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/module.h>
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/smp_lock.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/selinux.h>
59 #include <linux/vs_context.h>
60 #include <linux/vs_network.h>
61 #include <linux/vs_limit.h>
65 #include <net/netlink.h>
67 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
70 /* struct sock has to be the first member of netlink_sock */
78 unsigned long *groups;
80 wait_queue_head_t wait;
81 struct netlink_callback *cb;
83 void (*data_ready)(struct sock *sk, int bytes);
84 struct module *module;
87 #define NETLINK_KERNEL_SOCKET 0x1
88 #define NETLINK_RECV_PKTINFO 0x2
90 static inline struct netlink_sock *nlk_sk(struct sock *sk)
92 return (struct netlink_sock *)sk;
96 struct hlist_head *table;
97 unsigned long rehash_time;
102 unsigned int entries;
103 unsigned int max_shift;
108 struct netlink_table {
109 struct nl_pid_hash hash;
110 struct hlist_head mc_list;
111 unsigned long *listeners;
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 ATOMIC_NOTIFIER_HEAD(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_irq(&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_irq(&nl_table_lock);
174 write_lock_irq(&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_irq(&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;
303 netlink_update_listeners(struct sock *sk)
305 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
306 struct hlist_node *node;
310 for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) {
312 sk_for_each_bound(sk, node, &tbl->mc_list)
313 mask |= nlk_sk(sk)->groups[i];
314 tbl->listeners[i] = mask;
316 /* this function is only called with the netlink table "grabbed", which
317 * makes sure updates are visible before bind or setsockopt return. */
320 static int netlink_insert(struct sock *sk, u32 pid)
322 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
323 struct hlist_head *head;
324 int err = -EADDRINUSE;
326 struct hlist_node *node;
329 netlink_table_grab();
330 head = nl_pid_hashfn(hash, pid);
332 sk_for_each(osk, node, head) {
333 if (nlk_sk(osk)->pid == pid)
345 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
348 if (len && nl_pid_hash_dilute(hash, len))
349 head = nl_pid_hashfn(hash, pid);
351 nlk_sk(sk)->pid = pid;
352 sk_add_node(sk, head);
356 netlink_table_ungrab();
360 static void netlink_remove(struct sock *sk)
362 netlink_table_grab();
363 if (sk_del_node_init(sk))
364 nl_table[sk->sk_protocol].hash.entries--;
365 if (nlk_sk(sk)->subscriptions)
366 __sk_del_bind_node(sk);
367 netlink_table_ungrab();
370 static struct proto netlink_proto = {
372 .owner = THIS_MODULE,
373 .obj_size = sizeof(struct netlink_sock),
376 static int __netlink_create(struct socket *sock, int protocol)
379 struct netlink_sock *nlk;
381 sock->ops = &netlink_ops;
383 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
387 sock_init_data(sock, sk);
390 spin_lock_init(&nlk->cb_lock);
391 init_waitqueue_head(&nlk->wait);
393 sk->sk_destruct = netlink_sock_destruct;
394 sk->sk_protocol = protocol;
398 static int netlink_create(struct socket *sock, int protocol)
400 struct module *module = NULL;
401 struct netlink_sock *nlk;
405 sock->state = SS_UNCONNECTED;
407 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
408 return -ESOCKTNOSUPPORT;
410 if (protocol<0 || protocol >= MAX_LINKS)
411 return -EPROTONOSUPPORT;
413 netlink_lock_table();
415 if (!nl_table[protocol].registered) {
416 netlink_unlock_table();
417 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
418 netlink_lock_table();
421 if (nl_table[protocol].registered &&
422 try_module_get(nl_table[protocol].module))
423 module = nl_table[protocol].module;
424 groups = nl_table[protocol].groups;
425 netlink_unlock_table();
427 if ((err = __netlink_create(sock, protocol)) < 0)
430 nlk = nlk_sk(sock->sk);
431 nlk->module = module;
440 static int netlink_release(struct socket *sock)
442 struct sock *sk = sock->sk;
443 struct netlink_sock *nlk;
451 spin_lock(&nlk->cb_lock);
454 nlk->cb->done(nlk->cb);
455 netlink_destroy_callback(nlk->cb);
458 spin_unlock(&nlk->cb_lock);
460 /* OK. Socket is unlinked, and, therefore,
461 no new packets will arrive */
465 wake_up_interruptible_all(&nlk->wait);
467 skb_queue_purge(&sk->sk_write_queue);
469 if (nlk->pid && !nlk->subscriptions) {
470 struct netlink_notify n = {
471 .protocol = sk->sk_protocol,
474 atomic_notifier_call_chain(&netlink_chain,
475 NETLINK_URELEASE, &n);
479 module_put(nlk->module);
481 netlink_table_grab();
482 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
