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/mutex.h>
63 #include <net/netlink.h>
65 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
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);
145 BUG_ON(mutex_is_locked(nlk_sk(sk)->cb_mutex));
148 nlk->cb->done(nlk->cb);
149 netlink_destroy_callback(nlk->cb);
152 skb_queue_purge(&sk->sk_receive_queue);
154 if (!sock_flag(sk, SOCK_DEAD)) {
155 printk("Freeing alive netlink socket %p\n", sk);
158 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
159 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
160 BUG_TRAP(!nlk_sk(sk)->groups);
163 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
164 * Look, when several writers sleep and reader wakes them up, all but one
165 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
166 * this, _but_ remember, it adds useless work on UP machines.
169 static void netlink_table_grab(void)
171 write_lock_irq(&nl_table_lock);
173 if (atomic_read(&nl_table_users)) {
174 DECLARE_WAITQUEUE(wait, current);
176 add_wait_queue_exclusive(&nl_table_wait, &wait);
178 set_current_state(TASK_UNINTERRUPTIBLE);
179 if (atomic_read(&nl_table_users) == 0)
181 write_unlock_irq(&nl_table_lock);
183 write_lock_irq(&nl_table_lock);
186 __set_current_state(TASK_RUNNING);
187 remove_wait_queue(&nl_table_wait, &wait);
191 static __inline__ void netlink_table_ungrab(void)
193 write_unlock_irq(&nl_table_lock);
194 wake_up(&nl_table_wait);
197 static __inline__ void
198 netlink_lock_table(void)
200 /* read_lock() synchronizes us to netlink_table_grab */
202 read_lock(&nl_table_lock);
203 atomic_inc(&nl_table_users);
204 read_unlock(&nl_table_lock);
207 static __inline__ void
208 netlink_unlock_table(void)
210 if (atomic_dec_and_test(&nl_table_users))
211 wake_up(&nl_table_wait);
214 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
216 struct nl_pid_hash *hash = &nl_table[protocol].hash;
217 struct hlist_head *head;
219 struct hlist_node *node;
221 read_lock(&nl_table_lock);
222 head = nl_pid_hashfn(hash, pid);
223 sk_for_each(sk, node, head) {
224 if (nlk_sk(sk)->pid == pid) {
231 read_unlock(&nl_table_lock);
235 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
237 if (size <= PAGE_SIZE)
238 return kmalloc(size, GFP_ATOMIC);
240 return (struct hlist_head *)
241 __get_free_pages(GFP_ATOMIC, get_order(size));
244 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
246 if (size <= PAGE_SIZE)
249 free_pages((unsigned long)table, get_order(size));
252 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
254 unsigned int omask, mask, shift;
256 struct hlist_head *otable, *table;
259 omask = mask = hash->mask;
260 osize = size = (mask + 1) * sizeof(*table);
264 if (++shift > hash->max_shift)
270 table = nl_pid_hash_alloc(size);
274 memset(table, 0, size);
275 otable = hash->table;
279 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
281 for (i = 0; i <= omask; i++) {
283 struct hlist_node *node, *tmp;
285 sk_for_each_safe(sk, node, tmp, &otable[i])
286 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
289 nl_pid_hash_free(otable, osize);
290 hash->rehash_time = jiffies + 10 * 60 * HZ;
294 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
296 int avg = hash->entries >> hash->shift;
298 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
301 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
302 nl_pid_hash_rehash(hash, 0);
309 static const struct proto_ops netlink_ops;
312 netlink_update_listeners(struct sock *sk)
314 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
315 struct hlist_node *node;
319 for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) {
321 sk_for_each_bound(sk, node, &tbl->mc_list)
322 mask |= nlk_sk(sk)->groups[i];
323 tbl->listeners[i] = mask;
325 /* this function is only called with the netlink table "grabbed", which
326 * makes sure updates are visible before bind or setsockopt return. */
329 static int netlink_insert(struct sock *sk, u32 pid)
331 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
332 struct hlist_head *head;
333 int err = -EADDRINUSE;
335 struct hlist_node *node;
338 netlink_table_grab();
339 head = nl_pid_hashfn(hash, pid);
341 sk_for_each(osk, node, head) {
342 if (nlk_sk(osk)->pid == pid)
354 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
357 if (len && nl_pid_hash_dilute(hash, len))
358 head = nl_pid_hashfn(hash, pid);
360 nlk_sk(sk)->pid = pid;
361 sk_add_node(sk, head);
