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/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/smp_lock.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
61 #include <net/netlink.h>
64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
67 /* struct sock has to be the first member of netlink_sock */
75 unsigned long *groups;
77 wait_queue_head_t wait;
78 struct netlink_callback *cb;
80 void (*data_ready)(struct sock *sk, int bytes);
81 struct module *module;
84 #define NETLINK_KERNEL_SOCKET 0x1
85 #define NETLINK_RECV_PKTINFO 0x2
87 static inline struct netlink_sock *nlk_sk(struct sock *sk)
89 return (struct netlink_sock *)sk;
93 struct hlist_head *table;
94 unsigned long rehash_time;
100 unsigned int max_shift;
105 struct netlink_table {
106 struct nl_pid_hash hash;
107 struct hlist_head mc_list;
108 unsigned int nl_nonroot;
110 struct module *module;
114 static struct netlink_table *nl_table;
116 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
118 static int netlink_dump(struct sock *sk);
119 static void netlink_destroy_callback(struct netlink_callback *cb);
121 static DEFINE_RWLOCK(nl_table_lock);
122 static atomic_t nl_table_users = ATOMIC_INIT(0);
124 static struct notifier_block *netlink_chain;
126 static u32 netlink_group_mask(u32 group)
128 return group ? 1 << (group - 1) : 0;
131 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
133 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
136 static void netlink_sock_destruct(struct sock *sk)
138 skb_queue_purge(&sk->sk_receive_queue);
140 if (!sock_flag(sk, SOCK_DEAD)) {
141 printk("Freeing alive netlink socket %p\n", sk);
144 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
145 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
146 BUG_TRAP(!nlk_sk(sk)->cb);
147 BUG_TRAP(!nlk_sk(sk)->groups);
150 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
151 * Look, when several writers sleep and reader wakes them up, all but one
152 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
153 * this, _but_ remember, it adds useless work on UP machines.
156 static void netlink_table_grab(void)
158 write_lock_bh(&nl_table_lock);
160 if (atomic_read(&nl_table_users)) {
161 DECLARE_WAITQUEUE(wait, current);
163 add_wait_queue_exclusive(&nl_table_wait, &wait);
165 set_current_state(TASK_UNINTERRUPTIBLE);
166 if (atomic_read(&nl_table_users) == 0)
168 write_unlock_bh(&nl_table_lock);
170 write_lock_bh(&nl_table_lock);
173 __set_current_state(TASK_RUNNING);
174 remove_wait_queue(&nl_table_wait, &wait);
178 static __inline__ void netlink_table_ungrab(void)
180 write_unlock_bh(&nl_table_lock);
181 wake_up(&nl_table_wait);
184 static __inline__ void
185 netlink_lock_table(void)
187 /* read_lock() synchronizes us to netlink_table_grab */
189 read_lock(&nl_table_lock);
190 atomic_inc(&nl_table_users);
191 read_unlock(&nl_table_lock);
194 static __inline__ void
195 netlink_unlock_table(void)
197 if (atomic_dec_and_test(&nl_table_users))
198 wake_up(&nl_table_wait);
201 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
203 struct nl_pid_hash *hash = &nl_table[protocol].hash;
204 struct hlist_head *head;
206 struct hlist_node *node;
208 read_lock(&nl_table_lock);
209 head = nl_pid_hashfn(hash, pid);
210 sk_for_each(sk, node, head) {
211 if (nlk_sk(sk)->pid == pid) {
218 read_unlock(&nl_table_lock);
222 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
224 if (size <= PAGE_SIZE)
225 return kmalloc(size, GFP_ATOMIC);
227 return (struct hlist_head *)
228 __get_free_pages(GFP_ATOMIC, get_order(size));
231 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
233 if (size <= PAGE_SIZE)
236 free_pages((unsigned long)table, get_order(size));
239 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
241 unsigned int omask, mask, shift;
243 struct hlist_head *otable, *table;
246 omask = mask = hash->mask;
247 osize = size = (mask + 1) * sizeof(*table);
251 if (++shift > hash->max_shift)
