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>
60 /* struct sock has to be the first member of netlink_sock */
65 unsigned int dst_groups;
67 wait_queue_head_t wait;
68 struct netlink_callback *cb;
70 void (*data_ready)(struct sock *sk, int bytes);
73 static inline struct netlink_sock *nlk_sk(struct sock *sk)
75 return (struct netlink_sock *)sk;
79 struct hlist_head *table;
80 unsigned long rehash_time;
86 unsigned int max_shift;
91 struct netlink_table {
92 struct nl_pid_hash hash;
93 struct hlist_head mc_list;
94 unsigned int nl_nonroot;
97 static struct netlink_table *nl_table;
99 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
101 static int netlink_dump(struct sock *sk);
102 static void netlink_destroy_callback(struct netlink_callback *cb);
104 static DEFINE_RWLOCK(nl_table_lock);
105 static atomic_t nl_table_users = ATOMIC_INIT(0);
107 static struct notifier_block *netlink_chain;
109 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
111 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
114 static void netlink_sock_destruct(struct sock *sk)
116 skb_queue_purge(&sk->sk_receive_queue);
118 if (!sock_flag(sk, SOCK_DEAD)) {
119 printk("Freeing alive netlink socket %p\n", sk);
122 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
123 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
124 BUG_TRAP(!nlk_sk(sk)->cb);
127 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
128 * Look, when several writers sleep and reader wakes them up, all but one
129 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
130 * this, _but_ remember, it adds useless work on UP machines.
133 static void netlink_table_grab(void)
135 write_lock_bh(&nl_table_lock);
137 if (atomic_read(&nl_table_users)) {
138 DECLARE_WAITQUEUE(wait, current);
140 add_wait_queue_exclusive(&nl_table_wait, &wait);
142 set_current_state(TASK_UNINTERRUPTIBLE);
143 if (atomic_read(&nl_table_users) == 0)
145 write_unlock_bh(&nl_table_lock);
147 write_lock_bh(&nl_table_lock);
150 __set_current_state(TASK_RUNNING);
151 remove_wait_queue(&nl_table_wait, &wait);
155 static __inline__ void netlink_table_ungrab(void)
157 write_unlock_bh(&nl_table_lock);
158 wake_up(&nl_table_wait);
161 static __inline__ void
162 netlink_lock_table(void)
164 /* read_lock() synchronizes us to netlink_table_grab */
166 read_lock(&nl_table_lock);
167 atomic_inc(&nl_table_users);
168 read_unlock(&nl_table_lock);
171 static __inline__ void
172 netlink_unlock_table(void)
174 if (atomic_dec_and_test(&nl_table_users))
175 wake_up(&nl_table_wait);
178 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
180 struct nl_pid_hash *hash = &nl_table[protocol].hash;
181 struct hlist_head *head;
183 struct hlist_node *node;
185 read_lock(&nl_table_lock);
186 head = nl_pid_hashfn(hash, pid);
187 sk_for_each(sk, node, head) {
188 if (nlk_sk(sk)->pid == pid) {
195 read_unlock(&nl_table_lock);
199 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
201 if (size <= PAGE_SIZE)
202 return kmalloc(size, GFP_ATOMIC);
204 return (struct hlist_head *)
205 __get_free_pages(GFP_ATOMIC, get_order(size));
208 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
210 if (size <= PAGE_SIZE)
213 free_pages((unsigned long)table, get_order(size));
216 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
218 unsigned int omask, mask, shift;
220 struct hlist_head *otable, *table;
223 omask = mask = hash->mask;
224 osize = size = (mask + 1) * sizeof(*table);
228 if (++shift > hash->max_shift)
234 table = nl_pid_hash_alloc(size);
238 memset(table, 0, size);
239 otable = hash->table;
243 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
245 for (i = 0; i <= omask; i++) {
