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>
59 #include <linux/selinux.h>
60 #include <linux/vs_context.h>
61 #include <linux/vs_network.h>
62 #include <linux/vs_limit.h>
66 #include <net/netlink.h>
68 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
71 /* struct sock has to be the first member of netlink_sock */
79 unsigned long *groups;
81 wait_queue_head_t wait;
82 struct netlink_callback *cb;
84 void (*data_ready)(struct sock *sk, int bytes);
85 struct module *module;
88 #define NETLINK_KERNEL_SOCKET 0x1
89 #define NETLINK_RECV_PKTINFO 0x2
91 static inline struct netlink_sock *nlk_sk(struct sock *sk)
93 return (struct netlink_sock *)sk;
97 struct hlist_head *table;
98 unsigned long rehash_time;
103 unsigned int entries;
104 unsigned int max_shift;
109 struct netlink_table {
110 struct nl_pid_hash hash;
111 struct hlist_head mc_list;
112 unsigned long *listeners;
113 unsigned int nl_nonroot;
115 struct module *module;
119 static struct netlink_table *nl_table;
121 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
123 static int netlink_dump(struct sock *sk);
124 static void netlink_destroy_callback(struct netlink_callback *cb);
126 static DEFINE_RWLOCK(nl_table_lock);
127 static atomic_t nl_table_users = ATOMIC_INIT(0);
129 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
131 static u32 netlink_group_mask(u32 group)
133 return group ? 1 << (group - 1) : 0;
136 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
138 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
141 static void netlink_sock_destruct(struct sock *sk)
143 skb_queue_purge(&sk->sk_receive_queue);
145 if (!sock_flag(sk, SOCK_DEAD)) {
146 printk("Freeing alive netlink socket %p\n", sk);
149 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
150 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
151 BUG_TRAP(!nlk_sk(sk)->cb);
152 BUG_TRAP(!nlk_sk(sk)->groups);
155 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
156 * Look, when several writers sleep and reader wakes them up, all but one
157 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
158 * this, _but_ remember, it adds useless work on UP machines.
161 static void netlink_table_grab(void)
163 write_lock_bh(&nl_table_lock);
165 if (atomic_read(&nl_table_users)) {
166 DECLARE_WAITQUEUE(wait, current);
168 add_wait_queue_exclusive(&nl_table_wait, &wait);
170 set_current_state(TASK_UNINTERRUPTIBLE);
171 if (atomic_read(&nl_table_users) == 0)
173 write_unlock_bh(&nl_table_lock);
175 write_lock_bh(&nl_table_lock);
178 __set_current_state(TASK_RUNNING);
179 remove_wait_queue(&nl_table_wait, &wait);
183 static __inline__ void netlink_table_ungrab(void)
185 write_unlock_bh(&nl_table_lock);
186 wake_up(&nl_table_wait);
189 static __inline__ void
190 netlink_lock_table(void)
192 /* read_lock() synchronizes us to netlink_table_grab */
194 read_lock(&nl_table_lock);
195 atomic_inc(&nl_table_users);
196 read_unlock(&nl_table_lock);
199 static __inline__ void
200 netlink_unlock_table(void)
202 if (atomic_dec_and_test(&nl_table_users))
203 wake_up(&nl_table_wait);
206 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
208 struct nl_pid_hash *hash = &nl_table[protocol].hash;
209 struct hlist_head *head;
211 struct hlist_node *node;
213 read_lock(&nl_table_lock);
214 head = nl_pid_hashfn(hash, pid);
215 sk_for_each(sk, node, head) {
216 if (nlk_sk(sk)->pid == pid) {
223 read_unlock(&nl_table_lock);
227 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
229 if (size <= PAGE_SIZE)
230 return kmalloc(size, GFP_ATOMIC);
232 return (struct hlist_head *)
233 __get_free_pages(GFP_ATOMIC, get_order(size));
236 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
238 if (size <= PAGE_SIZE)
241 free_pages((unsigned long)table, get_order(size));
244 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
246 unsigned int omask, mask, shift;
248 struct hlist_head *otable, *table;
251 omask = mask = hash->mask;
252 osize = size = (mask + 1) * sizeof(*table);
256 if (++shift > hash->max_shift)
262 table = nl_pid_hash_alloc(size);
266 memset(table, 0, size);
267 otable = hash->table;
271 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
273 for (i = 0; i <= omask; i++) {