365 netlink_table_ungrab();
369 static void netlink_remove(struct sock *sk)
371 netlink_table_grab();
372 if (sk_del_node_init(sk))
373 nl_table[sk->sk_protocol].hash.entries--;
374 if (nlk_sk(sk)->subscriptions)
375 __sk_del_bind_node(sk);
376 netlink_table_ungrab();
379 static struct proto netlink_proto = {
381 .owner = THIS_MODULE,
382 .obj_size = sizeof(struct netlink_sock),
385 static int __netlink_create(struct socket *sock, struct mutex *cb_mutex,
389 struct netlink_sock *nlk;
391 sock->ops = &netlink_ops;
393 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
397 sock_init_data(sock, sk);
401 nlk->cb_mutex = cb_mutex;
403 nlk->cb_mutex = &nlk->cb_def_mutex;
404 mutex_init(nlk->cb_mutex);
406 init_waitqueue_head(&nlk->wait);
408 sk->sk_destruct = netlink_sock_destruct;
409 sk->sk_protocol = protocol;
413 static int netlink_create(struct socket *sock, int protocol)
415 struct module *module = NULL;
416 struct mutex *cb_mutex;
417 struct netlink_sock *nlk;
420 sock->state = SS_UNCONNECTED;
422 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
423 return -ESOCKTNOSUPPORT;
425 if (protocol<0 || protocol >= MAX_LINKS)
426 return -EPROTONOSUPPORT;
428 netlink_lock_table();
430 if (!nl_table[protocol].registered) {
431 netlink_unlock_table();
432 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
433 netlink_lock_table();
436 if (nl_table[protocol].registered &&
437 try_module_get(nl_table[protocol].module))
438 module = nl_table[protocol].module;
439 cb_mutex = nl_table[protocol].cb_mutex;
440 netlink_unlock_table();
442 if ((err = __netlink_create(sock, cb_mutex, protocol)) < 0)
445 nlk = nlk_sk(sock->sk);
446 nlk->module = module;
455 static int netlink_release(struct socket *sock)
457 struct sock *sk = sock->sk;
458 struct netlink_sock *nlk;
468 * OK. Socket is unlinked, any packets that arrive now
473 wake_up_interruptible_all(&nlk->wait);
475 skb_queue_purge(&sk->sk_write_queue);
477 if (nlk->pid && !nlk->subscriptions) {
478 struct netlink_notify n = {
479 .protocol = sk->sk_protocol,
482 atomic_notifier_call_chain(&netlink_chain,
483 NETLINK_URELEASE, &n);
486 module_put(nlk->module);
488 netlink_table_grab();
489 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
490 kfree(nl_table[sk->sk_protocol].listeners);
491 nl_table[sk->sk_protocol].module = NULL;
492 nl_table[sk->sk_protocol].registered = 0;
493 } else if (nlk->subscriptions)
494 netlink_update_listeners(sk);
495 netlink_table_ungrab();
504 static int netlink_autobind(struct socket *sock)
506 struct sock *sk = sock->sk;
507 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
508 struct hlist_head *head;
510 struct hlist_node *node;
511 s32 pid = current->tgid;
513 static s32 rover = -4097;
517 netlink_table_grab();
518 head = nl_pid_hashfn(hash, pid);
519 sk_for_each(osk, node, head) {
520 if (nlk_sk(osk)->pid == pid) {
521 /* Bind collision, search negative pid values. */
525 netlink_table_ungrab();
529 netlink_table_ungrab();
531 err = netlink_insert(sk, pid);
532 if (err == -EADDRINUSE)
535 /* If 2 threads race to autobind, that is fine. */
542 static inline int netlink_capable(struct socket *sock, unsigned int flag)
544 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
545 capable(CAP_NET_ADMIN);
549 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
551 struct netlink_sock *nlk = nlk_sk(sk);
553 if (nlk->subscriptions && !subscriptions)
554 __sk_del_bind_node(sk);
555 else if (!nlk->subscriptions && subscriptions)
556 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
557 nlk->subscriptions = subscriptions;
560 static int netlink_alloc_groups(struct sock *sk)
562 struct netlink_sock *nlk = nlk_sk(sk);
566 netlink_lock_table();
567 groups = nl_table[sk->sk_protocol].groups;
568 if (!nl_table[sk->sk_protocol].registered)
570 netlink_unlock_table();
575 nlk->groups = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
576 if (nlk->groups == NULL)
578 nlk->ngroups = groups;
582 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
584 struct sock *sk = sock->sk;
585 struct netlink_sock *nlk = nlk_sk(sk);
586 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
589 if (nladdr->nl_family != AF_NETLINK)
592 /* Only superuser is allowed to listen multicasts */
593 if (nladdr->nl_groups) {
594 if (!netlink_capable(sock, NL_NONROOT_RECV))
596 if (nlk->groups == NULL) {
597 err = netlink_alloc_groups(sk);
604 if (nladdr->nl_pid != nlk->pid)
607 err = nladdr->nl_pid ?
