2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@redhat.com>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
24 #include <linux/config.h>
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <asm/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.h>
49 #include <linux/smp_lock.h>
50 #include <linux/notifier.h>
51 #include <linux/security.h>
52 #include <linux/jhash.h>
53 #include <linux/jiffies.h>
54 #include <linux/random.h>
55 #include <linux/bitops.h>
57 #include <linux/types.h>
58 #include <linux/audit.h>
59 #include <linux/vs_context.h>
60 #include <linux/vs_network.h>
61 #include <linux/vs_limit.h>
65 #include <net/netlink.h>
68 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
71 /* struct sock has to be the first member of netlink_sock */
79 unsigned long *groups;
81 wait_queue_head_t wait;
82 struct netlink_callback *cb;
84 void (*data_ready)(struct sock *sk, int bytes);
85 struct module *module;
88 #define NETLINK_KERNEL_SOCKET 0x1
89 #define NETLINK_RECV_PKTINFO 0x2
91 static inline struct netlink_sock *nlk_sk(struct sock *sk)
93 return (struct netlink_sock *)sk;
97 struct hlist_head *table;
98 unsigned long rehash_time;
103 unsigned int entries;
104 unsigned int max_shift;
109 struct netlink_table {
110 struct nl_pid_hash hash;
111 struct hlist_head mc_list;
112 unsigned int nl_nonroot;
114 struct module *module;
118 static struct netlink_table *nl_table;
120 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
122 static int netlink_dump(struct sock *sk);
123 static void netlink_destroy_callback(struct netlink_callback *cb);
125 static DEFINE_RWLOCK(nl_table_lock);
126 static atomic_t nl_table_users = ATOMIC_INIT(0);
128 static struct notifier_block *netlink_chain;
130 static u32 netlink_group_mask(u32 group)
132 return group ? 1 << (group - 1) : 0;
135 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
137 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
140 static void netlink_sock_destruct(struct sock *sk)
142 skb_queue_purge(&sk->sk_receive_queue);
144 if (!sock_flag(sk, SOCK_DEAD)) {
145 printk("Freeing alive netlink socket %p\n", sk);
148 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
149 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
150 BUG_TRAP(!nlk_sk(sk)->cb);
151 BUG_TRAP(!nlk_sk(sk)->groups);
154 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
155 * Look, when several writers sleep and reader wakes them up, all but one
156 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
157 * this, _but_ remember, it adds useless work on UP machines.
160 static void netlink_table_grab(void)
162 write_lock_bh(&nl_table_lock);
164 if (atomic_read(&nl_table_users)) {
165 DECLARE_WAITQUEUE(wait, current);
167 add_wait_queue_exclusive(&nl_table_wait, &wait);
169 set_current_state(TASK_UNINTERRUPTIBLE);
170 if (atomic_read(&nl_table_users) == 0)
172 write_unlock_bh(&nl_table_lock);
174 write_lock_bh(&nl_table_lock);
177 __set_current_state(TASK_RUNNING);
178 remove_wait_queue(&nl_table_wait, &wait);
182 static __inline__ void netlink_table_ungrab(void)
184 write_unlock_bh(&nl_table_lock);
185 wake_up(&nl_table_wait);
188 static __inline__ void
189 netlink_lock_table(void)
191 /* read_lock() synchronizes us to netlink_table_grab */
193 read_lock(&nl_table_lock);
194 atomic_inc(&nl_table_users);
195 read_unlock(&nl_table_lock);
198 static __inline__ void
199 netlink_unlock_table(void)
201 if (atomic_dec_and_test(&nl_table_users))
202 wake_up(&nl_table_wait);
205 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
207 struct nl_pid_hash *hash = &nl_table[protocol].hash;
208 struct hlist_head *head;
210 struct hlist_node *node;
212 read_lock(&nl_table_lock);
213 head = nl_pid_hashfn(hash, pid);
214 sk_for_each(sk, node, head) {
215 if (nlk_sk(sk)->pid == pid) {
222 read_unlock(&nl_table_lock);
226 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
228 if (size <= PAGE_SIZE)
229 return kmalloc(size, GFP_ATOMIC);
231 return (struct hlist_head *)
232 __get_free_pages(GFP_ATOMIC, get_order(size));
235 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
237 if (size <= PAGE_SIZE)
