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>
62 #include <net/netlink.h>
64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
67 /* struct sock has to be the first member of netlink_sock */
75 unsigned long *groups;
77 wait_queue_head_t wait;
78 struct netlink_callback *cb;
80 void (*data_ready)(struct sock *sk, int bytes);
81 struct module *module;
84 #define NETLINK_KERNEL_SOCKET 0x1
85 #define NETLINK_RECV_PKTINFO 0x2
87 static inline struct netlink_sock *nlk_sk(struct sock *sk)
89 return (struct netlink_sock *)sk;
93 struct hlist_head *table;
94 unsigned long rehash_time;
100 unsigned int max_shift;
105 struct netlink_table {
106 struct nl_pid_hash hash;
107 struct hlist_head mc_list;
108 unsigned long *listeners;
109 unsigned int nl_nonroot;
111 struct module *module;
115 static struct netlink_table *nl_table;
117 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
119 static int netlink_dump(struct sock *sk);
120 static void netlink_destroy_callback(struct netlink_callback *cb);
122 static DEFINE_RWLOCK(nl_table_lock);
123 static atomic_t nl_table_users = ATOMIC_INIT(0);
125 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
127 static u32 netlink_group_mask(u32 group)
129 return group ? 1 << (group - 1) : 0;
132 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
134 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
137 static void netlink_sock_destruct(struct sock *sk)
139 skb_queue_purge(&sk->sk_receive_queue);
141 if (!sock_flag(sk, SOCK_DEAD)) {
142 printk("Freeing alive netlink socket %p\n", sk);
145 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
146 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
147 BUG_TRAP(!nlk_sk(sk)->cb);
148 BUG_TRAP(!nlk_sk(sk)->groups);
151 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
152 * Look, when several writers sleep and reader wakes them up, all but one
153 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
154 * this, _but_ remember, it adds useless work on UP machines.
157 static void netlink_table_grab(void)
159 write_lock_irq(&nl_table_lock);
161 if (atomic_read(&nl_table_users)) {
162 DECLARE_WAITQUEUE(wait, current);
164 add_wait_queue_exclusive(&nl_table_wait, &wait);
166 set_current_state(TASK_UNINTERRUPTIBLE);
167 if (atomic_read(&nl_table_users) == 0)
169 write_unlock_irq(&nl_table_lock);
171 write_lock_irq(&nl_table_lock);
174 __set_current_state(TASK_RUNNING);
175 remove_wait_queue(&nl_table_wait, &wait);
179 static __inline__ void netlink_table_ungrab(void)
181 write_unlock_irq(&nl_table_lock);
182 wake_up(&nl_table_wait);
185 static __inline__ void
186 netlink_lock_table(void)
188 /* read_lock() synchronizes us to netlink_table_grab */
190 read_lock(&nl_table_lock);
191 atomic_inc(&nl_table_users);
192 read_unlock(&nl_table_lock);
195 static __inline__ void
196 netlink_unlock_table(void)
198 if (atomic_dec_and_test(&nl_table_users))
199 wake_up(&nl_table_wait);
202 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
204 struct nl_pid_hash *hash = &nl_table[protocol].hash;
205 struct hlist_head *head;
207 struct hlist_node *node;
209 read_lock(&nl_table_lock);
210 head = nl_pid_hashfn(hash, pid);
211 sk_for_each(sk, node, head) {
212 if (nlk_sk(sk)->pid == pid) {
219 read_unlock(&nl_table_lock);
223 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
225 if (size <= PAGE_SIZE)
226 return kmalloc(size, GFP_ATOMIC);
228 return (struct hlist_head *)
229 __get_free_pages(GFP_ATOMIC, get_order(size));
232 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
234 if (size <= PAGE_SIZE)
237 free_pages((unsigned long)table, get_order(size));
240 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
242 unsigned int omask, mask, shift;
244 struct hlist_head *otable, *table;
247 omask = mask = hash->mask;
248 osize = size = (mask + 1) * sizeof(*table);
252 if (++shift > hash->max_shift)
258 table = nl_pid_hash_alloc(size);
262 memset(table, 0, size);
263 otable = hash->table;
267 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
269 for (i = 0; i <= omask; i++) {
271 struct hlist_node *node, *tmp;
273 sk_for_each_safe(sk, node, tmp, &otable[i])