608 netlink_insert(sk, nladdr->nl_pid) :
609 netlink_autobind(sock);
614 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
617 netlink_table_grab();
618 netlink_update_subscriptions(sk, nlk->subscriptions +
619 hweight32(nladdr->nl_groups) -
620 hweight32(nlk->groups[0]));
621 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
622 netlink_update_listeners(sk);
623 netlink_table_ungrab();
628 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
632 struct sock *sk = sock->sk;
633 struct netlink_sock *nlk = nlk_sk(sk);
634 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
636 if (addr->sa_family == AF_UNSPEC) {
637 sk->sk_state = NETLINK_UNCONNECTED;
642 if (addr->sa_family != AF_NETLINK)
645 /* Only superuser is allowed to send multicasts */
646 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
650 err = netlink_autobind(sock);
653 sk->sk_state = NETLINK_CONNECTED;
654 nlk->dst_pid = nladdr->nl_pid;
655 nlk->dst_group = ffs(nladdr->nl_groups);
661 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
663 struct sock *sk = sock->sk;
664 struct netlink_sock *nlk = nlk_sk(sk);
665 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
667 nladdr->nl_family = AF_NETLINK;
669 *addr_len = sizeof(*nladdr);
672 nladdr->nl_pid = nlk->dst_pid;
673 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
675 nladdr->nl_pid = nlk->pid;
676 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
681 static void netlink_overrun(struct sock *sk)
683 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
684 sk->sk_err = ENOBUFS;
685 sk->sk_error_report(sk);
689 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
691 int protocol = ssk->sk_protocol;
693 struct netlink_sock *nlk;
695 sock = netlink_lookup(protocol, pid);
697 return ERR_PTR(-ECONNREFUSED);
699 /* Don't bother queuing skb if kernel socket has no input function */
701 if ((nlk->pid == 0 && !nlk->data_ready) ||
702 (sock->sk_state == NETLINK_CONNECTED &&
703 nlk->dst_pid != nlk_sk(ssk)->pid)) {
705 return ERR_PTR(-ECONNREFUSED);
710 struct sock *netlink_getsockbyfilp(struct file *filp)
712 struct inode *inode = filp->f_path.dentry->d_inode;
715 if (!S_ISSOCK(inode->i_mode))
716 return ERR_PTR(-ENOTSOCK);
718 sock = SOCKET_I(inode)->sk;
719 if (sock->sk_family != AF_NETLINK)
720 return ERR_PTR(-EINVAL);
727 * Attach a skb to a netlink socket.
728 * The caller must hold a reference to the destination socket. On error, the
729 * reference is dropped. The skb is not send to the destination, just all
730 * all error checks are performed and memory in the queue is reserved.
732 * < 0: error. skb freed, reference to sock dropped.
734 * 1: repeat lookup - reference dropped while waiting for socket memory.