257 table = nl_pid_hash_alloc(size);
261 memset(table, 0, size);
262 otable = hash->table;
266 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
268 for (i = 0; i <= omask; i++) {
270 struct hlist_node *node, *tmp;
272 sk_for_each_safe(sk, node, tmp, &otable[i])
273 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
276 nl_pid_hash_free(otable, osize);
277 hash->rehash_time = jiffies + 10 * 60 * HZ;
281 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
283 int avg = hash->entries >> hash->shift;
285 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
288 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
289 nl_pid_hash_rehash(hash, 0);
296 static struct proto_ops netlink_ops;
298 static int netlink_insert(struct sock *sk, u32 pid)
300 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
301 struct hlist_head *head;
302 int err = -EADDRINUSE;
304 struct hlist_node *node;
307 netlink_table_grab();
308 head = nl_pid_hashfn(hash, pid);
310 sk_for_each(osk, node, head) {
311 if (nlk_sk(osk)->pid == pid)
323 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
326 if (len && nl_pid_hash_dilute(hash, len))
327 head = nl_pid_hashfn(hash, pid);
329 nlk_sk(sk)->pid = pid;
330 sk_add_node(sk, head);
334 netlink_table_ungrab();
338 static void netlink_remove(struct sock *sk)
340 netlink_table_grab();
341 if (sk_del_node_init(sk))
342 nl_table[sk->sk_protocol].hash.entries--;
343 if (nlk_sk(sk)->subscriptions)
344 __sk_del_bind_node(sk);
345 netlink_table_ungrab();
348 static struct proto netlink_proto = {
350 .owner = THIS_MODULE,
351 .obj_size = sizeof(struct netlink_sock),
354 static int __netlink_create(struct socket *sock, int protocol)
357 struct netlink_sock *nlk;
359 sock->ops = &netlink_ops;
361 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
365 sock_init_data(sock, sk);
368 spin_lock_init(&nlk->cb_lock);
369 init_waitqueue_head(&nlk->wait);
371 sk->sk_destruct = netlink_sock_destruct;
372 sk->sk_protocol = protocol;
376 static int netlink_create(struct socket *sock, int protocol)
378 struct module *module = NULL;
379 struct netlink_sock *nlk;
383 sock->state = SS_UNCONNECTED;
385 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
386 return -ESOCKTNOSUPPORT;
388 if (protocol<0 || protocol >= MAX_LINKS)
389 return -EPROTONOSUPPORT;
391 netlink_lock_table();
393 if (!nl_table[protocol].registered) {
394 netlink_unlock_table();
395 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
396 netlink_lock_table();
399 if (nl_table[protocol].registered &&
400 try_module_get(nl_table[protocol].module))
401 module = nl_table[protocol].module;
402 groups = nl_table[protocol].groups;
403 netlink_unlock_table();
405 if ((err = __netlink_create(sock, protocol) < 0))
408 nlk = nlk_sk(sock->sk);
409 nlk->module = module;
418 static int netlink_release(struct socket *sock)
420 struct sock *sk = sock->sk;
421 struct netlink_sock *nlk;
429 spin_lock(&nlk->cb_lock);
432 nlk->cb->done(nlk->cb);
433 netlink_destroy_callback(nlk->cb);
436 spin_unlock(&nlk->cb_lock);
438 /* OK. Socket is unlinked, and, therefore,
439 no new packets will arrive */
443 wake_up_interruptible_all(&nlk->wait);
445 skb_queue_purge(&sk->sk_write_queue);
447 if (nlk->pid && !nlk->subscriptions) {
448 struct netlink_notify n = {
449 .protocol = sk->sk_protocol,
452 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
456 module_put(nlk->module);
458 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
459 netlink_table_grab();
460 nl_table[sk->sk_protocol].module = NULL;
461 nl_table[sk->sk_protocol].registered = 0;
462 netlink_table_ungrab();
472 static int netlink_autobind(struct socket *sock)
474 struct sock *sk = sock->sk;
475 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
476 struct hlist_head *head;
478 struct hlist_node *node;
479 s32 pid = current->tgid;
481 static s32 rover = -4097;
485 netlink_table_grab();
486 head = nl_pid_hashfn(hash, pid);
487 sk_for_each(osk, node, head) {
