scm, flags);
1247 netlink_rcv_wake(sk);
1248 return err ? : copied;
1251 static void netlink_data_ready(struct sock *sk, int len)
1253 struct netlink_sock *nlk = nlk_sk(sk);
1255 if (nlk->data_ready)
1256 nlk->data_ready(sk, len);
1257 netlink_rcv_wake(sk);
1261 * We export these functions to other modules. They provide a
1262 * complete set of kernel non-blocking support for message
1267 netlink_kernel_create(int unit, unsigned int groups,
1268 void (*input)(struct sock *sk, int len),
1269 struct module *module)
1271 struct socket *sock;
1273 struct netlink_sock *nlk;
1274 unsigned long *listeners = NULL;
1279 if (unit<0 || unit>=MAX_LINKS)
1282 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1285 if (__netlink_create(sock, unit) < 0)
1286 goto out_sock_release;
1291 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1293 goto out_sock_release;
1296 sk->sk_data_ready = netlink_data_ready;
1298 nlk_sk(sk)->data_ready = input;
1300 if (netlink_insert(sk, 0))
1301 goto out_sock_release;
1304 nlk->flags |= NETLINK_KERNEL_SOCKET;
1306 netlink_table_grab();
1307 nl_table[unit].groups = groups;
1308 nl_table[unit].listeners = listeners;
1309 nl_table[unit].module = module;
1310 nl_table[unit].registered = 1;
1311 netlink_table_ungrab();
1321 void netlink_set_nonroot(int protocol, unsigned int flags)
1323 if ((unsigned int)protocol < MAX_LINKS)
1324 nl_table[protocol].nl_nonroot = flags;
1327 static void netlink_destroy_callback(struct netlink_callback *cb)
1335 * It looks a bit ugly.
1336 * It would be better to create kernel thread.
1339 static int netlink_dump(struct sock *sk)
1341 struct netlink_sock *nlk = nlk_sk(sk);
1342 struct netlink_callback *cb;
1343 struct sk_buff *skb;
1344 struct nlmsghdr *nlh;
1347 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1351 spin_lock(&nlk->cb_lock);
1355 spin_unlock(&nlk->cb_lock);
1360 len = cb->dump(skb, cb);
1363 spin_unlock(&nlk->cb_lock);
1364 skb_queue_tail(&sk->sk_receive_queue, skb);
1365 sk->sk_data_ready(sk, len);
1369 nlh = NLMSG_NEW_ANSWER(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1370 memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1371 skb_queue_tail(&sk->sk_receive_queue, skb);
1372 sk->sk_data_ready(sk, skb->len);
1377 spin_unlock(&nlk->cb_lock);
1379 netlink_destroy_callback(cb);
1386 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1387 struct nlmsghdr *nlh,
1388 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1389 int (*done)(struct netlink_callback*))
1391 struct netlink_callback *cb;
1393 struct netlink_sock *nlk;
1395 cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1399 memset(cb, 0, sizeof(*cb));
1403 atomic_inc(&skb->users);
1406 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1408 netlink_destroy_callback(cb);
1409 return -ECONNREFUSED;
1412 /* A dump is in progress... */
1413 spin_lock(&nlk->cb_lock);
1415 spin_unlock(&nlk->cb_lock);
1416 netlink_destroy_callback(cb);
1421 spin_unlock(&nlk->cb_lock);
1428 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1430 struct sk_buff *skb;
1431 struct nlmsghdr *rep;
1432 struct nlmsgerr *errmsg;
1436 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1438 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1440 skb = alloc_skb(size, GFP_KERNEL);
1444 sk = netlink_lookup(in_skb->sk->sk_protocol,
1445 NETLINK_CB(in_skb).pid);
1447 sk->sk_err = ENOBUFS;
1448 sk->sk_error_report(sk);
1454 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1455 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1456 errmsg = NLMSG_DATA(rep);
1457 errmsg->error = err;
1458 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1459 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1462 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1463 struct nlmsghdr *, int *))
1465 unsigned int total_len;
1466 struct nlmsghdr *nlh;
1469 while (skb->len >= nlmsg_total_size(0)) {
1470 nlh = (struct nlmsghdr *) skb->data;
1472 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1475 total_len = min(NLMSG_ALIGN(nlh->nlmsg_len), skb->len);
1477 if (cb(skb, nlh, &err) < 0) {
1478 /* Not an error, but we have to interrupt processing
1479 * here. Note: that in this case we do not pull
1480 * message from skb, it will be processed later.