240 free_pages((unsigned long)table, get_order(size));
243 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
245 unsigned int omask, mask, shift;
247 struct hlist_head *otable, *table;
250 omask = mask = hash->mask;
251 osize = size = (mask + 1) * sizeof(*table);
255 if (++shift > hash->max_shift)
261 table = nl_pid_hash_alloc(size);
265 memset(table, 0, size);
266 otable = hash->table;
270 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
272 for (i = 0; i <= omask; i++) {
274 struct hlist_node *node, *tmp;
276 sk_for_each_safe(sk, node, tmp, &otable[i])
277 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
280 nl_pid_hash_free(otable, osize);
281 hash->rehash_time = jiffies + 10 * 60 * HZ;
285 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
287 int avg = hash->entries >> hash->shift;
289 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
292 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
293 nl_pid_hash_rehash(hash, 0);
300 static const struct proto_ops netlink_ops;
302 static int netlink_insert(struct sock *sk, u32 pid)
304 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
305 struct hlist_head *head;
306 int err = -EADDRINUSE;
308 struct hlist_node *node;
311 netlink_table_grab();
312 head = nl_pid_hashfn(hash, pid);
314 sk_for_each(osk, node, head) {
315 if (nlk_sk(osk)->pid == pid)
327 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
330 if (len && nl_pid_hash_dilute(hash, len))
331 head = nl_pid_hashfn(hash, pid);
333 nlk_sk(sk)->pid = pid;
334 sk_add_node(sk, head);
338 netlink_table_ungrab();
342 static void netlink_remove(struct sock *sk)
344 netlink_table_grab();
345 if (sk_del_node_init(sk))
346 nl_table[sk->sk_protocol].hash.entries--;
347 if (nlk_sk(sk)->subscriptions)
348 __sk_del_bind_node(sk);
349 netlink_table_ungrab();
352 static struct proto netlink_proto = {
354 .owner = THIS_MODULE,
355 .obj_size = sizeof(struct netlink_sock),
358 static int __netlink_create(struct socket *sock, int protocol)
361 struct netlink_sock *nlk;
363 sock->ops = &netlink_ops;
365 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
369 sock_init_data(sock, sk);
372 spin_lock_init(&nlk->cb_lock);
373 init_waitqueue_head(&nlk->wait);
375 sk->sk_destruct = netlink_sock_destruct;
376 sk->sk_protocol = protocol;
380 static int netlink_create(struct socket *sock, int protocol)
382 struct module *module = NULL;
383 struct netlink_sock *nlk;
387 sock->state = SS_UNCONNECTED;
389 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
390 return -ESOCKTNOSUPPORT;
392 if (protocol<0 || protocol >= MAX_LINKS)
393 return -EPROTONOSUPPORT;
395 netlink_lock_table();
397 if (!nl_table[protocol].registered) {
398 netlink_unlock_table();
399 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
400 netlink_lock_table();
403 if (nl_table[protocol].registered &&
404 try_module_get(nl_table[protocol].module))
405 module = nl_table[protocol].module;
406 groups = nl_table[protocol].groups;
407 netlink_unlock_table();
409 if ((err = __netlink_create(sock, protocol)) < 0)
412 nlk = nlk_sk(sock->sk);
413 nlk->module = module;
422 static int netlink_release(struct socket *sock)
424 struct sock *sk = sock->sk;
425 struct netlink_sock *nlk;
433 spin_lock(&nlk->cb_lock);
436 nlk->cb->done(nlk->cb);
437 netlink_destroy_callback(nlk->cb);
440 spin_unlock(&nlk->cb_lock);
442 /* OK. Socket is unlinked, and, therefore,
443 no new packets will arrive */
447 wake_up_interruptible_all(&nlk->wait);
449 skb_queue_purge(&sk->sk_write_queue);
451 if (nlk->pid && !nlk->subscriptions) {
452 struct netlink_notify n = {
453 .protocol = sk->sk_protocol,
456 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
460 module_put(nlk->module);
462 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
463 netlink_table_grab();
464 nl_table[sk->sk_protocol].module = NULL;
465 nl_table[sk->sk_protocol].registered = 0;
466 netlink_table_ungrab();
476 static int netlink_autobind(struct socket *sock)
478 struct sock *sk = sock->sk;
479 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
480 struct hlist_head *head;
482 struct hlist_node *node;