275 struct hlist_node *node, *tmp;
277 sk_for_each_safe(sk, node, tmp, &otable[i])
278 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
281 nl_pid_hash_free(otable, osize);
282 hash->rehash_time = jiffies + 10 * 60 * HZ;
286 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
288 int avg = hash->entries >> hash->shift;
290 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
293 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
294 nl_pid_hash_rehash(hash, 0);
301 static const struct proto_ops netlink_ops;
304 netlink_update_listeners(struct sock *sk)
306 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
307 struct hlist_node *node;
311 for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) {
313 sk_for_each_bound(sk, node, &tbl->mc_list)
314 mask |= nlk_sk(sk)->groups[i];
315 tbl->listeners[i] = mask;
317 /* this function is only called with the netlink table "grabbed", which
318 * makes sure updates are visible before bind or setsockopt return. */
321 static int netlink_insert(struct sock *sk, u32 pid)
323 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
324 struct hlist_head *head;
325 int err = -EADDRINUSE;
327 struct hlist_node *node;
330 netlink_table_grab();
331 head = nl_pid_hashfn(hash, pid);
333 sk_for_each(osk, node, head) {
334 if (nlk_sk(osk)->pid == pid)
346 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
349 if (len && nl_pid_hash_dilute(hash, len))
350 head = nl_pid_hashfn(hash, pid);
352 nlk_sk(sk)->pid = pid;
353 sk_add_node(sk, head);
357 netlink_table_ungrab();
361 static void netlink_remove(struct sock *sk)
363 netlink_table_grab();
364 if (sk_del_node_init(sk))
365 nl_table[sk->sk_protocol].hash.entries--;
366 if (nlk_sk(sk)->subscriptions)
367 __sk_del_bind_node(sk);
368 netlink_table_ungrab();
371 static struct proto netlink_proto = {
373 .owner = THIS_MODULE,
374 .obj_size = sizeof(struct netlink_sock),
377 static int __netlink_create(struct socket *sock, int protocol)
380 struct netlink_sock *nlk;
382 sock->ops = &netlink_ops;
384 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
388 sock_init_data(sock, sk);
391 spin_lock_init(&nlk->cb_lock);
392 init_waitqueue_head(&nlk->wait);
394 sk->sk_destruct = netlink_sock_destruct;
395 sk->sk_protocol = protocol;
399 static int netlink_create(struct socket *sock, int protocol)
401 struct module *module = NULL;
402 struct netlink_sock *nlk;
406 sock->state = SS_UNCONNECTED;
408 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
409 return -ESOCKTNOSUPPORT;
411 if (protocol<0 || protocol >= MAX_LINKS)
412 return -EPROTONOSUPPORT;
414 netlink_lock_table();
416 if (!nl_table[protocol].registered) {
417 netlink_unlock_table();
418 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
419 netlink_lock_table();
422 if (nl_table[protocol].registered &&
423 try_module_get(nl_table[protocol].module))
424 module = nl_table[protocol].module;
425 groups = nl_table[protocol].groups;
426 netlink_unlock_table();
428 if ((err = __netlink_create(sock, protocol)) < 0)
431 nlk = nlk_sk(sock->sk);
432 nlk->module = module;
441 static int netlink_release(struct socket *sock)
443 struct sock *sk = sock->sk;
444 struct netlink_sock *nlk;
452 spin_lock(&nlk->cb_lock);
455 nlk->cb->done(nlk->cb);
456 netlink_destroy_callback(nlk->cb);
459 spin_unlock(&nlk->cb_lock);
461 /* OK. Socket is unlinked, and, therefore,
462 no new packets will arrive */
466 wake_up_interruptible_all(&nlk->wait);
468 skb_queue_purge(&sk->sk_write_queue);
470 if (nlk->pid && !nlk->subscriptions) {
471 struct netlink_notify n = {
472 .protocol = sk->sk_protocol,
475 atomic_notifier_call_chain(&netlink_chain,
476 NETLINK_URELEASE, &n);
480 module_put(nlk->module);
482 netlink_table_grab();
483 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
484 kfree(nl_table[sk->sk_protocol].listeners);
485 nl_table[sk->sk_protocol].module = NULL;
486 nl_table[sk->sk_protocol].registered = 0;
487 } else if (nlk->subscriptions)
488 netlink_update_listeners(sk);
489 netlink_table_ungrab();
498 static int netlink_autobind(struct socket *sock)
500 struct sock *sk = sock->sk;
501 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
502 struct hlist_head *head;