1484 netlink_ack(skb, nlh, err);
1485 } else if (nlh->nlmsg_flags & NLM_F_ACK)
1486 netlink_ack(skb, nlh, 0);
1488 skb_pull(skb, total_len);
1495 * nelink_run_queue - Process netlink receive queue.
1496 * @sk: Netlink socket containing the queue
1497 * @qlen: Place to store queue length upon entry
1498 * @cb: Callback function invoked for each netlink message found
1500 * Processes as much as there was in the queue upon entry and invokes
1501 * a callback function for each netlink message found. The callback
1502 * function may refuse a message by returning a negative error code
1503 * but setting the error pointer to 0 in which case this function
1504 * returns with a qlen != 0.
1506 * qlen must be initialized to 0 before the initial entry, afterwards
1507 * the function may be called repeatedly until qlen reaches 0.
1509 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1510 int (*cb)(struct sk_buff *, struct nlmsghdr *, int *))
1512 struct sk_buff *skb;
1514 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1515 *qlen = skb_queue_len(&sk->sk_receive_queue);
1517 for (; *qlen; (*qlen)--) {
1518 skb = skb_dequeue(&sk->sk_receive_queue);
1519 if (netlink_rcv_skb(skb, cb)) {
1521 skb_queue_head(&sk->sk_receive_queue, skb);
1534 * netlink_queue_skip - Skip netlink message while processing queue.
1535 * @nlh: Netlink message to be skipped
1536 * @skb: Socket buffer containing the netlink messages.
1538 * Pulls the given netlink message off the socket buffer so the next
1539 * call to netlink_queue_run() will not reconsider the message.
1541 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1543 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1545 if (msglen > skb->len)
1548 skb_pull(skb, msglen);
1551 #ifdef CONFIG_PROC_FS
1552 struct nl_seq_iter {
1557 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1559 struct nl_seq_iter *iter = seq->private;
1562 struct hlist_node *node;
1565 for (i=0; i<MAX_LINKS; i++) {
1566 struct nl_pid_hash *hash = &nl_table[i].hash;
1568 for (j = 0; j <= hash->mask; j++) {
1569 sk_for_each(s, node, &hash->table[j]) {
1582 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1584 read_lock(&nl_table_lock);
1585 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1588 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1591 struct nl_seq_iter *iter;
1596 if (v == SEQ_START_TOKEN)
1597 return netlink_seq_socket_idx(seq, 0);
1603 iter = seq->private;
1605 j = iter->hash_idx + 1;
1608 struct nl_pid_hash *hash = &nl_table[i].hash;
1610 for (; j <= hash->mask; j++) {
1611 s = sk_head(&hash->table[j]);
1620 } while (++i < MAX_LINKS);
1625 static void netlink_seq_stop(struct seq_file *seq, void *v)
1627 read_unlock(&nl_table_lock);
1631 static int netlink_seq_show(struct seq_file *seq, void *v)
1633 if (v == SEQ_START_TOKEN)
1635 "sk Eth Pid Groups "
1636 "Rmem Wmem Dump Locks\n");
1639 struct netlink_sock *nlk = nlk_sk(s);
1641 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1645 nlk->groups ? (u32)nlk->groups[0] : 0,
1646 atomic_read(&s->sk_rmem_alloc),
1647 atomic_read(&s->sk_wmem_alloc),
1649 atomic_read(&s->sk_refcnt)
1656 static struct seq_operations netlink_seq_ops = {
1657 .start = netlink_seq_start,
1658 .next = netlink_seq_next,
1659 .stop = netlink_seq_stop,
1660 .show = netlink_seq_show,
