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_base.h>
61 #include <linux/vs_context.h>
62 #include <linux/vs_network.h>
63 #include <linux/vs_limit.h>
67 #include <net/netlink.h>
69 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
72 /* struct sock has to be the first member of netlink_sock */
80 unsigned long *groups;
82 wait_queue_head_t wait;
83 struct netlink_callback *cb;
85 void (*data_ready)(struct sock *sk, int bytes);
86 struct module *module;
89 #define NETLINK_KERNEL_SOCKET 0x1
90 #define NETLINK_RECV_PKTINFO 0x2
92 static inline struct netlink_sock *nlk_sk(struct sock *sk)
94 return (struct netlink_sock *)sk;
98 struct hlist_head *table;
99 unsigned long rehash_time;
104 unsigned int entries;
105 unsigned int max_shift;
110 struct netlink_table {
111 struct nl_pid_hash hash;
112 struct hlist_head mc_list;
113 unsigned long *listeners;
114 unsigned int nl_nonroot;
116 struct module *module;
120 static struct netlink_table *nl_table;
122 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
124 static int netlink_dump(struct sock *sk);
125 static void netlink_destroy_callback(struct netlink_callback *cb);
127 static DEFINE_RWLOCK(nl_table_lock);
128 static atomic_t nl_table_users = ATOMIC_INIT(0);
130 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
132 static u32 netlink_group_mask(u32 group)
134 return group ? 1 << (group - 1) : 0;
137 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
139 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
142 static void netlink_sock_destruct(struct sock *sk)
144 skb_queue_purge(&sk->sk_receive_queue);
146 if (!sock_flag(sk, SOCK_DEAD)) {
147 printk("Freeing alive netlink socket %p\n", sk);
150 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
151 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
152 BUG_TRAP(!nlk_sk(sk)->cb);
153 BUG_TRAP(!nlk_sk(sk)->groups);
156 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
157 * Look, when several writers sleep and reader wakes them up, all but one
158 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
159 * this, _but_ remember, it adds useless work on UP machines.
162 static void netlink_table_grab(void)
164 write_lock_bh(&nl_table_lock);
166 if (atomic_read(&nl_table_users)) {
167 DECLARE_WAITQUEUE(wait, current);
169 add_wait_queue_exclusive(&nl_table_wait, &wait);
171 set_current_state(TASK_UNINTERRUPTIBLE);
172 if (atomic_read(&nl_table_users) == 0)
174 write_unlock_bh(&nl_table_lock);
176 write_lock_bh(&nl_table_lock);
179 __set_current_state(TASK_RUNNING);
180 remove_wait_queue(&nl_table_wait, &wait);
184 static __inline__ void netlink_table_ungrab(void)
186 write_unlock_bh(&nl_table_lock);
187 wake_up(&nl_table_wait);
190 static __inline__ void
191 netlink_lock_table(void)
193 /* read_lock() synchronizes us to netlink_table_grab */
195 read_lock(&nl_table_lock);
196 atomic_inc(&nl_table_users);
197 read_unlock(&nl_table_lock);
200 static __inline__ void
201 netlink_unlock_table(void)
203 if (atomic_dec_and_test(&nl_table_users))
204 wake_up(&nl_table_wait);
207 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
209 struct nl_pid_hash *hash = &nl_table[protocol].hash;
210 struct hlist_head *head;
212 struct hlist_node *node;
214 read_lock(&nl_table_lock);
215 head = nl_pid_hashfn(hash, pid);
216 sk_for_each(sk, node, head) {
217 if (nlk_sk(sk)->pid == pid) {
224 read_unlock(&nl_table_lock);
228 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
230 if (size <= PAGE_SIZE)
231 return kmalloc(size, GFP_ATOMIC);
233 return (struct hlist_head *)
234 __get_free_pages(GFP_ATOMIC, get_order(size));
237 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
239 if (size <= PAGE_SIZE)
242 free_pages((unsigned long)table, get_order(size));
245 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
247 unsigned int omask, mask, shift;
249 struct hlist_head *otable, *table;
252 omask = mask = hash->mask;
253 osize = size = (mask + 1) * sizeof(*table);
257 if (++shift > hash->max_shift)
263 table = nl_pid_hash_alloc(size);
267 memset(table, 0, size);
268 otable = hash->table;
272 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
274 for (i = 0; i <= omask; i++) {