483 s32 pid = current->tgid;
485 static s32 rover = -4097;
489 netlink_table_grab();
490 head = nl_pid_hashfn(hash, pid);
491 sk_for_each(osk, node, head) {
492 if (nlk_sk(osk)->pid == pid) {
493 /* Bind collision, search negative pid values. */
497 netlink_table_ungrab();
501 netlink_table_ungrab();
503 err = netlink_insert(sk, pid);
504 if (err == -EADDRINUSE)
507 /* If 2 threads race to autobind, that is fine. */
514 static inline int netlink_capable(struct socket *sock, unsigned int flag)
516 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
517 capable(CAP_NET_ADMIN);
521 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
523 struct netlink_sock *nlk = nlk_sk(sk);
525 if (nlk->subscriptions && !subscriptions)
526 __sk_del_bind_node(sk);
527 else if (!nlk->subscriptions && subscriptions)
528 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
529 nlk->subscriptions = subscriptions;
532 static int netlink_alloc_groups(struct sock *sk)
534 struct netlink_sock *nlk = nlk_sk(sk);
538 netlink_lock_table();
539 groups = nl_table[sk->sk_protocol].groups;
540 if (!nl_table[sk->sk_protocol].registered)
542 netlink_unlock_table();
547 nlk->groups = kmalloc(NLGRPSZ(groups), GFP_KERNEL);
548 if (nlk->groups == NULL)
550 memset(nlk->groups, 0, NLGRPSZ(groups));
551 nlk->ngroups = groups;
555 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
557 struct sock *sk = sock->sk;
558 struct netlink_sock *nlk = nlk_sk(sk);
559 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
562 if (nladdr->nl_family != AF_NETLINK)
565 /* Only superuser is allowed to listen multicasts */
566 if (nladdr->nl_groups) {
567 if (!netlink_capable(sock, NL_NONROOT_RECV))
569 if (nlk->groups == NULL) {
570 err = netlink_alloc_groups(sk);
577 if (nladdr->nl_pid != nlk->pid)
580 err = nladdr->nl_pid ?
581 netlink_insert(sk, nladdr->nl_pid) :
582 netlink_autobind(sock);
587 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
590 netlink_table_grab();
591 netlink_update_subscriptions(sk, nlk->subscriptions +
592 hweight32(nladdr->nl_groups) -
593 hweight32(nlk->groups[0]));
594 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
595 netlink_table_ungrab();
600 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
604 struct sock *sk = sock->sk;
605 struct netlink_sock *nlk = nlk_sk(sk);
606 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
608 if (addr->sa_family == AF_UNSPEC) {
609 sk->sk_state = NETLINK_UNCONNECTED;
614 if (addr->sa_family != AF_NETLINK)
617 /* Only superuser is allowed to send multicasts */
618 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
622 err = netlink_autobind(sock);
625 sk->sk_state = NETLINK_CONNECTED;
626 nlk->dst_pid = nladdr->nl_pid;
627 nlk->dst_group = ffs(nladdr->nl_groups);
633 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
635 struct sock *sk = sock->sk;
636 struct netlink_sock *nlk = nlk_sk(sk);
637 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
639 nladdr->nl_family = AF_NETLINK;
641 *addr_len = sizeof(*nladdr);
644 nladdr->nl_pid = nlk->dst_pid;
645 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
647 nladdr->nl_pid = nlk->pid;
648 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
653 static void netlink_overrun(struct sock *sk)
655 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
656 sk->sk_err = ENOBUFS;
657 sk->sk_error_report(sk);
661 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
663 int protocol = ssk->sk_protocol;
665 struct netlink_sock *nlk;
667 sock = netlink_lookup(protocol, pid);
669 return ERR_PTR(-ECONNREFUSED);
671 /* Don't bother queuing skb if kernel socket has no input function */
673 if ((nlk->pid == 0 && !nlk->data_ready) ||
674 (sock->sk_state == NETLINK_CONNECTED &&
675 nlk->dst_pid != nlk_sk(ssk)->pid)) {
677 return ERR_PTR(-ECONNREFUSED);
682 struct sock *netlink_getsockbyfilp(struct file *filp)
684 struct inode *inode = filp->f_dentry->d_inode;
687 if (!S_ISSOCK(inode->i_mode))
688 return ERR_PTR(-ENOTSOCK);
690 sock = SOCKET_I(inode)->sk;
691 if (sock->sk_family != AF_NETLINK)
692 return ERR_PTR(-EINVAL);
699 * Attach a skb to a netlink socket.
