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)
19 #include <linux/config.h>
20 #include <linux/module.h>
22 #include <linux/kernel.h>
23 #include <linux/init.h>
24 #include <linux/signal.h>
25 #include <linux/sched.h>
26 #include <linux/errno.h>
27 #include <linux/string.h>
28 #include <linux/stat.h>
29 #include <linux/socket.h>
31 #include <linux/fcntl.h>
32 #include <linux/termios.h>
33 #include <linux/sockios.h>
34 #include <linux/net.h>
36 #include <linux/slab.h>
37 #include <asm/uaccess.h>
38 #include <linux/skbuff.h>
39 #include <linux/netdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/proc_fs.h>
42 #include <linux/seq_file.h>
43 #include <linux/smp_lock.h>
44 #include <linux/notifier.h>
45 #include <linux/security.h>
46 #include <linux/jhash.h>
47 #include <linux/jiffies.h>
48 #include <linux/random.h>
49 #include <linux/bitops.h>
51 #include <linux/types.h>
52 #include <linux/vs_context.h>
53 #include <linux/vs_network.h>
54 #include <linux/vs_limit.h>
55 #include <linux/audit.h>
56 #include <linux/vs_context.h>
57 #include <linux/vs_network.h>
58 #include <linux/vs_limit.h>
66 /* struct sock has to be the first member of netlink_sock */
71 unsigned int dst_groups;
73 wait_queue_head_t wait;
74 struct netlink_callback *cb;
76 void (*data_ready)(struct sock *sk, int bytes);
79 static inline struct netlink_sock *nlk_sk(struct sock *sk)
81 return (struct netlink_sock *)sk;
85 struct hlist_head *table;
86 unsigned long rehash_time;
92 unsigned int max_shift;
97 struct netlink_table {
98 struct nl_pid_hash hash;
99 struct hlist_head mc_list;
100 unsigned int nl_nonroot;
103 static struct netlink_table *nl_table;
105 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
107 static int netlink_dump(struct sock *sk);
108 static void netlink_destroy_callback(struct netlink_callback *cb);
110 static DEFINE_RWLOCK(nl_table_lock);
111 static atomic_t nl_table_users = ATOMIC_INIT(0);
113 static struct notifier_block *netlink_chain;
115 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
117 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
120 static void netlink_sock_destruct(struct sock *sk)
122 skb_queue_purge(&sk->sk_receive_queue);
124 if (!sock_flag(sk, SOCK_DEAD)) {
125 printk("Freeing alive netlink socket %p\n", sk);
128 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
129 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
130 BUG_TRAP(!nlk_sk(sk)->cb);
133 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
134 * Look, when several writers sleep and reader wakes them up, all but one
135 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
136 * this, _but_ remember, it adds useless work on UP machines.
139 static void netlink_table_grab(void)
141 write_lock_bh(&nl_table_lock);
143 if (atomic_read(&nl_table_users)) {
144 DECLARE_WAITQUEUE(wait, current);
146 add_wait_queue_exclusive(&nl_table_wait, &wait);
148 set_current_state(TASK_UNINTERRUPTIBLE);
149 if (atomic_read(&nl_table_users) == 0)
151 write_unlock_bh(&nl_table_lock);
153 write_lock_bh(&nl_table_lock);
156 __set_current_state(TASK_RUNNING);
157 remove_wait_queue(&nl_table_wait, &wait);
161 static __inline__ void netlink_table_ungrab(void)
163 write_unlock_bh(&nl_table_lock);
164 wake_up(&nl_table_wait);
167 static __inline__ void
168 netlink_lock_table(void)
170 /* read_lock() synchronizes us to netlink_table_grab */
172 read_lock(&nl_table_lock);
173 atomic_inc(&nl_table_users);
174 read_unlock(&nl_table_lock);
177 static __inline__ void
178 netlink_unlock_table(void)
180 if (atomic_dec_and_test(&nl_table_users))
181 wake_up(&nl_table_wait);
184 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
186 struct nl_pid_hash *hash = &nl_table[protocol].hash;
187 struct hlist_head *head;
189 struct hlist_node *node;
191 read_lock(&nl_table_lock);
192 head = nl_pid_hashfn(hash, pid);
193 sk_for_each(sk, node, head) {
194 if (nlk_sk(sk)->pid == pid) {
201 read_unlock(&nl_table_lock);
205 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
207 if (size <= PAGE_SIZE)
208 return kmalloc(size, GFP_ATOMIC);
210 return (struct hlist_head *)
211 __get_free_pages(GFP_ATOMIC, get_order(size));
214 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
216 if (size <= PAGE_SIZE)
