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/audit.h>
53 #include <linux/vs_context.h>
54 #include <linux/vs_network.h>
55 #include <linux/vs_limit.h>
63 /* struct sock has to be the first member of netlink_sock */
68 unsigned int dst_groups;
70 wait_queue_head_t wait;
71 struct netlink_callback *cb;
73 void (*data_ready)(struct sock *sk, int bytes);
76 static inline struct netlink_sock *nlk_sk(struct sock *sk)
78 return (struct netlink_sock *)sk;
82 struct hlist_head *table;
83 unsigned long rehash_time;
89 unsigned int max_shift;
94 struct netlink_table {
95 struct nl_pid_hash hash;
96 struct hlist_head mc_list;
97 unsigned int nl_nonroot;
100 static struct netlink_table *nl_table;
102 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
104 static int netlink_dump(struct sock *sk);
105 static void netlink_destroy_callback(struct netlink_callback *cb);
107 static DEFINE_RWLOCK(nl_table_lock);
108 static atomic_t nl_table_users = ATOMIC_INIT(0);
110 static struct notifier_block *netlink_chain;
112 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
114 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
117 static void netlink_sock_destruct(struct sock *sk)
119 skb_queue_purge(&sk->sk_receive_queue);
121 if (!sock_flag(sk, SOCK_DEAD)) {
122 printk("Freeing alive netlink socket %p\n", sk);
125 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
126 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
127 BUG_TRAP(!nlk_sk(sk)->cb);
130 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
131 * Look, when several writers sleep and reader wakes them up, all but one
132 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
133 * this, _but_ remember, it adds useless work on UP machines.
136 static void netlink_table_grab(void)
138 write_lock_bh(&nl_table_lock);
140 if (atomic_read(&nl_table_users)) {
141 DECLARE_WAITQUEUE(wait, current);
143 add_wait_queue_exclusive(&nl_table_wait, &wait);
145 set_current_state(TASK_UNINTERRUPTIBLE);
146 if (atomic_read(&nl_table_users) == 0)
148 write_unlock_bh(&nl_table_lock);
150 write_lock_bh(&nl_table_lock);
153 __set_current_state(TASK_RUNNING);
154 remove_wait_queue(&nl_table_wait, &wait);
158 static __inline__ void netlink_table_ungrab(void)
160 write_unlock_bh(&nl_table_lock);
161 wake_up(&nl_table_wait);
164 static __inline__ void
165 netlink_lock_table(void)
167 /* read_lock() synchronizes us to netlink_table_grab */
169 read_lock(&nl_table_lock);
170 atomic_inc(&nl_table_users);
171 read_unlock(&nl_table_lock);
174 static __inline__ void
175 netlink_unlock_table(void)
177 if (atomic_dec_and_test(&nl_table_users))
178 wake_up(&nl_table_wait);
181 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
183 struct nl_pid_hash *hash = &nl_table[protocol].hash;
184 struct hlist_head *head;
186 struct hlist_node *node;
188 read_lock(&nl_table_lock);
189 head = nl_pid_hashfn(hash, pid);
190 sk_for_each(sk, node, head) {
191 if (nlk_sk(sk)->pid == pid) {
198 read_unlock(&nl_table_lock);
202 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
204 if (size <= PAGE_SIZE)
205 return kmalloc(size, GFP_ATOMIC);
207 return (struct hlist_head *)
208 __get_free_pages(GFP_ATOMIC, get_order(size));
211 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
213 if (size <= PAGE_SIZE)
216 free_pages((unsigned long)table, get_order(size));
219 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
221 unsigned int omask, mask, shift;
223 struct hlist_head *otable, *table;
226 omask = mask = hash->mask;
227 osize = size = (mask + 1) * sizeof(*table);
231 if (++shift > hash->max_shift)
237 table = nl_pid_hash_alloc(size);
241 memset(table, 0, size);
242 otable = hash->table;
246 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
248 for (i = 0; i <= omask; i++) {