274 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
277 nl_pid_hash_free(otable, osize);
278 hash->rehash_time = jiffies + 10 * 60 * HZ;
282 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
284 int avg = hash->entries >> hash->shift;
286 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
289 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
290 nl_pid_hash_rehash(hash, 0);
297 static const struct proto_ops netlink_ops;
300 netlink_update_listeners(struct sock *sk)
302 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
303 struct hlist_node *node;
307 for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) {
309 sk_for_each_bound(sk, node, &tbl->mc_list)
310 mask |= nlk_sk(sk)->groups[i];
311 tbl->listeners[i] = mask;
313 /* this function is only called with the netlink table "grabbed", which
314 * makes sure updates are visible before bind or setsockopt return. */
317 static int netlink_insert(struct sock *sk, u32 pid)
319 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
320 struct hlist_head *head;
321 int err = -EADDRINUSE;
323 struct hlist_node *node;
326 netlink_table_grab();
327 head = nl_pid_hashfn(hash, pid);
329 sk_for_each(osk, node, head) {
330 if (nlk_sk(osk)->pid == pid)
342 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
345 if (len && nl_pid_hash_dilute(hash, len))
346 head = nl_pid_hashfn(hash, pid);
348 nlk_sk(sk)->pid = pid;
349 sk_add_node(sk, head);
353 netlink_table_ungrab();
357 static void netlink_remove(struct sock *sk)
359 netlink_table_grab();
360 if (sk_del_node_init(sk))
361 nl_table[sk->sk_protocol].hash.entries--;
362 if (nlk_sk(sk)->subscriptions)
363 __sk_del_bind_node(sk);
364 netlink_table_ungrab();
367 static struct proto netlink_proto = {
369 .owner = THIS_MODULE,
370 .obj_size = sizeof(struct netlink_sock),
373 static int __netlink_create(struct socket *sock, int protocol)
376 struct netlink_sock *nlk;
378 sock->ops = &netlink_ops;
380 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
384 sock_init_data(sock, sk);
387 spin_lock_init(&nlk->cb_lock);
388 init_waitqueue_head(&nlk->wait);
390 sk->sk_destruct = netlink_sock_destruct;
391 sk->sk_protocol = protocol;
395 static int netlink_create(struct socket *sock, int protocol)
397 struct module *module = NULL;
398 struct netlink_sock *nlk;
402 sock->state = SS_UNCONNECTED;
404 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
405 return -ESOCKTNOSUPPORT;
407 if (protocol<0 || protocol >= MAX_LINKS)
408 return -EPROTONOSUPPORT;
410 netlink_lock_table();
412 if (!nl_table[protocol].registered) {
413 netlink_unlock_table();
414 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
415 netlink_lock_table();
418 if (nl_table[protocol].registered &&
419 try_module_get(nl_table[protocol].module))
420 module = nl_table[protocol].module;
421 groups = nl_table[protocol].groups;
422 netlink_unlock_table();
424 if ((err = __netlink_create(sock, protocol)) < 0)
427 nlk = nlk_sk(sock->sk);
428 nlk->module = module;
437 static int netlink_release(struct socket *sock)
439 struct sock *sk = sock->sk;
440 struct netlink_sock *nlk;
448 spin_lock(&nlk->cb_lock);
451 nlk->cb->done(nlk->cb);
452 netlink_destroy_callback(nlk->cb);
455 spin_unlock(&nlk->cb_lock);
457 /* OK. Socket is unlinked, and, therefore,
458 no new packets will arrive */
462 wake_up_interruptible_all(&nlk->wait);
464 skb_queue_purge(&sk->sk_write_queue);
466 if (nlk->pid && !nlk->subscriptions) {
467 struct netlink_notify n = {
468 .protocol = sk->sk_protocol,
471 atomic_notifier_call_chain(&netlink_chain,
472 NETLINK_URELEASE, &n);
476 module_put(nlk->module);
478 netlink_table_grab();
479 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
480 kfree(nl_table[sk->sk_protocol].listeners);
481 nl_table[sk->sk_protocol].module = NULL;
482 nl_table[sk->sk_protocol].registered = 0;
483 } else if (nlk->subscriptions)
484 netlink_update_listeners(sk);
485 netlink_table_ungrab();
494 static int netlink_autobind(struct socket *sock)
496 struct sock *sk = sock->sk;
497 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
498 struct hlist_head *head;
500 struct hlist_node *node;