736 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
737 long timeo, struct sock *ssk)
739 struct netlink_sock *nlk;
743 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
744 test_bit(0, &nlk->state)) {
745 DECLARE_WAITQUEUE(wait, current);
747 if (!ssk || nlk_sk(ssk)->pid == 0)
754 __set_current_state(TASK_INTERRUPTIBLE);
755 add_wait_queue(&nlk->wait, &wait);
757 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
758 test_bit(0, &nlk->state)) &&
759 !sock_flag(sk, SOCK_DEAD))
760 timeo = schedule_timeout(timeo);
762 __set_current_state(TASK_RUNNING);
763 remove_wait_queue(&nlk->wait, &wait);
766 if (signal_pending(current)) {
768 return sock_intr_errno(timeo);
772 skb_set_owner_r(skb, sk);
776 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
780 skb_queue_tail(&sk->sk_receive_queue, skb);
781 sk->sk_data_ready(sk, len);
786 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
792 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
799 delta = skb->end - skb->tail;
800 if (delta * 2 < skb->truesize)
803 if (skb_shared(skb)) {
804 struct sk_buff *nskb = skb_clone(skb, allocation);
811 if (!pskb_expand_head(skb, 0, -delta, allocation))
812 skb->truesize -= delta;
817 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
823 skb = netlink_trim(skb, gfp_any());
825 timeo = sock_sndtimeo(ssk, nonblock);
827 sk = netlink_getsockbypid(ssk, pid);
832 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
838 return netlink_sendskb(sk, skb, ssk->sk_protocol);
841 int netlink_has_listeners(struct sock *sk, unsigned int group)
845 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
846 if (group - 1 < nl_table[sk->sk_protocol].groups)
847 res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners);
850 EXPORT_SYMBOL_GPL(netlink_has_listeners);
852 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
854 struct netlink_sock *nlk = nlk_sk(sk);
856 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
857 !test_bit(0, &nlk->state)) {
858 skb_set_owner_r(skb, sk);
859 skb_queue_tail(&sk->sk_receive_queue, skb);
860 sk->sk_data_ready(sk, skb->len);
861 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
866 struct netlink_broadcast_data {
867 struct sock *exclude_sk;
874 struct sk_buff *skb, *skb2;
877 static inline int do_one_broadcast(struct sock *sk,
878 struct netlink_broadcast_data *p)
880 struct netlink_sock *nlk = nlk_sk(sk);
883 if (p->exclude_sk == sk)
886 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
887 !test_bit(p->group - 1, nlk->groups))
896 if (p->skb2 == NULL) {
897 if (skb_shared(p->skb)) {
898 p->skb2 = skb_clone(p->skb, p->allocation);
900 p->skb2 = skb_get(p->skb);
902 * skb ownership may have been set when
903 * delivered to a previous socket.
908 if (p->skb2 == NULL) {
910 /* Clone failed. Notify ALL listeners. */
912 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
925 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
926 u32 group, gfp_t allocation)
928 struct netlink_broadcast_data info;
929 struct hlist_node *node;
932 skb = netlink_trim(skb, allocation);
934 info.exclude_sk = ssk;
940 info.allocation = allocation;
944 /* While we sleep in clone, do not allow to change socket list */
946 netlink_lock_table();
948 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
949 do_one_broadcast(sk, &info);
953 netlink_unlock_table();
956 kfree_skb(info.skb2);
958 if (info.delivered) {
959 if (info.congested && (allocation & __GFP_WAIT))
968 struct netlink_set_err_data {
969 struct sock *exclude_sk;
975 static inline int do_one_set_err(struct sock *sk,
976 struct netlink_set_err_data *p)
978 struct netlink_sock *nlk = nlk_sk(sk);
980 if (sk == p->exclude_sk)
983 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
984 !test_bit(p->group - 1, nlk->groups))
987 sk->sk_err = p->code;
988 sk->sk_error_report(sk);
993 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
995 struct netlink_set_err_data info;
996 struct hlist_node *node;
999 info.exclude_sk = ssk;
1004 read_lock(&nl_table_lock);
1006 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1007 do_one_set_err(sk, &info);
1009 read_unlock(&nl_table_lock);
1012 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1013 char __user *optval, int optlen)
1015 struct sock *sk = sock->sk;
1016 struct netlink_sock *nlk = nlk_sk(sk);
1019 if (level != SOL_NETLINK)
1020 return -ENOPROTOOPT;
1022 if (optlen >= sizeof(int) &&
1023 get_user(val, (int __user *)optval))
1027 case NETLINK_PKTINFO:
1029 nlk->flags |= NETLINK_RECV_PKTINFO;
1031 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1034 case NETLINK_ADD_MEMBERSHIP:
1035 case NETLINK_DROP_MEMBERSHIP: {
1036 unsigned int subscriptions;
1037 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1039 if (!netlink_capable(sock, NL_NONROOT_RECV))
1041 if (nlk->groups == NULL) {
1042 err = netlink_alloc_groups(sk);
1046 if (!val || val - 1 >= nlk->ngroups)
1048 netlink_table_grab();
1049 old = test_bit(val - 1, nlk->groups);
1050 subscriptions = nlk->subscriptions - old + new;
1052 __set_bit(val - 1, nlk->groups);
1054 __clear_bit(val - 1, nlk->groups);
1055 netlink_update_subscriptions(sk, subscriptions);
1056 netlink_update_listeners(sk);
1057 netlink_table_ungrab();
1067 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1068 char __user *optval, int __user *optlen)
1070 struct sock *sk = sock->sk;
1071 struct netlink_sock *nlk = nlk_sk(sk);
1074 if (level != SOL_NETLINK)
1075 return -ENOPROTOOPT;
1077 if (get_user(len, optlen))
1083 case NETLINK_PKTINFO:
1084 if (len < sizeof(int))
1087 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1088 if (put_user(len, optlen) ||
1089 put_user(val, optval))
1099 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1101 struct nl_pktinfo info;
1103 info.group = NETLINK_CB(skb).dst_group;
1104 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1107 static inline void netlink_rcv_wake(struct sock *sk)
1109 struct netlink_sock *nlk = nlk_sk(sk);
1111 if (skb_queue_empty(&sk->sk_receive_queue))
1112 clear_bit(0, &nlk->state);
1113 if (!test_bit(0, &nlk->state))
1114 wake_up_interruptible(&nlk->wait);
1117 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1118 struct msghdr *msg, size_t len)
1120 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1121 struct sock *sk = sock->sk;
1122 struct netlink_sock *nlk = nlk_sk(sk);
1123 struct sockaddr_nl *addr=msg->msg_name;
1126 struct sk_buff *skb;
1128 struct scm_cookie scm;
1130 if (msg->msg_flags&MSG_OOB)
1133 if (NULL == siocb->scm)
1135 err = scm_send(sock, msg, siocb->scm);
1139 if (msg->msg_namelen) {
1140 if (addr->nl_family != AF_NETLINK)
1142 dst_pid = addr->nl_pid;
1143 dst_group = ffs(addr->nl_groups);
1144 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1147 dst_pid = nlk->dst_pid;
1148 dst_group = nlk->dst_group;
1152 err = netlink_autobind(sock);
1158 if (len > sk->sk_sndbuf - 32)
1161 skb = alloc_skb(len, GFP_KERNEL);
1165 NETLINK_CB(skb).pid = nlk->pid;
1166 NETLINK_CB(skb).dst_group = dst_group;
1167 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1168 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1169 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1171 /* What can I do? Netlink is asynchronous, so that
1172 we will have to save current capabilities to
1173 check them, when this message will be delivered
1174 to corresponding kernel module. --ANK (980802)
1178 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1183 err = security_netlink_send(sk, skb);
1190 atomic_inc(&skb->users);
1191 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1193 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1199 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1200 struct msghdr *msg, size_t len,
1203 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1204 struct scm_cookie scm;
1205 struct sock *sk = sock->sk;
1206 struct netlink_sock *nlk = nlk_sk(sk);
1207 int noblock = flags&MSG_DONTWAIT;
1209 struct sk_buff *skb;
1217 skb = skb_recv_datagram(sk,flags,noblock,&err);
1221 msg->msg_namelen = 0;
1225 msg->msg_flags |= MSG_TRUNC;
1229 skb_reset_transport_header(skb);
1230 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1232 if (msg->msg_name) {
1233 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1234 addr->nl_family = AF_NETLINK;
1236 addr->nl_pid = NETLINK_CB(skb).pid;
1237 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1238 msg->msg_namelen = sizeof(*addr);
1241 if (nlk->flags & NETLINK_RECV_PKTINFO)
1242 netlink_cmsg_recv_pktinfo(msg, skb);
1244 if (NULL == siocb->scm) {
1245 memset(&scm, 0, sizeof(scm));
1248 siocb->scm->creds = *NETLINK_CREDS(skb);
1249 skb_free_datagram(sk, skb);
1251 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1254 scm_recv(sock, msg, siocb->scm, flags);
1256 if (flags & MSG_TRUNC)
1260 netlink_rcv_wake(sk);
1261 return err ? : copied;
1264 static void netlink_data_ready(struct sock *sk, int len)
1266 struct netlink_sock *nlk = nlk_sk(sk);