504 struct hlist_node *node;
505 s32 pid = current->tgid;
507 static s32 rover = -4097;
511 netlink_table_grab();
512 head = nl_pid_hashfn(hash, pid);
513 sk_for_each(osk, node, head) {
514 if (nlk_sk(osk)->pid == pid) {
515 /* Bind collision, search negative pid values. */
519 netlink_table_ungrab();
523 netlink_table_ungrab();
525 err = netlink_insert(sk, pid);
526 if (err == -EADDRINUSE)
529 /* If 2 threads race to autobind, that is fine. */
536 static inline int netlink_capable(struct socket *sock, unsigned int flag)
538 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
539 capable(CAP_NET_ADMIN);
543 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
545 struct netlink_sock *nlk = nlk_sk(sk);
547 if (nlk->subscriptions && !subscriptions)
548 __sk_del_bind_node(sk);
549 else if (!nlk->subscriptions && subscriptions)
550 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
551 nlk->subscriptions = subscriptions;
554 static int netlink_alloc_groups(struct sock *sk)
556 struct netlink_sock *nlk = nlk_sk(sk);
560 netlink_lock_table();
561 groups = nl_table[sk->sk_protocol].groups;
562 if (!nl_table[sk->sk_protocol].registered)
564 netlink_unlock_table();
569 nlk->groups = kmalloc(NLGRPSZ(groups), GFP_KERNEL);
570 if (nlk->groups == NULL)
572 memset(nlk->groups, 0, NLGRPSZ(groups));
573 nlk->ngroups = groups;
577 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
579 struct sock *sk = sock->sk;
580 struct netlink_sock *nlk = nlk_sk(sk);
581 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
584 if (nladdr->nl_family != AF_NETLINK)
587 /* Only superuser is allowed to listen multicasts */
588 if (nladdr->nl_groups) {
589 if (!netlink_capable(sock, NL_NONROOT_RECV))
591 if (nlk->groups == NULL) {
592 err = netlink_alloc_groups(sk);
599 if (nladdr->nl_pid != nlk->pid)
602 err = nladdr->nl_pid ?
603 netlink_insert(sk, nladdr->nl_pid) :
604 netlink_autobind(sock);
609 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
612 netlink_table_grab();
613 netlink_update_subscriptions(sk, nlk->subscriptions +
614 hweight32(nladdr->nl_groups) -
615 hweight32(nlk->groups[0]));
616 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
617 netlink_update_listeners(sk);
618 netlink_table_ungrab();
623 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
627 struct sock *sk = sock->sk;
628 struct netlink_sock *nlk = nlk_sk(sk);
629 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
631 if (addr->sa_family == AF_UNSPEC) {
632 sk->sk_state = NETLINK_UNCONNECTED;
637 if (addr->sa_family != AF_NETLINK)
640 /* Only superuser is allowed to send multicasts */
641 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
645 err = netlink_autobind(sock);
648 sk->sk_state = NETLINK_CONNECTED;
649 nlk->dst_pid = nladdr->nl_pid;
650 nlk->dst_group = ffs(nladdr->nl_groups);
656 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
658 struct sock *sk = sock->sk;
659 struct netlink_sock *nlk = nlk_sk(sk);
660 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
662 nladdr->nl_family = AF_NETLINK;
664 *addr_len = sizeof(*nladdr);
667 nladdr->nl_pid = nlk->dst_pid;
668 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
670 nladdr->nl_pid = nlk->pid;
671 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
676 static void netlink_overrun(struct sock *sk)
678 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
679 sk->sk_err = ENOBUFS;
680 sk->sk_error_report(sk);
684 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
686 int protocol = ssk->sk_protocol;
688 struct netlink_sock *nlk;
690 sock = netlink_lookup(protocol, pid);
692 return ERR_PTR(-ECONNREFUSED);
694 /* Don't bother queuing skb if kernel socket has no input function */
696 if ((nlk->pid == 0 && !nlk->data_ready) ||
697 (sock->sk_state == NETLINK_CONNECTED &&
698 nlk->dst_pid != nlk_sk(ssk)->pid)) {
700 return ERR_PTR(-ECONNREFUSED);
705 struct sock *netlink_getsockbyfilp(struct file *filp)
707 struct inode *inode = filp->f_dentry->d_inode;
710 if (!S_ISSOCK(inode->i_mode))
711 return ERR_PTR(-ENOTSOCK);
713 sock = SOCKET_I(inode)->sk;
714 if (sock->sk_family != AF_NETLINK)
715 return ERR_PTR(-EINVAL);
722 * Attach a skb to a netlink socket.