488 if (nlk_sk(osk)->pid == pid) {
489 /* Bind collision, search negative pid values. */
493 netlink_table_ungrab();
497 netlink_table_ungrab();
499 err = netlink_insert(sk, pid);
500 if (err == -EADDRINUSE)
503 /* If 2 threads race to autobind, that is fine. */
510 static inline int netlink_capable(struct socket *sock, unsigned int flag)
512 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
513 capable(CAP_NET_ADMIN);
517 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
519 struct netlink_sock *nlk = nlk_sk(sk);
521 if (nlk->subscriptions && !subscriptions)
522 __sk_del_bind_node(sk);
523 else if (!nlk->subscriptions && subscriptions)
524 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
525 nlk->subscriptions = subscriptions;
528 static int netlink_alloc_groups(struct sock *sk)
530 struct netlink_sock *nlk = nlk_sk(sk);
534 netlink_lock_table();
535 groups = nl_table[sk->sk_protocol].groups;
536 if (!nl_table[sk->sk_protocol].registered)
538 netlink_unlock_table();
543 nlk->groups = kmalloc(NLGRPSZ(groups), GFP_KERNEL);
544 if (nlk->groups == NULL)
546 memset(nlk->groups, 0, NLGRPSZ(groups));
547 nlk->ngroups = groups;
551 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
553 struct sock *sk = sock->sk;
554 struct netlink_sock *nlk = nlk_sk(sk);
555 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
558 if (nladdr->nl_family != AF_NETLINK)
561 /* Only superuser is allowed to listen multicasts */
562 if (nladdr->nl_groups) {
563 if (!netlink_capable(sock, NL_NONROOT_RECV))
565 if (nlk->groups == NULL) {
566 err = netlink_alloc_groups(sk);
573 if (nladdr->nl_pid != nlk->pid)
576 err = nladdr->nl_pid ?
577 netlink_insert(sk, nladdr->nl_pid) :
578 netlink_autobind(sock);
583 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
586 netlink_table_grab();
587 netlink_update_subscriptions(sk, nlk->subscriptions +
588 hweight32(nladdr->nl_groups) -
589 hweight32(nlk->groups[0]));
590 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
591 netlink_table_ungrab();
596 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
600 struct sock *sk = sock->sk;
601 struct netlink_sock *nlk = nlk_sk(sk);
602 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
604 if (addr->sa_family == AF_UNSPEC) {
605 sk->sk_state = NETLINK_UNCONNECTED;
610 if (addr->sa_family != AF_NETLINK)
613 /* Only superuser is allowed to send multicasts */
614 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
618 err = netlink_autobind(sock);
621 sk->sk_state = NETLINK_CONNECTED;
622 nlk->dst_pid = nladdr->nl_pid;
623 nlk->dst_group = ffs(nladdr->nl_groups);
629 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
631 struct sock *sk = sock->sk;
632 struct netlink_sock *nlk = nlk_sk(sk);
633 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
635 nladdr->nl_family = AF_NETLINK;
637 *addr_len = sizeof(*nladdr);
640 nladdr->nl_pid = nlk->dst_pid;
641 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
643 nladdr->nl_pid = nlk->pid;
644 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
649 static void netlink_overrun(struct sock *sk)
651 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
652 sk->sk_err = ENOBUFS;
653 sk->sk_error_report(sk);
657 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
659 int protocol = ssk->sk_protocol;
661 struct netlink_sock *nlk;
663 sock = netlink_lookup(protocol, pid);
665 return ERR_PTR(-ECONNREFUSED);
667 /* Don't bother queuing skb if kernel socket has no input function */
669 if ((nlk->pid == 0 && !nlk->data_ready) ||
670 (sock->sk_state == NETLINK_CONNECTED &&
671 nlk->dst_pid != nlk_sk(ssk)->pid)) {
673 return ERR_PTR(-ECONNREFUSED);
678 struct sock *netlink_getsockbyfilp(struct file *filp)
680 struct inode *inode = filp->f_dentry->d_inode;
683 if (!S_ISSOCK(inode->i_mode))
684 return ERR_PTR(-ENOTSOCK);
686 sock = SOCKET_I(inode)->sk;
687 if (sock->sk_family != AF_NETLINK)
688 return ERR_PTR(-EINVAL);
695 * Attach a skb to a netlink socket.