250 struct hlist_node *node, *tmp;
252 sk_for_each_safe(sk, node, tmp, &otable[i])
253 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
256 nl_pid_hash_free(otable, osize);
257 hash->rehash_time = jiffies + 10 * 60 * HZ;
261 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
263 int avg = hash->entries >> hash->shift;
265 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
268 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
269 nl_pid_hash_rehash(hash, 0);
276 static struct proto_ops netlink_ops;
278 static int netlink_insert(struct sock *sk, u32 pid)
280 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
281 struct hlist_head *head;
282 int err = -EADDRINUSE;
284 struct hlist_node *node;
287 netlink_table_grab();
288 head = nl_pid_hashfn(hash, pid);
290 sk_for_each(osk, node, head) {
291 if (nlk_sk(osk)->pid == pid)
303 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
306 if (len && nl_pid_hash_dilute(hash, len))
307 head = nl_pid_hashfn(hash, pid);
309 nlk_sk(sk)->pid = pid;
310 sk_add_node(sk, head);
314 netlink_table_ungrab();
318 static void netlink_remove(struct sock *sk)
320 netlink_table_grab();
321 if (sk_del_node_init(sk))
322 nl_table[sk->sk_protocol].hash.entries--;
323 if (nlk_sk(sk)->groups)
324 __sk_del_bind_node(sk);
325 netlink_table_ungrab();
328 static struct proto netlink_proto = {
330 .owner = THIS_MODULE,
331 .obj_size = sizeof(struct netlink_sock),
334 static int netlink_create(struct socket *sock, int protocol)
337 struct netlink_sock *nlk;
339 sock->state = SS_UNCONNECTED;
341 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
342 return -ESOCKTNOSUPPORT;
344 if (protocol<0 || protocol >= MAX_LINKS)
345 return -EPROTONOSUPPORT;
347 sock->ops = &netlink_ops;
349 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
353 sock_init_data(sock, sk);
357 spin_lock_init(&nlk->cb_lock);
358 init_waitqueue_head(&nlk->wait);
359 sk->sk_destruct = netlink_sock_destruct;
361 sk->sk_protocol = protocol;
365 static int netlink_release(struct socket *sock)
367 struct sock *sk = sock->sk;
368 struct netlink_sock *nlk;
376 spin_lock(&nlk->cb_lock);
378 nlk->cb->done(nlk->cb);
379 netlink_destroy_callback(nlk->cb);
382 spin_unlock(&nlk->cb_lock);
384 /* OK. Socket is unlinked, and, therefore,
385 no new packets will arrive */
389 wake_up_interruptible_all(&nlk->wait);
391 skb_queue_purge(&sk->sk_write_queue);
393 if (nlk->pid && !nlk->groups) {
394 struct netlink_notify n = {
395 .protocol = sk->sk_protocol,
398 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
405 static int netlink_autobind(struct socket *sock)
407 struct sock *sk = sock->sk;
408 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
409 struct hlist_head *head;
411 struct hlist_node *node;
412 s32 pid = current->pid;
414 static s32 rover = -4097;
418 netlink_table_grab();
419 head = nl_pid_hashfn(hash, pid);
420 sk_for_each(osk, node, head) {
421 if (nlk_sk(osk)->pid == pid) {
422 /* Bind collision, search negative pid values. */
426 netlink_table_ungrab();
430 netlink_table_ungrab();
432 err = netlink_insert(sk, pid);
433 if (err == -EADDRINUSE)
436 /* If 2 threads race to autobind, that is fine. */
443 static inline int netlink_capable(struct socket *sock, unsigned int flag)
445 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
446 capable(CAP_NET_ADMIN);
449 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
451 struct sock *sk = sock->sk;
452 struct netlink_sock *nlk = nlk_sk(sk);
453 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
456 if (nladdr->nl_family != AF_NETLINK)
459 /* Only superuser is allowed to listen multicasts */
460 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_RECV))
464 if (nladdr->nl_pid != nlk->pid)
467 err = nladdr->nl_pid ?