723 * The caller must hold a reference to the destination socket. On error, the
724 * reference is dropped. The skb is not send to the destination, just all
725 * all error checks are performed and memory in the queue is reserved.
727 * < 0: error. skb freed, reference to sock dropped.
729 * 1: repeat lookup - reference dropped while waiting for socket memory.
731 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
732 long timeo, struct sock *ssk)
734 struct netlink_sock *nlk;
738 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
739 test_bit(0, &nlk->state)) {
740 DECLARE_WAITQUEUE(wait, current);
742 if (!ssk || nlk_sk(ssk)->pid == 0)
749 __set_current_state(TASK_INTERRUPTIBLE);
750 add_wait_queue(&nlk->wait, &wait);
752 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
753 test_bit(0, &nlk->state)) &&
754 !sock_flag(sk, SOCK_DEAD))
755 timeo = schedule_timeout(timeo);
757 __set_current_state(TASK_RUNNING);
758 remove_wait_queue(&nlk->wait, &wait);
761 if (signal_pending(current)) {
763 return sock_intr_errno(timeo);
767 skb_set_owner_r(skb, sk);
771 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
775 skb_queue_tail(&sk->sk_receive_queue, skb);
776 sk->sk_data_ready(sk, len);
781 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
787 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
794 delta = skb->end - skb->tail;
795 if (delta * 2 < skb->truesize)
798 if (skb_shared(skb)) {
799 struct sk_buff *nskb = skb_clone(skb, allocation);
806 if (!pskb_expand_head(skb, 0, -delta, allocation))
807 skb->truesize -= delta;
812 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
818 skb = netlink_trim(skb, gfp_any());
820 timeo = sock_sndtimeo(ssk, nonblock);
822 sk = netlink_getsockbypid(ssk, pid);
827 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
833 return netlink_sendskb(sk, skb, ssk->sk_protocol);
836 int netlink_has_listeners(struct sock *sk, unsigned int group)
840 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
841 if (group - 1 < nl_table[sk->sk_protocol].groups)
842 res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners);
845 EXPORT_SYMBOL_GPL(netlink_has_listeners);
847 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
849 struct netlink_sock *nlk = nlk_sk(sk);
851 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
852 !test_bit(0, &nlk->state)) {
853 skb_set_owner_r(skb, sk);
854 skb_queue_tail(&sk->sk_receive_queue, skb);
855 sk->sk_data_ready(sk, skb->len);
856 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
861 struct netlink_broadcast_data {
862 struct sock *exclude_sk;
869 struct sk_buff *skb, *skb2;
872 static inline int do_one_broadcast(struct sock *sk,
873 struct netlink_broadcast_data *p)
875 struct netlink_sock *nlk = nlk_sk(sk);
878 if (p->exclude_sk == sk)
881 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
882 !test_bit(p->group - 1, nlk->groups))
891 if (p->skb2 == NULL) {
892 if (skb_shared(p->skb)) {
893 p->skb2 = skb_clone(p->skb, p->allocation);
895 p->skb2 = skb_get(p->skb);
897 * skb ownership may have been set when
898 * delivered to a previous socket.