696 * The caller must hold a reference to the destination socket. On error, the
697 * reference is dropped. The skb is not send to the destination, just all
698 * all error checks are performed and memory in the queue is reserved.
700 * < 0: error. skb freed, reference to sock dropped.
702 * 1: repeat lookup - reference dropped while waiting for socket memory.
704 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo)
706 struct netlink_sock *nlk;
710 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
711 test_bit(0, &nlk->state)) {
712 DECLARE_WAITQUEUE(wait, current);
721 __set_current_state(TASK_INTERRUPTIBLE);
722 add_wait_queue(&nlk->wait, &wait);
724 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
725 test_bit(0, &nlk->state)) &&
726 !sock_flag(sk, SOCK_DEAD))
727 timeo = schedule_timeout(timeo);
729 __set_current_state(TASK_RUNNING);
730 remove_wait_queue(&nlk->wait, &wait);
733 if (signal_pending(current)) {
735 return sock_intr_errno(timeo);
739 skb_set_owner_r(skb, sk);
743 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
747 skb_queue_tail(&sk->sk_receive_queue, skb);
748 sk->sk_data_ready(sk, len);
753 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
759 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
766 delta = skb->end - skb->tail;
767 if (delta * 2 < skb->truesize)
770 if (skb_shared(skb)) {
771 struct sk_buff *nskb = skb_clone(skb, allocation);
778 if (!pskb_expand_head(skb, 0, -delta, allocation))
779 skb->truesize -= delta;
784 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
790 skb = netlink_trim(skb, gfp_any());
792 timeo = sock_sndtimeo(ssk, nonblock);
794 sk = netlink_getsockbypid(ssk, pid);
799 err = netlink_attachskb(sk, skb, nonblock, timeo);
805 return netlink_sendskb(sk, skb, ssk->sk_protocol);
808 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
810 struct netlink_sock *nlk = nlk_sk(sk);
812 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
813 !test_bit(0, &nlk->state)) {
814 skb_set_owner_r(skb, sk);
815 skb_queue_tail(&sk->sk_receive_queue, skb);
816 sk->sk_data_ready(sk, skb->len);
817 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
822 struct netlink_broadcast_data {
823 struct sock *exclude_sk;
830 struct sk_buff *skb, *skb2;
833 static inline int do_one_broadcast(struct sock *sk,
834 struct netlink_broadcast_data *p)
836 struct netlink_sock *nlk = nlk_sk(sk);
839 if (p->exclude_sk == sk)
842 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
843 !test_bit(p->group - 1, nlk->groups))
852 if (p->skb2 == NULL) {
853 if (skb_shared(p->skb)) {
854 p->skb2 = skb_clone(p->skb, p->allocation);
856 p->skb2 = skb_get(p->skb);
858 * skb ownership may have been set when
859 * delivered to a previous socket.
864 if (p->skb2 == NULL) {
866 /* Clone failed. Notify ALL listeners. */
868 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
881 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
882 u32 group, gfp_t allocation)
884 struct netlink_broadcast_data info;
885 struct hlist_node *node;
888 skb = netlink_trim(skb, allocation);
890 info.exclude_sk = ssk;
896 info.allocation = allocation;
900 /* While we sleep in clone, do not allow to change socket list */
902 netlink_lock_table();
904 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
905 do_one_broadcast(sk, &info);
909 netlink_unlock_table();
912 kfree_skb(info.skb2);
914 if (info.delivered) {
915 if (info.congested && (allocation & __GFP_WAIT))
924 struct netlink_set_err_data {
925 struct sock *exclude_sk;
931 static inline int do_one_set_err(struct sock *sk,
932 struct netlink_set_err_data *p)
934 struct netlink_sock *nlk = nlk_sk(sk);
936 if (sk == p->exclude_sk)
939 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
940 !test_bit(p->group - 1, nlk->groups))
943 sk->sk_err = p->code;
944 sk->sk_error_report(sk);
949 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
951 struct netlink_set_err_data info;
952 struct hlist_node *node;
955 info.exclude_sk = ssk;
960 read_lock(&nl_table_lock);
962 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
963 do_one_set_err(sk, &info);
965 read_unlock(&nl_table_lock);
968 static int netlink_setsockopt(struct socket *sock, int level, int optname,
969 char __user *optval, int optlen)