501 s32 pid = current->tgid;
503 static s32 rover = -4097;
507 netlink_table_grab();
508 head = nl_pid_hashfn(hash, pid);
509 sk_for_each(osk, node, head) {
510 if (nlk_sk(osk)->pid == pid) {
511 /* Bind collision, search negative pid values. */
515 netlink_table_ungrab();
519 netlink_table_ungrab();
521 err = netlink_insert(sk, pid);
522 if (err == -EADDRINUSE)
525 /* If 2 threads race to autobind, that is fine. */
532 static inline int netlink_capable(struct socket *sock, unsigned int flag)
534 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
535 capable(CAP_NET_ADMIN);
539 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
541 struct netlink_sock *nlk = nlk_sk(sk);
543 if (nlk->subscriptions && !subscriptions)
544 __sk_del_bind_node(sk);
545 else if (!nlk->subscriptions && subscriptions)
546 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
547 nlk->subscriptions = subscriptions;
550 static int netlink_alloc_groups(struct sock *sk)
552 struct netlink_sock *nlk = nlk_sk(sk);
556 netlink_lock_table();
557 groups = nl_table[sk->sk_protocol].groups;
558 if (!nl_table[sk->sk_protocol].registered)
560 netlink_unlock_table();
565 nlk->groups = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
566 if (nlk->groups == NULL)
568 nlk->ngroups = groups;
572 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
574 struct sock *sk = sock->sk;
575 struct netlink_sock *nlk = nlk_sk(sk);
576 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
579 if (nladdr->nl_family != AF_NETLINK)
582 /* Only superuser is allowed to listen multicasts */
583 if (nladdr->nl_groups) {
584 if (!netlink_capable(sock, NL_NONROOT_RECV))
586 if (nlk->groups == NULL) {
587 err = netlink_alloc_groups(sk);
594 if (nladdr->nl_pid != nlk->pid)
597 err = nladdr->nl_pid ?
598 netlink_insert(sk, nladdr->nl_pid) :
599 netlink_autobind(sock);
604 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
607 netlink_table_grab();
608 netlink_update_subscriptions(sk, nlk->subscriptions +
609 hweight32(nladdr->nl_groups) -
610 hweight32(nlk->groups[0]));
611 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
612 netlink_update_listeners(sk);
613 netlink_table_ungrab();
618 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
622 struct sock *sk = sock->sk;
623 struct netlink_sock *nlk = nlk_sk(sk);
624 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
626 if (addr->sa_family == AF_UNSPEC) {
627 sk->sk_state = NETLINK_UNCONNECTED;
632 if (addr->sa_family != AF_NETLINK)
635 /* Only superuser is allowed to send multicasts */
636 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
640 err = netlink_autobind(sock);
643 sk->sk_state = NETLINK_CONNECTED;
644 nlk->dst_pid = nladdr->nl_pid;
645 nlk->dst_group = ffs(nladdr->nl_groups);
651 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
653 struct sock *sk = sock->sk;
654 struct netlink_sock *nlk = nlk_sk(sk);
655 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
657 nladdr->nl_family = AF_NETLINK;
659 *addr_len = sizeof(*nladdr);
662 nladdr->nl_pid = nlk->dst_pid;
663 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
665 nladdr->nl_pid = nlk->pid;
666 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
671 static void netlink_overrun(struct sock *sk)
673 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
674 sk->sk_err = ENOBUFS;
675 sk->sk_error_report(sk);
679 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
681 int protocol = ssk->sk_protocol;
683 struct netlink_sock *nlk;
685 sock = netlink_lookup(protocol, pid);
687 return ERR_PTR(-ECONNREFUSED);
689 /* Don't bother queuing skb if kernel socket has no input function */
691 if ((nlk->pid == 0 && !nlk->data_ready) ||
692 (sock->sk_state == NETLINK_CONNECTED &&
693 nlk->dst_pid != nlk_sk(ssk)->pid)) {
695 return ERR_PTR(-ECONNREFUSED);
700 struct sock *netlink_getsockbyfilp(struct file *filp)
702 struct inode *inode = filp->f_dentry->d_inode;
705 if (!S_ISSOCK(inode->i_mode))
706 return ERR_PTR(-ENOTSOCK);
708 sock = SOCKET_I(inode)->sk;
709 if (sock->sk_family != AF_NETLINK)
710 return ERR_PTR(-EINVAL);
717 * Attach a skb to a netlink socket.