903 if (p->skb2 == NULL) {
905 /* Clone failed. Notify ALL listeners. */
907 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
920 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
921 u32 group, gfp_t allocation)
923 struct netlink_broadcast_data info;
924 struct hlist_node *node;
927 skb = netlink_trim(skb, allocation);
929 info.exclude_sk = ssk;
935 info.allocation = allocation;
939 /* While we sleep in clone, do not allow to change socket list */
941 netlink_lock_table();
943 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
944 do_one_broadcast(sk, &info);
948 netlink_unlock_table();
951 kfree_skb(info.skb2);
953 if (info.delivered) {
954 if (info.congested && (allocation & __GFP_WAIT))
963 struct netlink_set_err_data {
964 struct sock *exclude_sk;
970 static inline int do_one_set_err(struct sock *sk,
971 struct netlink_set_err_data *p)
973 struct netlink_sock *nlk = nlk_sk(sk);
975 if (sk == p->exclude_sk)
978 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
979 !test_bit(p->group - 1, nlk->groups))
982 sk->sk_err = p->code;
983 sk->sk_error_report(sk);
988 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
990 struct netlink_set_err_data info;
991 struct hlist_node *node;
994 info.exclude_sk = ssk;
999 read_lock(&nl_table_lock);
1001 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1002 do_one_set_err(sk, &info);
1004 read_unlock(&nl_table_lock);
1007 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1008 char __user *optval, int optlen)
1010 struct sock *sk = sock->sk;
1011 struct netlink_sock *nlk = nlk_sk(sk);
1014 if (level != SOL_NETLINK)
1015 return -ENOPROTOOPT;
1017 if (optlen >= sizeof(int) &&
1018 get_user(val, (int __user *)optval))
1022 case NETLINK_PKTINFO:
1024 nlk->flags |= NETLINK_RECV_PKTINFO;
1026 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1029 case NETLINK_ADD_MEMBERSHIP:
1030 case NETLINK_DROP_MEMBERSHIP: {
1031 unsigned int subscriptions;
1032 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1034 if (!netlink_capable(sock, NL_NONROOT_RECV))
1036 if (nlk->groups == NULL) {
1037 err = netlink_alloc_groups(sk);
1041 if (!val || val - 1 >= nlk->ngroups)
1043 netlink_table_grab();
1044 old = test_bit(val - 1, nlk->groups);
1045 subscriptions = nlk->subscriptions - old + new;
1047 __set_bit(val - 1, nlk->groups);
1049 __clear_bit(val - 1, nlk->groups);
1050 netlink_update_subscriptions(sk, subscriptions);
1051 netlink_update_listeners(sk);
1052 netlink_table_ungrab();
1062 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1063 char __user *optval, int __user *optlen)
1065 struct sock *sk = sock->sk;
1066 struct netlink_sock *nlk = nlk_sk(sk);
1069 if (level != SOL_NETLINK)
1070 return -ENOPROTOOPT;
1072 if (get_user(len, optlen))
1078 case NETLINK_PKTINFO:
1079 if (len < sizeof(int))
1082 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1083 put_user(len, optlen);
1084 put_user(val, optval);
1093 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1095 struct nl_pktinfo info;
1097 info.group = NETLINK_CB(skb).dst_group;
1098 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1101 static inline void netlink_rcv_wake(struct sock *sk)
1103 struct netlink_sock *nlk = nlk_sk(sk);
1105 if (skb_queue_empty(&sk->sk_receive_queue))
1106 clear_bit(0, &nlk->state);
1107 if (!test_bit(0, &nlk->state))
1108 wake_up_interruptible(&nlk->wait);
1111 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1112 struct msghdr *msg, size_t len)
1114 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1115 struct sock *sk = sock->sk;
1116 struct netlink_sock *nlk = nlk_sk(sk);
1117 struct sockaddr_nl *addr=msg->msg_name;
1120 struct sk_buff *skb;
1122 struct scm_cookie scm;
1124 if (msg->msg_flags&MSG_OOB)
1127 if (NULL == siocb->scm)
1129 err = scm_send(sock, msg, siocb->scm);
1133 if (msg->msg_namelen) {
1134 if (addr->nl_family != AF_NETLINK)
1136 dst_pid = addr->nl_pid;
1137 dst_group = ffs(addr->nl_groups);
1138 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1141 dst_pid = nlk->dst_pid;
1142 dst_group = nlk->dst_group;
1146 err = netlink_autobind(sock);
1152 if (len > sk->sk_sndbuf - 32)
1155 skb = alloc_skb(len, GFP_KERNEL);
1159 NETLINK_CB(skb).pid = nlk->pid;