219 free_pages((unsigned long)table, get_order(size));
222 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
224 unsigned int omask, mask, shift;
226 struct hlist_head *otable, *table;
229 omask = mask = hash->mask;
230 osize = size = (mask + 1) * sizeof(*table);
234 if (++shift > hash->max_shift)
240 table = nl_pid_hash_alloc(size);
244 memset(table, 0, size);
245 otable = hash->table;
249 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
251 for (i = 0; i <= omask; i++) {
253 struct hlist_node *node, *tmp;
255 sk_for_each_safe(sk, node, tmp, &otable[i])
256 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
259 nl_pid_hash_free(otable, osize);
260 hash->rehash_time = jiffies + 10 * 60 * HZ;
264 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
266 int avg = hash->entries >> hash->shift;
268 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
271 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
272 nl_pid_hash_rehash(hash, 0);
279 static struct proto_ops netlink_ops;
281 static int netlink_insert(struct sock *sk, u32 pid)
283 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
284 struct hlist_head *head;
285 int err = -EADDRINUSE;
287 struct hlist_node *node;
290 netlink_table_grab();
291 head = nl_pid_hashfn(hash, pid);
293 sk_for_each(osk, node, head) {
294 if (nlk_sk(osk)->pid == pid)
306 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
309 if (len && nl_pid_hash_dilute(hash, len))
310 head = nl_pid_hashfn(hash, pid);
312 nlk_sk(sk)->pid = pid;
313 sk_add_node(sk, head);
317 netlink_table_ungrab();
321 static void netlink_remove(struct sock *sk)
323 netlink_table_grab();
324 if (sk_del_node_init(sk))
325 nl_table[sk->sk_protocol].hash.entries--;
326 if (nlk_sk(sk)->groups)
327 __sk_del_bind_node(sk);
328 netlink_table_ungrab();
331 static struct proto netlink_proto = {
333 .owner = THIS_MODULE,
334 .obj_size = sizeof(struct netlink_sock),
337 static int netlink_create(struct socket *sock, int protocol)
340 struct netlink_sock *nlk;
342 sock->state = SS_UNCONNECTED;
344 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
345 return -ESOCKTNOSUPPORT;
347 if (protocol<0 || protocol >= MAX_LINKS)
348 return -EPROTONOSUPPORT;
350 sock->ops = &netlink_ops;
352 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
356 sock_init_data(sock, sk);
360 spin_lock_init(&nlk->cb_lock);
361 init_waitqueue_head(&nlk->wait);
362 sk->sk_destruct = netlink_sock_destruct;
364 sk->sk_protocol = protocol;
368 static int netlink_release(struct socket *sock)
370 struct sock *sk = sock->sk;
371 struct netlink_sock *nlk;
379 spin_lock(&nlk->cb_lock);
381 nlk->cb->done(nlk->cb);
382 netlink_destroy_callback(nlk->cb);
385 spin_unlock(&nlk->cb_lock);
387 /* OK. Socket is unlinked, and, therefore,
388 no new packets will arrive */
392 wake_up_interruptible_all(&nlk->wait);
394 skb_queue_purge(&sk->sk_write_queue);
396 if (nlk->pid && !nlk->groups) {
397 struct netlink_notify n = {
398 .protocol = sk->sk_protocol,
401 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
408 static int netlink_autobind(struct socket *sock)
410 struct sock *sk = sock->sk;
411 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
412 struct hlist_head *head;
414 struct hlist_node *node;
415 s32 pid = current->pid;
417 static s32 rover = -4097;
421 netlink_table_grab();
422 head = nl_pid_hashfn(hash, pid);
423 sk_for_each(osk, node, head) {
424 if (nlk_sk(osk)->pid == pid) {
425 /* Bind collision, search negative pid values. */
429 netlink_table_ungrab();
433 netlink_table_ungrab();
435 err = netlink_insert(sk, pid);
436 if (err == -EADDRINUSE)
439 /* If 2 threads race to autobind, that is fine. */
446 static inline int netlink_capable(struct socket *sock, unsigned int flag)
448 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
449 capable(CAP_NET_ADMIN);
452 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
454 struct sock *sk = sock->sk;
455 struct netlink_sock *nlk = nlk_sk(sk);
456 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
459 if (nladdr->nl_family != AF_NETLINK)
462 /* Only superuser is allowed to listen multicasts */
463 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_RECV))
467 if (nladdr->nl_pid != nlk->pid)
470 err = nladdr->nl_pid ?