718 * The caller must hold a reference to the destination socket. On error, the
719 * reference is dropped. The skb is not send to the destination, just all
720 * all error checks are performed and memory in the queue is reserved.
722 * < 0: error. skb freed, reference to sock dropped.
724 * 1: repeat lookup - reference dropped while waiting for socket memory.
726 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
727 long timeo, struct sock *ssk)
729 struct netlink_sock *nlk;
733 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
734 test_bit(0, &nlk->state)) {
735 DECLARE_WAITQUEUE(wait, current);
737 if (!ssk || nlk_sk(ssk)->pid == 0)
744 __set_current_state(TASK_INTERRUPTIBLE);
745 add_wait_queue(&nlk->wait, &wait);
747 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
748 test_bit(0, &nlk->state)) &&
749 !sock_flag(sk, SOCK_DEAD))
750 timeo = schedule_timeout(timeo);
752 __set_current_state(TASK_RUNNING);
753 remove_wait_queue(&nlk->wait, &wait);
756 if (signal_pending(current)) {
758 return sock_intr_errno(timeo);
762 skb_set_owner_r(skb, sk);
766 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
770 skb_queue_tail(&sk->sk_receive_queue, skb);
771 sk->sk_data_ready(sk, len);
776 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
782 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
789 delta = skb->end - skb->tail;
790 if (delta * 2 < skb->truesize)
793 if (skb_shared(skb)) {
794 struct sk_buff *nskb = skb_clone(skb, allocation);
801 if (!pskb_expand_head(skb, 0, -delta, allocation))
802 skb->truesize -= delta;
807 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
813 skb = netlink_trim(skb, gfp_any());
815 timeo = sock_sndtimeo(ssk, nonblock);
817 sk = netlink_getsockbypid(ssk, pid);
822 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
828 return netlink_sendskb(sk, skb, ssk->sk_protocol);
831 int netlink_has_listeners(struct sock *sk, unsigned int group)
835 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
836 if (group - 1 < nl_table[sk->sk_protocol].groups)
837 res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners);
840 EXPORT_SYMBOL_GPL(netlink_has_listeners);
842 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
844 struct netlink_sock *nlk = nlk_sk(sk);
846 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
847 !test_bit(0, &nlk->state)) {
848 skb_set_owner_r(skb, sk);
849 skb_queue_tail(&sk->sk_receive_queue, skb);
850 sk->sk_data_ready(sk, skb->len);
851 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
856 struct netlink_broadcast_data {
857 struct sock *exclude_sk;
864 struct sk_buff *skb, *skb2;
867 static inline int do_one_broadcast(struct sock *sk,
868 struct netlink_broadcast_data *p)
870 struct netlink_sock *nlk = nlk_sk(sk);
873 if (p->exclude_sk == sk)
876 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
877 !test_bit(p->group - 1, nlk->groups))
886 if (p->skb2 == NULL) {
887 if (skb_shared(p->skb)) {
888 p->skb2 = skb_clone(p->skb, p->allocation);
890 p->skb2 = skb_get(p->skb);
892 * skb ownership may have been set when
893 * delivered to a previous socket.