276 struct hlist_node *node, *tmp;
278 sk_for_each_safe(sk, node, tmp, &otable[i])
279 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
282 nl_pid_hash_free(otable, osize);
283 hash->rehash_time = jiffies + 10 * 60 * HZ;
287 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
289 int avg = hash->entries >> hash->shift;
291 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
294 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
295 nl_pid_hash_rehash(hash, 0);
302 static const struct proto_ops netlink_ops;
305 netlink_update_listeners(struct sock *sk)
307 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
308 struct hlist_node *node;
312 for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) {
314 sk_for_each_bound(sk, node, &tbl->mc_list)
315 mask |= nlk_sk(sk)->groups[i];
316 tbl->listeners[i] = mask;
318 /* this function is only called with the netlink table "grabbed", which
319 * makes sure updates are visible before bind or setsockopt return. */
322 static int netlink_insert(struct sock *sk, u32 pid)
324 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
325 struct hlist_head *head;
326 int err = -EADDRINUSE;
328 struct hlist_node *node;
331 netlink_table_grab();
332 head = nl_pid_hashfn(hash, pid);
334 sk_for_each(osk, node, head) {
335 if (nlk_sk(osk)->pid == pid)
347 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
350 if (len && nl_pid_hash_dilute(hash, len))
351 head = nl_pid_hashfn(hash, pid);
353 nlk_sk(sk)->pid = pid;
354 sk_add_node(sk, head);
358 netlink_table_ungrab();
362 static void netlink_remove(struct sock *sk)
364 netlink_table_grab();
365 if (sk_del_node_init(sk))
366 nl_table[sk->sk_protocol].hash.entries--;
367 if (nlk_sk(sk)->subscriptions)
368 __sk_del_bind_node(sk);
369 netlink_table_ungrab();
372 static struct proto netlink_proto = {
374 .owner = THIS_MODULE,
375 .obj_size = sizeof(struct netlink_sock),
378 static int __netlink_create(struct socket *sock, int protocol)
381 struct netlink_sock *nlk;
383 sock->ops = &netlink_ops;
385 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
389 sock_init_data(sock, sk);
392 spin_lock_init(&nlk->cb_lock);
393 init_waitqueue_head(&nlk->wait);
395 sk->sk_destruct = netlink_sock_destruct;
396 sk->sk_protocol = protocol;
400 static int netlink_create(struct socket *sock, int protocol)
402 struct module *module = NULL;
403 struct netlink_sock *nlk;
407 sock->state = SS_UNCONNECTED;
409 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
410 return -ESOCKTNOSUPPORT;
412 if (protocol<0 || protocol >= MAX_LINKS)
413 return -EPROTONOSUPPORT;
415 netlink_lock_table();
417 if (!nl_table[protocol].registered) {
418 netlink_unlock_table();
419 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
420 netlink_lock_table();
423 if (nl_table[protocol].registered &&
424 try_module_get(nl_table[protocol].module))
425 module = nl_table[protocol].module;
426 groups = nl_table[protocol].groups;
427 netlink_unlock_table();
429 if ((err = __netlink_create(sock, protocol)) < 0)
432 nlk = nlk_sk(sock->sk);
433 nlk->module = module;
442 static int netlink_release(struct socket *sock)
444 struct sock *sk = sock->sk;
445 struct netlink_sock *nlk;
453 spin_lock(&nlk->cb_lock);
456 nlk->cb->done(nlk->cb);
457 netlink_destroy_callback(nlk->cb);
460 spin_unlock(&nlk->cb_lock);
462 /* OK. Socket is unlinked, and, therefore,
463 no new packets will arrive */
467 wake_up_interruptible_all(&nlk->wait);
469 skb_queue_purge(&sk->sk_write_queue);
471 if (nlk->pid && !nlk->subscriptions) {
472 struct netlink_notify n = {
473 .protocol = sk->sk_protocol,
476 atomic_notifier_call_chain(&netlink_chain,
477 NETLINK_URELEASE, &n);
481 module_put(nlk->module);
483 netlink_table_grab();
484 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
485 kfree(nl_table[sk->sk_protocol].listeners);
486 nl_table[sk->sk_protocol].module = NULL;
487 nl_table[sk->sk_protocol].registered = 0;
488 } else if (nlk->subscriptions)
489 netlink_update_listeners(sk);
490 netlink_table_ungrab();
499 static int netlink_autobind(struct socket *sock)
501 struct sock *sk = sock->sk;
502 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
503 struct hlist_head *head;