471 netlink_insert(sk, nladdr->nl_pid) :
472 netlink_autobind(sock);
477 if (!nladdr->nl_groups && !nlk->groups)
480 netlink_table_grab();
481 if (nlk->groups && !nladdr->nl_groups)
482 __sk_del_bind_node(sk);
483 else if (!nlk->groups && nladdr->nl_groups)
484 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
485 nlk->groups = nladdr->nl_groups;
486 netlink_table_ungrab();
491 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
495 struct sock *sk = sock->sk;
496 struct netlink_sock *nlk = nlk_sk(sk);
497 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
499 if (addr->sa_family == AF_UNSPEC) {
500 sk->sk_state = NETLINK_UNCONNECTED;
505 if (addr->sa_family != AF_NETLINK)
508 /* Only superuser is allowed to send multicasts */
509 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
513 err = netlink_autobind(sock);
516 sk->sk_state = NETLINK_CONNECTED;
517 nlk->dst_pid = nladdr->nl_pid;
518 nlk->dst_groups = nladdr->nl_groups;
524 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
526 struct sock *sk = sock->sk;
527 struct netlink_sock *nlk = nlk_sk(sk);
528 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
530 nladdr->nl_family = AF_NETLINK;
532 *addr_len = sizeof(*nladdr);
535 nladdr->nl_pid = nlk->dst_pid;
536 nladdr->nl_groups = nlk->dst_groups;
538 nladdr->nl_pid = nlk->pid;
539 nladdr->nl_groups = nlk->groups;
544 static void netlink_overrun(struct sock *sk)
546 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
547 sk->sk_err = ENOBUFS;
548 sk->sk_error_report(sk);
552 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
554 int protocol = ssk->sk_protocol;
556 struct netlink_sock *nlk;
558 sock = netlink_lookup(protocol, pid);
560 return ERR_PTR(-ECONNREFUSED);
562 /* Don't bother queuing skb if kernel socket has no input function */
564 if ((nlk->pid == 0 && !nlk->data_ready) ||
565 (sock->sk_state == NETLINK_CONNECTED &&
566 nlk->dst_pid != nlk_sk(ssk)->pid)) {
568 return ERR_PTR(-ECONNREFUSED);
573 struct sock *netlink_getsockbyfilp(struct file *filp)
575 struct inode *inode = filp->f_dentry->d_inode;
578 if (!S_ISSOCK(inode->i_mode))
579 return ERR_PTR(-ENOTSOCK);
581 sock = SOCKET_I(inode)->sk;
582 if (sock->sk_family != AF_NETLINK)
583 return ERR_PTR(-EINVAL);
590 * Attach a skb to a netlink socket.
591 * The caller must hold a reference to the destination socket. On error, the
592 * reference is dropped. The skb is not send to the destination, just all
593 * all error checks are performed and memory in the queue is reserved.
595 * < 0: error. skb freed, reference to sock dropped.
597 * 1: repeat lookup - reference dropped while waiting for socket memory.
