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/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/selinux.h>
58 #include <linux/mutex.h>
62 #include <net/netlink.h>
64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
65 #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
68 /* struct sock has to be the first member of netlink_sock */
76 unsigned long *groups;
78 wait_queue_head_t wait;
79 struct netlink_callback *cb;
80 struct mutex *cb_mutex;
81 struct mutex cb_def_mutex;
82 void (*data_ready)(struct sock *sk, int bytes);
83 struct module *module;
86 #define NETLINK_KERNEL_SOCKET 0x1
87 #define NETLINK_RECV_PKTINFO 0x2
89 static inline struct netlink_sock *nlk_sk(struct sock *sk)
91 return (struct netlink_sock *)sk;
95 struct hlist_head *table;
96 unsigned long rehash_time;
101 unsigned int entries;
102 unsigned int max_shift;
107 struct netlink_table {
108 struct nl_pid_hash hash;
109 struct hlist_head mc_list;
110 unsigned long *listeners;
111 unsigned int nl_nonroot;
113 struct mutex *cb_mutex;
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);
124 static void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb);
126 static DEFINE_RWLOCK(nl_table_lock);
127 static atomic_t nl_table_users = ATOMIC_INIT(0);
129 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
131 static u32 netlink_group_mask(u32 group)
133 return group ? 1 << (group - 1) : 0;
136 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
138 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
141 static void netlink_sock_destruct(struct sock *sk)
143 struct netlink_sock *nlk = nlk_sk(sk);
147 nlk->cb->done(nlk->cb);
148 netlink_destroy_callback(nlk->cb);
151 skb_queue_purge(&sk->sk_receive_queue);
153 if (!sock_flag(sk, SOCK_DEAD)) {
154 printk("Freeing alive netlink socket %p\n", sk);
157 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
158 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
159 BUG_TRAP(!nlk_sk(sk)->groups);
162 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
163 * Look, when several writers sleep and reader wakes them up, all but one
164 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
165 * this, _but_ remember, it adds useless work on UP machines.
168 static void netlink_table_grab(void)
170 write_lock_irq(&nl_table_lock);
172 if (atomic_read(&nl_table_users)) {
173 DECLARE_WAITQUEUE(wait, current);
175 add_wait_queue_exclusive(&nl_table_wait, &wait);
177 set_current_state(TASK_UNINTERRUPTIBLE);
178 if (atomic_read(&nl_table_users) == 0)
180 write_unlock_irq(&nl_table_lock);
182 write_lock_irq(&nl_table_lock);
185 __set_current_state(TASK_RUNNING);
186 remove_wait_queue(&nl_table_wait, &wait);
190 static __inline__ void netlink_table_ungrab(void)
192 write_unlock_irq(&nl_table_lock);
193 wake_up(&nl_table_wait);
196 static __inline__ void
197 netlink_lock_table(void)
199 /* read_lock() synchronizes us to netlink_table_grab */
201 read_lock(&nl_table_lock);
202 atomic_inc(&nl_table_users);
203 read_unlock(&nl_table_lock);
206 static __inline__ void
207 netlink_unlock_table(void)
209 if (atomic_dec_and_test(&nl_table_users))
210 wake_up(&nl_table_wait);
213 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
215 struct nl_pid_hash *hash = &nl_table[protocol].hash;
216 struct hlist_head *head;
218 struct hlist_node *node;
220 read_lock(&nl_table_lock);
221 head = nl_pid_hashfn(hash, pid);
222 sk_for_each(sk, node, head) {
223 if (nlk_sk(sk)->pid == pid) {
230 read_unlock(&nl_table_lock);
234 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
236 if (size <= PAGE_SIZE)
237 return kmalloc(size, GFP_ATOMIC);
239 return (struct hlist_head *)
240 __get_free_pages(GFP_ATOMIC, get_order(size));
243 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
245 if (size <= PAGE_SIZE)
248 free_pages((unsigned long)table, get_order(size));
251 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
253 unsigned int omask, mask, shift;
255 struct hlist_head *otable, *table;
258 omask = mask = hash->mask;
259 osize = size = (mask + 1) * sizeof(*table);
263 if (++shift > hash->max_shift)
269 table = nl_pid_hash_alloc(size);
273 memset(table, 0, size);
274 otable = hash->table;
278 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
280 for (i = 0; i <= omask; i++) {
282 struct hlist_node *node, *tmp;
284 sk_for_each_safe(sk, node, tmp, &otable[i])
285 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