1664 static int netlink_seq_open(struct inode *inode, struct file *file)
1666 struct seq_file *seq;
1667 struct nl_seq_iter *iter;
1670 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1674 err = seq_open(file, &netlink_seq_ops);
1680 memset(iter, 0, sizeof(*iter));
1681 seq = file->private_data;
1682 seq->private = iter;
1686 static struct file_operations netlink_seq_fops = {
1687 .owner = THIS_MODULE,
1688 .open = netlink_seq_open,
1690 .llseek = seq_lseek,
1691 .release = seq_release_private,
1696 int netlink_register_notifier(struct notifier_block *nb)
1698 return notifier_chain_register(&netlink_chain, nb);
1701 int netlink_unregister_notifier(struct notifier_block *nb)
1703 return notifier_chain_unregister(&netlink_chain, nb);
1706 static const struct proto_ops netlink_ops = {
1707 .family = PF_NETLINK,
1708 .owner = THIS_MODULE,
1709 .release = netlink_release,
1710 .bind = netlink_bind,
1711 .connect = netlink_connect,
1712 .socketpair = sock_no_socketpair,
1713 .accept = sock_no_accept,
1714 .getname = netlink_getname,
1715 .poll = datagram_poll,
1716 .ioctl = sock_no_ioctl,
1717 .listen = sock_no_listen,
1718 .shutdown = sock_no_shutdown,
1719 .setsockopt = netlink_setsockopt,
1720 .getsockopt = netlink_getsockopt,
1721 .sendmsg = netlink_sendmsg,
1722 .recvmsg = netlink_recvmsg,
1723 .mmap = sock_no_mmap,
1724 .sendpage = sock_no_sendpage,
1727 static struct net_proto_family netlink_family_ops = {
1728 .family = PF_NETLINK,
1729 .create = netlink_create,
1730 .owner = THIS_MODULE, /* for consistency 8) */
1733 extern void netlink_skb_parms_too_large(void);
1735 static int __init netlink_proto_init(void)
1737 struct sk_buff *dummy_skb;
1741 int err = proto_register(&netlink_proto, 0);
1746 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1747 netlink_skb_parms_too_large();
1749 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1752 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1756 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1758 if (num_physpages >= (128 * 1024))
1759 max = num_physpages >> (21 - PAGE_SHIFT);
1761 max = num_physpages >> (23 - PAGE_SHIFT);
1763 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1764 max = (1UL << order) / sizeof(struct hlist_head);
1765 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1767 for (i = 0; i < MAX_LINKS; i++) {
1768 struct nl_pid_hash *hash = &nl_table[i].hash;
1770 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1773 nl_pid_hash_free(nl_table[i].hash.table,
1774 1 * sizeof(*hash->table));
1778 memset(hash->table, 0, 1 * sizeof(*hash->table));
1779 hash->max_shift = order;
1782 hash->rehash_time = jiffies;
1785 sock_register(&netlink_family_ops);
1786 #ifdef CONFIG_PROC_FS
1787 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1789 /* The netlink device handler may be needed early. */
1795 core_initcall(netlink_proto_init);
1797 EXPORT_SYMBOL(netlink_ack);
1798 EXPORT_SYMBOL(netlink_run_queue);
1799 EXPORT_SYMBOL(netlink_queue_skip);
1800 EXPORT_SYMBOL(netlink_broadcast);
1801 EXPORT_SYMBOL(netlink_dump_start);
1802 EXPORT_SYMBOL(netlink_kernel_create);
1803 EXPORT_SYMBOL(netlink_register_notifier);
1804 EXPORT_SYMBOL(netlink_set_err);
1805 EXPORT_SYMBOL(netlink_set_nonroot);
1806 EXPORT_SYMBOL(netlink_unicast);
1807 EXPORT_SYMBOL(netlink_unregister_notifier);