898 if (p->skb2 == NULL) {
900 /* Clone failed. Notify ALL listeners. */
902 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
915 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
916 u32 group, gfp_t allocation)
918 struct netlink_broadcast_data info;
919 struct hlist_node *node;
922 skb = netlink_trim(skb, allocation);
924 info.exclude_sk = ssk;
930 info.allocation = allocation;
934 /* While we sleep in clone, do not allow to change socket list */
936 netlink_lock_table();
938 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
939 do_one_broadcast(sk, &info);
943 netlink_unlock_table();
946 kfree_skb(info.skb2);
948 if (info.delivered) {
949 if (info.congested && (allocation & __GFP_WAIT))
958 struct netlink_set_err_data {
959 struct sock *exclude_sk;
965 static inline int do_one_set_err(struct sock *sk,
966 struct netlink_set_err_data *p)
968 struct netlink_sock *nlk = nlk_sk(sk);
970 if (sk == p->exclude_sk)
973 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
974 !test_bit(p->group - 1, nlk->groups))
977 sk->sk_err = p->code;
978 sk->sk_error_report(sk);
983 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
985 struct netlink_set_err_data info;
986 struct hlist_node *node;
989 info.exclude_sk = ssk;
994 read_lock(&nl_table_lock);
996 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
997 do_one_set_err(sk, &info);
999 read_unlock(&nl_table_lock);
1002 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1003 char __user *optval, int optlen)
1005 struct sock *sk = sock->sk;
1006 struct netlink_sock *nlk = nlk_sk(sk);
1009 if (level != SOL_NETLINK)
1010 return -ENOPROTOOPT;
1012 if (optlen >= sizeof(int) &&
1013 get_user(val, (int __user *)optval))
1017 case NETLINK_PKTINFO:
1019 nlk->flags |= NETLINK_RECV_PKTINFO;
1021 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1024 case NETLINK_ADD_MEMBERSHIP:
1025 case NETLINK_DROP_MEMBERSHIP: {
1026 unsigned int subscriptions;
1027 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1029 if (!netlink_capable(sock, NL_NONROOT_RECV))
1031 if (nlk->groups == NULL) {
1032 err = netlink_alloc_groups(sk);
1036 if (!val || val - 1 >= nlk->ngroups)
1038 netlink_table_grab();
1039 old = test_bit(val - 1, nlk->groups);
1040 subscriptions = nlk->subscriptions - old + new;
1042 __set_bit(val - 1, nlk->groups);
1044 __clear_bit(val - 1, nlk->groups);
1045 netlink_update_subscriptions(sk, subscriptions);
1046 netlink_update_listeners(sk);
1047 netlink_table_ungrab();
1057 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1058 char __user *optval, int __user *optlen)
1060 struct sock *sk = sock->sk;
1061 struct netlink_sock *nlk = nlk_sk(sk);
1064 if (level != SOL_NETLINK)
1065 return -ENOPROTOOPT;
1067 if (get_user(len, optlen))
1073 case NETLINK_PKTINFO:
1074 if (len < sizeof(int))
1077 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1078 put_user(len, optlen);
1079 put_user(val, optval);
1088 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1090 struct nl_pktinfo info;
1092 info.group = NETLINK_CB(skb).dst_group;
1093 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1096 static inline void netlink_rcv_wake(struct sock *sk)
1098 struct netlink_sock *nlk = nlk_sk(sk);
1100 if (skb_queue_empty(&sk->sk_receive_queue))
1101 clear_bit(0, &nlk->state);
1102 if (!test_bit(0, &nlk->state))
1103 wake_up_interruptible(&nlk->wait);
1106 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1107 struct msghdr *msg, size_t len)
1109 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1110 struct sock *sk = sock->sk;
1111 struct netlink_sock *nlk = nlk_sk(sk);
1112 struct sockaddr_nl *addr=msg->msg_name;
1115 struct sk_buff *skb;
1117 struct scm_cookie scm;
1119 if (msg->msg_flags&MSG_OOB)
1122 if (NULL == siocb->scm)
1124 err = scm_send(sock, msg, siocb->scm);