1160 NETLINK_CB(skb).dst_pid = dst_pid;
1161 NETLINK_CB(skb).dst_group = dst_group;
1162 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1163 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1164 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1166 /* What can I do? Netlink is asynchronous, so that
1167 we will have to save current capabilities to
1168 check them, when this message will be delivered
1169 to corresponding kernel module. --ANK (980802)
1173 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1178 err = security_netlink_send(sk, skb);
1185 atomic_inc(&skb->users);
1186 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1188 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1194 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1195 struct msghdr *msg, size_t len,
1198 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1199 struct scm_cookie scm;
1200 struct sock *sk = sock->sk;
1201 struct netlink_sock *nlk = nlk_sk(sk);
1202 int noblock = flags&MSG_DONTWAIT;
1204 struct sk_buff *skb;
1212 skb = skb_recv_datagram(sk,flags,noblock,&err);
1216 msg->msg_namelen = 0;
1220 msg->msg_flags |= MSG_TRUNC;
1224 skb->h.raw = skb->data;
1225 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1227 if (msg->msg_name) {
1228 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1229 addr->nl_family = AF_NETLINK;
1231 addr->nl_pid = NETLINK_CB(skb).pid;
1232 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1233 msg->msg_namelen = sizeof(*addr);
1236 if (nlk->flags & NETLINK_RECV_PKTINFO)
1237 netlink_cmsg_recv_pktinfo(msg, skb);
1239 if (NULL == siocb->scm) {
1240 memset(&scm, 0, sizeof(scm));
1243 siocb->scm->creds = *NETLINK_CREDS(skb);
1244 skb_free_datagram(sk, skb);
1246 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1249 scm_recv(sock, msg, siocb->scm, flags);
1252 netlink_rcv_wake(sk);
1253 return err ? : copied;
1256 static void netlink_data_ready(struct sock *sk, int len)
1258 struct netlink_sock *nlk = nlk_sk(sk);
1260 if (nlk->data_ready)
1261 nlk->data_ready(sk, len);
1262 netlink_rcv_wake(sk);
1266 * We export these functions to other modules. They provide a
1267 * complete set of kernel non-blocking support for message
1272 netlink_kernel_create(int unit, unsigned int groups,
1273 void (*input)(struct sock *sk, int len),
1274 struct module *module)
1276 struct socket *sock;
1278 struct netlink_sock *nlk;
1279 unsigned long *listeners = NULL;
1284 if (unit<0 || unit>=MAX_LINKS)
1287 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1290 if (__netlink_create(sock, unit) < 0)
1291 goto out_sock_release;
1296 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1298 goto out_sock_release;
1301 sk->sk_data_ready = netlink_data_ready;
1303 nlk_sk(sk)->data_ready = input;
1305 if (netlink_insert(sk, 0))
1306 goto out_sock_release;
1309 nlk->flags |= NETLINK_KERNEL_SOCKET;
1311 netlink_table_grab();
1312 nl_table[unit].groups = groups;
1313 nl_table[unit].listeners = listeners;
1314 nl_table[unit].module = module;
1315 nl_table[unit].registered = 1;
1316 netlink_table_ungrab();
1326 void netlink_set_nonroot(int protocol, unsigned int flags)
1328 if ((unsigned int)protocol < MAX_LINKS)
1329 nl_table[protocol].nl_nonroot = flags;
1332 static void netlink_destroy_callback(struct netlink_callback *cb)
1340 * It looks a bit ugly.
1341 * It would be better to create kernel thread.
1344 static int netlink_dump(struct sock *sk)
1346 struct netlink_sock *nlk = nlk_sk(sk);
1347 struct netlink_callback *cb;
1348 struct sk_buff *skb;
1349 struct nlmsghdr *nlh;
1352 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1356 spin_lock(&nlk->cb_lock);
1360 spin_unlock(&nlk->cb_lock);
1365 len = cb->dump(skb, cb);
1368 spin_unlock(&nlk->cb_lock);
1369 skb_queue_tail(&sk->sk_receive_queue, skb);
1370 sk->sk_data_ready(sk, len);
1374 nlh = NLMSG_NEW_ANSWER(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1375 memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1376 skb_queue_tail(&sk->sk_receive_queue, skb);
1377 sk->sk_data_ready(sk, skb->len);
1382 spin_unlock(&nlk->cb_lock);
1384 netlink_destroy_callback(cb);
1391 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1392 struct nlmsghdr *nlh,