247 struct hlist_node *node, *tmp;
249 sk_for_each_safe(sk, node, tmp, &otable[i])
250 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
253 nl_pid_hash_free(otable, osize);
254 hash->rehash_time = jiffies + 10 * 60 * HZ;
258 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
260 int avg = hash->entries >> hash->shift;
262 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
265 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
266 nl_pid_hash_rehash(hash, 0);
273 static struct proto_ops netlink_ops;
275 static int netlink_insert(struct sock *sk, u32 pid)
277 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
278 struct hlist_head *head;
279 int err = -EADDRINUSE;
281 struct hlist_node *node;
284 netlink_table_grab();
285 head = nl_pid_hashfn(hash, pid);
287 sk_for_each(osk, node, head) {
288 if (nlk_sk(osk)->pid == pid)
300 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
303 if (len && nl_pid_hash_dilute(hash, len))
304 head = nl_pid_hashfn(hash, pid);
306 nlk_sk(sk)->pid = pid;
307 sk_add_node(sk, head);
311 netlink_table_ungrab();
315 static void netlink_remove(struct sock *sk)
317 netlink_table_grab();
318 nl_table[sk->sk_protocol].hash.entries--;
319 sk_del_node_init(sk);
320 if (nlk_sk(sk)->groups)
321 __sk_del_bind_node(sk);
322 netlink_table_ungrab();
325 static struct proto netlink_proto = {
327 .owner = THIS_MODULE,
328 .obj_size = sizeof(struct netlink_sock),
331 static int netlink_create(struct socket *sock, int protocol)
334 struct netlink_sock *nlk;
336 sock->state = SS_UNCONNECTED;
338 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
339 return -ESOCKTNOSUPPORT;
341 if (protocol<0 || protocol >= MAX_LINKS)
342 return -EPROTONOSUPPORT;
344 sock->ops = &netlink_ops;
346 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
350 sock_init_data(sock, sk);
354 spin_lock_init(&nlk->cb_lock);
355 init_waitqueue_head(&nlk->wait);
356 sk->sk_destruct = netlink_sock_destruct;
358 sk->sk_protocol = protocol;
362 static int netlink_release(struct socket *sock)
364 struct sock *sk = sock->sk;
365 struct netlink_sock *nlk;
373 spin_lock(&nlk->cb_lock);
375 nlk->cb->done(nlk->cb);
376 netlink_destroy_callback(nlk->cb);
379 spin_unlock(&nlk->cb_lock);
381 /* OK. Socket is unlinked, and, therefore,
382 no new packets will arrive */
386 wake_up_interruptible_all(&nlk->wait);
388 skb_queue_purge(&sk->sk_write_queue);
390 if (nlk->pid && !nlk->groups) {
391 struct netlink_notify n = {
392 .protocol = sk->sk_protocol,
395 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
402 static int netlink_autobind(struct socket *sock)
404 struct sock *sk = sock->sk;
405 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
406 struct hlist_head *head;
408 struct hlist_node *node;
409 s32 pid = current->pid;
411 static s32 rover = -4097;
415 netlink_table_grab();
416 head = nl_pid_hashfn(hash, pid);
417 sk_for_each(osk, node, head) {
418 if (nlk_sk(osk)->pid == pid) {
419 /* Bind collision, search negative pid values. */
423 netlink_table_ungrab();
427 netlink_table_ungrab();
429 err = netlink_insert(sk, pid);
430 if (err == -EADDRINUSE)
435 static inline int netlink_capable(struct socket *sock, unsigned int flag)
437 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
438 capable(CAP_NET_ADMIN);
441 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
443 struct sock *sk = sock->sk;
444 struct netlink_sock *nlk = nlk_sk(sk);
445 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
448 if (nladdr->nl_family != AF_NETLINK)
451 /* Only superuser is allowed to listen multicasts */
452 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_RECV))
456 if (nladdr->nl_pid != nlk->pid)
459 err = nladdr->nl_pid ?