288 nl_pid_hash_free(otable, osize);
289 hash->rehash_time = jiffies + 10 * 60 * HZ;
293 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
295 int avg = hash->entries >> hash->shift;
297 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
300 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
301 nl_pid_hash_rehash(hash, 0);
308 static const struct proto_ops netlink_ops;
311 netlink_update_listeners(struct sock *sk)
313 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
314 struct hlist_node *node;
318 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
320 sk_for_each_bound(sk, node, &tbl->mc_list) {
321 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
322 mask |= nlk_sk(sk)->groups[i];
324 tbl->listeners[i] = mask;
326 /* this function is only called with the netlink table "grabbed", which
327 * makes sure updates are visible before bind or setsockopt return. */
330 static int netlink_insert(struct sock *sk, u32 pid)
332 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
333 struct hlist_head *head;
334 int err = -EADDRINUSE;
336 struct hlist_node *node;
339 netlink_table_grab();
340 head = nl_pid_hashfn(hash, pid);
342 sk_for_each(osk, node, head) {
343 if (nlk_sk(osk)->pid == pid)
355 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
358 if (len && nl_pid_hash_dilute(hash, len))
359 head = nl_pid_hashfn(hash, pid);
361 nlk_sk(sk)->pid = pid;
362 sk_add_node(sk, head);
366 netlink_table_ungrab();
370 static void netlink_remove(struct sock *sk)
372 netlink_table_grab();
373 if (sk_del_node_init(sk))
374 nl_table[sk->sk_protocol].hash.entries--;
375 if (nlk_sk(sk)->subscriptions)
376 __sk_del_bind_node(sk);
377 netlink_table_ungrab();
380 static struct proto netlink_proto = {
382 .owner = THIS_MODULE,
383 .obj_size = sizeof(struct netlink_sock),
386 static int __netlink_create(struct socket *sock, struct mutex *cb_mutex,
390 struct netlink_sock *nlk;
392 sock->ops = &netlink_ops;
394 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
398 sock_init_data(sock, sk);
402 nlk->cb_mutex = cb_mutex;
404 nlk->cb_mutex = &nlk->cb_def_mutex;
405 mutex_init(nlk->cb_mutex);
407 init_waitqueue_head(&nlk->wait);
409 sk->sk_destruct = netlink_sock_destruct;
410 sk->sk_protocol = protocol;
414 static int netlink_create(struct socket *sock, int protocol)
416 struct module *module = NULL;
417 struct mutex *cb_mutex;
418 struct netlink_sock *nlk;
421 sock->state = SS_UNCONNECTED;
423 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
424 return -ESOCKTNOSUPPORT;
426 if (protocol<0 || protocol >= MAX_LINKS)
427 return -EPROTONOSUPPORT;
429 netlink_lock_table();
431 if (!nl_table[protocol].registered) {
432 netlink_unlock_table();
433 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
434 netlink_lock_table();
437 if (nl_table[protocol].registered &&
438 try_module_get(nl_table[protocol].module))
439 module = nl_table[protocol].module;
440 cb_mutex = nl_table[protocol].cb_mutex;
441 netlink_unlock_table();
443 if ((err = __netlink_create(sock, cb_mutex, protocol)) < 0)
446 nlk = nlk_sk(sock->sk);
447 nlk->module = module;
456 static int netlink_release(struct socket *sock)
458 struct sock *sk = sock->sk;
459 struct netlink_sock *nlk;
469 * OK. Socket is unlinked, any packets that arrive now
474 wake_up_interruptible_all(&nlk->wait);
476 skb_queue_purge(&sk->sk_write_queue);
478 if (nlk->pid && !nlk->subscriptions) {
479 struct netlink_notify n = {
480 .protocol = sk->sk_protocol,
483 atomic_notifier_call_chain(&netlink_chain,
484 NETLINK_URELEASE, &n);
487 module_put(nlk->module);
489 netlink_table_grab();
490 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
491 kfree(nl_table[sk->sk_protocol].listeners);
492 nl_table[sk->sk_protocol].module = NULL;
493 nl_table[sk->sk_protocol].registered = 0;
494 } else if (nlk->subscriptions)
495 netlink_update_listeners(sk);
496 netlink_table_ungrab();
505 static int netlink_autobind(struct socket *sock)
507 struct sock *sk = sock->sk;
508 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
509 struct hlist_head *head;
511 struct hlist_node *node;
512 s32 pid = current->tgid;
514 static s32 rover = -4097;
518 netlink_table_grab();
519 head = nl_pid_hashfn(hash, pid);
520 sk_for_each(osk, node, head) {
521 if (nlk_sk(osk)->pid == pid) {
522 /* Bind collision, search negative pid values. */
526 netlink_table_ungrab();