599 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo)
601 struct netlink_sock *nlk;
605 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
606 test_bit(0, &nlk->state)) {
607 DECLARE_WAITQUEUE(wait, current);
616 __set_current_state(TASK_INTERRUPTIBLE);
617 add_wait_queue(&nlk->wait, &wait);
619 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
620 test_bit(0, &nlk->state)) &&
621 !sock_flag(sk, SOCK_DEAD))
622 timeo = schedule_timeout(timeo);
624 __set_current_state(TASK_RUNNING);
625 remove_wait_queue(&nlk->wait, &wait);
628 if (signal_pending(current)) {
630 return sock_intr_errno(timeo);
634 skb_set_owner_r(skb, sk);
638 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
640 struct netlink_sock *nlk;
645 skb_queue_tail(&sk->sk_receive_queue, skb);
646 sk->sk_data_ready(sk, len);
651 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
657 static inline struct sk_buff *netlink_trim(struct sk_buff *skb, int allocation)
663 delta = skb->end - skb->tail;
664 if (delta * 2 < skb->truesize)
667 if (skb_shared(skb)) {
668 struct sk_buff *nskb = skb_clone(skb, allocation);
675 if (!pskb_expand_head(skb, 0, -delta, allocation))
676 skb->truesize -= delta;
681 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
687 skb = netlink_trim(skb, gfp_any());
689 timeo = sock_sndtimeo(ssk, nonblock);
691 sk = netlink_getsockbypid(ssk, pid);
696 err = netlink_attachskb(sk, skb, nonblock, timeo);
702 return netlink_sendskb(sk, skb, ssk->sk_protocol);
705 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
707 struct netlink_sock *nlk = nlk_sk(sk);
709 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
710 !test_bit(0, &nlk->state)) {
711 skb_set_owner_r(skb, sk);
712 skb_queue_tail(&sk->sk_receive_queue, skb);
713 sk->sk_data_ready(sk, skb->len);
714 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
719 struct netlink_broadcast_data {
720 struct sock *exclude_sk;
727 struct sk_buff *skb, *skb2;
730 static inline int do_one_broadcast(struct sock *sk,
731 struct netlink_broadcast_data *p)
733 struct netlink_sock *nlk = nlk_sk(sk);
736 if (p->exclude_sk == sk)
739 if (nlk->pid == p->pid || !(nlk->groups & p->group))
748 if (p->skb2 == NULL) {
749 if (skb_shared(p->skb)) {
750 p->skb2 = skb_clone(p->skb, p->allocation);
752 p->skb2 = skb_get(p->skb);
754 * skb ownership may have been set when
755 * delivered to a previous socket.
760 if (p->skb2 == NULL) {
762 /* Clone failed. Notify ALL listeners. */
764 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
777 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
778 u32 group, int allocation)
780 struct netlink_broadcast_data info;
781 struct hlist_node *node;
784 skb = netlink_trim(skb, allocation);
786 info.exclude_sk = ssk;
792 info.allocation = allocation;
796 /* While we sleep in clone, do not allow to change socket list */
798 netlink_lock_table();
800 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
801 do_one_broadcast(sk, &info);
805 netlink_unlock_table();
808 kfree_skb(info.skb2);
810 if (info.delivered) {
811 if (info.congested && (allocation & __GFP_WAIT))
820 struct netlink_set_err_data {
821 struct sock *exclude_sk;
827 static inline int do_one_set_err(struct sock *sk,
828 struct netlink_set_err_data *p)
830 struct netlink_sock *nlk = nlk_sk(sk);
832 if (sk == p->exclude_sk)
835 if (nlk->pid == p->pid || !(nlk->groups & p->group))
838 sk->sk_err = p->code;
839 sk->sk_error_report(sk);
844 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
846 struct netlink_set_err_data info;
847 struct hlist_node *node;
850 info.exclude_sk = ssk;
855 read_lock(&nl_table_lock);
857 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
858 do_one_set_err(sk, &info);
860 read_unlock(&nl_table_lock);