971 struct sock *sk = sock->sk;
972 struct netlink_sock *nlk = nlk_sk(sk);
975 if (level != SOL_NETLINK)
978 if (optlen >= sizeof(int) &&
979 get_user(val, (int __user *)optval))
983 case NETLINK_PKTINFO:
985 nlk->flags |= NETLINK_RECV_PKTINFO;
987 nlk->flags &= ~NETLINK_RECV_PKTINFO;
990 case NETLINK_ADD_MEMBERSHIP:
991 case NETLINK_DROP_MEMBERSHIP: {
992 unsigned int subscriptions;
993 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
995 if (!netlink_capable(sock, NL_NONROOT_RECV))
997 if (nlk->groups == NULL) {
998 err = netlink_alloc_groups(sk);
1002 if (!val || val - 1 >= nlk->ngroups)
1004 netlink_table_grab();
1005 old = test_bit(val - 1, nlk->groups);
1006 subscriptions = nlk->subscriptions - old + new;
1008 __set_bit(val - 1, nlk->groups);
1010 __clear_bit(val - 1, nlk->groups);
1011 netlink_update_subscriptions(sk, subscriptions);
1012 netlink_table_ungrab();
1022 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1023 char __user *optval, int __user *optlen)
1025 struct sock *sk = sock->sk;
1026 struct netlink_sock *nlk = nlk_sk(sk);
1029 if (level != SOL_NETLINK)
1030 return -ENOPROTOOPT;
1032 if (get_user(len, optlen))
1038 case NETLINK_PKTINFO:
1039 if (len < sizeof(int))
1042 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1043 put_user(len, optlen);
1044 put_user(val, optval);
1053 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1055 struct nl_pktinfo info;
1057 info.group = NETLINK_CB(skb).dst_group;
1058 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1061 static inline void netlink_rcv_wake(struct sock *sk)
1063 struct netlink_sock *nlk = nlk_sk(sk);
1065 if (skb_queue_empty(&sk->sk_receive_queue))
1066 clear_bit(0, &nlk->state);
1067 if (!test_bit(0, &nlk->state))
1068 wake_up_interruptible(&nlk->wait);
1071 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1072 struct msghdr *msg, size_t len)
1074 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1075 struct sock *sk = sock->sk;
1076 struct netlink_sock *nlk = nlk_sk(sk);
1077 struct sockaddr_nl *addr=msg->msg_name;
1080 struct sk_buff *skb;
1082 struct scm_cookie scm;
1084 if (msg->msg_flags&MSG_OOB)
1087 if (NULL == siocb->scm)
1089 err = scm_send(sock, msg, siocb->scm);
1093 if (msg->msg_namelen) {
1094 if (addr->nl_family != AF_NETLINK)
1096 dst_pid = addr->nl_pid;
1097 dst_group = ffs(addr->nl_groups);
1098 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1101 dst_pid = nlk->dst_pid;
1102 dst_group = nlk->dst_group;
1106 err = netlink_autobind(sock);
1112 if (len > sk->sk_sndbuf - 32)
1115 skb = alloc_skb(len, GFP_KERNEL);
1119 NETLINK_CB(skb).pid = nlk->pid;
1120 NETLINK_CB(skb).dst_pid = dst_pid;
1121 NETLINK_CB(skb).dst_group = dst_group;
1122 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1123 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1125 /* What can I do? Netlink is asynchronous, so that
1126 we will have to save current capabilities to
1127 check them, when this message will be delivered
1128 to corresponding kernel module. --ANK (980802)
1132 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1137 err = security_netlink_send(sk, skb);
1144 atomic_inc(&skb->users);
1145 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1147 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1153 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1154 struct msghdr *msg, size_t len,
1157 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1158 struct scm_cookie scm;
1159 struct sock *sk = sock->sk;
1160 struct netlink_sock *nlk = nlk_sk(sk);
1161 int noblock = flags&MSG_DONTWAIT;
1163 struct sk_buff *skb;
1171 skb = skb_recv_datagram(sk,flags,noblock,&err);
1175 msg->msg_namelen = 0;
1179 msg->msg_flags |= MSG_TRUNC;
1183 skb->h.raw = skb->data;
1184 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1186 if (msg->msg_name) {
1187 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1188 addr->nl_family = AF_NETLINK;
1190 addr->nl_pid = NETLINK_CB(skb).pid;
1191 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1192 msg->msg_namelen = sizeof(*addr);
1195 if (NULL == siocb->scm) {
1196 memset(&scm, 0, sizeof(scm));
1199 siocb->scm->creds = *NETLINK_CREDS(skb);
1200 skb_free_datagram(sk, skb);