530 netlink_table_ungrab();
532 err = netlink_insert(sk, pid);
533 if (err == -EADDRINUSE)
536 /* If 2 threads race to autobind, that is fine. */
543 static inline int netlink_capable(struct socket *sock, unsigned int flag)
545 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
546 capable(CAP_NET_ADMIN);
550 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
552 struct netlink_sock *nlk = nlk_sk(sk);
554 if (nlk->subscriptions && !subscriptions)
555 __sk_del_bind_node(sk);
556 else if (!nlk->subscriptions && subscriptions)
557 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
558 nlk->subscriptions = subscriptions;
561 static int netlink_realloc_groups(struct sock *sk)
563 struct netlink_sock *nlk = nlk_sk(sk);
565 unsigned long *new_groups;
568 netlink_table_grab();
570 groups = nl_table[sk->sk_protocol].groups;
571 if (!nl_table[sk->sk_protocol].registered) {
576 if (nlk->ngroups >= groups)
579 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
580 if (new_groups == NULL) {
584 memset((char*)new_groups + NLGRPSZ(nlk->ngroups), 0,
585 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
587 nlk->groups = new_groups;
588 nlk->ngroups = groups;
590 netlink_table_ungrab();
594 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
596 struct sock *sk = sock->sk;
597 struct netlink_sock *nlk = nlk_sk(sk);
598 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
601 if (nladdr->nl_family != AF_NETLINK)
604 /* Only superuser is allowed to listen multicasts */
605 if (nladdr->nl_groups) {
606 if (!netlink_capable(sock, NL_NONROOT_RECV))
608 err = netlink_realloc_groups(sk);
614 if (nladdr->nl_pid != nlk->pid)
617 err = nladdr->nl_pid ?
618 netlink_insert(sk, nladdr->nl_pid) :
619 netlink_autobind(sock);
624 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
627 netlink_table_grab();
628 netlink_update_subscriptions(sk, nlk->subscriptions +
629 hweight32(nladdr->nl_groups) -
630 hweight32(nlk->groups[0]));
631 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
632 netlink_update_listeners(sk);
633 netlink_table_ungrab();
638 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
642 struct sock *sk = sock->sk;
643 struct netlink_sock *nlk = nlk_sk(sk);
644 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
646 if (addr->sa_family == AF_UNSPEC) {
647 sk->sk_state = NETLINK_UNCONNECTED;
652 if (addr->sa_family != AF_NETLINK)
655 /* Only superuser is allowed to send multicasts */
656 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
660 err = netlink_autobind(sock);
663 sk->sk_state = NETLINK_CONNECTED;
664 nlk->dst_pid = nladdr->nl_pid;
665 nlk->dst_group = ffs(nladdr->nl_groups);
671 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
673 struct sock *sk = sock->sk;
674 struct netlink_sock *nlk = nlk_sk(sk);
675 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
677 nladdr->nl_family = AF_NETLINK;
679 *addr_len = sizeof(*nladdr);
682 nladdr->nl_pid = nlk->dst_pid;
683 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
685 nladdr->nl_pid = nlk->pid;
686 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
691 static void netlink_overrun(struct sock *sk)
693 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
694 sk->sk_err = ENOBUFS;
695 sk->sk_error_report(sk);
699 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
701 int protocol = ssk->sk_protocol;
703 struct netlink_sock *nlk;
705 sock = netlink_lookup(protocol, pid);
707 return ERR_PTR(-ECONNREFUSED);
709 /* Don't bother queuing skb if kernel socket has no input function */
711 if ((nlk->pid == 0 && !nlk->data_ready) ||
712 (sock->sk_state == NETLINK_CONNECTED &&
713 nlk->dst_pid != nlk_sk(ssk)->pid)) {
715 return ERR_PTR(-ECONNREFUSED);
720 struct sock *netlink_getsockbyfilp(struct file *filp)
722 struct inode *inode = filp->f_path.dentry->d_inode;
725 if (!S_ISSOCK(inode->i_mode))
726 return ERR_PTR(-ENOTSOCK);
728 sock = SOCKET_I(inode)->sk;
729 if (sock->sk_family != AF_NETLINK)
730 return ERR_PTR(-EINVAL);
737 * Attach a skb to a netlink socket.
738 * The caller must hold a reference to the destination socket. On error, the
739 * reference is dropped. The skb is not send to the destination, just all
740 * all error checks are performed and memory in the queue is reserved.
742 * < 0: error. skb freed, reference to sock dropped.
744 * 1: repeat lookup - reference dropped while waiting for socket memory.