468 netlink_insert(sk, nladdr->nl_pid) :
469 netlink_autobind(sock);
474 if (!nladdr->nl_groups && !nlk->groups)
477 netlink_table_grab();
478 if (nlk->groups && !nladdr->nl_groups)
479 __sk_del_bind_node(sk);
480 else if (!nlk->groups && nladdr->nl_groups)
481 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
482 nlk->groups = nladdr->nl_groups;
483 netlink_table_ungrab();
488 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
492 struct sock *sk = sock->sk;
493 struct netlink_sock *nlk = nlk_sk(sk);
494 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
496 if (addr->sa_family == AF_UNSPEC) {
497 sk->sk_state = NETLINK_UNCONNECTED;
502 if (addr->sa_family != AF_NETLINK)
505 /* Only superuser is allowed to send multicasts */
506 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
510 err = netlink_autobind(sock);
513 sk->sk_state = NETLINK_CONNECTED;
514 nlk->dst_pid = nladdr->nl_pid;
515 nlk->dst_groups = nladdr->nl_groups;
521 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
523 struct sock *sk = sock->sk;
524 struct netlink_sock *nlk = nlk_sk(sk);
525 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
527 nladdr->nl_family = AF_NETLINK;
529 *addr_len = sizeof(*nladdr);
532 nladdr->nl_pid = nlk->dst_pid;
533 nladdr->nl_groups = nlk->dst_groups;
535 nladdr->nl_pid = nlk->pid;
536 nladdr->nl_groups = nlk->groups;
541 static void netlink_overrun(struct sock *sk)
543 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
544 sk->sk_err = ENOBUFS;
545 sk->sk_error_report(sk);
549 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
551 int protocol = ssk->sk_protocol;
553 struct netlink_sock *nlk;
555 sock = netlink_lookup(protocol, pid);
557 return ERR_PTR(-ECONNREFUSED);
559 /* Don't bother queuing skb if kernel socket has no input function */
561 if ((nlk->pid == 0 && !nlk->data_ready) ||
562 (sock->sk_state == NETLINK_CONNECTED &&
563 nlk->dst_pid != nlk_sk(ssk)->pid)) {
565 return ERR_PTR(-ECONNREFUSED);
570 struct sock *netlink_getsockbyfilp(struct file *filp)
572 struct inode *inode = filp->f_dentry->d_inode;
575 if (!S_ISSOCK(inode->i_mode))
576 return ERR_PTR(-ENOTSOCK);
578 sock = SOCKET_I(inode)->sk;
579 if (sock->sk_family != AF_NETLINK)
580 return ERR_PTR(-EINVAL);
587 * Attach a skb to a netlink socket.
588 * The caller must hold a reference to the destination socket. On error, the
589 * reference is dropped. The skb is not send to the destination, just all
590 * all error checks are performed and memory in the queue is reserved.
592 * < 0: error. skb freed, reference to sock dropped.
594 * 1: repeat lookup - reference dropped while waiting for socket memory.