1128 if (msg->msg_namelen) {
1129 if (addr->nl_family != AF_NETLINK)
1131 dst_pid = addr->nl_pid;
1132 dst_group = ffs(addr->nl_groups);
1133 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1136 dst_pid = nlk->dst_pid;
1137 dst_group = nlk->dst_group;
1141 err = netlink_autobind(sock);
1147 if (len > sk->sk_sndbuf - 32)
1150 skb = nlmsg_new(len, GFP_KERNEL);
1154 NETLINK_CB(skb).pid = nlk->pid;
1155 NETLINK_CB(skb).dst_pid = dst_pid;
1156 NETLINK_CB(skb).dst_group = dst_group;
1157 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1158 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1159 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1161 /* What can I do? Netlink is asynchronous, so that
1162 we will have to save current capabilities to
1163 check them, when this message will be delivered
1164 to corresponding kernel module. --ANK (980802)
1168 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1173 err = security_netlink_send(sk, skb);
1180 atomic_inc(&skb->users);
1181 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1183 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1189 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1190 struct msghdr *msg, size_t len,
1193 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1194 struct scm_cookie scm;
1195 struct sock *sk = sock->sk;
1196 struct netlink_sock *nlk = nlk_sk(sk);
1197 int noblock = flags&MSG_DONTWAIT;
1199 struct sk_buff *skb;
1207 skb = skb_recv_datagram(sk,flags,noblock,&err);
1211 msg->msg_namelen = 0;
1215 msg->msg_flags |= MSG_TRUNC;
1219 skb->h.raw = skb->data;
1220 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1222 if (msg->msg_name) {
1223 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1224 addr->nl_family = AF_NETLINK;
1226 addr->nl_pid = NETLINK_CB(skb).pid;
1227 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1228 msg->msg_namelen = sizeof(*addr);
1231 if (nlk->flags & NETLINK_RECV_PKTINFO)
1232 netlink_cmsg_recv_pktinfo(msg, skb);
1234 if (NULL == siocb->scm) {
1235 memset(&scm, 0, sizeof(scm));
1238 siocb->scm->creds = *NETLINK_CREDS(skb);
1239 skb_free_datagram(sk, skb);
1241 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1244 scm_recv(sock, msg, siocb->scm, flags);
1247 netlink_rcv_wake(sk);
1248 return err ? : copied;
1251 static void netlink_data_ready(struct sock *sk, int len)
1253 struct netlink_sock *nlk = nlk_sk(sk);
1255 if (nlk->data_ready)
1256 nlk->data_ready(sk, len);
1257 netlink_rcv_wake(sk);
1261 * We export these functions to other modules. They provide a
1262 * complete set of kernel non-blocking support for message
1267 netlink_kernel_create(int unit, unsigned int groups,
1268 void (*input)(struct sock *sk, int len),
1269 struct module *module)
1271 struct socket *sock;
1273 struct netlink_sock *nlk;
1274 unsigned long *listeners = NULL;
1278 if (unit<0 || unit>=MAX_LINKS)
1281 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1284 if (__netlink_create(sock, unit) < 0)
1285 goto out_sock_release;
1290 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1292 goto out_sock_release;
1295 sk->sk_data_ready = netlink_data_ready;
1297 nlk_sk(sk)->data_ready = input;
1299 if (netlink_insert(sk, 0))
1300 goto out_sock_release;
1303 nlk->flags |= NETLINK_KERNEL_SOCKET;
1305 netlink_table_grab();
1306 nl_table[unit].groups = groups;
1307 nl_table[unit].listeners = listeners;
1308 nl_table[unit].module = module;
1309 nl_table[unit].registered = 1;
1310 netlink_table_ungrab();
1320 void netlink_set_nonroot(int protocol, unsigned int flags)
1322 if ((unsigned int)protocol < MAX_LINKS)
1323 nl_table[protocol].nl_nonroot = flags;
1326 static void netlink_destroy_callback(struct netlink_callback *cb)
1334 * It looks a bit ugly.
1335 * It would be better to create kernel thread.