1268 if (nlk->data_ready)
1269 nlk->data_ready(sk, len);
1270 netlink_rcv_wake(sk);
1274 * We export these functions to other modules. They provide a
1275 * complete set of kernel non-blocking support for message
1280 netlink_kernel_create(int unit, unsigned int groups,
1281 void (*input)(struct sock *sk, int len),
1282 struct mutex *cb_mutex, struct module *module)
1284 struct socket *sock;
1286 struct netlink_sock *nlk;
1287 unsigned long *listeners = NULL;
1291 if (unit<0 || unit>=MAX_LINKS)
1294 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1297 if (__netlink_create(sock, cb_mutex, unit) < 0)
1298 goto out_sock_release;
1303 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1305 goto out_sock_release;
1308 sk->sk_data_ready = netlink_data_ready;
1310 nlk_sk(sk)->data_ready = input;
1312 if (netlink_insert(sk, 0))
1313 goto out_sock_release;
1316 nlk->flags |= NETLINK_KERNEL_SOCKET;
1318 netlink_table_grab();
1319 nl_table[unit].groups = groups;
1320 nl_table[unit].listeners = listeners;
1321 nl_table[unit].cb_mutex = cb_mutex;
1322 nl_table[unit].module = module;
1323 nl_table[unit].registered = 1;
1324 netlink_table_ungrab();
1334 void netlink_set_nonroot(int protocol, unsigned int flags)
1336 if ((unsigned int)protocol < MAX_LINKS)
1337 nl_table[protocol].nl_nonroot = flags;
1340 static void netlink_destroy_callback(struct netlink_callback *cb)
1348 * It looks a bit ugly.
1349 * It would be better to create kernel thread.
1352 static int netlink_dump(struct sock *sk)
1354 struct netlink_sock *nlk = nlk_sk(sk);
1355 struct netlink_callback *cb;
1356 struct sk_buff *skb;
1357 struct nlmsghdr *nlh;
1358 int len, err = -ENOBUFS;
1360 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1364 mutex_lock(nlk->cb_mutex);
1372 len = cb->dump(skb, cb);
1375 mutex_unlock(nlk->cb_mutex);
1376 skb_queue_tail(&sk->sk_receive_queue, skb);
1377 sk->sk_data_ready(sk, len);
1381 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1385 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1387 skb_queue_tail(&sk->sk_receive_queue, skb);
1388 sk->sk_data_ready(sk, skb->len);
1393 mutex_unlock(nlk->cb_mutex);
1395 netlink_destroy_callback(cb);
1399 mutex_unlock(nlk->cb_mutex);
1405 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1406 struct nlmsghdr *nlh,
1407 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1408 int (*done)(struct netlink_callback*))
1410 struct netlink_callback *cb;
1412 struct netlink_sock *nlk;
1414 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1421 atomic_inc(&skb->users);
1424 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1426 netlink_destroy_callback(cb);
1427 return -ECONNREFUSED;
1430 /* A dump is in progress... */
1431 mutex_lock(nlk->cb_mutex);
1433 mutex_unlock(nlk->cb_mutex);
1434 netlink_destroy_callback(cb);
1439 mutex_unlock(nlk->cb_mutex);
1444 /* We successfully started a dump, by returning -EINTR we
1445 * signal the queue mangement to interrupt processing of
1446 * any netlink messages so userspace gets a chance to read
1451 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1453 struct sk_buff *skb;
1454 struct nlmsghdr *rep;
1455 struct nlmsgerr *errmsg;
1456 size_t payload = sizeof(*errmsg);
1458 /* error messages get the original request appened */
1460 payload += nlmsg_len(nlh);
1462 skb = nlmsg_new(payload, GFP_KERNEL);
1466 sk = netlink_lookup(in_skb->sk->sk_protocol,
1467 NETLINK_CB(in_skb).pid);
1469 sk->sk_err = ENOBUFS;
1470 sk->sk_error_report(sk);
1476 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1477 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1478 errmsg = nlmsg_data(rep);
1479 errmsg->error = err;
1480 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1481 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1484 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1487 struct nlmsghdr *nlh;
1490 while (skb->len >= nlmsg_total_size(0)) {
1491 nlh = nlmsg_hdr(skb);
1494 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1497 /* Only requests are handled by the kernel */
1498 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1501 /* Skip control messages */
1502 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1506 if (err == -EINTR) {
1507 /* Not an error, but we interrupt processing */
1508 netlink_queue_skip(nlh, skb);
1512 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1513 netlink_ack(skb, nlh, err);
1515 netlink_queue_skip(nlh, skb);
1522 * nelink_run_queue - Process netlink receive queue.