596 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo)
598 struct netlink_sock *nlk;
602 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
603 test_bit(0, &nlk->state)) {
604 DECLARE_WAITQUEUE(wait, current);
613 __set_current_state(TASK_INTERRUPTIBLE);
614 add_wait_queue(&nlk->wait, &wait);
616 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
617 test_bit(0, &nlk->state)) &&
618 !sock_flag(sk, SOCK_DEAD))
619 timeo = schedule_timeout(timeo);
621 __set_current_state(TASK_RUNNING);
622 remove_wait_queue(&nlk->wait, &wait);
625 if (signal_pending(current)) {
627 return sock_intr_errno(timeo);
631 skb_set_owner_r(skb, sk);
635 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
637 struct netlink_sock *nlk;
642 skb_queue_tail(&sk->sk_receive_queue, skb);
643 sk->sk_data_ready(sk, len);
648 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
654 static inline struct sk_buff *netlink_trim(struct sk_buff *skb, int allocation)
660 delta = skb->end - skb->tail;
661 if (delta * 2 < skb->truesize)
664 if (skb_shared(skb)) {
665 struct sk_buff *nskb = skb_clone(skb, allocation);
672 if (!pskb_expand_head(skb, 0, -delta, allocation))
673 skb->truesize -= delta;
678 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
684 skb = netlink_trim(skb, gfp_any());
686 timeo = sock_sndtimeo(ssk, nonblock);
688 sk = netlink_getsockbypid(ssk, pid);
693 err = netlink_attachskb(sk, skb, nonblock, timeo);
699 return netlink_sendskb(sk, skb, ssk->sk_protocol);
702 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
704 struct netlink_sock *nlk = nlk_sk(sk);
706 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
707 !test_bit(0, &nlk->state)) {
708 skb_set_owner_r(skb, sk);
709 skb_queue_tail(&sk->sk_receive_queue, skb);
710 sk->sk_data_ready(sk, skb->len);
711 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
716 struct netlink_broadcast_data {
717 struct sock *exclude_sk;
724 struct sk_buff *skb, *skb2;
727 static inline int do_one_broadcast(struct sock *sk,
728 struct netlink_broadcast_data *p)
730 struct netlink_sock *nlk = nlk_sk(sk);
733 if (p->exclude_sk == sk)
736 if (nlk->pid == p->pid || !(nlk->groups & p->group))
745 if (p->skb2 == NULL) {
746 if (skb_shared(p->skb)) {
747 p->skb2 = skb_clone(p->skb, p->allocation);
749 p->skb2 = skb_get(p->skb);
751 * skb ownership may have been set when
752 * delivered to a previous socket.
757 if (p->skb2 == NULL) {
759 /* Clone failed. Notify ALL listeners. */
761 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
774 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
775 u32 group, int allocation)
777 struct netlink_broadcast_data info;
778 struct hlist_node *node;
781 skb = netlink_trim(skb, allocation);
783 info.exclude_sk = ssk;
789 info.allocation = allocation;
793 /* While we sleep in clone, do not allow to change socket list */
795 netlink_lock_table();
797 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
798 do_one_broadcast(sk, &info);
802 netlink_unlock_table();
805 kfree_skb(info.skb2);
807 if (info.delivered) {
808 if (info.congested && (allocation & __GFP_WAIT))
817 struct netlink_set_err_data {
818 struct sock *exclude_sk;
824 static inline int do_one_set_err(struct sock *sk,
825 struct netlink_set_err_data *p)
827 struct netlink_sock *nlk = nlk_sk(sk);
829 if (sk == p->exclude_sk)
832 if (nlk->pid == p->pid || !(nlk->groups & p->group))
835 sk->sk_err = p->code;
836 sk->sk_error_report(sk);
841 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
843 struct netlink_set_err_data info;
844 struct hlist_node *node;
847 info.exclude_sk = ssk;
852 read_lock(&nl_table_lock);
854 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
855 do_one_set_err(sk, &info);
857 read_unlock(&nl_table_lock);