746 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
747 long timeo, struct sock *ssk)
749 struct netlink_sock *nlk;
753 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
754 test_bit(0, &nlk->state)) {
755 DECLARE_WAITQUEUE(wait, current);
757 if (!ssk || nlk_sk(ssk)->pid == 0)
764 __set_current_state(TASK_INTERRUPTIBLE);
765 add_wait_queue(&nlk->wait, &wait);
767 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
768 test_bit(0, &nlk->state)) &&
769 !sock_flag(sk, SOCK_DEAD))
770 timeo = schedule_timeout(timeo);
772 __set_current_state(TASK_RUNNING);
773 remove_wait_queue(&nlk->wait, &wait);
776 if (signal_pending(current)) {
778 return sock_intr_errno(timeo);
782 skb_set_owner_r(skb, sk);
786 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
790 skb_queue_tail(&sk->sk_receive_queue, skb);
791 sk->sk_data_ready(sk, len);
796 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
802 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
809 delta = skb->end - skb->tail;
810 if (delta * 2 < skb->truesize)
813 if (skb_shared(skb)) {
814 struct sk_buff *nskb = skb_clone(skb, allocation);
821 if (!pskb_expand_head(skb, 0, -delta, allocation))
822 skb->truesize -= delta;
827 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
833 skb = netlink_trim(skb, gfp_any());
835 timeo = sock_sndtimeo(ssk, nonblock);
837 sk = netlink_getsockbypid(ssk, pid);
842 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
848 return netlink_sendskb(sk, skb, ssk->sk_protocol);
851 int netlink_has_listeners(struct sock *sk, unsigned int group)
854 unsigned long *listeners;
856 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
859 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
861 if (group - 1 < nl_table[sk->sk_protocol].groups)
862 res = test_bit(group - 1, listeners);
868 EXPORT_SYMBOL_GPL(netlink_has_listeners);
870 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
872 struct netlink_sock *nlk = nlk_sk(sk);
874 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
875 !test_bit(0, &nlk->state)) {
876 skb_set_owner_r(skb, sk);
877 skb_queue_tail(&sk->sk_receive_queue, skb);
878 sk->sk_data_ready(sk, skb->len);
879 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
884 struct netlink_broadcast_data {
885 struct sock *exclude_sk;
892 struct sk_buff *skb, *skb2;
895 static inline int do_one_broadcast(struct sock *sk,
896 struct netlink_broadcast_data *p)
898 struct netlink_sock *nlk = nlk_sk(sk);
901 if (p->exclude_sk == sk)
904 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
905 !test_bit(p->group - 1, nlk->groups))
914 if (p->skb2 == NULL) {
915 if (skb_shared(p->skb)) {
916 p->skb2 = skb_clone(p->skb, p->allocation);
918 p->skb2 = skb_get(p->skb);
920 * skb ownership may have been set when
921 * delivered to a previous socket.
926 if (p->skb2 == NULL) {
928 /* Clone failed. Notify ALL listeners. */
930 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
943 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
944 u32 group, gfp_t allocation)
946 struct netlink_broadcast_data info;
947 struct hlist_node *node;
950 skb = netlink_trim(skb, allocation);
952 info.exclude_sk = ssk;
958 info.allocation = allocation;
962 /* While we sleep in clone, do not allow to change socket list */
964 netlink_lock_table();
966 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
967 do_one_broadcast(sk, &info);
971 netlink_unlock_table();
974 kfree_skb(info.skb2);
976 if (info.delivered) {
977 if (info.congested && (allocation & __GFP_WAIT))
986 struct netlink_set_err_data {
987 struct sock *exclude_sk;
993 static inline int do_one_set_err(struct sock *sk,
994 struct netlink_set_err_data *p)
996 struct netlink_sock *nlk = nlk_sk(sk);
998 if (sk == p->exclude_sk)
1001 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1002 !test_bit(p->group - 1, nlk->groups))
1005 sk->sk_err = p->code;
1006 sk->sk_error_report(sk);
1011 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1013 struct netlink_set_err_data info;
1014 struct hlist_node *node;
1017 info.exclude_sk = ssk;
1022 read_lock(&nl_table_lock);