460 netlink_insert(sk, nladdr->nl_pid) :
461 netlink_autobind(sock);
466 if (!nladdr->nl_groups && !nlk->groups)
469 netlink_table_grab();
470 if (nlk->groups && !nladdr->nl_groups)
471 __sk_del_bind_node(sk);
472 else if (!nlk->groups && nladdr->nl_groups)
473 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
474 nlk->groups = nladdr->nl_groups;
475 netlink_table_ungrab();
480 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
484 struct sock *sk = sock->sk;
485 struct netlink_sock *nlk = nlk_sk(sk);
486 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
488 if (addr->sa_family == AF_UNSPEC) {
489 sk->sk_state = NETLINK_UNCONNECTED;
494 if (addr->sa_family != AF_NETLINK)
497 /* Only superuser is allowed to send multicasts */
498 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
502 err = netlink_autobind(sock);
505 sk->sk_state = NETLINK_CONNECTED;
506 nlk->dst_pid = nladdr->nl_pid;
507 nlk->dst_groups = nladdr->nl_groups;
513 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
515 struct sock *sk = sock->sk;
516 struct netlink_sock *nlk = nlk_sk(sk);
517 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
519 nladdr->nl_family = AF_NETLINK;
521 *addr_len = sizeof(*nladdr);
524 nladdr->nl_pid = nlk->dst_pid;
525 nladdr->nl_groups = nlk->dst_groups;
527 nladdr->nl_pid = nlk->pid;
528 nladdr->nl_groups = nlk->groups;
533 static void netlink_overrun(struct sock *sk)
535 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
536 sk->sk_err = ENOBUFS;
537 sk->sk_error_report(sk);
541 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
543 int protocol = ssk->sk_protocol;
545 struct netlink_sock *nlk;
547 sock = netlink_lookup(protocol, pid);
549 return ERR_PTR(-ECONNREFUSED);
551 /* Don't bother queuing skb if kernel socket has no input function */
553 if ((nlk->pid == 0 && !nlk->data_ready) ||
554 (sock->sk_state == NETLINK_CONNECTED &&
555 nlk->dst_pid != nlk_sk(ssk)->pid)) {
557 return ERR_PTR(-ECONNREFUSED);
562 struct sock *netlink_getsockbyfilp(struct file *filp)
564 struct inode *inode = filp->f_dentry->d_inode;
567 if (!S_ISSOCK(inode->i_mode))
568 return ERR_PTR(-ENOTSOCK);
570 sock = SOCKET_I(inode)->sk;
571 if (sock->sk_family != AF_NETLINK)
572 return ERR_PTR(-EINVAL);
579 * Attach a skb to a netlink socket.
580 * The caller must hold a reference to the destination socket. On error, the
581 * reference is dropped. The skb is not send to the destination, just all
582 * all error checks are performed and memory in the queue is reserved.
584 * < 0: error. skb freed, reference to sock dropped.
586 * 1: repeat lookup - reference dropped while waiting for socket memory.