863 static inline void netlink_rcv_wake(struct sock *sk)
865 struct netlink_sock *nlk = nlk_sk(sk);
867 if (!skb_queue_len(&sk->sk_receive_queue))
868 clear_bit(0, &nlk->state);
869 if (!test_bit(0, &nlk->state))
870 wake_up_interruptible(&nlk->wait);
873 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
874 struct msghdr *msg, size_t len)
876 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
877 struct sock *sk = sock->sk;
878 struct netlink_sock *nlk = nlk_sk(sk);
879 struct sockaddr_nl *addr=msg->msg_name;
884 struct scm_cookie scm;
886 if (msg->msg_flags&MSG_OOB)
889 if (NULL == siocb->scm)
891 err = scm_send(sock, msg, siocb->scm);
895 if (msg->msg_namelen) {
896 if (addr->nl_family != AF_NETLINK)
898 dst_pid = addr->nl_pid;
899 dst_groups = addr->nl_groups;
900 if (dst_groups && !netlink_capable(sock, NL_NONROOT_SEND))
903 dst_pid = nlk->dst_pid;
904 dst_groups = nlk->dst_groups;
908 err = netlink_autobind(sock);
914 if (len > sk->sk_sndbuf - 32)
917 skb = alloc_skb(len, GFP_KERNEL);
921 NETLINK_CB(skb).pid = nlk->pid;
922 NETLINK_CB(skb).groups = nlk->groups;
923 NETLINK_CB(skb).dst_pid = dst_pid;
924 NETLINK_CB(skb).dst_groups = dst_groups;
925 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
926 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
928 /* What can I do? Netlink is asynchronous, so that
929 we will have to save current capabilities to
930 check them, when this message will be delivered
931 to corresponding kernel module. --ANK (980802)
935 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
940 err = security_netlink_send(sk, skb);
947 atomic_inc(&skb->users);
948 netlink_broadcast(sk, skb, dst_pid, dst_groups, GFP_KERNEL);
950 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
956 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
957 struct msghdr *msg, size_t len,
960 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
961 struct scm_cookie scm;
962 struct sock *sk = sock->sk;
963 struct netlink_sock *nlk = nlk_sk(sk);
964 int noblock = flags&MSG_DONTWAIT;
974 skb = skb_recv_datagram(sk,flags,noblock,&err);
978 msg->msg_namelen = 0;
982 msg->msg_flags |= MSG_TRUNC;
986 skb->h.raw = skb->data;
987 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
990 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
991 addr->nl_family = AF_NETLINK;
993 addr->nl_pid = NETLINK_CB(skb).pid;
994 addr->nl_groups = NETLINK_CB(skb).dst_groups;
995 msg->msg_namelen = sizeof(*addr);
998 if (NULL == siocb->scm) {
999 memset(&scm, 0, sizeof(scm));
1002 siocb->scm->creds = *NETLINK_CREDS(skb);
1003 skb_free_datagram(sk, skb);
1005 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1008 scm_recv(sock, msg, siocb->scm, flags);
1011 netlink_rcv_wake(sk);
1012 return err ? : copied;
1015 static void netlink_data_ready(struct sock *sk, int len)
1017 struct netlink_sock *nlk = nlk_sk(sk);
1019 if (nlk->data_ready)
1020 nlk->data_ready(sk, len);
1021 netlink_rcv_wake(sk);
1025 * We export these functions to other modules. They provide a
1026 * complete set of kernel non-blocking support for message
1031 netlink_kernel_create(int unit, void (*input)(struct sock *sk, int len))
1033 struct socket *sock;
1039 if (unit<0 || unit>=MAX_LINKS)
1042 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1045 if (netlink_create(sock, unit) < 0) {
1050 sk->sk_data_ready = netlink_data_ready;
1052 nlk_sk(sk)->data_ready = input;
1054 if (netlink_insert(sk, 0)) {
1061 void netlink_set_nonroot(int protocol, unsigned int flags)
1063 if ((unsigned int)protocol < MAX_LINKS)
1064 nl_table[protocol].nl_nonroot = flags;
1067 static void netlink_destroy_callback(struct netlink_callback *cb)
1075 * It looks a bit ugly.
1076 * It would be better to create kernel thread.