505 struct hlist_node *node;
506 s32 pid = current->tgid;
508 static s32 rover = -4097;
512 netlink_table_grab();
513 head = nl_pid_hashfn(hash, pid);
514 sk_for_each(osk, node, head) {
515 if (nlk_sk(osk)->pid == pid) {
516 /* Bind collision, search negative pid values. */
520 netlink_table_ungrab();
524 netlink_table_ungrab();
526 err = netlink_insert(sk, pid);
527 if (err == -EADDRINUSE)
530 /* If 2 threads race to autobind, that is fine. */
537 static inline int netlink_capable(struct socket *sock, unsigned int flag)
539 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
540 capable(CAP_NET_ADMIN);
544 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
546 struct netlink_sock *nlk = nlk_sk(sk);
548 if (nlk->subscriptions && !subscriptions)
549 __sk_del_bind_node(sk);
550 else if (!nlk->subscriptions && subscriptions)
551 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
552 nlk->subscriptions = subscriptions;
555 static int netlink_alloc_groups(struct sock *sk)
557 struct netlink_sock *nlk = nlk_sk(sk);
561 netlink_lock_table();
562 groups = nl_table[sk->sk_protocol].groups;
563 if (!nl_table[sk->sk_protocol].registered)
565 netlink_unlock_table();
570 nlk->groups = kmalloc(NLGRPSZ(groups), GFP_KERNEL);
571 if (nlk->groups == NULL)
573 memset(nlk->groups, 0, NLGRPSZ(groups));
574 nlk->ngroups = groups;
578 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
580 struct sock *sk = sock->sk;
581 struct netlink_sock *nlk = nlk_sk(sk);
582 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
585 if (nladdr->nl_family != AF_NETLINK)
588 /* Only superuser is allowed to listen multicasts */
589 if (nladdr->nl_groups) {
590 if (!netlink_capable(sock, NL_NONROOT_RECV))
592 if (nlk->groups == NULL) {
593 err = netlink_alloc_groups(sk);
600 if (nladdr->nl_pid != nlk->pid)
603 err = nladdr->nl_pid ?
604 netlink_insert(sk, nladdr->nl_pid) :
605 netlink_autobind(sock);
610 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
613 netlink_table_grab();
614 netlink_update_subscriptions(sk, nlk->subscriptions +
615 hweight32(nladdr->nl_groups) -
616 hweight32(nlk->groups[0]));
617 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
618 netlink_update_listeners(sk);
619 netlink_table_ungrab();
624 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
628 struct sock *sk = sock->sk;
629 struct netlink_sock *nlk = nlk_sk(sk);
630 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
632 if (addr->sa_family == AF_UNSPEC) {
633 sk->sk_state = NETLINK_UNCONNECTED;
638 if (addr->sa_family != AF_NETLINK)
641 /* Only superuser is allowed to send multicasts */
642 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
646 err = netlink_autobind(sock);
649 sk->sk_state = NETLINK_CONNECTED;
650 nlk->dst_pid = nladdr->nl_pid;
651 nlk->dst_group = ffs(nladdr->nl_groups);
657 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
659 struct sock *sk = sock->sk;
660 struct netlink_sock *nlk = nlk_sk(sk);
661 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
663 nladdr->nl_family = AF_NETLINK;
665 *addr_len = sizeof(*nladdr);
668 nladdr->nl_pid = nlk->dst_pid;
669 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
671 nladdr->nl_pid = nlk->pid;
672 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
677 static void netlink_overrun(struct sock *sk)
679 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
680 sk->sk_err = ENOBUFS;
681 sk->sk_error_report(sk);
685 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
687 int protocol = ssk->sk_protocol;
689 struct netlink_sock *nlk;
691 sock = netlink_lookup(protocol, pid);
693 return ERR_PTR(-ECONNREFUSED);
695 /* Don't bother queuing skb if kernel socket has no input function */
697 if ((nlk->pid == 0 && !nlk->data_ready) ||
698 (sock->sk_state == NETLINK_CONNECTED &&
699 nlk->dst_pid != nlk_sk(ssk)->pid)) {
701 return ERR_PTR(-ECONNREFUSED);
706 struct sock *netlink_getsockbyfilp(struct file *filp)
708 struct inode *inode = filp->f_dentry->d_inode;
711 if (!S_ISSOCK(inode->i_mode))
712 return ERR_PTR(-ENOTSOCK);
714 sock = SOCKET_I(inode)->sk;
715 if (sock->sk_family != AF_NETLINK)
716 return ERR_PTR(-EINVAL);
723 * Attach a skb to a netlink socket.