1523 * @sk: Netlink socket containing the queue
1524 * @qlen: Place to store queue length upon entry
1525 * @cb: Callback function invoked for each netlink message found
1527 * Processes as much as there was in the queue upon entry and invokes
1528 * a callback function for each netlink message found. The callback
1529 * function may refuse a message by returning a negative error code
1530 * but setting the error pointer to 0 in which case this function
1531 * returns with a qlen != 0.
1533 * qlen must be initialized to 0 before the initial entry, afterwards
1534 * the function may be called repeatedly until qlen reaches 0.
1536 * The callback function may return -EINTR to signal that processing
1537 * of netlink messages shall be interrupted. In this case the message
1538 * currently being processed will NOT be requeued onto the receive
1541 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1542 int (*cb)(struct sk_buff *, struct nlmsghdr *))
1544 struct sk_buff *skb;
1546 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1547 *qlen = skb_queue_len(&sk->sk_receive_queue);
1549 for (; *qlen; (*qlen)--) {
1550 skb = skb_dequeue(&sk->sk_receive_queue);
1551 if (netlink_rcv_skb(skb, cb)) {
1553 skb_queue_head(&sk->sk_receive_queue, skb);
1566 * netlink_queue_skip - Skip netlink message while processing queue.
1567 * @nlh: Netlink message to be skipped
1568 * @skb: Socket buffer containing the netlink messages.
1570 * Pulls the given netlink message off the socket buffer so the next
1571 * call to netlink_queue_run() will not reconsider the message.
1573 static void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1575 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1577 if (msglen > skb->len)
1580 skb_pull(skb, msglen);
1584 * nlmsg_notify - send a notification netlink message
1585 * @sk: netlink socket to use
1586 * @skb: notification message
1587 * @pid: destination netlink pid for reports or 0
1588 * @group: destination multicast group or 0
1589 * @report: 1 to report back, 0 to disable
1590 * @flags: allocation flags
1592 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1593 unsigned int group, int report, gfp_t flags)
1598 int exclude_pid = 0;
1601 atomic_inc(&skb->users);
1605 /* errors reported via destination sk->sk_err */
1606 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1610 err = nlmsg_unicast(sk, skb, pid);
1615 #ifdef CONFIG_PROC_FS
1616 struct nl_seq_iter {
1621 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1623 struct nl_seq_iter *iter = seq->private;
1626 struct hlist_node *node;
1629 for (i=0; i<MAX_LINKS; i++) {
1630 struct nl_pid_hash *hash = &nl_table[i].hash;
1632 for (j = 0; j <= hash->mask; j++) {
1633 sk_for_each(s, node, &hash->table[j]) {
1646 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1648 read_lock(&nl_table_lock);
1649 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1652 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1655 struct nl_seq_iter *iter;
1660 if (v == SEQ_START_TOKEN)
1661 return netlink_seq_socket_idx(seq, 0);
1667 iter = seq->private;
1669 j = iter->hash_idx + 1;
1672 struct nl_pid_hash *hash = &nl_table[i].hash;
1674 for (; j <= hash->mask; j++) {
1675 s = sk_head(&hash->table[j]);
1684 } while (++i < MAX_LINKS);
1689 static void netlink_seq_stop(struct seq_file *seq, void *v)
1691 read_unlock(&nl_table_lock);
1695 static int netlink_seq_show(struct seq_file *seq, void *v)
1697 if (v == SEQ_START_TOKEN)
1699 "sk Eth Pid Groups "
1700 "Rmem Wmem Dump Locks\n");