1079 static int netlink_dump(struct sock *sk)
1081 struct netlink_sock *nlk = nlk_sk(sk);
1082 struct netlink_callback *cb;
1083 struct sk_buff *skb;
1084 struct nlmsghdr *nlh;
1087 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1091 spin_lock(&nlk->cb_lock);
1095 spin_unlock(&nlk->cb_lock);
1100 len = cb->dump(skb, cb);
1103 spin_unlock(&nlk->cb_lock);
1104 skb_queue_tail(&sk->sk_receive_queue, skb);
1105 sk->sk_data_ready(sk, len);
1109 nlh = __nlmsg_put(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq, NLMSG_DONE, sizeof(int));
1110 nlh->nlmsg_flags |= NLM_F_MULTI;
1111 memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1112 skb_queue_tail(&sk->sk_receive_queue, skb);
1113 sk->sk_data_ready(sk, skb->len);
1117 spin_unlock(&nlk->cb_lock);
1119 netlink_destroy_callback(cb);
1123 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1124 struct nlmsghdr *nlh,
1125 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1126 int (*done)(struct netlink_callback*))
1128 struct netlink_callback *cb;
1130 struct netlink_sock *nlk;
1132 cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1136 memset(cb, 0, sizeof(*cb));
1140 atomic_inc(&skb->users);
1143 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1145 netlink_destroy_callback(cb);
1146 return -ECONNREFUSED;
1149 /* A dump is in progress... */
1150 spin_lock(&nlk->cb_lock);
1152 spin_unlock(&nlk->cb_lock);
1153 netlink_destroy_callback(cb);
1158 spin_unlock(&nlk->cb_lock);
1165 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1167 struct sk_buff *skb;
1168 struct nlmsghdr *rep;
1169 struct nlmsgerr *errmsg;
1173 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1175 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1177 skb = alloc_skb(size, GFP_KERNEL);
1181 sk = netlink_lookup(in_skb->sk->sk_protocol,
1182 NETLINK_CB(in_skb).pid);
1184 sk->sk_err = ENOBUFS;
1185 sk->sk_error_report(sk);
1191 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1192 NLMSG_ERROR, sizeof(struct nlmsgerr));
1193 errmsg = NLMSG_DATA(rep);
1194 errmsg->error = err;
1195 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1196 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1200 #ifdef CONFIG_PROC_FS
1201 struct nl_seq_iter {
1206 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1208 struct nl_seq_iter *iter = seq->private;
1211 struct hlist_node *node;
1214 for (i=0; i<MAX_LINKS; i++) {
1215 struct nl_pid_hash *hash = &nl_table[i].hash;
1217 for (j = 0; j <= hash->mask; j++) {
1218 sk_for_each(s, node, &hash->table[j]) {
1231 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1233 read_lock(&nl_table_lock);
1234 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1237 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1240 struct nl_seq_iter *iter;
1245 if (v == SEQ_START_TOKEN)
1246 return netlink_seq_socket_idx(seq, 0);
1252 iter = seq->private;
1254 j = iter->hash_idx + 1;
1257 struct nl_pid_hash *hash = &nl_table[i].hash;
1259 for (; j <= hash->mask; j++) {
1260 s = sk_head(&hash->table[j]);
1269 } while (++i < MAX_LINKS);
1274 static void netlink_seq_stop(struct seq_file *seq, void *v)
1276 read_unlock(&nl_table_lock);
1280 static int netlink_seq_show(struct seq_file *seq, void *v)
1282 if (v == SEQ_START_TOKEN)
1284 "sk Eth Pid Groups "
1285 "Rmem Wmem Dump Locks\n");
1288 struct netlink_sock *nlk = nlk_sk(s);
1290 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1295 atomic_read(&s->sk_rmem_alloc),
1296 atomic_read(&s->sk_wmem_alloc),
1298 atomic_read(&s->sk_refcnt)
1305 static struct seq_operations netlink_seq_ops = {
1306 .start = netlink_seq_start,