1202 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1205 scm_recv(sock, msg, siocb->scm, flags);
1206 if (nlk->flags & NETLINK_RECV_PKTINFO)
1207 netlink_cmsg_recv_pktinfo(msg, skb);
1210 netlink_rcv_wake(sk);
1211 return err ? : copied;
1214 static void netlink_data_ready(struct sock *sk, int len)
1216 struct netlink_sock *nlk = nlk_sk(sk);
1218 if (nlk->data_ready)
1219 nlk->data_ready(sk, len);
1220 netlink_rcv_wake(sk);
1224 * We export these functions to other modules. They provide a
1225 * complete set of kernel non-blocking support for message
1230 netlink_kernel_create(int unit, unsigned int groups,
1231 void (*input)(struct sock *sk, int len),
1232 struct module *module)
1234 struct socket *sock;
1236 struct netlink_sock *nlk;
1241 if (unit<0 || unit>=MAX_LINKS)
1244 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1247 if (__netlink_create(sock, unit) < 0)
1248 goto out_sock_release;
1251 sk->sk_data_ready = netlink_data_ready;
1253 nlk_sk(sk)->data_ready = input;
1255 if (netlink_insert(sk, 0))
1256 goto out_sock_release;
1259 nlk->flags |= NETLINK_KERNEL_SOCKET;
1261 netlink_table_grab();
1262 nl_table[unit].groups = groups < 32 ? 32 : groups;
1263 nl_table[unit].module = module;
1264 nl_table[unit].registered = 1;
1265 netlink_table_ungrab();
1274 void netlink_set_nonroot(int protocol, unsigned int flags)
1276 if ((unsigned int)protocol < MAX_LINKS)
1277 nl_table[protocol].nl_nonroot = flags;
1280 static void netlink_destroy_callback(struct netlink_callback *cb)
1288 * It looks a bit ugly.
1289 * It would be better to create kernel thread.
1292 static int netlink_dump(struct sock *sk)
1294 struct netlink_sock *nlk = nlk_sk(sk);
1295 struct netlink_callback *cb;
1296 struct sk_buff *skb;
1297 struct nlmsghdr *nlh;
1300 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1304 spin_lock(&nlk->cb_lock);
1308 spin_unlock(&nlk->cb_lock);
1313 len = cb->dump(skb, cb);
1316 spin_unlock(&nlk->cb_lock);
1317 skb_queue_tail(&sk->sk_receive_queue, skb);
1318 sk->sk_data_ready(sk, len);
1322 nlh = NLMSG_NEW_ANSWER(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1323 memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1324 skb_queue_tail(&sk->sk_receive_queue, skb);
1325 sk->sk_data_ready(sk, skb->len);
1330 spin_unlock(&nlk->cb_lock);
1332 netlink_destroy_callback(cb);
1339 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1340 struct nlmsghdr *nlh,
1341 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1342 int (*done)(struct netlink_callback*))
1344 struct netlink_callback *cb;
1346 struct netlink_sock *nlk;
1348 cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1352 memset(cb, 0, sizeof(*cb));
1356 atomic_inc(&skb->users);
1359 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1361 netlink_destroy_callback(cb);
1362 return -ECONNREFUSED;
1365 /* A dump is in progress... */
1366 spin_lock(&nlk->cb_lock);
1368 spin_unlock(&nlk->cb_lock);
1369 netlink_destroy_callback(cb);
1374 spin_unlock(&nlk->cb_lock);
1381 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1383 struct sk_buff *skb;
1384 struct nlmsghdr *rep;
1385 struct nlmsgerr *errmsg;
1389 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1391 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1393 skb = alloc_skb(size, GFP_KERNEL);
1397 sk = netlink_lookup(in_skb->sk->sk_protocol,
1398 NETLINK_CB(in_skb).pid);
1400 sk->sk_err = ENOBUFS;
1401 sk->sk_error_report(sk);
1407 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1408 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1409 errmsg = NLMSG_DATA(rep);
1410 errmsg->error = err;
1411 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1412 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1415 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1416 struct nlmsghdr *, int *))
1418 unsigned int total_len;
1419 struct nlmsghdr *nlh;
1422 while (skb->len >= nlmsg_total_size(0)) {
1423 nlh = (struct nlmsghdr *) skb->data;
1425 if (skb->len < nlh->nlmsg_len)
1428 total_len = min(NLMSG_ALIGN(nlh->nlmsg_len), skb->len);
1430 if (cb(skb, nlh, &err) < 0) {
1431 /* Not an error, but we have to interrupt processing
1432 * here. Note: that in this case we do not pull
1433 * message from skb, it will be processed later.