860 static inline void netlink_rcv_wake(struct sock *sk)
862 struct netlink_sock *nlk = nlk_sk(sk);
864 if (!skb_queue_len(&sk->sk_receive_queue))
865 clear_bit(0, &nlk->state);
866 if (!test_bit(0, &nlk->state))
867 wake_up_interruptible(&nlk->wait);
870 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
871 struct msghdr *msg, size_t len)
873 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
874 struct sock *sk = sock->sk;
875 struct netlink_sock *nlk = nlk_sk(sk);
876 struct sockaddr_nl *addr=msg->msg_name;
881 struct scm_cookie scm;
883 if (msg->msg_flags&MSG_OOB)
886 if (NULL == siocb->scm)
888 err = scm_send(sock, msg, siocb->scm);
892 if (msg->msg_namelen) {
893 if (addr->nl_family != AF_NETLINK)
895 dst_pid = addr->nl_pid;
896 dst_groups = addr->nl_groups;
897 if (dst_groups && !netlink_capable(sock, NL_NONROOT_SEND))
900 dst_pid = nlk->dst_pid;
901 dst_groups = nlk->dst_groups;
905 err = netlink_autobind(sock);
911 if (len > sk->sk_sndbuf - 32)
914 skb = alloc_skb(len, GFP_KERNEL);
918 NETLINK_CB(skb).pid = nlk->pid;
919 NETLINK_CB(skb).groups = nlk->groups;
920 NETLINK_CB(skb).dst_pid = dst_pid;
921 NETLINK_CB(skb).dst_groups = dst_groups;
922 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
923 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
925 /* What can I do? Netlink is asynchronous, so that
926 we will have to save current capabilities to
927 check them, when this message will be delivered
928 to corresponding kernel module. --ANK (980802)
932 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
937 err = security_netlink_send(sk, skb);
944 atomic_inc(&skb->users);
945 netlink_broadcast(sk, skb, dst_pid, dst_groups, GFP_KERNEL);
947 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
953 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
954 struct msghdr *msg, size_t len,
957 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
958 struct scm_cookie scm;
959 struct sock *sk = sock->sk;
960 struct netlink_sock *nlk = nlk_sk(sk);
961 int noblock = flags&MSG_DONTWAIT;
971 skb = skb_recv_datagram(sk,flags,noblock,&err);
975 msg->msg_namelen = 0;
979 msg->msg_flags |= MSG_TRUNC;
983 skb->h.raw = skb->data;
984 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
987 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
988 addr->nl_family = AF_NETLINK;
990 addr->nl_pid = NETLINK_CB(skb).pid;
991 addr->nl_groups = NETLINK_CB(skb).dst_groups;
992 msg->msg_namelen = sizeof(*addr);
995 if (NULL == siocb->scm) {
996 memset(&scm, 0, sizeof(scm));
999 siocb->scm->creds = *NETLINK_CREDS(skb);
1000 skb_free_datagram(sk, skb);
1002 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1005 scm_recv(sock, msg, siocb->scm, flags);
1008 netlink_rcv_wake(sk);
1009 return err ? : copied;
1012 static void netlink_data_ready(struct sock *sk, int len)
1014 struct netlink_sock *nlk = nlk_sk(sk);
1016 if (nlk->data_ready)
1017 nlk->data_ready(sk, len);
1018 netlink_rcv_wake(sk);
1022 * We export these functions to other modules. They provide a
1023 * complete set of kernel non-blocking support for message
1028 netlink_kernel_create(int unit, void (*input)(struct sock *sk, int len))
1030 struct socket *sock;
1036 if (unit<0 || unit>=MAX_LINKS)
1039 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1042 if (netlink_create(sock, unit) < 0) {
1047 sk->sk_data_ready = netlink_data_ready;
1049 nlk_sk(sk)->data_ready = input;
1051 if (netlink_insert(sk, 0)) {
1058 void netlink_set_nonroot(int protocol, unsigned int flags)
1060 if ((unsigned int)protocol < MAX_LINKS)
1061 nl_table[protocol].nl_nonroot = flags;
1064 static void netlink_destroy_callback(struct netlink_callback *cb)
1072 * It looks a bit ugly.
1073 * It would be better to create kernel thread.