1024 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1025 do_one_set_err(sk, &info);
1027 read_unlock(&nl_table_lock);
1030 /* must be called with netlink table grabbed */
1031 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1035 int old, new = !!is_new, subscriptions;
1037 old = test_bit(group - 1, nlk->groups);
1038 subscriptions = nlk->subscriptions - old + new;
1040 __set_bit(group - 1, nlk->groups);
1042 __clear_bit(group - 1, nlk->groups);
1043 netlink_update_subscriptions(&nlk->sk, subscriptions);
1044 netlink_update_listeners(&nlk->sk);
1047 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1048 char __user *optval, int optlen)
1050 struct sock *sk = sock->sk;
1051 struct netlink_sock *nlk = nlk_sk(sk);
1052 unsigned int val = 0;
1055 if (level != SOL_NETLINK)
1056 return -ENOPROTOOPT;
1058 if (optlen >= sizeof(int) &&
1059 get_user(val, (unsigned int __user *)optval))
1063 case NETLINK_PKTINFO:
1065 nlk->flags |= NETLINK_RECV_PKTINFO;
1067 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1070 case NETLINK_ADD_MEMBERSHIP:
1071 case NETLINK_DROP_MEMBERSHIP: {
1072 if (!netlink_capable(sock, NL_NONROOT_RECV))
1074 err = netlink_realloc_groups(sk);
1077 if (!val || val - 1 >= nlk->ngroups)
1079 netlink_table_grab();
1080 netlink_update_socket_mc(nlk, val,
1081 optname == NETLINK_ADD_MEMBERSHIP);
1082 netlink_table_ungrab();
1092 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1093 char __user *optval, int __user *optlen)
1095 struct sock *sk = sock->sk;
1096 struct netlink_sock *nlk = nlk_sk(sk);
1099 if (level != SOL_NETLINK)
1100 return -ENOPROTOOPT;
1102 if (get_user(len, optlen))
1108 case NETLINK_PKTINFO:
1109 if (len < sizeof(int))
1112 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1113 if (put_user(len, optlen) ||
1114 put_user(val, optval))
1124 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1126 struct nl_pktinfo info;
1128 info.group = NETLINK_CB(skb).dst_group;
1129 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1132 static inline void netlink_rcv_wake(struct sock *sk)
1134 struct netlink_sock *nlk = nlk_sk(sk);
1136 if (skb_queue_empty(&sk->sk_receive_queue))
1137 clear_bit(0, &nlk->state);
1138 if (!test_bit(0, &nlk->state))
1139 wake_up_interruptible(&nlk->wait);
1142 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1143 struct msghdr *msg, size_t len)
1145 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1146 struct sock *sk = sock->sk;
1147 struct netlink_sock *nlk = nlk_sk(sk);
1148 struct sockaddr_nl *addr=msg->msg_name;
1151 struct sk_buff *skb;
1153 struct scm_cookie scm;
1155 if (msg->msg_flags&MSG_OOB)
1158 if (NULL == siocb->scm)
1160 err = scm_send(sock, msg, siocb->scm);
1164 if (msg->msg_namelen) {
1165 if (addr->nl_family != AF_NETLINK)
1167 dst_pid = addr->nl_pid;
1168 dst_group = ffs(addr->nl_groups);
1169 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1172 dst_pid = nlk->dst_pid;
1173 dst_group = nlk->dst_group;
1177 err = netlink_autobind(sock);
1183 if (len > sk->sk_sndbuf - 32)
1186 skb = alloc_skb(len, GFP_KERNEL);
1190 NETLINK_CB(skb).pid = nlk->pid;
1191 NETLINK_CB(skb).dst_group = dst_group;
1192 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1193 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1194 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1196 /* What can I do? Netlink is asynchronous, so that
1197 we will have to save current capabilities to
1198 check them, when this message will be delivered
1199 to corresponding kernel module. --ANK (980802)
1203 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1208 err = security_netlink_send(sk, skb);
1215 atomic_inc(&skb->users);
1216 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1218 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1224 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1225 struct msghdr *msg, size_t len,
1228 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1229 struct scm_cookie scm;
1230 struct sock *sk = sock->sk;
1231 struct netlink_sock *nlk = nlk_sk(sk);
1232 int noblock = flags&MSG_DONTWAIT;
1234 struct sk_buff *skb;