724 * The caller must hold a reference to the destination socket. On error, the
725 * reference is dropped. The skb is not send to the destination, just all
726 * all error checks are performed and memory in the queue is reserved.
728 * < 0: error. skb freed, reference to sock dropped.
730 * 1: repeat lookup - reference dropped while waiting for socket memory.
732 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
733 long timeo, struct sock *ssk)
735 struct netlink_sock *nlk;
739 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
740 test_bit(0, &nlk->state)) {
741 DECLARE_WAITQUEUE(wait, current);
743 if (!ssk || nlk_sk(ssk)->pid == 0)
750 __set_current_state(TASK_INTERRUPTIBLE);
751 add_wait_queue(&nlk->wait, &wait);
753 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
754 test_bit(0, &nlk->state)) &&
755 !sock_flag(sk, SOCK_DEAD))
756 timeo = schedule_timeout(timeo);
758 __set_current_state(TASK_RUNNING);
759 remove_wait_queue(&nlk->wait, &wait);
762 if (signal_pending(current)) {
764 return sock_intr_errno(timeo);
768 skb_set_owner_r(skb, sk);
772 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
776 skb_queue_tail(&sk->sk_receive_queue, skb);
777 sk->sk_data_ready(sk, len);
782 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
788 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
795 delta = skb->end - skb->tail;
796 if (delta * 2 < skb->truesize)
799 if (skb_shared(skb)) {
800 struct sk_buff *nskb = skb_clone(skb, allocation);
807 if (!pskb_expand_head(skb, 0, -delta, allocation))
808 skb->truesize -= delta;
813 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
819 skb = netlink_trim(skb, gfp_any());
821 timeo = sock_sndtimeo(ssk, nonblock);
823 sk = netlink_getsockbypid(ssk, pid);
828 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
834 return netlink_sendskb(sk, skb, ssk->sk_protocol);
837 int netlink_has_listeners(struct sock *sk, unsigned int group)
841 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
842 if (group - 1 < nl_table[sk->sk_protocol].groups)
843 res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners);
846 EXPORT_SYMBOL_GPL(netlink_has_listeners);
848 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
850 struct netlink_sock *nlk = nlk_sk(sk);
852 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
853 !test_bit(0, &nlk->state)) {
854 skb_set_owner_r(skb, sk);
855 skb_queue_tail(&sk->sk_receive_queue, skb);
856 sk->sk_data_ready(sk, skb->len);
857 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
862 struct netlink_broadcast_data {
863 struct sock *exclude_sk;
870 struct sk_buff *skb, *skb2;
873 static inline int do_one_broadcast(struct sock *sk,
874 struct netlink_broadcast_data *p)
876 struct netlink_sock *nlk = nlk_sk(sk);
879 if (p->exclude_sk == sk)
882 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
883 !test_bit(p->group - 1, nlk->groups))
892 if (p->skb2 == NULL) {
893 if (skb_shared(p->skb)) {
894 p->skb2 = skb_clone(p->skb, p->allocation);
896 p->skb2 = skb_get(p->skb);
898 * skb ownership may have been set when
899 * delivered to a previous socket.