1393 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1394 int (*done)(struct netlink_callback*))
1396 struct netlink_callback *cb;
1398 struct netlink_sock *nlk;
1400 cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1404 memset(cb, 0, sizeof(*cb));
1408 atomic_inc(&skb->users);
1411 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1413 netlink_destroy_callback(cb);
1414 return -ECONNREFUSED;
1417 /* A dump is in progress... */
1418 spin_lock(&nlk->cb_lock);
1420 spin_unlock(&nlk->cb_lock);
1421 netlink_destroy_callback(cb);
1426 spin_unlock(&nlk->cb_lock);
1433 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1435 struct sk_buff *skb;
1436 struct nlmsghdr *rep;
1437 struct nlmsgerr *errmsg;
1441 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1443 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1445 skb = alloc_skb(size, GFP_KERNEL);
1449 sk = netlink_lookup(in_skb->sk->sk_protocol,
1450 NETLINK_CB(in_skb).pid);
1452 sk->sk_err = ENOBUFS;
1453 sk->sk_error_report(sk);
1459 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1460 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1461 errmsg = NLMSG_DATA(rep);
1462 errmsg->error = err;
1463 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1464 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1467 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1468 struct nlmsghdr *, int *))
1470 unsigned int total_len;
1471 struct nlmsghdr *nlh;
1474 while (skb->len >= nlmsg_total_size(0)) {
1475 nlh = (struct nlmsghdr *) skb->data;
1477 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1480 total_len = min(NLMSG_ALIGN(nlh->nlmsg_len), skb->len);
1482 if (cb(skb, nlh, &err) < 0) {
1483 /* Not an error, but we have to interrupt processing
1484 * here. Note: that in this case we do not pull
1485 * message from skb, it will be processed later.
1489 netlink_ack(skb, nlh, err);
1490 } else if (nlh->nlmsg_flags & NLM_F_ACK)
1491 netlink_ack(skb, nlh, 0);
1493 skb_pull(skb, total_len);
1500 * nelink_run_queue - Process netlink receive queue.
1501 * @sk: Netlink socket containing the queue
1502 * @qlen: Place to store queue length upon entry
1503 * @cb: Callback function invoked for each netlink message found
1505 * Processes as much as there was in the queue upon entry and invokes
1506 * a callback function for each netlink message found. The callback
1507 * function may refuse a message by returning a negative error code
1508 * but setting the error pointer to 0 in which case this function
1509 * returns with a qlen != 0.
1511 * qlen must be initialized to 0 before the initial entry, afterwards
1512 * the function may be called repeatedly until qlen reaches 0.
1514 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1515 int (*cb)(struct sk_buff *, struct nlmsghdr *, int *))
1517 struct sk_buff *skb;
1519 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1520 *qlen = skb_queue_len(&sk->sk_receive_queue);
1522 for (; *qlen; (*qlen)--) {
1523 skb = skb_dequeue(&sk->sk_receive_queue);
1524 if (netlink_rcv_skb(skb, cb)) {
1526 skb_queue_head(&sk->sk_receive_queue, skb);
1539 * netlink_queue_skip - Skip netlink message while processing queue.
1540 * @nlh: Netlink message to be skipped
1541 * @skb: Socket buffer containing the netlink messages.
1543 * Pulls the given netlink message off the socket buffer so the next
1544 * call to netlink_queue_run() will not reconsider the message.
1546 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1548 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1550 if (msglen > skb->len)
1553 skb_pull(skb, msglen);
1556 #ifdef CONFIG_PROC_FS
1557 struct nl_seq_iter {
1562 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1564 struct nl_seq_iter *iter = seq->private;
1567 struct hlist_node *node;
1570 for (i=0; i<MAX_LINKS; i++) {
1571 struct nl_pid_hash *hash = &nl_table[i].hash;
1573 for (j = 0; j <= hash->mask; j++) {
1574 sk_for_each(s, node, &hash->table[j]) {
1587 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1589 read_lock(&nl_table_lock);
1590 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1593 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1596 struct nl_seq_iter *iter;
1601 if (v == SEQ_START_TOKEN)
1602 return netlink_seq_socket_idx(seq, 0);