1076 static int netlink_dump(struct sock *sk)
1078 struct netlink_sock *nlk = nlk_sk(sk);
1079 struct netlink_callback *cb;
1080 struct sk_buff *skb;
1081 struct nlmsghdr *nlh;
1084 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1088 spin_lock(&nlk->cb_lock);
1092 spin_unlock(&nlk->cb_lock);
1097 len = cb->dump(skb, cb);
1100 spin_unlock(&nlk->cb_lock);
1101 skb_queue_tail(&sk->sk_receive_queue, skb);
1102 sk->sk_data_ready(sk, len);
1106 nlh = __nlmsg_put(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq, NLMSG_DONE, sizeof(int));
1107 nlh->nlmsg_flags |= NLM_F_MULTI;
1108 memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1109 skb_queue_tail(&sk->sk_receive_queue, skb);
1110 sk->sk_data_ready(sk, skb->len);
1114 spin_unlock(&nlk->cb_lock);
1116 netlink_destroy_callback(cb);
1120 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1121 struct nlmsghdr *nlh,
1122 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1123 int (*done)(struct netlink_callback*))
1125 struct netlink_callback *cb;
1127 struct netlink_sock *nlk;
1129 cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1133 memset(cb, 0, sizeof(*cb));
1137 atomic_inc(&skb->users);
1140 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1142 netlink_destroy_callback(cb);
1143 return -ECONNREFUSED;
1146 /* A dump is in progress... */
1147 spin_lock(&nlk->cb_lock);
1149 spin_unlock(&nlk->cb_lock);
1150 netlink_destroy_callback(cb);
1155 spin_unlock(&nlk->cb_lock);
1162 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1164 struct sk_buff *skb;
1165 struct nlmsghdr *rep;
1166 struct nlmsgerr *errmsg;
1170 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1172 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1174 skb = alloc_skb(size, GFP_KERNEL);
1178 sk = netlink_lookup(in_skb->sk->sk_protocol,
1179 NETLINK_CB(in_skb).pid);
1181 sk->sk_err = ENOBUFS;
1182 sk->sk_error_report(sk);
1188 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1189 NLMSG_ERROR, sizeof(struct nlmsgerr));
1190 errmsg = NLMSG_DATA(rep);
1191 errmsg->error = err;
1192 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1193 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1197 #ifdef CONFIG_PROC_FS
1198 struct nl_seq_iter {
1203 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1205 struct nl_seq_iter *iter = seq->private;
1208 struct hlist_node *node;
1211 for (i=0; i<MAX_LINKS; i++) {
1212 struct nl_pid_hash *hash = &nl_table[i].hash;
1214 for (j = 0; j <= hash->mask; j++) {
1215 sk_for_each(s, node, &hash->table[j]) {
1228 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1230 read_lock(&nl_table_lock);
1231 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1234 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1237 struct nl_seq_iter *iter;
1242 if (v == SEQ_START_TOKEN)
1243 return netlink_seq_socket_idx(seq, 0);
1249 iter = seq->private;
1251 j = iter->hash_idx + 1;
1254 struct nl_pid_hash *hash = &nl_table[i].hash;
1256 for (; j <= hash->mask; j++) {
1257 s = sk_head(&hash->table[j]);
1266 } while (++i < MAX_LINKS);
1271 static void netlink_seq_stop(struct seq_file *seq, void *v)
1273 read_unlock(&nl_table_lock);
1277 static int netlink_seq_show(struct seq_file *seq, void *v)
1279 if (v == SEQ_START_TOKEN)
1281 "sk Eth Pid Groups "
1282 "Rmem Wmem Dump Locks\n");
1285 struct netlink_sock *nlk = nlk_sk(s);
1287 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1292 atomic_read(&s->sk_rmem_alloc),
1293 atomic_read(&s->sk_wmem_alloc),
1295 atomic_read(&s->sk_refcnt)
1302 static struct seq_operations netlink_seq_ops = {
1303 .start = netlink_seq_start,