588 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo)
590 struct netlink_sock *nlk;
594 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
595 test_bit(0, &nlk->state)) {
596 DECLARE_WAITQUEUE(wait, current);
605 __set_current_state(TASK_INTERRUPTIBLE);
606 add_wait_queue(&nlk->wait, &wait);
608 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
609 test_bit(0, &nlk->state)) &&
610 !sock_flag(sk, SOCK_DEAD))
611 timeo = schedule_timeout(timeo);
613 __set_current_state(TASK_RUNNING);
614 remove_wait_queue(&nlk->wait, &wait);
617 if (signal_pending(current)) {
619 return sock_intr_errno(timeo);
623 skb_set_owner_r(skb, sk);
627 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
629 struct netlink_sock *nlk;
634 skb_queue_tail(&sk->sk_receive_queue, skb);
635 sk->sk_data_ready(sk, len);
640 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
646 static inline struct sk_buff *netlink_trim(struct sk_buff *skb, int allocation)
652 delta = skb->end - skb->tail;
653 if (delta * 2 < skb->truesize)
656 if (skb_shared(skb)) {
657 struct sk_buff *nskb = skb_clone(skb, allocation);
664 if (!pskb_expand_head(skb, 0, -delta, allocation))
665 skb->truesize -= delta;
670 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
676 skb = netlink_trim(skb, gfp_any());
678 timeo = sock_sndtimeo(ssk, nonblock);
680 sk = netlink_getsockbypid(ssk, pid);
685 err = netlink_attachskb(sk, skb, nonblock, timeo);
691 return netlink_sendskb(sk, skb, ssk->sk_protocol);
694 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
696 struct netlink_sock *nlk = nlk_sk(sk);
698 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
699 !test_bit(0, &nlk->state)) {
700 skb_set_owner_r(skb, sk);
701 skb_queue_tail(&sk->sk_receive_queue, skb);
702 sk->sk_data_ready(sk, skb->len);
703 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
708 struct netlink_broadcast_data {
709 struct sock *exclude_sk;
716 struct sk_buff *skb, *skb2;
719 static inline int do_one_broadcast(struct sock *sk,
720 struct netlink_broadcast_data *p)
722 struct netlink_sock *nlk = nlk_sk(sk);
725 if (p->exclude_sk == sk)
728 if (nlk->pid == p->pid || !(nlk->groups & p->group))
737 if (p->skb2 == NULL) {
738 if (skb_shared(p->skb)) {
739 p->skb2 = skb_clone(p->skb, p->allocation);
741 p->skb2 = skb_get(p->skb);
743 * skb ownership may have been set when
744 * delivered to a previous socket.
749 if (p->skb2 == NULL) {
751 /* Clone failed. Notify ALL listeners. */
753 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
766 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
767 u32 group, int allocation)
769 struct netlink_broadcast_data info;
770 struct hlist_node *node;
773 skb = netlink_trim(skb, allocation);
775 info.exclude_sk = ssk;
781 info.allocation = allocation;
785 /* While we sleep in clone, do not allow to change socket list */
787 netlink_lock_table();
789 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
790 do_one_broadcast(sk, &info);
794 netlink_unlock_table();
797 kfree_skb(info.skb2);
799 if (info.delivered) {
800 if (info.congested && (allocation & __GFP_WAIT))
809 struct netlink_set_err_data {
810 struct sock *exclude_sk;
816 static inline int do_one_set_err(struct sock *sk,
817 struct netlink_set_err_data *p)
819 struct netlink_sock *nlk = nlk_sk(sk);
821 if (sk == p->exclude_sk)
824 if (nlk->pid == p->pid || !(nlk->groups & p->group))
827 sk->sk_err = p->code;
828 sk->sk_error_report(sk);
833 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
835 struct netlink_set_err_data info;
836 struct hlist_node *node;
839 info.exclude_sk = ssk;
844 read_lock(&nl_table_lock);
846 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
847 do_one_set_err(sk, &info);
849 read_unlock(&nl_table_lock);