700 * The caller must hold a reference to the destination socket. On error, the
701 * reference is dropped. The skb is not send to the destination, just all
702 * all error checks are performed and memory in the queue is reserved.
704 * < 0: error. skb freed, reference to sock dropped.
706 * 1: repeat lookup - reference dropped while waiting for socket memory.
708 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
709 long timeo, struct sock *ssk)
711 struct netlink_sock *nlk;
715 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
716 test_bit(0, &nlk->state)) {
717 DECLARE_WAITQUEUE(wait, current);
719 if (!ssk || nlk_sk(ssk)->pid == 0)
726 __set_current_state(TASK_INTERRUPTIBLE);
727 add_wait_queue(&nlk->wait, &wait);
729 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
730 test_bit(0, &nlk->state)) &&
731 !sock_flag(sk, SOCK_DEAD))
732 timeo = schedule_timeout(timeo);
734 __set_current_state(TASK_RUNNING);
735 remove_wait_queue(&nlk->wait, &wait);
738 if (signal_pending(current)) {
740 return sock_intr_errno(timeo);
744 skb_set_owner_r(skb, sk);
748 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
752 skb_queue_tail(&sk->sk_receive_queue, skb);
753 sk->sk_data_ready(sk, len);
758 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
764 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
771 delta = skb->end - skb->tail;
772 if (delta * 2 < skb->truesize)
775 if (skb_shared(skb)) {
776 struct sk_buff *nskb = skb_clone(skb, allocation);
783 if (!pskb_expand_head(skb, 0, -delta, allocation))
784 skb->truesize -= delta;
789 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
795 skb = netlink_trim(skb, gfp_any());
797 timeo = sock_sndtimeo(ssk, nonblock);
799 sk = netlink_getsockbypid(ssk, pid);
804 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
810 return netlink_sendskb(sk, skb, ssk->sk_protocol);
813 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
815 struct netlink_sock *nlk = nlk_sk(sk);
817 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
818 !test_bit(0, &nlk->state)) {
819 skb_set_owner_r(skb, sk);
820 skb_queue_tail(&sk->sk_receive_queue, skb);
821 sk->sk_data_ready(sk, skb->len);
822 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
827 struct netlink_broadcast_data {
828 struct sock *exclude_sk;
835 struct sk_buff *skb, *skb2;
838 static inline int do_one_broadcast(struct sock *sk,
839 struct netlink_broadcast_data *p)
841 struct netlink_sock *nlk = nlk_sk(sk);
844 if (p->exclude_sk == sk)
847 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
848 !test_bit(p->group - 1, nlk->groups))
857 if (p->skb2 == NULL) {
858 if (skb_shared(p->skb)) {
859 p->skb2 = skb_clone(p->skb, p->allocation);
861 p->skb2 = skb_get(p->skb);
863 * skb ownership may have been set when
864 * delivered to a previous socket.