904 if (p->skb2 == NULL) {
906 /* Clone failed. Notify ALL listeners. */
908 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
921 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
922 u32 group, gfp_t allocation)
924 struct netlink_broadcast_data info;
925 struct hlist_node *node;
928 skb = netlink_trim(skb, allocation);
930 info.exclude_sk = ssk;
936 info.allocation = allocation;
940 /* While we sleep in clone, do not allow to change socket list */
942 netlink_lock_table();
944 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
945 do_one_broadcast(sk, &info);
949 netlink_unlock_table();
952 kfree_skb(info.skb2);
954 if (info.delivered) {
955 if (info.congested && (allocation & __GFP_WAIT))
964 struct netlink_set_err_data {
965 struct sock *exclude_sk;
971 static inline int do_one_set_err(struct sock *sk,
972 struct netlink_set_err_data *p)
974 struct netlink_sock *nlk = nlk_sk(sk);
976 if (sk == p->exclude_sk)
979 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
980 !test_bit(p->group - 1, nlk->groups))
983 sk->sk_err = p->code;
984 sk->sk_error_report(sk);
989 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
991 struct netlink_set_err_data info;
992 struct hlist_node *node;
995 info.exclude_sk = ssk;
1000 read_lock(&nl_table_lock);
1002 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1003 do_one_set_err(sk, &info);
1005 read_unlock(&nl_table_lock);
1008 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1009 char __user *optval, int optlen)
1011 struct sock *sk = sock->sk;
1012 struct netlink_sock *nlk = nlk_sk(sk);
1015 if (level != SOL_NETLINK)
1016 return -ENOPROTOOPT;
1018 if (optlen >= sizeof(int) &&
1019 get_user(val, (int __user *)optval))
1023 case NETLINK_PKTINFO:
1025 nlk->flags |= NETLINK_RECV_PKTINFO;
1027 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1030 case NETLINK_ADD_MEMBERSHIP:
1031 case NETLINK_DROP_MEMBERSHIP: {
1032 unsigned int subscriptions;
1033 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1035 if (!netlink_capable(sock, NL_NONROOT_RECV))
1037 if (nlk->groups == NULL) {
1038 err = netlink_alloc_groups(sk);
1042 if (!val || val - 1 >= nlk->ngroups)
1044 netlink_table_grab();
1045 old = test_bit(val - 1, nlk->groups);
1046 subscriptions = nlk->subscriptions - old + new;
1048 __set_bit(val - 1, nlk->groups);
1050 __clear_bit(val - 1, nlk->groups);
1051 netlink_update_subscriptions(sk, subscriptions);
1052 netlink_update_listeners(sk);
1053 netlink_table_ungrab();
1063 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1064 char __user *optval, int __user *optlen)
1066 struct sock *sk = sock->sk;
1067 struct netlink_sock *nlk = nlk_sk(sk);
1070 if (level != SOL_NETLINK)
1071 return -ENOPROTOOPT;
1073 if (get_user(len, optlen))
1079 case NETLINK_PKTINFO:
1080 if (len < sizeof(int))
1083 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1084 put_user(len, optlen);
1085 put_user(val, optval);
1094 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1096 struct nl_pktinfo info;
1098 info.group = NETLINK_CB(skb).dst_group;
1099 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1102 static inline void netlink_rcv_wake(struct sock *sk)
1104 struct netlink_sock *nlk = nlk_sk(sk);
1106 if (skb_queue_empty(&sk->sk_receive_queue))
1107 clear_bit(0, &nlk->state);
1108 if (!test_bit(0, &nlk->state))
1109 wake_up_interruptible(&nlk->wait);
1112 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1113 struct msghdr *msg, size_t len)
1115 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1116 struct sock *sk = sock->sk;
1117 struct netlink_sock *nlk = nlk_sk(sk);
1118 struct sockaddr_nl *addr=msg->msg_name;
1121 struct sk_buff *skb;
1123 struct scm_cookie scm;
1125 if (msg->msg_flags&MSG_OOB)
1128 if (NULL == siocb->scm)
1130 err = scm_send(sock, msg, siocb->scm);
1134 if (msg->msg_namelen) {
1135 if (addr->nl_family != AF_NETLINK)
1137 dst_pid = addr->nl_pid;
1138 dst_group = ffs(addr->nl_groups);
1139 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1142 dst_pid = nlk->dst_pid;
1143 dst_group = nlk->dst_group;
1147 err = netlink_autobind(sock);
1153 if (len > sk->sk_sndbuf - 32)
1156 skb = alloc_skb(len, GFP_KERNEL);
1160 NETLINK_CB(skb).pid = nlk->pid;