1437 netlink_ack(skb, nlh, err);
1438 } else if (nlh->nlmsg_flags & NLM_F_ACK)
1439 netlink_ack(skb, nlh, 0);
1441 skb_pull(skb, total_len);
1448 * nelink_run_queue - Process netlink receive queue.
1449 * @sk: Netlink socket containing the queue
1450 * @qlen: Place to store queue length upon entry
1451 * @cb: Callback function invoked for each netlink message found
1453 * Processes as much as there was in the queue upon entry and invokes
1454 * a callback function for each netlink message found. The callback
1455 * function may refuse a message by returning a negative error code
1456 * but setting the error pointer to 0 in which case this function
1457 * returns with a qlen != 0.
1459 * qlen must be initialized to 0 before the initial entry, afterwards
1460 * the function may be called repeatedly until qlen reaches 0.
1462 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1463 int (*cb)(struct sk_buff *, struct nlmsghdr *, int *))
1465 struct sk_buff *skb;
1467 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1468 *qlen = skb_queue_len(&sk->sk_receive_queue);
1470 for (; *qlen; (*qlen)--) {
1471 skb = skb_dequeue(&sk->sk_receive_queue);
1472 if (netlink_rcv_skb(skb, cb)) {
1474 skb_queue_head(&sk->sk_receive_queue, skb);
1487 * netlink_queue_skip - Skip netlink message while processing queue.
1488 * @nlh: Netlink message to be skipped
1489 * @skb: Socket buffer containing the netlink messages.
1491 * Pulls the given netlink message off the socket buffer so the next
1492 * call to netlink_queue_run() will not reconsider the message.
1494 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1496 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1498 if (msglen > skb->len)
1501 skb_pull(skb, msglen);
1504 #ifdef CONFIG_PROC_FS
1505 struct nl_seq_iter {
1510 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1512 struct nl_seq_iter *iter = seq->private;
1515 struct hlist_node *node;
1518 for (i=0; i<MAX_LINKS; i++) {
1519 struct nl_pid_hash *hash = &nl_table[i].hash;
1521 for (j = 0; j <= hash->mask; j++) {
1522 sk_for_each(s, node, &hash->table[j]) {
1535 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1537 read_lock(&nl_table_lock);
1538 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1541 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1544 struct nl_seq_iter *iter;
1549 if (v == SEQ_START_TOKEN)
1550 return netlink_seq_socket_idx(seq, 0);
1556 iter = seq->private;
1558 j = iter->hash_idx + 1;
1561 struct nl_pid_hash *hash = &nl_table[i].hash;
1563 for (; j <= hash->mask; j++) {
1564 s = sk_head(&hash->table[j]);
1573 } while (++i < MAX_LINKS);
1578 static void netlink_seq_stop(struct seq_file *seq, void *v)
1580 read_unlock(&nl_table_lock);
1584 static int netlink_seq_show(struct seq_file *seq, void *v)
1586 if (v == SEQ_START_TOKEN)
1588 "sk Eth Pid Groups "
1589 "Rmem Wmem Dump Locks\n");
1592 struct netlink_sock *nlk = nlk_sk(s);
1594 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1598 nlk->groups ? (u32)nlk->groups[0] : 0,
1599 atomic_read(&s->sk_rmem_alloc),
1600 atomic_read(&s->sk_wmem_alloc),
1602 atomic_read(&s->sk_refcnt)
1609 static struct seq_operations netlink_seq_ops = {
1610 .start = netlink_seq_start,
1611 .next = netlink_seq_next,
1612 .stop = netlink_seq_stop,
1613 .show = netlink_seq_show,