483 kfree(nl_table[sk->sk_protocol].listeners);
484 nl_table[sk->sk_protocol].module = NULL;
485 nl_table[sk->sk_protocol].registered = 0;
486 } else if (nlk->subscriptions)
487 netlink_update_listeners(sk);
488 netlink_table_ungrab();
497 static int netlink_autobind(struct socket *sock)
499 struct sock *sk = sock->sk;
500 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
501 struct hlist_head *head;
503 struct hlist_node *node;
504 s32 pid = current->tgid;
506 static s32 rover = -4097;
510 netlink_table_grab();
511 head = nl_pid_hashfn(hash, pid);
512 sk_for_each(osk, node, head) {
513 if (nlk_sk(osk)->pid == pid) {
514 /* Bind collision, search negative pid values. */
518 netlink_table_ungrab();
522 netlink_table_ungrab();
524 err = netlink_insert(sk, pid);
525 if (err == -EADDRINUSE)
528 /* If 2 threads race to autobind, that is fine. */
535 static inline int netlink_capable(struct socket *sock, unsigned int flag)
537 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
538 capable(CAP_NET_ADMIN);
542 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
544 struct netlink_sock *nlk = nlk_sk(sk);
546 if (nlk->subscriptions && !subscriptions)
547 __sk_del_bind_node(sk);
548 else if (!nlk->subscriptions && subscriptions)
549 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
550 nlk->subscriptions = subscriptions;
553 static int netlink_alloc_groups(struct sock *sk)
555 struct netlink_sock *nlk = nlk_sk(sk);
559 netlink_lock_table();
560 groups = nl_table[sk->sk_protocol].groups;
561 if (!nl_table[sk->sk_protocol].registered)
563 netlink_unlock_table();
568 nlk->groups = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
569 if (nlk->groups == NULL)
571 nlk->ngroups = groups;
575 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
577 struct sock *sk = sock->sk;
578 struct netlink_sock *nlk = nlk_sk(sk);
579 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
582 if (nladdr->nl_family != AF_NETLINK)
585 /* Only superuser is allowed to listen multicasts */
586 if (nladdr->nl_groups) {
587 if (!netlink_capable(sock, NL_NONROOT_RECV))
589 if (nlk->groups == NULL) {
590 err = netlink_alloc_groups(sk);
597 if (nladdr->nl_pid != nlk->pid)
600 err = nladdr->nl_pid ?
601 netlink_insert(sk, nladdr->nl_pid) :
602 netlink_autobind(sock);
607 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
610 netlink_table_grab();
611 netlink_update_subscriptions(sk, nlk->subscriptions +
612 hweight32(nladdr->nl_groups) -
613 hweight32(nlk->groups[0]));
614 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
615 netlink_update_listeners(sk);
616 netlink_table_ungrab();
621 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
625 struct sock *sk = sock->sk;
626 struct netlink_sock *nlk = nlk_sk(sk);
627 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
629 if (addr->sa_family == AF_UNSPEC) {
630 sk->sk_state = NETLINK_UNCONNECTED;
635 if (addr->sa_family != AF_NETLINK)
638 /* Only superuser is allowed to send multicasts */
639 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
643 err = netlink_autobind(sock);
646 sk->sk_state = NETLINK_CONNECTED;
647 nlk->dst_pid = nladdr->nl_pid;
648 nlk->dst_group = ffs(nladdr->nl_groups);
654 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
656 struct sock *sk = sock->sk;
657 struct netlink_sock *nlk = nlk_sk(sk);
658 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
660 nladdr->nl_family = AF_NETLINK;
662 *addr_len = sizeof(*nladdr);
665 nladdr->nl_pid = nlk->dst_pid;
666 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
668 nladdr->nl_pid = nlk->pid;
669 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
674 static void netlink_overrun(struct sock *sk)
676 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
677 sk->sk_err = ENOBUFS;
678 sk->sk_error_report(sk);
682 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
684 int protocol = ssk->sk_protocol;
686 struct netlink_sock *nlk;
688 sock = netlink_lookup(protocol, pid);
690 return ERR_PTR(-ECONNREFUSED);
692 /* Don't bother queuing skb if kernel socket has no input function */
694 if ((nlk->pid == 0 && !nlk->data_ready) ||
695 (sock->sk_state == NETLINK_CONNECTED &&
696 nlk->dst_pid != nlk_sk(ssk)->pid)) {
698 return ERR_PTR(-ECONNREFUSED);
703 struct sock *netlink_getsockbyfilp(struct file *filp)
705 struct inode *inode = filp->f_dentry->d_inode;
708 if (!S_ISSOCK(inode->i_mode))
709 return ERR_PTR(-ENOTSOCK);
711 sock = SOCKET_I(inode)->sk;
712 if (sock->sk_family != AF_NETLINK)
713 return ERR_PTR(-EINVAL);
720 * Attach a skb to a netlink socket.