1161 NETLINK_CB(skb).dst_pid = dst_pid;
1162 NETLINK_CB(skb).dst_group = dst_group;
1163 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1164 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1165 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1167 /* What can I do? Netlink is asynchronous, so that
1168 we will have to save current capabilities to
1169 check them, when this message will be delivered
1170 to corresponding kernel module. --ANK (980802)
1174 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1179 err = security_netlink_send(sk, skb);
1186 atomic_inc(&skb->users);
1187 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1189 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1195 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1196 struct msghdr *msg, size_t len,
1199 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1200 struct scm_cookie scm;
1201 struct sock *sk = sock->sk;
1202 struct netlink_sock *nlk = nlk_sk(sk);
1203 int noblock = flags&MSG_DONTWAIT;
1205 struct sk_buff *skb;
1213 skb = skb_recv_datagram(sk,flags,noblock,&err);
1217 msg->msg_namelen = 0;
1221 msg->msg_flags |= MSG_TRUNC;
1225 skb->h.raw = skb->data;
1226 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1228 if (msg->msg_name) {
1229 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1230 addr->nl_family = AF_NETLINK;
1232 addr->nl_pid = NETLINK_CB(skb).pid;
1233 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1234 msg->msg_namelen = sizeof(*addr);
1237 if (nlk->flags & NETLINK_RECV_PKTINFO)
1238 netlink_cmsg_recv_pktinfo(msg, skb);
1240 if (NULL == siocb->scm) {
1241 memset(&scm, 0, sizeof(scm));
1244 siocb->scm->creds = *NETLINK_CREDS(skb);
1245 skb_free_datagram(sk, skb);
1247 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1250 scm_recv(sock, msg, siocb->scm, flags);
1253 netlink_rcv_wake(sk);
1254 return err ? : copied;
1257 static void netlink_data_ready(struct sock *sk, int len)
1259 struct netlink_sock *nlk = nlk_sk(sk);
1261 if (nlk->data_ready)
1262 nlk->data_ready(sk, len);
1263 netlink_rcv_wake(sk);
1267 * We export these functions to other modules. They provide a
1268 * complete set of kernel non-blocking support for message
1273 netlink_kernel_create(int unit, unsigned int groups,
1274 void (*input)(struct sock *sk, int len),
1275 struct module *module)
1277 struct socket *sock;
1279 struct netlink_sock *nlk;
1280 unsigned long *listeners = NULL;
1285 if (unit<0 || unit>=MAX_LINKS)
1288 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1291 if (__netlink_create(sock, unit) < 0)
1292 goto out_sock_release;
1297 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1299 goto out_sock_release;
1302 sk->sk_data_ready = netlink_data_ready;
1304 nlk_sk(sk)->data_ready = input;
1306 if (netlink_insert(sk, 0))
1307 goto out_sock_release;
1310 nlk->flags |= NETLINK_KERNEL_SOCKET;
1312 netlink_table_grab();
1313 nl_table[unit].groups = groups;
1314 nl_table[unit].listeners = listeners;
1315 nl_table[unit].module = module;
1316 nl_table[unit].registered = 1;
1317 netlink_table_ungrab();
1327 void netlink_set_nonroot(int protocol, unsigned int flags)
1329 if ((unsigned int)protocol < MAX_LINKS)
1330 nl_table[protocol].nl_nonroot = flags;
1333 static void netlink_destroy_callback(struct netlink_callback *cb)
1341 * It looks a bit ugly.
1342 * It would be better to create kernel thread.
1345 static int netlink_dump(struct sock *sk)
1347 struct netlink_sock *nlk = nlk_sk(sk);
1348 struct netlink_callback *cb;
1349 struct sk_buff *skb;
1350 struct nlmsghdr *nlh;
1353 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1357 spin_lock(&nlk->cb_lock);
1361 spin_unlock(&nlk->cb_lock);
1366 len = cb->dump(skb, cb);
1369 spin_unlock(&nlk->cb_lock);
1370 skb_queue_tail(&sk->sk_receive_queue, skb);
1371 sk->sk_data_ready(sk, len);
1375 nlh = NLMSG_NEW_ANSWER(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1376 memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1377 skb_queue_tail(&sk->sk_receive_queue, skb);
1378 sk->sk_data_ready(sk, skb->len);
1383 spin_unlock(&nlk->cb_lock);
1385 netlink_destroy_callback(cb);
1392 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1393 struct nlmsghdr *nlh,