869 if (p->skb2 == NULL) {
871 /* Clone failed. Notify ALL listeners. */
873 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
886 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
887 u32 group, gfp_t allocation)
889 struct netlink_broadcast_data info;
890 struct hlist_node *node;
893 skb = netlink_trim(skb, allocation);
895 info.exclude_sk = ssk;
901 info.allocation = allocation;
905 /* While we sleep in clone, do not allow to change socket list */
907 netlink_lock_table();
909 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
910 do_one_broadcast(sk, &info);
914 netlink_unlock_table();
917 kfree_skb(info.skb2);
919 if (info.delivered) {
920 if (info.congested && (allocation & __GFP_WAIT))
929 struct netlink_set_err_data {
930 struct sock *exclude_sk;
936 static inline int do_one_set_err(struct sock *sk,
937 struct netlink_set_err_data *p)
939 struct netlink_sock *nlk = nlk_sk(sk);
941 if (sk == p->exclude_sk)
944 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
945 !test_bit(p->group - 1, nlk->groups))
948 sk->sk_err = p->code;
949 sk->sk_error_report(sk);
954 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
956 struct netlink_set_err_data info;
957 struct hlist_node *node;
960 info.exclude_sk = ssk;
965 read_lock(&nl_table_lock);
967 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
968 do_one_set_err(sk, &info);
970 read_unlock(&nl_table_lock);
973 static int netlink_setsockopt(struct socket *sock, int level, int optname,
974 char __user *optval, int optlen)
976 struct sock *sk = sock->sk;
977 struct netlink_sock *nlk = nlk_sk(sk);
980 if (level != SOL_NETLINK)
983 if (optlen >= sizeof(int) &&
984 get_user(val, (int __user *)optval))
988 case NETLINK_PKTINFO:
990 nlk->flags |= NETLINK_RECV_PKTINFO;
992 nlk->flags &= ~NETLINK_RECV_PKTINFO;
995 case NETLINK_ADD_MEMBERSHIP:
996 case NETLINK_DROP_MEMBERSHIP: {
997 unsigned int subscriptions;
998 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1000 if (!netlink_capable(sock, NL_NONROOT_RECV))
1002 if (nlk->groups == NULL) {
1003 err = netlink_alloc_groups(sk);
1007 if (!val || val - 1 >= nlk->ngroups)
1009 netlink_table_grab();
1010 old = test_bit(val - 1, nlk->groups);
1011 subscriptions = nlk->subscriptions - old + new;
1013 __set_bit(val - 1, nlk->groups);
1015 __clear_bit(val - 1, nlk->groups);
1016 netlink_update_subscriptions(sk, subscriptions);
1017 netlink_table_ungrab();
1027 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1028 char __user *optval, int __user *optlen)
1030 struct sock *sk = sock->sk;
1031 struct netlink_sock *nlk = nlk_sk(sk);
1034 if (level != SOL_NETLINK)
1035 return -ENOPROTOOPT;
1037 if (get_user(len, optlen))
1043 case NETLINK_PKTINFO:
1044 if (len < sizeof(int))
1047 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1048 put_user(len, optlen);
1049 put_user(val, optval);
1058 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1060 struct nl_pktinfo info;
1062 info.group = NETLINK_CB(skb).dst_group;
1063 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1066 static inline void netlink_rcv_wake(struct sock *sk)
1068 struct netlink_sock *nlk = nlk_sk(sk);
1070 if (skb_queue_empty(&sk->sk_receive_queue))
1071 clear_bit(0, &nlk->state);
1072 if (!test_bit(0, &nlk->state))
1073 wake_up_interruptible(&nlk->wait);
1076 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1077 struct msghdr *msg, size_t len)
1079 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1080 struct sock *sk = sock->sk;
1081 struct netlink_sock *nlk = nlk_sk(sk);
1082 struct sockaddr_nl *addr=msg->msg_name;
1085 struct sk_buff *skb;
1087 struct scm_cookie scm;
1089 if (msg->msg_flags&MSG_OOB)
1092 if (NULL == siocb->scm)
1094 err = scm_send(sock, msg, siocb->scm);
1098 if (msg->msg_namelen) {
1099 if (addr->nl_family != AF_NETLINK)
1101 dst_pid = addr->nl_pid;
1102 dst_group = ffs(addr->nl_groups);
1103 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1106 dst_pid = nlk->dst_pid;
1107 dst_group = nlk->dst_group;
1111 err = netlink_autobind(sock);
1117 if (len > sk->sk_sndbuf - 32)
1120 skb = alloc_skb(len, GFP_KERNEL);
1124 NETLINK_CB(skb).pid = nlk->pid;
1125 NETLINK_CB(skb).dst_pid = dst_pid;
1126 NETLINK_CB(skb).dst_group = dst_group;
1127 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1128 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1130 /* What can I do? Netlink is asynchronous, so that
1131 we will have to save current capabilities to
1132 check them, when this message will be delivered
1133 to corresponding kernel module. --ANK (980802)
1137 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1142 err = security_netlink_send(sk, skb);
1149 atomic_inc(&skb->users);
1150 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1152 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1158 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1159 struct msghdr *msg, size_t len,
1162 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1163 struct scm_cookie scm;
1164 struct sock *sk = sock->sk;