721 * The caller must hold a reference to the destination socket. On error, the
722 * reference is dropped. The skb is not send to the destination, just all
723 * all error checks are performed and memory in the queue is reserved.
725 * < 0: error. skb freed, reference to sock dropped.
727 * 1: repeat lookup - reference dropped while waiting for socket memory.
729 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
730 long timeo, struct sock *ssk)
732 struct netlink_sock *nlk;
736 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
737 test_bit(0, &nlk->state)) {
738 DECLARE_WAITQUEUE(wait, current);
740 if (!ssk || nlk_sk(ssk)->pid == 0)
747 __set_current_state(TASK_INTERRUPTIBLE);
748 add_wait_queue(&nlk->wait, &wait);
750 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
751 test_bit(0, &nlk->state)) &&
752 !sock_flag(sk, SOCK_DEAD))
753 timeo = schedule_timeout(timeo);
755 __set_current_state(TASK_RUNNING);
756 remove_wait_queue(&nlk->wait, &wait);
759 if (signal_pending(current)) {
761 return sock_intr_errno(timeo);
765 skb_set_owner_r(skb, sk);
769 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
773 skb_queue_tail(&sk->sk_receive_queue, skb);
774 sk->sk_data_ready(sk, len);
779 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
785 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
792 delta = skb->end - skb->tail;
793 if (delta * 2 < skb->truesize)
796 if (skb_shared(skb)) {
797 struct sk_buff *nskb = skb_clone(skb, allocation);
804 if (!pskb_expand_head(skb, 0, -delta, allocation))
805 skb->truesize -= delta;
810 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
816 skb = netlink_trim(skb, gfp_any());
818 timeo = sock_sndtimeo(ssk, nonblock);
820 sk = netlink_getsockbypid(ssk, pid);
825 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
831 return netlink_sendskb(sk, skb, ssk->sk_protocol);
834 int netlink_has_listeners(struct sock *sk, unsigned int group)
838 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
839 if (group - 1 < nl_table[sk->sk_protocol].groups)
840 res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners);
843 EXPORT_SYMBOL_GPL(netlink_has_listeners);
845 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
847 struct netlink_sock *nlk = nlk_sk(sk);
849 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
850 !test_bit(0, &nlk->state)) {
851 skb_set_owner_r(skb, sk);
852 skb_queue_tail(&sk->sk_receive_queue, skb);
853 sk->sk_data_ready(sk, skb->len);
854 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
859 struct netlink_broadcast_data {
860 struct sock *exclude_sk;
867 struct sk_buff *skb, *skb2;
870 static inline int do_one_broadcast(struct sock *sk,
871 struct netlink_broadcast_data *p)
873 struct netlink_sock *nlk = nlk_sk(sk);
876 if (p->exclude_sk == sk)
879 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
880 !test_bit(p->group - 1, nlk->groups))
889 if (p->skb2 == NULL) {
890 if (skb_shared(p->skb)) {
891 p->skb2 = skb_clone(p->skb, p->allocation);
893 p->skb2 = skb_get(p->skb);
895 * skb ownership may have been set when
896 * delivered to a previous socket.