1307 .next = netlink_seq_next,
1308 .stop = netlink_seq_stop,
1309 .show = netlink_seq_show,
1313 static int netlink_seq_open(struct inode *inode, struct file *file)
1315 struct seq_file *seq;
1316 struct nl_seq_iter *iter;
1319 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1323 err = seq_open(file, &netlink_seq_ops);
1329 memset(iter, 0, sizeof(*iter));
1330 seq = file->private_data;
1331 seq->private = iter;
1335 static struct file_operations netlink_seq_fops = {
1336 .owner = THIS_MODULE,
1337 .open = netlink_seq_open,
1339 .llseek = seq_lseek,
1340 .release = seq_release_private,
1345 int netlink_register_notifier(struct notifier_block *nb)
1347 return notifier_chain_register(&netlink_chain, nb);
1350 int netlink_unregister_notifier(struct notifier_block *nb)
1352 return notifier_chain_unregister(&netlink_chain, nb);
1355 static struct proto_ops netlink_ops = {
1356 .family = PF_NETLINK,
1357 .owner = THIS_MODULE,
1358 .release = netlink_release,
1359 .bind = netlink_bind,
1360 .connect = netlink_connect,
1361 .socketpair = sock_no_socketpair,
1362 .accept = sock_no_accept,
1363 .getname = netlink_getname,
1364 .poll = datagram_poll,
1365 .ioctl = sock_no_ioctl,
1366 .listen = sock_no_listen,
1367 .shutdown = sock_no_shutdown,
1368 .setsockopt = sock_no_setsockopt,
1369 .getsockopt = sock_no_getsockopt,
1370 .sendmsg = netlink_sendmsg,
1371 .recvmsg = netlink_recvmsg,
1372 .mmap = sock_no_mmap,
1373 .sendpage = sock_no_sendpage,
1376 static struct net_proto_family netlink_family_ops = {
1377 .family = PF_NETLINK,
1378 .create = netlink_create,
1379 .owner = THIS_MODULE, /* for consistency 8) */
1382 extern void netlink_skb_parms_too_large(void);
1384 static int __init netlink_proto_init(void)
1386 struct sk_buff *dummy_skb;
1390 int err = proto_register(&netlink_proto, 0);
1395 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1396 netlink_skb_parms_too_large();
1398 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1401 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1405 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1407 if (num_physpages >= (128 * 1024))
1408 max = num_physpages >> (21 - PAGE_SHIFT);
1410 max = num_physpages >> (23 - PAGE_SHIFT);
1412 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1413 max = (1UL << order) / sizeof(struct hlist_head);
1414 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1416 for (i = 0; i < MAX_LINKS; i++) {
1417 struct nl_pid_hash *hash = &nl_table[i].hash;
1419 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1422 nl_pid_hash_free(nl_table[i].hash.table,
1423 1 * sizeof(*hash->table));
1427 memset(hash->table, 0, 1 * sizeof(*hash->table));
1428 hash->max_shift = order;
1431 hash->rehash_time = jiffies;
1434 sock_register(&netlink_family_ops);
1435 #ifdef CONFIG_PROC_FS
1436 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1438 /* The netlink device handler may be needed early. */
1444 static void __exit netlink_proto_exit(void)
1446 sock_unregister(PF_NETLINK);
1447 proc_net_remove("netlink");
1450 proto_unregister(&netlink_proto);
1453 core_initcall(netlink_proto_init);
1454 module_exit(netlink_proto_exit);
1456 MODULE_LICENSE("GPL");
1458 MODULE_ALIAS_NETPROTO(PF_NETLINK);
1460 EXPORT_SYMBOL(netlink_ack);
1461 EXPORT_SYMBOL(netlink_broadcast);
1462 EXPORT_SYMBOL(netlink_dump_start);
1463 EXPORT_SYMBOL(netlink_kernel_create);
1464 EXPORT_SYMBOL(netlink_register_notifier);
1465 EXPORT_SYMBOL(netlink_set_err);
1466 EXPORT_SYMBOL(netlink_set_nonroot);
1467 EXPORT_SYMBOL(netlink_unicast);
1468 EXPORT_SYMBOL(netlink_unregister_notifier);