1242 skb = skb_recv_datagram(sk,flags,noblock,&err);
1246 msg->msg_namelen = 0;
1250 msg->msg_flags |= MSG_TRUNC;
1254 skb_reset_transport_header(skb);
1255 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1257 if (msg->msg_name) {
1258 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1259 addr->nl_family = AF_NETLINK;
1261 addr->nl_pid = NETLINK_CB(skb).pid;
1262 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1263 msg->msg_namelen = sizeof(*addr);
1266 if (nlk->flags & NETLINK_RECV_PKTINFO)
1267 netlink_cmsg_recv_pktinfo(msg, skb);
1269 if (NULL == siocb->scm) {
1270 memset(&scm, 0, sizeof(scm));
1273 siocb->scm->creds = *NETLINK_CREDS(skb);
1274 if (flags & MSG_TRUNC)
1276 skb_free_datagram(sk, skb);
1278 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1281 scm_recv(sock, msg, siocb->scm, flags);
1283 netlink_rcv_wake(sk);
1284 return err ? : copied;
1287 static void netlink_data_ready(struct sock *sk, int len)
1289 struct netlink_sock *nlk = nlk_sk(sk);
1291 if (nlk->data_ready)
1292 nlk->data_ready(sk, len);
1293 netlink_rcv_wake(sk);
1297 * We export these functions to other modules. They provide a
1298 * complete set of kernel non-blocking support for message
1303 netlink_kernel_create(int unit, unsigned int groups,
1304 void (*input)(struct sock *sk, int len),
1305 struct mutex *cb_mutex, struct module *module)
1307 struct socket *sock;
1309 struct netlink_sock *nlk;
1310 unsigned long *listeners = NULL;
1314 if (unit<0 || unit>=MAX_LINKS)
1317 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1320 if (__netlink_create(sock, cb_mutex, unit) < 0)
1321 goto out_sock_release;
1326 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1328 goto out_sock_release;
1331 sk->sk_data_ready = netlink_data_ready;
1333 nlk_sk(sk)->data_ready = input;
1335 if (netlink_insert(sk, 0))
1336 goto out_sock_release;
1339 nlk->flags |= NETLINK_KERNEL_SOCKET;
1341 netlink_table_grab();
1342 nl_table[unit].groups = groups;
1343 nl_table[unit].listeners = listeners;
1344 nl_table[unit].cb_mutex = cb_mutex;
1345 nl_table[unit].module = module;
1346 nl_table[unit].registered = 1;
1347 netlink_table_ungrab();
1358 * netlink_change_ngroups - change number of multicast groups
1360 * This changes the number of multicast groups that are available
1361 * on a certain netlink family. Note that it is not possible to
1362 * change the number of groups to below 32. Also note that it does
1363 * not implicitly call netlink_clear_multicast_users() when the
1364 * number of groups is reduced.
1366 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1367 * @groups: The new number of groups.
1369 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1371 unsigned long *listeners, *old = NULL;
1372 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1378 netlink_table_grab();
1379 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1380 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1385 old = tbl->listeners;
1386 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1387 rcu_assign_pointer(tbl->listeners, listeners);
1389 tbl->groups = groups;
1392 netlink_table_ungrab();
1397 EXPORT_SYMBOL(netlink_change_ngroups);
1400 * netlink_clear_multicast_users - kick off multicast listeners
1402 * This function removes all listeners from the given group.
1403 * @ksk: The kernel netlink socket, as returned by
1404 * netlink_kernel_create().
1405 * @group: The multicast group to clear.
1407 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1410 struct hlist_node *node;
1411 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1413 netlink_table_grab();
1415 sk_for_each_bound(sk, node, &tbl->mc_list)
1416 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1418 netlink_table_ungrab();
1420 EXPORT_SYMBOL(netlink_clear_multicast_users);
1422 void netlink_set_nonroot(int protocol, unsigned int flags)
1424 if ((unsigned int)protocol < MAX_LINKS)
1425 nl_table[protocol].nl_nonroot = flags;
1428 static void netlink_destroy_callback(struct netlink_callback *cb)
1436 * It looks a bit ugly.
1437 * It would be better to create kernel thread.