1394 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1395 int (*done)(struct netlink_callback*))
1397 struct netlink_callback *cb;
1399 struct netlink_sock *nlk;
1401 cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1405 memset(cb, 0, sizeof(*cb));
1409 atomic_inc(&skb->users);
1412 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1414 netlink_destroy_callback(cb);
1415 return -ECONNREFUSED;
1418 /* A dump is in progress... */
1419 spin_lock(&nlk->cb_lock);
1421 spin_unlock(&nlk->cb_lock);
1422 netlink_destroy_callback(cb);
1427 spin_unlock(&nlk->cb_lock);
1434 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1436 struct sk_buff *skb;
1437 struct nlmsghdr *rep;
1438 struct nlmsgerr *errmsg;
1442 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1444 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1446 skb = alloc_skb(size, GFP_KERNEL);
1450 sk = netlink_lookup(in_skb->sk->sk_protocol,
1451 NETLINK_CB(in_skb).pid);
1453 sk->sk_err = ENOBUFS;
1454 sk->sk_error_report(sk);
1460 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1461 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1462 errmsg = NLMSG_DATA(rep);
1463 errmsg->error = err;
1464 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1465 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1468 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1469 struct nlmsghdr *, int *))
1471 unsigned int total_len;
1472 struct nlmsghdr *nlh;
1475 while (skb->len >= nlmsg_total_size(0)) {
1476 nlh = (struct nlmsghdr *) skb->data;
1478 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1481 total_len = min(NLMSG_ALIGN(nlh->nlmsg_len), skb->len);
1483 if (cb(skb, nlh, &err) < 0) {
1484 /* Not an error, but we have to interrupt processing
1485 * here. Note: that in this case we do not pull
1486 * message from skb, it will be processed later.
1490 netlink_ack(skb, nlh, err);
1491 } else if (nlh->nlmsg_flags & NLM_F_ACK)
1492 netlink_ack(skb, nlh, 0);
1494 skb_pull(skb, total_len);
1501 * nelink_run_queue - Process netlink receive queue.
1502 * @sk: Netlink socket containing the queue
1503 * @qlen: Place to store queue length upon entry
1504 * @cb: Callback function invoked for each netlink message found
1506 * Processes as much as there was in the queue upon entry and invokes
1507 * a callback function for each netlink message found. The callback
1508 * function may refuse a message by returning a negative error code
1509 * but setting the error pointer to 0 in which case this function
1510 * returns with a qlen != 0.
1512 * qlen must be initialized to 0 before the initial entry, afterwards
1513 * the function may be called repeatedly until qlen reaches 0.
1515 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1516 int (*cb)(struct sk_buff *, struct nlmsghdr *, int *))
1518 struct sk_buff *skb;
1520 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1521 *qlen = skb_queue_len(&sk->sk_receive_queue);
1523 for (; *qlen; (*qlen)--) {
1524 skb = skb_dequeue(&sk->sk_receive_queue);
1525 if (netlink_rcv_skb(skb, cb)) {
1527 skb_queue_head(&sk->sk_receive_queue, skb);
1540 * netlink_queue_skip - Skip netlink message while processing queue.
1541 * @nlh: Netlink message to be skipped
1542 * @skb: Socket buffer containing the netlink messages.
1544 * Pulls the given netlink message off the socket buffer so the next
1545 * call to netlink_queue_run() will not reconsider the message.
1547 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1549 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1551 if (msglen > skb->len)
1554 skb_pull(skb, msglen);
1557 #ifdef CONFIG_PROC_FS
1558 struct nl_seq_iter {
1563 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1565 struct nl_seq_iter *iter = seq->private;
1568 struct hlist_node *node;
1571 for (i=0; i<MAX_LINKS; i++) {
1572 struct nl_pid_hash *hash = &nl_table[i].hash;
1574 for (j = 0; j <= hash->mask; j++) {
1575 sk_for_each(s, node, &hash->table[j]) {
1588 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1590 read_lock(&nl_table_lock);
1591 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1594 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1597 struct nl_seq_iter *iter;
1602 if (v == SEQ_START_TOKEN)
1603 return netlink_seq_socket_idx(seq, 0);