1304 .next = netlink_seq_next,
1305 .stop = netlink_seq_stop,
1306 .show = netlink_seq_show,
1310 static int netlink_seq_open(struct inode *inode, struct file *file)
1312 struct seq_file *seq;
1313 struct nl_seq_iter *iter;
1316 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1320 err = seq_open(file, &netlink_seq_ops);
1326 memset(iter, 0, sizeof(*iter));
1327 seq = file->private_data;
1328 seq->private = iter;
1332 static struct file_operations netlink_seq_fops = {
1333 .owner = THIS_MODULE,
1334 .open = netlink_seq_open,
1336 .llseek = seq_lseek,
1337 .release = seq_release_private,
1342 int netlink_register_notifier(struct notifier_block *nb)
1344 return notifier_chain_register(&netlink_chain, nb);
1347 int netlink_unregister_notifier(struct notifier_block *nb)
1349 return notifier_chain_unregister(&netlink_chain, nb);
1352 static struct proto_ops netlink_ops = {
1353 .family = PF_NETLINK,
1354 .owner = THIS_MODULE,
1355 .release = netlink_release,
1356 .bind = netlink_bind,
1357 .connect = netlink_connect,
1358 .socketpair = sock_no_socketpair,
1359 .accept = sock_no_accept,
1360 .getname = netlink_getname,
1361 .poll = datagram_poll,
1362 .ioctl = sock_no_ioctl,
1363 .listen = sock_no_listen,
1364 .shutdown = sock_no_shutdown,
1365 .setsockopt = sock_no_setsockopt,
1366 .getsockopt = sock_no_getsockopt,
1367 .sendmsg = netlink_sendmsg,
1368 .recvmsg = netlink_recvmsg,
1369 .mmap = sock_no_mmap,
1370 .sendpage = sock_no_sendpage,
1373 static struct net_proto_family netlink_family_ops = {
1374 .family = PF_NETLINK,
1375 .create = netlink_create,
1376 .owner = THIS_MODULE, /* for consistency 8) */
1379 extern void netlink_skb_parms_too_large(void);
1381 static int __init netlink_proto_init(void)
1383 struct sk_buff *dummy_skb;
1387 int err = proto_register(&netlink_proto, 0);
1392 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1393 netlink_skb_parms_too_large();
1395 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1398 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1402 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1404 if (num_physpages >= (128 * 1024))
1405 max = num_physpages >> (21 - PAGE_SHIFT);
1407 max = num_physpages >> (23 - PAGE_SHIFT);
1409 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1410 max = (1UL << order) / sizeof(struct hlist_head);
1411 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1413 for (i = 0; i < MAX_LINKS; i++) {
1414 struct nl_pid_hash *hash = &nl_table[i].hash;
1416 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1419 nl_pid_hash_free(nl_table[i].hash.table,
1420 1 * sizeof(*hash->table));
1424 memset(hash->table, 0, 1 * sizeof(*hash->table));
1425 hash->max_shift = order;
1428 hash->rehash_time = jiffies;
1431 sock_register(&netlink_family_ops);
1432 #ifdef CONFIG_PROC_FS
1433 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1435 /* The netlink device handler may be needed early. */
1441 static void __exit netlink_proto_exit(void)
1443 sock_unregister(PF_NETLINK);
1444 proc_net_remove("netlink");
1447 proto_unregister(&netlink_proto);
1450 core_initcall(netlink_proto_init);
1451 module_exit(netlink_proto_exit);
1453 MODULE_LICENSE("GPL");
1455 MODULE_ALIAS_NETPROTO(PF_NETLINK);
1457 EXPORT_SYMBOL(netlink_ack);
1458 EXPORT_SYMBOL(netlink_broadcast);
1459 EXPORT_SYMBOL(netlink_dump_start);
1460 EXPORT_SYMBOL(netlink_kernel_create);
1461 EXPORT_SYMBOL(netlink_register_notifier);
1462 EXPORT_SYMBOL(netlink_set_err);
1463 EXPORT_SYMBOL(netlink_set_nonroot);
1464 EXPORT_SYMBOL(netlink_unicast);
1465 EXPORT_SYMBOL(netlink_unregister_notifier);