1440 static int netlink_dump(struct sock *sk)
1442 struct netlink_sock *nlk = nlk_sk(sk);
1443 struct netlink_callback *cb;
1444 struct sk_buff *skb;
1445 struct nlmsghdr *nlh;
1446 int len, err = -ENOBUFS;
1448 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1452 mutex_lock(nlk->cb_mutex);
1460 len = cb->dump(skb, cb);
1463 mutex_unlock(nlk->cb_mutex);
1464 skb_queue_tail(&sk->sk_receive_queue, skb);
1465 sk->sk_data_ready(sk, len);
1469 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1473 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1475 skb_queue_tail(&sk->sk_receive_queue, skb);
1476 sk->sk_data_ready(sk, skb->len);
1481 mutex_unlock(nlk->cb_mutex);
1483 netlink_destroy_callback(cb);
1487 mutex_unlock(nlk->cb_mutex);
1493 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1494 struct nlmsghdr *nlh,
1495 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1496 int (*done)(struct netlink_callback*))
1498 struct netlink_callback *cb;
1500 struct netlink_sock *nlk;
1502 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1509 atomic_inc(&skb->users);
1512 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1514 netlink_destroy_callback(cb);
1515 return -ECONNREFUSED;
1518 /* A dump is in progress... */
1519 mutex_lock(nlk->cb_mutex);
1521 mutex_unlock(nlk->cb_mutex);
1522 netlink_destroy_callback(cb);
1527 mutex_unlock(nlk->cb_mutex);
1532 /* We successfully started a dump, by returning -EINTR we
1533 * signal the queue mangement to interrupt processing of
1534 * any netlink messages so userspace gets a chance to read
1539 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1541 struct sk_buff *skb;
1542 struct nlmsghdr *rep;
1543 struct nlmsgerr *errmsg;
1544 size_t payload = sizeof(*errmsg);
1546 /* error messages get the original request appened */
1548 payload += nlmsg_len(nlh);
1550 skb = nlmsg_new(payload, GFP_KERNEL);
1554 sk = netlink_lookup(in_skb->sk->sk_protocol,
1555 NETLINK_CB(in_skb).pid);
1557 sk->sk_err = ENOBUFS;
1558 sk->sk_error_report(sk);
1564 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1565 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1566 errmsg = nlmsg_data(rep);
1567 errmsg->error = err;
1568 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1569 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1572 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1575 struct nlmsghdr *nlh;
1578 while (skb->len >= nlmsg_total_size(0)) {
1579 nlh = nlmsg_hdr(skb);
1582 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1585 /* Only requests are handled by the kernel */
1586 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1589 /* Skip control messages */
1590 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1594 if (err == -EINTR) {
1595 /* Not an error, but we interrupt processing */
1596 netlink_queue_skip(nlh, skb);
1600 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1601 netlink_ack(skb, nlh, err);
1603 netlink_queue_skip(nlh, skb);
1610 * nelink_run_queue - Process netlink receive queue.
1611 * @sk: Netlink socket containing the queue
1612 * @qlen: Place to store queue length upon entry
1613 * @cb: Callback function invoked for each netlink message found
1615 * Processes as much as there was in the queue upon entry and invokes
1616 * a callback function for each netlink message found. The callback
1617 * function may refuse a message by returning a negative error code
1618 * but setting the error pointer to 0 in which case this function
1619 * returns with a qlen != 0.
1621 * qlen must be initialized to 0 before the initial entry, afterwards
1622 * the function may be called repeatedly until qlen reaches 0.
1624 * The callback function may return -EINTR to signal that processing
1625 * of netlink messages shall be interrupted. In this case the message
1626 * currently being processed will NOT be requeued onto the receive
1629 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1630 int (*cb)(struct sk_buff *, struct nlmsghdr *))
1632 struct sk_buff *skb;
1634 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1635 *qlen = skb_queue_len(&sk->sk_receive_queue);
1637 for (; *qlen; (*qlen)--) {
1638 skb = skb_dequeue(&sk->sk_receive_queue);
1639 if (netlink_rcv_skb(skb, cb)) {
1641 skb_queue_head(&sk->sk_receive_queue, skb);
1654 * netlink_queue_skip - Skip netlink message while processing queue.
1655 * @nlh: Netlink message to be skipped
1656 * @skb: Socket buffer containing the netlink messages.
1658 * Pulls the given netlink message off the socket buffer so the next
1659 * call to netlink_queue_run() will not reconsider the message.
1661 static void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1663 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1665 if (msglen > skb->len)
1668 skb_pull(skb, msglen);
1672 * nlmsg_notify - send a notification netlink message
1673 * @sk: netlink socket to use
1674 * @skb: notification message
1675 * @pid: destination netlink pid for reports or 0
1676 * @group: destination multicast group or 0
1677 * @report: 1 to report back, 0 to disable