1617 static int netlink_seq_open(struct inode *inode, struct file *file)
1619 struct seq_file *seq;
1620 struct nl_seq_iter *iter;
1623 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1627 err = seq_open(file, &netlink_seq_ops);
1633 memset(iter, 0, sizeof(*iter));
1634 seq = file->private_data;
1635 seq->private = iter;
1639 static struct file_operations netlink_seq_fops = {
1640 .owner = THIS_MODULE,
1641 .open = netlink_seq_open,
1643 .llseek = seq_lseek,
1644 .release = seq_release_private,
1649 int netlink_register_notifier(struct notifier_block *nb)
1651 return notifier_chain_register(&netlink_chain, nb);
1654 int netlink_unregister_notifier(struct notifier_block *nb)
1656 return notifier_chain_unregister(&netlink_chain, nb);
1659 static struct proto_ops netlink_ops = {
1660 .family = PF_NETLINK,
1661 .owner = THIS_MODULE,
1662 .release = netlink_release,
1663 .bind = netlink_bind,
1664 .connect = netlink_connect,
1665 .socketpair = sock_no_socketpair,
1666 .accept = sock_no_accept,
1667 .getname = netlink_getname,
1668 .poll = datagram_poll,
1669 .ioctl = sock_no_ioctl,
1670 .listen = sock_no_listen,
1671 .shutdown = sock_no_shutdown,
1672 .setsockopt = netlink_setsockopt,
1673 .getsockopt = netlink_getsockopt,
1674 .sendmsg = netlink_sendmsg,
1675 .recvmsg = netlink_recvmsg,
1676 .mmap = sock_no_mmap,
1677 .sendpage = sock_no_sendpage,
1680 static struct net_proto_family netlink_family_ops = {
1681 .family = PF_NETLINK,
1682 .create = netlink_create,
1683 .owner = THIS_MODULE, /* for consistency 8) */
1686 extern void netlink_skb_parms_too_large(void);
1688 static int __init netlink_proto_init(void)
1690 struct sk_buff *dummy_skb;
1694 int err = proto_register(&netlink_proto, 0);
1699 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1700 netlink_skb_parms_too_large();
1702 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1705 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1709 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1711 if (num_physpages >= (128 * 1024))
1712 max = num_physpages >> (21 - PAGE_SHIFT);
1714 max = num_physpages >> (23 - PAGE_SHIFT);
1716 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1717 max = (1UL << order) / sizeof(struct hlist_head);
1718 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1720 for (i = 0; i < MAX_LINKS; i++) {
1721 struct nl_pid_hash *hash = &nl_table[i].hash;
1723 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1726 nl_pid_hash_free(nl_table[i].hash.table,
1727 1 * sizeof(*hash->table));
1731 memset(hash->table, 0, 1 * sizeof(*hash->table));
1732 hash->max_shift = order;
1735 hash->rehash_time = jiffies;
1738 sock_register(&netlink_family_ops);
1739 #ifdef CONFIG_PROC_FS
1740 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1742 /* The netlink device handler may be needed early. */
1748 core_initcall(netlink_proto_init);
1750 EXPORT_SYMBOL(netlink_ack);
1751 EXPORT_SYMBOL(netlink_run_queue);
1752 EXPORT_SYMBOL(netlink_queue_skip);
1753 EXPORT_SYMBOL(netlink_broadcast);
1754 EXPORT_SYMBOL(netlink_dump_start);
1755 EXPORT_SYMBOL(netlink_kernel_create);
1756 EXPORT_SYMBOL(netlink_register_notifier);
1757 EXPORT_SYMBOL(netlink_set_err);
1758 EXPORT_SYMBOL(netlink_set_nonroot);
1759 EXPORT_SYMBOL(netlink_unicast);
1760 EXPORT_SYMBOL(netlink_unregister_notifier);