1165 struct netlink_sock *nlk = nlk_sk(sk);
1166 int noblock = flags&MSG_DONTWAIT;
1168 struct sk_buff *skb;
1176 skb = skb_recv_datagram(sk,flags,noblock,&err);
1180 msg->msg_namelen = 0;
1184 msg->msg_flags |= MSG_TRUNC;
1188 skb->h.raw = skb->data;
1189 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1191 if (msg->msg_name) {
1192 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1193 addr->nl_family = AF_NETLINK;
1195 addr->nl_pid = NETLINK_CB(skb).pid;
1196 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1197 msg->msg_namelen = sizeof(*addr);
1200 if (nlk->flags & NETLINK_RECV_PKTINFO)
1201 netlink_cmsg_recv_pktinfo(msg, skb);
1203 if (NULL == siocb->scm) {
1204 memset(&scm, 0, sizeof(scm));
1207 siocb->scm->creds = *NETLINK_CREDS(skb);
1208 skb_free_datagram(sk, skb);
1210 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1213 scm_recv(sock, msg, siocb->scm, flags);
1216 netlink_rcv_wake(sk);
1217 return err ? : copied;
1220 static void netlink_data_ready(struct sock *sk, int len)
1222 struct netlink_sock *nlk = nlk_sk(sk);
1224 if (nlk->data_ready)
1225 nlk->data_ready(sk, len);
1226 netlink_rcv_wake(sk);
1230 * We export these functions to other modules. They provide a
1231 * complete set of kernel non-blocking support for message
1236 netlink_kernel_create(int unit, unsigned int groups,
1237 void (*input)(struct sock *sk, int len),
1238 struct module *module)
1240 struct socket *sock;
1242 struct netlink_sock *nlk;
1247 if (unit<0 || unit>=MAX_LINKS)
1250 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1253 if (__netlink_create(sock, unit) < 0)
1254 goto out_sock_release;
1257 sk->sk_data_ready = netlink_data_ready;
1259 nlk_sk(sk)->data_ready = input;
1261 if (netlink_insert(sk, 0))
1262 goto out_sock_release;
1265 nlk->flags |= NETLINK_KERNEL_SOCKET;
1267 netlink_table_grab();
1268 nl_table[unit].groups = groups < 32 ? 32 : groups;
1269 nl_table[unit].module = module;
1270 nl_table[unit].registered = 1;
1271 netlink_table_ungrab();
1280 void netlink_set_nonroot(int protocol, unsigned int flags)
1282 if ((unsigned int)protocol < MAX_LINKS)
1283 nl_table[protocol].nl_nonroot = flags;
1286 static void netlink_destroy_callback(struct netlink_callback *cb)
1294 * It looks a bit ugly.
1295 * It would be better to create kernel thread.
1298 static int netlink_dump(struct sock *sk)
1300 struct netlink_sock *nlk = nlk_sk(sk);
1301 struct netlink_callback *cb;
1302 struct sk_buff *skb;
1303 struct nlmsghdr *nlh;
1306 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1310 spin_lock(&nlk->cb_lock);
1314 spin_unlock(&nlk->cb_lock);
1319 len = cb->dump(skb, cb);
1322 spin_unlock(&nlk->cb_lock);
1323 skb_queue_tail(&sk->sk_receive_queue, skb);
1324 sk->sk_data_ready(sk, len);
1328 nlh = NLMSG_NEW_ANSWER(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1329 memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1330 skb_queue_tail(&sk->sk_receive_queue, skb);
1331 sk->sk_data_ready(sk, skb->len);
1336 spin_unlock(&nlk->cb_lock);
1338 netlink_destroy_callback(cb);
1345 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1346 struct nlmsghdr *nlh,
1347 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1348 int (*done)(struct netlink_callback*))
1350 struct netlink_callback *cb;
1352 struct netlink_sock *nlk;
1354 cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1358 memset(cb, 0, sizeof(*cb));
1362 atomic_inc(&skb->users);
1365 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1367 netlink_destroy_callback(cb);
1368 return -ECONNREFUSED;
1371 /* A dump is in progress... */
1372 spin_lock(&nlk->cb_lock);
1374 spin_unlock(&nlk->cb_lock);
1375 netlink_destroy_callback(cb);
1380 spin_unlock(&nlk->cb_lock);
1387 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1389 struct sk_buff *skb;
1390 struct nlmsghdr *rep;
1391 struct nlmsgerr *errmsg;
1395 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1397 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1399 skb = alloc_skb(size, GFP_KERNEL);
1403 sk = netlink_lookup(in_skb->sk->sk_protocol,
1404 NETLINK_CB(in_skb).pid);
1406 sk->sk_err = ENOBUFS;
1407 sk->sk_error_report(sk);
1413 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1414 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1415 errmsg = NLMSG_DATA(rep);
1416 errmsg->error = err;
1417 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1418 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1421 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1422 struct nlmsghdr *, int *))
1424 unsigned int total_len;
1425 struct nlmsghdr *nlh;
1428 while (skb->len >= nlmsg_total_size(0)) {
1429 nlh = (struct nlmsghdr *) skb->data;
1431 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1434 total_len = min(NLMSG_ALIGN(nlh->nlmsg_len), skb->len);
1436 if (cb(skb, nlh, &err) < 0) {
1437 /* Not an error, but we have to interrupt processing
1438 * here. Note: that in this case we do not pull
1439 * message from skb, it will be processed later.