901 if (p->skb2 == NULL) {
903 /* Clone failed. Notify ALL listeners. */
905 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
918 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
919 u32 group, gfp_t allocation)
921 struct netlink_broadcast_data info;
922 struct hlist_node *node;
925 skb = netlink_trim(skb, allocation);
927 info.exclude_sk = ssk;
933 info.allocation = allocation;
937 /* While we sleep in clone, do not allow to change socket list */
939 netlink_lock_table();
941 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
942 do_one_broadcast(sk, &info);
946 netlink_unlock_table();
949 kfree_skb(info.skb2);
951 if (info.delivered) {
952 if (info.congested && (allocation & __GFP_WAIT))
961 struct netlink_set_err_data {
962 struct sock *exclude_sk;
968 static inline int do_one_set_err(struct sock *sk,
969 struct netlink_set_err_data *p)
971 struct netlink_sock *nlk = nlk_sk(sk);
973 if (sk == p->exclude_sk)
976 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
977 !test_bit(p->group - 1, nlk->groups))
980 sk->sk_err = p->code;
981 sk->sk_error_report(sk);
986 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
988 struct netlink_set_err_data info;
989 struct hlist_node *node;
992 info.exclude_sk = ssk;
997 read_lock(&nl_table_lock);
999 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1000 do_one_set_err(sk, &info);
1002 read_unlock(&nl_table_lock);
1005 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1006 char __user *optval, int optlen)
1008 struct sock *sk = sock->sk;
1009 struct netlink_sock *nlk = nlk_sk(sk);
1012 if (level != SOL_NETLINK)
1013 return -ENOPROTOOPT;
1015 if (optlen >= sizeof(int) &&
1016 get_user(val, (int __user *)optval))
1020 case NETLINK_PKTINFO:
1022 nlk->flags |= NETLINK_RECV_PKTINFO;
1024 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1027 case NETLINK_ADD_MEMBERSHIP:
1028 case NETLINK_DROP_MEMBERSHIP: {
1029 unsigned int subscriptions;
1030 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1032 if (!netlink_capable(sock, NL_NONROOT_RECV))
1034 if (nlk->groups == NULL) {
1035 err = netlink_alloc_groups(sk);
1039 if (!val || val - 1 >= nlk->ngroups)
1041 netlink_table_grab();
1042 old = test_bit(val - 1, nlk->groups);
1043 subscriptions = nlk->subscriptions - old + new;
1045 __set_bit(val - 1, nlk->groups);
1047 __clear_bit(val - 1, nlk->groups);
1048 netlink_update_subscriptions(sk, subscriptions);
1049 netlink_update_listeners(sk);
1050 netlink_table_ungrab();
1060 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1061 char __user *optval, int __user *optlen)
1063 struct sock *sk = sock->sk;
1064 struct netlink_sock *nlk = nlk_sk(sk);
1067 if (level != SOL_NETLINK)
1068 return -ENOPROTOOPT;
1070 if (get_user(len, optlen))
1076 case NETLINK_PKTINFO:
1077 if (len < sizeof(int))
1080 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1081 put_user(len, optlen);
1082 put_user(val, optval);
1091 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1093 struct nl_pktinfo info;
1095 info.group = NETLINK_CB(skb).dst_group;
1096 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1099 static inline void netlink_rcv_wake(struct sock *sk)
1101 struct netlink_sock *nlk = nlk_sk(sk);
1103 if (skb_queue_empty(&sk->sk_receive_queue))
1104 clear_bit(0, &nlk->state);
1105 if (!test_bit(0, &nlk->state))
1106 wake_up_interruptible(&nlk->wait);
1109 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1110 struct msghdr *msg, size_t len)
1112 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1113 struct sock *sk = sock->sk;
1114 struct netlink_sock *nlk = nlk_sk(sk);
1115 struct sockaddr_nl *addr=msg->msg_name;
1118 struct sk_buff *skb;
1120 struct scm_cookie scm;
1122 if (msg->msg_flags&MSG_OOB)
1125 if (NULL == siocb->scm)
1127 err = scm_send(sock, msg, siocb->scm);