1609 iter = seq->private;
1611 j = iter->hash_idx + 1;
1614 struct nl_pid_hash *hash = &nl_table[i].hash;
1616 for (; j <= hash->mask; j++) {
1617 s = sk_head(&hash->table[j]);
1626 } while (++i < MAX_LINKS);
1631 static void netlink_seq_stop(struct seq_file *seq, void *v)
1633 read_unlock(&nl_table_lock);
1637 static int netlink_seq_show(struct seq_file *seq, void *v)
1639 if (v == SEQ_START_TOKEN)
1641 "sk Eth Pid Groups "
1642 "Rmem Wmem Dump Locks\n");
1645 struct netlink_sock *nlk = nlk_sk(s);
1647 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1651 nlk->groups ? (u32)nlk->groups[0] : 0,
1652 atomic_read(&s->sk_rmem_alloc),
1653 atomic_read(&s->sk_wmem_alloc),
1655 atomic_read(&s->sk_refcnt)
1662 static struct seq_operations netlink_seq_ops = {
1663 .start = netlink_seq_start,
1664 .next = netlink_seq_next,
1665 .stop = netlink_seq_stop,
1666 .show = netlink_seq_show,
1670 static int netlink_seq_open(struct inode *inode, struct file *file)
1672 struct seq_file *seq;
1673 struct nl_seq_iter *iter;
1676 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1680 err = seq_open(file, &netlink_seq_ops);
1686 memset(iter, 0, sizeof(*iter));
1687 seq = file->private_data;
1688 seq->private = iter;
1692 static struct file_operations netlink_seq_fops = {
1693 .owner = THIS_MODULE,
1694 .open = netlink_seq_open,
1696 .llseek = seq_lseek,
1697 .release = seq_release_private,
1702 int netlink_register_notifier(struct notifier_block *nb)
1704 return atomic_notifier_chain_register(&netlink_chain, nb);
1707 int netlink_unregister_notifier(struct notifier_block *nb)
1709 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1712 static const struct proto_ops netlink_ops = {
1713 .family = PF_NETLINK,
1714 .owner = THIS_MODULE,
1715 .release = netlink_release,
1716 .bind = netlink_bind,
1717 .connect = netlink_connect,
1718 .socketpair = sock_no_socketpair,
1719 .accept = sock_no_accept,
1720 .getname = netlink_getname,
1721 .poll = datagram_poll,
1722 .ioctl = sock_no_ioctl,
1723 .listen = sock_no_listen,
1724 .shutdown = sock_no_shutdown,
1725 .setsockopt = netlink_setsockopt,
1726 .getsockopt = netlink_getsockopt,
1727 .sendmsg = netlink_sendmsg,
1728 .recvmsg = netlink_recvmsg,
1729 .mmap = sock_no_mmap,
1730 .sendpage = sock_no_sendpage,
1733 static struct net_proto_family netlink_family_ops = {
1734 .family = PF_NETLINK,
1735 .create = netlink_create,
1736 .owner = THIS_MODULE, /* for consistency 8) */
1739 extern void netlink_skb_parms_too_large(void);
1741 static int __init netlink_proto_init(void)
1743 struct sk_buff *dummy_skb;
1747 int err = proto_register(&netlink_proto, 0);
1752 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1753 netlink_skb_parms_too_large();
1755 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1758 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1762 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1764 if (num_physpages >= (128 * 1024))
1765 max = num_physpages >> (21 - PAGE_SHIFT);
1767 max = num_physpages >> (23 - PAGE_SHIFT);
1769 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1770 max = (1UL << order) / sizeof(struct hlist_head);
1771 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1773 for (i = 0; i < MAX_LINKS; i++) {
1774 struct nl_pid_hash *hash = &nl_table[i].hash;
1776 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1779 nl_pid_hash_free(nl_table[i].hash.table,
1780 1 * sizeof(*hash->table));
1784 memset(hash->table, 0, 1 * sizeof(*hash->table));
1785 hash->max_shift = order;
1788 hash->rehash_time = jiffies;
1791 sock_register(&netlink_family_ops);
1792 #ifdef CONFIG_PROC_FS
1793 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1795 /* The netlink device handler may be needed early. */
1801 core_initcall(netlink_proto_init);
1803 EXPORT_SYMBOL(netlink_ack);
1804 EXPORT_SYMBOL(netlink_run_queue);
1805 EXPORT_SYMBOL(netlink_queue_skip);
1806 EXPORT_SYMBOL(netlink_broadcast);
1807 EXPORT_SYMBOL(netlink_dump_start);
1808 EXPORT_SYMBOL(netlink_kernel_create);
1809 EXPORT_SYMBOL(netlink_register_notifier);
1810 EXPORT_SYMBOL(netlink_set_err);
1811 EXPORT_SYMBOL(netlink_set_nonroot);
1812 EXPORT_SYMBOL(netlink_unicast);
1813 EXPORT_SYMBOL(netlink_unregister_notifier);