852 static inline void netlink_rcv_wake(struct sock *sk)
854 struct netlink_sock *nlk = nlk_sk(sk);
856 if (!skb_queue_len(&sk->sk_receive_queue))
857 clear_bit(0, &nlk->state);
858 if (!test_bit(0, &nlk->state))
859 wake_up_interruptible(&nlk->wait);
862 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
863 struct msghdr *msg, size_t len)
865 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
866 struct sock *sk = sock->sk;
867 struct netlink_sock *nlk = nlk_sk(sk);
868 struct sockaddr_nl *addr=msg->msg_name;
873 struct scm_cookie scm;
875 if (msg->msg_flags&MSG_OOB)
878 if (NULL == siocb->scm)
880 err = scm_send(sock, msg, siocb->scm);
884 if (msg->msg_namelen) {
885 if (addr->nl_family != AF_NETLINK)
887 dst_pid = addr->nl_pid;
888 dst_groups = addr->nl_groups;
889 if (dst_groups && !netlink_capable(sock, NL_NONROOT_SEND))
892 dst_pid = nlk->dst_pid;
893 dst_groups = nlk->dst_groups;
897 err = netlink_autobind(sock);
903 if (len > sk->sk_sndbuf - 32)
906 skb = alloc_skb(len, GFP_KERNEL);
910 NETLINK_CB(skb).pid = nlk->pid;
911 NETLINK_CB(skb).groups = nlk->groups;
912 NETLINK_CB(skb).dst_pid = dst_pid;
913 NETLINK_CB(skb).dst_groups = dst_groups;
914 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
915 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
917 /* What can I do? Netlink is asynchronous, so that
918 we will have to save current capabilities to
919 check them, when this message will be delivered
920 to corresponding kernel module. --ANK (980802)
924 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
929 err = security_netlink_send(sk, skb);
936 atomic_inc(&skb->users);
937 netlink_broadcast(sk, skb, dst_pid, dst_groups, GFP_KERNEL);
939 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
945 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
946 struct msghdr *msg, size_t len,
949 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
950 struct scm_cookie scm;
951 struct sock *sk = sock->sk;
952 struct netlink_sock *nlk = nlk_sk(sk);
953 int noblock = flags&MSG_DONTWAIT;
963 skb = skb_recv_datagram(sk,flags,noblock,&err);
967 msg->msg_namelen = 0;
971 msg->msg_flags |= MSG_TRUNC;
975 skb->h.raw = skb->data;
976 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
979 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
980 addr->nl_family = AF_NETLINK;
982 addr->nl_pid = NETLINK_CB(skb).pid;
983 addr->nl_groups = NETLINK_CB(skb).dst_groups;
984 msg->msg_namelen = sizeof(*addr);
987 if (NULL == siocb->scm) {
988 memset(&scm, 0, sizeof(scm));
991 siocb->scm->creds = *NETLINK_CREDS(skb);
992 skb_free_datagram(sk, skb);
994 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
997 scm_recv(sock, msg, siocb->scm, flags);
1000 netlink_rcv_wake(sk);
1001 return err ? : copied;
1004 static void netlink_data_ready(struct sock *sk, int len)
1006 struct netlink_sock *nlk = nlk_sk(sk);
1008 if (nlk->data_ready)
1009 nlk->data_ready(sk, len);
1010 netlink_rcv_wake(sk);
1014 * We export these functions to other modules. They provide a
1015 * complete set of kernel non-blocking support for message
1020 netlink_kernel_create(int unit, void (*input)(struct sock *sk, int len))
1022 struct socket *sock;
1028 if (unit<0 || unit>=MAX_LINKS)
1031 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1034 if (netlink_create(sock, unit) < 0) {
1039 sk->sk_data_ready = netlink_data_ready;
1041 nlk_sk(sk)->data_ready = input;
1043 if (netlink_insert(sk, 0)) {
1050 void netlink_set_nonroot(int protocol, unsigned int flags)
1052 if ((unsigned int)protocol < MAX_LINKS)
1053 nl_table[protocol].nl_nonroot = flags;
1056 static void netlink_destroy_callback(struct netlink_callback *cb)
1064 * It looks a bit ugly.
1065 * It would be better to create kernel thread.