1131 if (msg->msg_namelen) {
1132 if (addr->nl_family != AF_NETLINK)
1134 dst_pid = addr->nl_pid;
1135 dst_group = ffs(addr->nl_groups);
1136 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1139 dst_pid = nlk->dst_pid;
1140 dst_group = nlk->dst_group;
1144 err = netlink_autobind(sock);
1150 if (len > sk->sk_sndbuf - 32)
1153 skb = alloc_skb(len, GFP_KERNEL);
1157 NETLINK_CB(skb).pid = nlk->pid;
1158 NETLINK_CB(skb).dst_pid = dst_pid;
1159 NETLINK_CB(skb).dst_group = dst_group;
1160 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1161 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1162 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1164 /* What can I do? Netlink is asynchronous, so that
1165 we will have to save current capabilities to
1166 check them, when this message will be delivered
1167 to corresponding kernel module. --ANK (980802)
1171 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1176 err = security_netlink_send(sk, skb);
1183 atomic_inc(&skb->users);
1184 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1186 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1192 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1193 struct msghdr *msg, size_t len,
1196 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1197 struct scm_cookie scm;
1198 struct sock *sk = sock->sk;
1199 struct netlink_sock *nlk = nlk_sk(sk);
1200 int noblock = flags&MSG_DONTWAIT;
1202 struct sk_buff *skb;
1210 skb = skb_recv_datagram(sk,flags,noblock,&err);
1214 msg->msg_namelen = 0;
1218 msg->msg_flags |= MSG_TRUNC;
1222 skb->h.raw = skb->data;
1223 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1225 if (msg->msg_name) {
1226 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1227 addr->nl_family = AF_NETLINK;
1229 addr->nl_pid = NETLINK_CB(skb).pid;
1230 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1231 msg->msg_namelen = sizeof(*addr);
1234 if (nlk->flags & NETLINK_RECV_PKTINFO)
1235 netlink_cmsg_recv_pktinfo(msg, skb);
1237 if (NULL == siocb->scm) {
1238 memset(&scm, 0, sizeof(scm));
1241 siocb->scm->creds = *NETLINK_CREDS(skb);
1242 skb_free_datagram(sk, skb);
1244 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1247 scm_recv(sock, msg, siocb->scm, flags);
1250 netlink_rcv_wake(sk);
1251 return err ? : copied;
1254 static void netlink_data_ready(struct sock *sk, int len)
1256 struct netlink_sock *nlk = nlk_sk(sk);
1258 if (nlk->data_ready)
1259 nlk->data_ready(sk, len);
1260 netlink_rcv_wake(sk);
1264 * We export these functions to other modules. They provide a
1265 * complete set of kernel non-blocking support for message
1270 netlink_kernel_create(int unit, unsigned int groups,
1271 void (*input)(struct sock *sk, int len),
1272 struct module *module)
1274 struct socket *sock;
1276 struct netlink_sock *nlk;
1277 unsigned long *listeners = NULL;
1281 if (unit<0 || unit>=MAX_LINKS)
1284 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1287 if (__netlink_create(sock, unit) < 0)
1288 goto out_sock_release;
1293 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1295 goto out_sock_release;
1298 sk->sk_data_ready = netlink_data_ready;
1300 nlk_sk(sk)->data_ready = input;
1302 if (netlink_insert(sk, 0))
1303 goto out_sock_release;
1306 nlk->flags |= NETLINK_KERNEL_SOCKET;
1308 netlink_table_grab();
1309 nl_table[unit].groups = groups;
1310 nl_table[unit].listeners = listeners;
1311 nl_table[unit].module = module;
1312 nl_table[unit].registered = 1;
1313 netlink_table_ungrab();
1323 void netlink_set_nonroot(int protocol, unsigned int flags)
1325 if ((unsigned int)protocol < MAX_LINKS)
1326 nl_table[protocol].nl_nonroot = flags;
1329 static void netlink_destroy_callback(struct netlink_callback *cb)
1337 * It looks a bit ugly.
1338 * It would be better to create kernel thread.