1678 * @flags: allocation flags
1680 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1681 unsigned int group, int report, gfp_t flags)
1686 int exclude_pid = 0;
1689 atomic_inc(&skb->users);
1693 /* errors reported via destination sk->sk_err */
1694 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1698 err = nlmsg_unicast(sk, skb, pid);
1703 #ifdef CONFIG_PROC_FS
1704 struct nl_seq_iter {
1709 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1711 struct nl_seq_iter *iter = seq->private;
1714 struct hlist_node *node;
1717 for (i=0; i<MAX_LINKS; i++) {
1718 struct nl_pid_hash *hash = &nl_table[i].hash;
1720 for (j = 0; j <= hash->mask; j++) {
1721 sk_for_each(s, node, &hash->table[j]) {
1734 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1736 read_lock(&nl_table_lock);
1737 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1740 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1743 struct nl_seq_iter *iter;
1748 if (v == SEQ_START_TOKEN)
1749 return netlink_seq_socket_idx(seq, 0);
1755 iter = seq->private;
1757 j = iter->hash_idx + 1;
1760 struct nl_pid_hash *hash = &nl_table[i].hash;
1762 for (; j <= hash->mask; j++) {
1763 s = sk_head(&hash->table[j]);
1772 } while (++i < MAX_LINKS);
1777 static void netlink_seq_stop(struct seq_file *seq, void *v)
1779 read_unlock(&nl_table_lock);
1783 static int netlink_seq_show(struct seq_file *seq, void *v)
1785 if (v == SEQ_START_TOKEN)
1787 "sk Eth Pid Groups "
1788 "Rmem Wmem Dump Locks\n");
1791 struct netlink_sock *nlk = nlk_sk(s);
1793 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1797 nlk->groups ? (u32)nlk->groups[0] : 0,
1798 atomic_read(&s->sk_rmem_alloc),
1799 atomic_read(&s->sk_wmem_alloc),
1801 atomic_read(&s->sk_refcnt)
1808 static const struct seq_operations netlink_seq_ops = {
1809 .start = netlink_seq_start,
1810 .next = netlink_seq_next,
1811 .stop = netlink_seq_stop,
1812 .show = netlink_seq_show,
1816 static int netlink_seq_open(struct inode *inode, struct file *file)
1818 struct seq_file *seq;
1819 struct nl_seq_iter *iter;
1822 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1826 err = seq_open(file, &netlink_seq_ops);
1832 seq = file->private_data;
1833 seq->private = iter;
1837 static const struct file_operations netlink_seq_fops = {
1838 .owner = THIS_MODULE,
1839 .open = netlink_seq_open,
1841 .llseek = seq_lseek,
1842 .release = seq_release_private,
1847 int netlink_register_notifier(struct notifier_block *nb)
1849 return atomic_notifier_chain_register(&netlink_chain, nb);
1852 int netlink_unregister_notifier(struct notifier_block *nb)
1854 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1857 static const struct proto_ops netlink_ops = {
1858 .family = PF_NETLINK,
1859 .owner = THIS_MODULE,
1860 .release = netlink_release,
1861 .bind = netlink_bind,
1862 .connect = netlink_connect,
1863 .socketpair = sock_no_socketpair,
1864 .accept = sock_no_accept,
1865 .getname = netlink_getname,
1866 .poll = datagram_poll,
1867 .ioctl = sock_no_ioctl,
1868 .listen = sock_no_listen,
1869 .shutdown = sock_no_shutdown,
1870 .setsockopt = netlink_setsockopt,
1871 .getsockopt = netlink_getsockopt,
1872 .sendmsg = netlink_sendmsg,
1873 .recvmsg = netlink_recvmsg,
1874 .mmap = sock_no_mmap,
1875 .sendpage = sock_no_sendpage,
1878 static struct net_proto_family netlink_family_ops = {
1879 .family = PF_NETLINK,
1880 .create = netlink_create,
1881 .owner = THIS_MODULE, /* for consistency 8) */
1884 static int __init netlink_proto_init(void)
1886 struct sk_buff *dummy_skb;
1890 int err = proto_register(&netlink_proto, 0);
1895 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1897 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1901 if (num_physpages >= (128 * 1024))
1902 max = num_physpages >> (21 - PAGE_SHIFT);
1904 max = num_physpages >> (23 - PAGE_SHIFT);
1906 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1907 max = (1UL << order) / sizeof(struct hlist_head);
1908 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1910 for (i = 0; i < MAX_LINKS; i++) {
1911 struct nl_pid_hash *hash = &nl_table[i].hash;
1913 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1916 nl_pid_hash_free(nl_table[i].hash.table,
1917 1 * sizeof(*hash->table));
1921 memset(hash->table, 0, 1 * sizeof(*hash->table));
1922 hash->max_shift = order;
1925 hash->rehash_time = jiffies;
1928 sock_register(&netlink_family_ops);
1929 #ifdef CONFIG_PROC_FS
1930 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1932 /* The netlink device handler may be needed early. */
1937 panic("netlink_init: Cannot allocate nl_table\n");
1940 core_initcall(netlink_proto_init);
1942 EXPORT_SYMBOL(netlink_ack);
1943 EXPORT_SYMBOL(netlink_run_queue);
1944 EXPORT_SYMBOL(netlink_broadcast);
1945 EXPORT_SYMBOL(netlink_dump_start);
1946 EXPORT_SYMBOL(netlink_kernel_create);
1947 EXPORT_SYMBOL(netlink_register_notifier);
1948 EXPORT_SYMBOL(netlink_set_nonroot);
1949 EXPORT_SYMBOL(netlink_unicast);
1950 EXPORT_SYMBOL(netlink_unregister_notifier);
1951 EXPORT_SYMBOL(nlmsg_notify);