1443 netlink_ack(skb, nlh, err);
1444 } else if (nlh->nlmsg_flags & NLM_F_ACK)
1445 netlink_ack(skb, nlh, 0);
1447 skb_pull(skb, total_len);
1454 * nelink_run_queue - Process netlink receive queue.
1455 * @sk: Netlink socket containing the queue
1456 * @qlen: Place to store queue length upon entry
1457 * @cb: Callback function invoked for each netlink message found
1459 * Processes as much as there was in the queue upon entry and invokes
1460 * a callback function for each netlink message found. The callback
1461 * function may refuse a message by returning a negative error code
1462 * but setting the error pointer to 0 in which case this function
1463 * returns with a qlen != 0.
1465 * qlen must be initialized to 0 before the initial entry, afterwards
1466 * the function may be called repeatedly until qlen reaches 0.
1468 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1469 int (*cb)(struct sk_buff *, struct nlmsghdr *, int *))
1471 struct sk_buff *skb;
1473 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1474 *qlen = skb_queue_len(&sk->sk_receive_queue);
1476 for (; *qlen; (*qlen)--) {
1477 skb = skb_dequeue(&sk->sk_receive_queue);
1478 if (netlink_rcv_skb(skb, cb)) {
1480 skb_queue_head(&sk->sk_receive_queue, skb);
1493 * netlink_queue_skip - Skip netlink message while processing queue.
1494 * @nlh: Netlink message to be skipped
1495 * @skb: Socket buffer containing the netlink messages.
1497 * Pulls the given netlink message off the socket buffer so the next
1498 * call to netlink_queue_run() will not reconsider the message.
1500 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1502 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1504 if (msglen > skb->len)
1507 skb_pull(skb, msglen);
1510 #ifdef CONFIG_PROC_FS
1511 struct nl_seq_iter {
1516 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1518 struct nl_seq_iter *iter = seq->private;
1521 struct hlist_node *node;
1524 for (i=0; i<MAX_LINKS; i++) {
1525 struct nl_pid_hash *hash = &nl_table[i].hash;
1527 for (j = 0; j <= hash->mask; j++) {
1528 sk_for_each(s, node, &hash->table[j]) {
1541 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1543 read_lock(&nl_table_lock);
1544 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1547 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1550 struct nl_seq_iter *iter;
1555 if (v == SEQ_START_TOKEN)
1556 return netlink_seq_socket_idx(seq, 0);
1562 iter = seq->private;
1564 j = iter->hash_idx + 1;
1567 struct nl_pid_hash *hash = &nl_table[i].hash;
1569 for (; j <= hash->mask; j++) {
1570 s = sk_head(&hash->table[j]);
1579 } while (++i < MAX_LINKS);
1584 static void netlink_seq_stop(struct seq_file *seq, void *v)
1586 read_unlock(&nl_table_lock);
1590 static int netlink_seq_show(struct seq_file *seq, void *v)
1592 if (v == SEQ_START_TOKEN)
1594 "sk Eth Pid Groups "
1595 "Rmem Wmem Dump Locks\n");
1598 struct netlink_sock *nlk = nlk_sk(s);
1600 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1604 nlk->groups ? (u32)nlk->groups[0] : 0,
1605 atomic_read(&s->sk_rmem_alloc),
1606 atomic_read(&s->sk_wmem_alloc),
1608 atomic_read(&s->sk_refcnt)
1615 static struct seq_operations netlink_seq_ops = {
1616 .start = netlink_seq_start,
1617 .next = netlink_seq_next,
1618 .stop = netlink_seq_stop,
1619 .show = netlink_seq_show,