1068 static int netlink_dump(struct sock *sk)
1070 struct netlink_sock *nlk = nlk_sk(sk);
1071 struct netlink_callback *cb;
1072 struct sk_buff *skb;
1073 struct nlmsghdr *nlh;
1076 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1080 spin_lock(&nlk->cb_lock);
1084 spin_unlock(&nlk->cb_lock);
1089 len = cb->dump(skb, cb);
1092 spin_unlock(&nlk->cb_lock);
1093 skb_queue_tail(&sk->sk_receive_queue, skb);
1094 sk->sk_data_ready(sk, len);
1098 nlh = __nlmsg_put(skb, NETLINK_CB(cb->skb).pid, cb->nlh->nlmsg_seq, NLMSG_DONE, sizeof(int));
1099 nlh->nlmsg_flags |= NLM_F_MULTI;
1100 memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1101 skb_queue_tail(&sk->sk_receive_queue, skb);
1102 sk->sk_data_ready(sk, skb->len);
1106 spin_unlock(&nlk->cb_lock);
1108 netlink_destroy_callback(cb);
1112 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1113 struct nlmsghdr *nlh,
1114 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1115 int (*done)(struct netlink_callback*))
1117 struct netlink_callback *cb;
1119 struct netlink_sock *nlk;
1121 cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1125 memset(cb, 0, sizeof(*cb));
1129 atomic_inc(&skb->users);
1132 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1134 netlink_destroy_callback(cb);
1135 return -ECONNREFUSED;
1138 /* A dump is in progress... */
1139 spin_lock(&nlk->cb_lock);
1141 spin_unlock(&nlk->cb_lock);
1142 netlink_destroy_callback(cb);
1147 spin_unlock(&nlk->cb_lock);
1154 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1156 struct sk_buff *skb;
1157 struct nlmsghdr *rep;
1158 struct nlmsgerr *errmsg;
1162 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1164 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1166 skb = alloc_skb(size, GFP_KERNEL);
1170 sk = netlink_lookup(in_skb->sk->sk_protocol,
1171 NETLINK_CB(in_skb).pid);
1173 sk->sk_err = ENOBUFS;
1174 sk->sk_error_report(sk);
1180 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1181 NLMSG_ERROR, sizeof(struct nlmsgerr));
1182 errmsg = NLMSG_DATA(rep);
1183 errmsg->error = err;
1184 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1185 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1189 #ifdef CONFIG_PROC_FS
1190 struct nl_seq_iter {
1195 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1197 struct nl_seq_iter *iter = seq->private;
1200 struct hlist_node *node;
1203 for (i=0; i<MAX_LINKS; i++) {
1204 struct nl_pid_hash *hash = &nl_table[i].hash;
1206 for (j = 0; j <= hash->mask; j++) {
1207 sk_for_each(s, node, &hash->table[j]) {
1220 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1222 read_lock(&nl_table_lock);
1223 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1226 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1229 struct nl_seq_iter *iter;
1234 if (v == SEQ_START_TOKEN)
1235 return netlink_seq_socket_idx(seq, 0);
1241 iter = seq->private;
1243 j = iter->hash_idx + 1;
1246 struct nl_pid_hash *hash = &nl_table[i].hash;
1248 for (; j <= hash->mask; j++) {
1249 s = sk_head(&hash->table[j]);
1258 } while (++i < MAX_LINKS);
1263 static void netlink_seq_stop(struct seq_file *seq, void *v)
1265 read_unlock(&nl_table_lock);
1269 static int netlink_seq_show(struct seq_file *seq, void *v)
1271 if (v == SEQ_START_TOKEN)
1273 "sk Eth Pid Groups "
1274 "Rmem Wmem Dump Locks\n");
1277 struct netlink_sock *nlk = nlk_sk(s);
1279 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1284 atomic_read(&s->sk_rmem_alloc),
1285 atomic_read(&s->sk_wmem_alloc),
1287 atomic_read(&s->sk_refcnt)
1294 static struct seq_operations netlink_seq_ops = {
1295 .start = netlink_seq_start,