1341 static int netlink_dump(struct sock *sk)
1343 struct netlink_sock *nlk = nlk_sk(sk);
1344 struct netlink_callback *cb;
1345 struct sk_buff *skb;
1346 struct nlmsghdr *nlh;
1349 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1353 spin_lock(&nlk->cb_lock);
1357 spin_unlock(&nlk->cb_lock);
1362 len = cb->dump(skb, cb);
1365 spin_unlock(&nlk->cb_lock);
1366 skb_queue_tail(&sk->sk_receive_queue, skb);
1367 sk->sk_data_ready(sk, len);
1371 nlh = NLMSG_NEW_ANSWER(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1372 memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1373 skb_queue_tail(&sk->sk_receive_queue, skb);
1374 sk->sk_data_ready(sk, skb->len);
1379 spin_unlock(&nlk->cb_lock);
1381 netlink_destroy_callback(cb);
1388 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1389 struct nlmsghdr *nlh,
1390 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1391 int (*done)(struct netlink_callback*))
1393 struct netlink_callback *cb;
1395 struct netlink_sock *nlk;
1397 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1404 atomic_inc(&skb->users);
1407 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1409 netlink_destroy_callback(cb);
1410 return -ECONNREFUSED;
1413 /* A dump is in progress... */
1414 spin_lock(&nlk->cb_lock);
1416 spin_unlock(&nlk->cb_lock);
1417 netlink_destroy_callback(cb);
1422 spin_unlock(&nlk->cb_lock);
1429 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1431 struct sk_buff *skb;
1432 struct nlmsghdr *rep;
1433 struct nlmsgerr *errmsg;
1437 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1439 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1441 skb = alloc_skb(size, GFP_KERNEL);
1445 sk = netlink_lookup(in_skb->sk->sk_protocol,
1446 NETLINK_CB(in_skb).pid);
1448 sk->sk_err = ENOBUFS;
1449 sk->sk_error_report(sk);
1455 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1456 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1457 errmsg = NLMSG_DATA(rep);
1458 errmsg->error = err;
1459 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1460 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1463 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1464 struct nlmsghdr *, int *))
1466 unsigned int total_len;
1467 struct nlmsghdr *nlh;
1470 while (skb->len >= nlmsg_total_size(0)) {
1471 nlh = (struct nlmsghdr *) skb->data;
1473 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1476 total_len = min(NLMSG_ALIGN(nlh->nlmsg_len), skb->len);
1478 if (cb(skb, nlh, &err) < 0) {
1479 /* Not an error, but we have to interrupt processing
1480 * here. Note: that in this case we do not pull
1481 * message from skb, it will be processed later.
1485 netlink_ack(skb, nlh, err);
1486 } else if (nlh->nlmsg_flags & NLM_F_ACK)
1487 netlink_ack(skb, nlh, 0);
1489 skb_pull(skb, total_len);
1496 * nelink_run_queue - Process netlink receive queue.
1497 * @sk: Netlink socket containing the queue
1498 * @qlen: Place to store queue length upon entry
1499 * @cb: Callback function invoked for each netlink message found
1501 * Processes as much as there was in the queue upon entry and invokes
1502 * a callback function for each netlink message found. The callback
1503 * function may refuse a message by returning a negative error code
1504 * but setting the error pointer to 0 in which case this function
1505 * returns with a qlen != 0.
1507 * qlen must be initialized to 0 before the initial entry, afterwards
1508 * the function may be called repeatedly until qlen reaches 0.
1510 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1511 int (*cb)(struct sk_buff *, struct nlmsghdr *, int *))
1513 struct sk_buff *skb;
1515 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1516 *qlen = skb_queue_len(&sk->sk_receive_queue);
1518 for (; *qlen; (*qlen)--) {
1519 skb = skb_dequeue(&sk->sk_receive_queue);
1520 if (netlink_rcv_skb(skb, cb)) {
1522 skb_queue_head(&sk->sk_receive_queue, skb);
1535 * netlink_queue_skip - Skip netlink message while processing queue.
1536 * @nlh: Netlink message to be skipped
1537 * @skb: Socket buffer containing the netlink messages.
1539 * Pulls the given netlink message off the socket buffer so the next
1540 * call to netlink_queue_run() will not reconsider the message.