1338 static int netlink_dump(struct sock *sk)
1340 struct netlink_sock *nlk = nlk_sk(sk);
1341 struct netlink_callback *cb;
1342 struct sk_buff *skb;
1343 struct nlmsghdr *nlh;
1344 int len, err = -ENOBUFS;
1346 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1350 spin_lock(&nlk->cb_lock);
1358 len = cb->dump(skb, cb);
1361 spin_unlock(&nlk->cb_lock);
1362 skb_queue_tail(&sk->sk_receive_queue, skb);
1363 sk->sk_data_ready(sk, len);
1367 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1371 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);
1385 spin_unlock(&nlk->cb_lock);
1391 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1392 struct nlmsghdr *nlh,
1393 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1394 int (*done)(struct netlink_callback*))
1396 struct netlink_callback *cb;
1398 struct netlink_sock *nlk;
1400 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1407 atomic_inc(&skb->users);
1410 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1412 netlink_destroy_callback(cb);
1413 return -ECONNREFUSED;
1416 /* A dump is in progress... */
1417 spin_lock(&nlk->cb_lock);
1419 spin_unlock(&nlk->cb_lock);
1420 netlink_destroy_callback(cb);
1425 spin_unlock(&nlk->cb_lock);
1432 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1434 struct sk_buff *skb;
1435 struct nlmsghdr *rep;
1436 struct nlmsgerr *errmsg;
1440 size = nlmsg_total_size(sizeof(*errmsg));
1442 size = nlmsg_total_size(sizeof(*errmsg) + nlmsg_len(nlh));
1444 skb = nlmsg_new(size, GFP_KERNEL);
1448 sk = netlink_lookup(in_skb->sk->sk_protocol,
1449 NETLINK_CB(in_skb).pid);
1451 sk->sk_err = ENOBUFS;
1452 sk->sk_error_report(sk);
1458 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1459 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1460 errmsg = nlmsg_data(rep);
1461 errmsg->error = err;
1462 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1463 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1466 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1467 struct nlmsghdr *, int *))
1469 struct nlmsghdr *nlh;
1472 while (skb->len >= nlmsg_total_size(0)) {
1473 nlh = (struct nlmsghdr *) skb->data;
1475 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_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 netlink_queue_skip(nlh, skb);
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);
1553 * nlmsg_notify - send a notification netlink message
1554 * @sk: netlink socket to use
1555 * @skb: notification message
1556 * @pid: destination netlink pid for reports or 0
1557 * @group: destination multicast group or 0
1558 * @report: 1 to report back, 0 to disable
1559 * @flags: allocation flags
1561 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1562 unsigned int group, int report, gfp_t flags)
1567 int exclude_pid = 0;
1570 atomic_inc(&skb->users);
1574 /* errors reported via destination sk->sk_err */
1575 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1579 err = nlmsg_unicast(sk, skb, pid);
1584 #ifdef CONFIG_PROC_FS
1585 struct nl_seq_iter {
1590 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1592 struct nl_seq_iter *iter = seq->private;
1595 struct hlist_node *node;
1598 for (i=0; i<MAX_LINKS; i++) {
1599 struct nl_pid_hash *hash = &nl_table[i].hash;
1601 for (j = 0; j <= hash->mask; j++) {
1602 sk_for_each(s, node, &hash->table[j]) {
1615 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1617 read_lock(&nl_table_lock);
1618 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1621 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1624 struct nl_seq_iter *iter;
1629 if (v == SEQ_START_TOKEN)
1630 return netlink_seq_socket_idx(seq, 0);
1636 iter = seq->private;
1638 j = iter->hash_idx + 1;
1641 struct nl_pid_hash *hash = &nl_table[i].hash;
1643 for (; j <= hash->mask; j++) {
1644 s = sk_head(&hash->table[j]);
1653 } while (++i < MAX_LINKS);