1296 .next = netlink_seq_next,
1297 .stop = netlink_seq_stop,
1298 .show = netlink_seq_show,
1302 static int netlink_seq_open(struct inode *inode, struct file *file)
1304 struct seq_file *seq;
1305 struct nl_seq_iter *iter;
1308 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1312 err = seq_open(file, &netlink_seq_ops);
1318 memset(iter, 0, sizeof(*iter));
1319 seq = file->private_data;
1320 seq->private = iter;
1324 static struct file_operations netlink_seq_fops = {
1325 .owner = THIS_MODULE,
1326 .open = netlink_seq_open,
1328 .llseek = seq_lseek,
1329 .release = seq_release_private,
1334 int netlink_register_notifier(struct notifier_block *nb)
1336 return notifier_chain_register(&netlink_chain, nb);
1339 int netlink_unregister_notifier(struct notifier_block *nb)
1341 return notifier_chain_unregister(&netlink_chain, nb);
1344 static struct proto_ops netlink_ops = {
1345 .family = PF_NETLINK,
1346 .owner = THIS_MODULE,
1347 .release = netlink_release,
1348 .bind = netlink_bind,
1349 .connect = netlink_connect,
1350 .socketpair = sock_no_socketpair,
1351 .accept = sock_no_accept,
1352 .getname = netlink_getname,
1353 .poll = datagram_poll,
1354 .ioctl = sock_no_ioctl,
1355 .listen = sock_no_listen,
1356 .shutdown = sock_no_shutdown,
1357 .setsockopt = sock_no_setsockopt,
1358 .getsockopt = sock_no_getsockopt,
1359 .sendmsg = netlink_sendmsg,
1360 .recvmsg = netlink_recvmsg,
1361 .mmap = sock_no_mmap,
1362 .sendpage = sock_no_sendpage,
1365 static struct net_proto_family netlink_family_ops = {
1366 .family = PF_NETLINK,
1367 .create = netlink_create,
1368 .owner = THIS_MODULE, /* for consistency 8) */
1371 extern void netlink_skb_parms_too_large(void);
1373 static int __init netlink_proto_init(void)
1375 struct sk_buff *dummy_skb;
1379 int err = proto_register(&netlink_proto, 0);
1384 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1385 netlink_skb_parms_too_large();
1387 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1390 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1394 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1396 if (num_physpages >= (128 * 1024))
1397 max = num_physpages >> (21 - PAGE_SHIFT);
1399 max = num_physpages >> (23 - PAGE_SHIFT);
1401 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1402 max = (1UL << order) / sizeof(struct hlist_head);
1403 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1405 for (i = 0; i < MAX_LINKS; i++) {
1406 struct nl_pid_hash *hash = &nl_table[i].hash;
1408 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1411 nl_pid_hash_free(nl_table[i].hash.table,
1412 1 * sizeof(*hash->table));
1416 memset(hash->table, 0, 1 * sizeof(*hash->table));
1417 hash->max_shift = order;
1420 hash->rehash_time = jiffies;
1423 sock_register(&netlink_family_ops);
1424 #ifdef CONFIG_PROC_FS
1425 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1427 /* The netlink device handler may be needed early. */
1433 static void __exit netlink_proto_exit(void)
1435 sock_unregister(PF_NETLINK);
1436 proc_net_remove("netlink");
1439 proto_unregister(&netlink_proto);
1442 core_initcall(netlink_proto_init);
1443 module_exit(netlink_proto_exit);
1445 MODULE_LICENSE("GPL");
1447 MODULE_ALIAS_NETPROTO(PF_NETLINK);
1449 EXPORT_SYMBOL(netlink_ack);
1450 EXPORT_SYMBOL(netlink_broadcast);
1451 EXPORT_SYMBOL(netlink_dump_start);
1452 EXPORT_SYMBOL(netlink_kernel_create);
1453 EXPORT_SYMBOL(netlink_register_notifier);
1454 EXPORT_SYMBOL(netlink_set_err);
1455 EXPORT_SYMBOL(netlink_set_nonroot);
1456 EXPORT_SYMBOL(netlink_unicast);
1457 EXPORT_SYMBOL(netlink_unregister_notifier);