1542 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1544 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1546 if (msglen > skb->len)
1549 skb_pull(skb, msglen);
1552 #ifdef CONFIG_PROC_FS
1553 struct nl_seq_iter {
1558 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1560 struct nl_seq_iter *iter = seq->private;
1563 struct hlist_node *node;
1566 for (i=0; i<MAX_LINKS; i++) {
1567 struct nl_pid_hash *hash = &nl_table[i].hash;
1569 for (j = 0; j <= hash->mask; j++) {
1570 sk_for_each(s, node, &hash->table[j]) {
1583 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1585 read_lock(&nl_table_lock);
1586 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1589 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1592 struct nl_seq_iter *iter;
1597 if (v == SEQ_START_TOKEN)
1598 return netlink_seq_socket_idx(seq, 0);
1604 iter = seq->private;
1606 j = iter->hash_idx + 1;
1609 struct nl_pid_hash *hash = &nl_table[i].hash;
1611 for (; j <= hash->mask; j++) {
1612 s = sk_head(&hash->table[j]);
1621 } while (++i < MAX_LINKS);
1626 static void netlink_seq_stop(struct seq_file *seq, void *v)
1628 read_unlock(&nl_table_lock);
1632 static int netlink_seq_show(struct seq_file *seq, void *v)
1634 if (v == SEQ_START_TOKEN)
1636 "sk Eth Pid Groups "
1637 "Rmem Wmem Dump Locks\n");
1640 struct netlink_sock *nlk = nlk_sk(s);
1642 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1646 nlk->groups ? (u32)nlk->groups[0] : 0,
1647 atomic_read(&s->sk_rmem_alloc),
1648 atomic_read(&s->sk_wmem_alloc),
1650 atomic_read(&s->sk_refcnt)
1657 static struct seq_operations netlink_seq_ops = {
1658 .start = netlink_seq_start,
1659 .next = netlink_seq_next,
1660 .stop = netlink_seq_stop,
1661 .show = netlink_seq_show,
1665 static int netlink_seq_open(struct inode *inode, struct file *file)
1667 struct seq_file *seq;
1668 struct nl_seq_iter *iter;
1671 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1675 err = seq_open(file, &netlink_seq_ops);
1681 seq = file->private_data;
1682 seq->private = iter;
1686 static struct file_operations netlink_seq_fops = {
1687 .owner = THIS_MODULE,
1688 .open = netlink_seq_open,
1690 .llseek = seq_lseek,
1691 .release = seq_release_private,
1696 int netlink_register_notifier(struct notifier_block *nb)
1698 return atomic_notifier_chain_register(&netlink_chain, nb);
1701 int netlink_unregister_notifier(struct notifier_block *nb)
1703 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1706 static const struct proto_ops netlink_ops = {
1707 .family = PF_NETLINK,
1708 .owner = THIS_MODULE,
1709 .release = netlink_release,
1710 .bind = netlink_bind,
1711 .connect = netlink_connect,
1712 .socketpair = sock_no_socketpair,
1713 .accept = sock_no_accept,
1714 .getname = netlink_getname,
1715 .poll = datagram_poll,
1716 .ioctl = sock_no_ioctl,
1717 .listen = sock_no_listen,
1718 .shutdown = sock_no_shutdown,
1719 .setsockopt = netlink_setsockopt,
1720 .getsockopt = netlink_getsockopt,
1721 .sendmsg = netlink_sendmsg,
1722 .recvmsg = netlink_recvmsg,
1723 .mmap = sock_no_mmap,
1724 .sendpage = sock_no_sendpage,
1727 static struct net_proto_family netlink_family_ops = {
1728 .family = PF_NETLINK,
1729 .create = netlink_create,
1730 .owner = THIS_MODULE, /* for consistency 8) */
1733 extern void netlink_skb_parms_too_large(void);
1735 static int __init netlink_proto_init(void)
1737 struct sk_buff *dummy_skb;
1741 int err = proto_register(&netlink_proto, 0);
1746 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1747 netlink_skb_parms_too_large();
1749 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1753 if (num_physpages >= (128 * 1024))
1754 max = num_physpages >> (21 - PAGE_SHIFT);
1756 max = num_physpages >> (23 - PAGE_SHIFT);
1758 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1759 max = (1UL << order) / sizeof(struct hlist_head);
1760 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1762 for (i = 0; i < MAX_LINKS; i++) {
1763 struct nl_pid_hash *hash = &nl_table[i].hash;
1765 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1768 nl_pid_hash_free(nl_table[i].hash.table,
1769 1 * sizeof(*hash->table));
1773 memset(hash->table, 0, 1 * sizeof(*hash->table));
1774 hash->max_shift = order;
1777 hash->rehash_time = jiffies;
1780 sock_register(&netlink_family_ops);
1781 #ifdef CONFIG_PROC_FS
1782 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1784 /* The netlink device handler may be needed early. */
1789 panic("netlink_init: Cannot allocate nl_table\n");
1792 core_initcall(netlink_proto_init);
1794 EXPORT_SYMBOL(netlink_ack);
1795 EXPORT_SYMBOL(netlink_run_queue);
1796 EXPORT_SYMBOL(netlink_queue_skip);
1797 EXPORT_SYMBOL(netlink_broadcast);
1798 EXPORT_SYMBOL(netlink_dump_start);
1799 EXPORT_SYMBOL(netlink_kernel_create);
1800 EXPORT_SYMBOL(netlink_register_notifier);
1801 EXPORT_SYMBOL(netlink_set_err);
1802 EXPORT_SYMBOL(netlink_set_nonroot);
1803 EXPORT_SYMBOL(netlink_unicast);
1804 EXPORT_SYMBOL(netlink_unregister_notifier);