1608 iter = seq->private;
1610 j = iter->hash_idx + 1;
1613 struct nl_pid_hash *hash = &nl_table[i].hash;
1615 for (; j <= hash->mask; j++) {
1616 s = sk_head(&hash->table[j]);
1625 } while (++i < MAX_LINKS);
1630 static void netlink_seq_stop(struct seq_file *seq, void *v)
1632 read_unlock(&nl_table_lock);
1636 static int netlink_seq_show(struct seq_file *seq, void *v)
1638 if (v == SEQ_START_TOKEN)
1640 "sk Eth Pid Groups "
1641 "Rmem Wmem Dump Locks\n");
1644 struct netlink_sock *nlk = nlk_sk(s);
1646 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1650 nlk->groups ? (u32)nlk->groups[0] : 0,
1651 atomic_read(&s->sk_rmem_alloc),
1652 atomic_read(&s->sk_wmem_alloc),
1654 atomic_read(&s->sk_refcnt)
1661 static struct seq_operations netlink_seq_ops = {
1662 .start = netlink_seq_start,
1663 .next = netlink_seq_next,
1664 .stop = netlink_seq_stop,
1665 .show = netlink_seq_show,
1669 static int netlink_seq_open(struct inode *inode, struct file *file)
1671 struct seq_file *seq;
1672 struct nl_seq_iter *iter;
1675 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1679 err = seq_open(file, &netlink_seq_ops);
1685 memset(iter, 0, sizeof(*iter));
1686 seq = file->private_data;
1687 seq->private = iter;
1691 static struct file_operations netlink_seq_fops = {
1692 .owner = THIS_MODULE,
1693 .open = netlink_seq_open,
1695 .llseek = seq_lseek,
1696 .release = seq_release_private,
1701 int netlink_register_notifier(struct notifier_block *nb)
1703 return atomic_notifier_chain_register(&netlink_chain, nb);
1706 int netlink_unregister_notifier(struct notifier_block *nb)
1708 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1711 static const struct proto_ops netlink_ops = {
1712 .family = PF_NETLINK,
1713 .owner = THIS_MODULE,
1714 .release = netlink_release,
1715 .bind = netlink_bind,
1716 .connect = netlink_connect,
1717 .socketpair = sock_no_socketpair,
1718 .accept = sock_no_accept,
1719 .getname = netlink_getname,
1720 .poll = datagram_poll,
1721 .ioctl = sock_no_ioctl,
1722 .listen = sock_no_listen,
1723 .shutdown = sock_no_shutdown,
1724 .setsockopt = netlink_setsockopt,
1725 .getsockopt = netlink_getsockopt,
1726 .sendmsg = netlink_sendmsg,
1727 .recvmsg = netlink_recvmsg,
1728 .mmap = sock_no_mmap,
1729 .sendpage = sock_no_sendpage,
1732 static struct net_proto_family netlink_family_ops = {
1733 .family = PF_NETLINK,
1734 .create = netlink_create,
1735 .owner = THIS_MODULE, /* for consistency 8) */
1738 extern void netlink_skb_parms_too_large(void);
1740 static int __init netlink_proto_init(void)
1742 struct sk_buff *dummy_skb;
1746 int err = proto_register(&netlink_proto, 0);
1751 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1752 netlink_skb_parms_too_large();
1754 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1757 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1761 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1763 if (num_physpages >= (128 * 1024))
1764 max = num_physpages >> (21 - PAGE_SHIFT);
1766 max = num_physpages >> (23 - PAGE_SHIFT);
1768 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1769 max = (1UL << order) / sizeof(struct hlist_head);
1770 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1772 for (i = 0; i < MAX_LINKS; i++) {
1773 struct nl_pid_hash *hash = &nl_table[i].hash;
1775 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1778 nl_pid_hash_free(nl_table[i].hash.table,
1779 1 * sizeof(*hash->table));
1783 memset(hash->table, 0, 1 * sizeof(*hash->table));
1784 hash->max_shift = order;
1787 hash->rehash_time = jiffies;
1790 sock_register(&netlink_family_ops);
1791 #ifdef CONFIG_PROC_FS
1792 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1794 /* The netlink device handler may be needed early. */
1800 core_initcall(netlink_proto_init);
1802 EXPORT_SYMBOL(netlink_ack);
1803 EXPORT_SYMBOL(netlink_run_queue);
1804 EXPORT_SYMBOL(netlink_queue_skip);
1805 EXPORT_SYMBOL(netlink_broadcast);
1806 EXPORT_SYMBOL(netlink_dump_start);
1807 EXPORT_SYMBOL(netlink_kernel_create);
1808 EXPORT_SYMBOL(netlink_register_notifier);
1809 EXPORT_SYMBOL(netlink_set_err);
1810 EXPORT_SYMBOL(netlink_set_nonroot);
1811 EXPORT_SYMBOL(netlink_unicast);
1812 EXPORT_SYMBOL(netlink_unregister_notifier);