1658 static void netlink_seq_stop(struct seq_file *seq, void *v)
1660 read_unlock(&nl_table_lock);
1664 static int netlink_seq_show(struct seq_file *seq, void *v)
1666 if (v == SEQ_START_TOKEN)
1668 "sk Eth Pid Groups "
1669 "Rmem Wmem Dump Locks\n");
1672 struct netlink_sock *nlk = nlk_sk(s);
1674 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1678 nlk->groups ? (u32)nlk->groups[0] : 0,
1679 atomic_read(&s->sk_rmem_alloc),
1680 atomic_read(&s->sk_wmem_alloc),
1682 atomic_read(&s->sk_refcnt)
1689 static struct seq_operations netlink_seq_ops = {
1690 .start = netlink_seq_start,
1691 .next = netlink_seq_next,
1692 .stop = netlink_seq_stop,
1693 .show = netlink_seq_show,
1697 static int netlink_seq_open(struct inode *inode, struct file *file)
1699 struct seq_file *seq;
1700 struct nl_seq_iter *iter;
1703 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1707 err = seq_open(file, &netlink_seq_ops);
1713 seq = file->private_data;
1714 seq->private = iter;
1718 static struct file_operations netlink_seq_fops = {
1719 .owner = THIS_MODULE,
1720 .open = netlink_seq_open,
1722 .llseek = seq_lseek,
1723 .release = seq_release_private,
1728 int netlink_register_notifier(struct notifier_block *nb)
1730 return atomic_notifier_chain_register(&netlink_chain, nb);
1733 int netlink_unregister_notifier(struct notifier_block *nb)
1735 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1738 static const struct proto_ops netlink_ops = {
1739 .family = PF_NETLINK,
1740 .owner = THIS_MODULE,
1741 .release = netlink_release,
1742 .bind = netlink_bind,
1743 .connect = netlink_connect,
1744 .socketpair = sock_no_socketpair,
1745 .accept = sock_no_accept,
1746 .getname = netlink_getname,
1747 .poll = datagram_poll,
1748 .ioctl = sock_no_ioctl,
1749 .listen = sock_no_listen,
1750 .shutdown = sock_no_shutdown,
1751 .setsockopt = netlink_setsockopt,
1752 .getsockopt = netlink_getsockopt,
1753 .sendmsg = netlink_sendmsg,
1754 .recvmsg = netlink_recvmsg,
1755 .mmap = sock_no_mmap,
1756 .sendpage = sock_no_sendpage,
1759 static struct net_proto_family netlink_family_ops = {
1760 .family = PF_NETLINK,
1761 .create = netlink_create,
1762 .owner = THIS_MODULE, /* for consistency 8) */
1765 static int __init netlink_proto_init(void)
1767 struct sk_buff *dummy_skb;
1771 int err = proto_register(&netlink_proto, 0);
1776 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1778 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1782 if (num_physpages >= (128 * 1024))
1783 max = num_physpages >> (21 - PAGE_SHIFT);
1785 max = num_physpages >> (23 - PAGE_SHIFT);
1787 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1788 max = (1UL << order) / sizeof(struct hlist_head);
1789 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1791 for (i = 0; i < MAX_LINKS; i++) {
1792 struct nl_pid_hash *hash = &nl_table[i].hash;
1794 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1797 nl_pid_hash_free(nl_table[i].hash.table,
1798 1 * sizeof(*hash->table));
1802 memset(hash->table, 0, 1 * sizeof(*hash->table));
1803 hash->max_shift = order;
1806 hash->rehash_time = jiffies;
1809 sock_register(&netlink_family_ops);
1810 #ifdef CONFIG_PROC_FS
1811 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1813 /* The netlink device handler may be needed early. */
1818 panic("netlink_init: Cannot allocate nl_table\n");
1821 core_initcall(netlink_proto_init);
1823 EXPORT_SYMBOL(netlink_ack);
1824 EXPORT_SYMBOL(netlink_run_queue);
1825 EXPORT_SYMBOL(netlink_queue_skip);
1826 EXPORT_SYMBOL(netlink_broadcast);
1827 EXPORT_SYMBOL(netlink_dump_start);
1828 EXPORT_SYMBOL(netlink_kernel_create);
1829 EXPORT_SYMBOL(netlink_register_notifier);
1830 EXPORT_SYMBOL(netlink_set_err);
1831 EXPORT_SYMBOL(netlink_set_nonroot);
1832 EXPORT_SYMBOL(netlink_unicast);
1833 EXPORT_SYMBOL(netlink_unregister_notifier);
1834 EXPORT_SYMBOL(nlmsg_notify);