1623 static int netlink_seq_open(struct inode *inode, struct file *file)
1625 struct seq_file *seq;
1626 struct nl_seq_iter *iter;
1629 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1633 err = seq_open(file, &netlink_seq_ops);
1639 memset(iter, 0, sizeof(*iter));
1640 seq = file->private_data;
1641 seq->private = iter;
1645 static struct file_operations netlink_seq_fops = {
1646 .owner = THIS_MODULE,
1647 .open = netlink_seq_open,
1649 .llseek = seq_lseek,
1650 .release = seq_release_private,
1655 int netlink_register_notifier(struct notifier_block *nb)
1657 return notifier_chain_register(&netlink_chain, nb);
1660 int netlink_unregister_notifier(struct notifier_block *nb)
1662 return notifier_chain_unregister(&netlink_chain, nb);
1665 static const struct proto_ops netlink_ops = {
1666 .family = PF_NETLINK,
1667 .owner = THIS_MODULE,
1668 .release = netlink_release,
1669 .bind = netlink_bind,
1670 .connect = netlink_connect,
1671 .socketpair = sock_no_socketpair,
1672 .accept = sock_no_accept,
1673 .getname = netlink_getname,
1674 .poll = datagram_poll,
1675 .ioctl = sock_no_ioctl,
1676 .listen = sock_no_listen,
1677 .shutdown = sock_no_shutdown,
1678 .setsockopt = netlink_setsockopt,
1679 .getsockopt = netlink_getsockopt,
1680 .sendmsg = netlink_sendmsg,
1681 .recvmsg = netlink_recvmsg,
1682 .mmap = sock_no_mmap,
1683 .sendpage = sock_no_sendpage,
1686 static struct net_proto_family netlink_family_ops = {
1687 .family = PF_NETLINK,
1688 .create = netlink_create,
1689 .owner = THIS_MODULE, /* for consistency 8) */
1692 extern void netlink_skb_parms_too_large(void);
1694 static int __init netlink_proto_init(void)
1696 struct sk_buff *dummy_skb;
1700 int err = proto_register(&netlink_proto, 0);
1705 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1706 netlink_skb_parms_too_large();
1708 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1711 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1715 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1717 if (num_physpages >= (128 * 1024))
1718 max = num_physpages >> (21 - PAGE_SHIFT);
1720 max = num_physpages >> (23 - PAGE_SHIFT);
1722 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1723 max = (1UL << order) / sizeof(struct hlist_head);
1724 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1726 for (i = 0; i < MAX_LINKS; i++) {
1727 struct nl_pid_hash *hash = &nl_table[i].hash;
1729 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1732 nl_pid_hash_free(nl_table[i].hash.table,
1733 1 * sizeof(*hash->table));
1737 memset(hash->table, 0, 1 * sizeof(*hash->table));
1738 hash->max_shift = order;
1741 hash->rehash_time = jiffies;
1744 sock_register(&netlink_family_ops);
1745 #ifdef CONFIG_PROC_FS
1746 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1748 /* The netlink device handler may be needed early. */
1754 core_initcall(netlink_proto_init);
1756 EXPORT_SYMBOL(netlink_ack);
1757 EXPORT_SYMBOL(netlink_run_queue);
1758 EXPORT_SYMBOL(netlink_queue_skip);
1759 EXPORT_SYMBOL(netlink_broadcast);
1760 EXPORT_SYMBOL(netlink_dump_start);
1761 EXPORT_SYMBOL(netlink_kernel_create);
1762 EXPORT_SYMBOL(netlink_register_notifier);
1763 EXPORT_SYMBOL(netlink_set_err);
1764 EXPORT_SYMBOL(netlink_set_nonroot);
1765 EXPORT_SYMBOL(netlink_unicast);
1766 EXPORT_SYMBOL(netlink_unregister_notifier);