1703 struct netlink_sock *nlk = nlk_sk(s);
1705 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1709 nlk->groups ? (u32)nlk->groups[0] : 0,
1710 atomic_read(&s->sk_rmem_alloc),
1711 atomic_read(&s->sk_wmem_alloc),
1713 atomic_read(&s->sk_refcnt)
1720 static struct seq_operations netlink_seq_ops = {
1721 .start = netlink_seq_start,
1722 .next = netlink_seq_next,
1723 .stop = netlink_seq_stop,
1724 .show = netlink_seq_show,
1728 static int netlink_seq_open(struct inode *inode, struct file *file)
1730 struct seq_file *seq;
1731 struct nl_seq_iter *iter;
1734 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1738 err = seq_open(file, &netlink_seq_ops);
1744 seq = file->private_data;
1745 seq->private = iter;
1749 static const struct file_operations netlink_seq_fops = {
1750 .owner = THIS_MODULE,
1751 .open = netlink_seq_open,
1753 .llseek = seq_lseek,
1754 .release = seq_release_private,
1759 int netlink_register_notifier(struct notifier_block *nb)
1761 return atomic_notifier_chain_register(&netlink_chain, nb);
1764 int netlink_unregister_notifier(struct notifier_block *nb)
1766 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1769 static const struct proto_ops netlink_ops = {
1770 .family = PF_NETLINK,
1771 .owner = THIS_MODULE,
1772 .release = netlink_release,
1773 .bind = netlink_bind,
1774 .connect = netlink_connect,
1775 .socketpair = sock_no_socketpair,
1776 .accept = sock_no_accept,
1777 .getname = netlink_getname,
1778 .poll = datagram_poll,
1779 .ioctl = sock_no_ioctl,
1780 .listen = sock_no_listen,
1781 .shutdown = sock_no_shutdown,
1782 .setsockopt = netlink_setsockopt,
1783 .getsockopt = netlink_getsockopt,
1784 .sendmsg = netlink_sendmsg,
1785 .recvmsg = netlink_recvmsg,
1786 .mmap = sock_no_mmap,
1787 .sendpage = sock_no_sendpage,
1790 static struct net_proto_family netlink_family_ops = {
1791 .family = PF_NETLINK,
1792 .create = netlink_create,
1793 .owner = THIS_MODULE, /* for consistency 8) */
1796 static int __init netlink_proto_init(void)
1798 struct sk_buff *dummy_skb;
1802 int err = proto_register(&netlink_proto, 0);
1807 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1809 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1813 if (num_physpages >= (128 * 1024))
1814 max = num_physpages >> (21 - PAGE_SHIFT);
1816 max = num_physpages >> (23 - PAGE_SHIFT);
1818 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1819 max = (1UL << order) / sizeof(struct hlist_head);
1820 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1822 for (i = 0; i < MAX_LINKS; i++) {
1823 struct nl_pid_hash *hash = &nl_table[i].hash;
1825 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1828 nl_pid_hash_free(nl_table[i].hash.table,
1829 1 * sizeof(*hash->table));
1833 memset(hash->table, 0, 1 * sizeof(*hash->table));
1834 hash->max_shift = order;
1837 hash->rehash_time = jiffies;
1840 sock_register(&netlink_family_ops);
1841 #ifdef CONFIG_PROC_FS
1842 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1844 /* The netlink device handler may be needed early. */
1849 panic("netlink_init: Cannot allocate nl_table\n");
1852 core_initcall(netlink_proto_init);
1854 EXPORT_SYMBOL(netlink_ack);
1855 EXPORT_SYMBOL(netlink_run_queue);
1856 EXPORT_SYMBOL(netlink_broadcast);
1857 EXPORT_SYMBOL(netlink_dump_start);
1858 EXPORT_SYMBOL(netlink_kernel_create);
1859 EXPORT_SYMBOL(netlink_register_notifier);
1860 EXPORT_SYMBOL(netlink_set_nonroot);
1861 EXPORT_SYMBOL(netlink_unicast);
1862 EXPORT_SYMBOL(netlink_unregister_notifier);
1863 EXPORT_SYMBOL(nlmsg_notify);