2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
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/mutex.h>
59 #include <net/net_namespace.h>
62 #include <net/netlink.h>
64 #include "af_netlink.h"
68 unsigned long masks[0];
72 #define NETLINK_CONGESTED 0x0
75 #define NETLINK_KERNEL_SOCKET 0x1
76 #define NETLINK_RECV_PKTINFO 0x2
77 #define NETLINK_BROADCAST_SEND_ERROR 0x4
78 #define NETLINK_RECV_NO_ENOBUFS 0x8
80 static inline int netlink_is_kernel(struct sock *sk)
82 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
85 struct netlink_table *nl_table;
86 EXPORT_SYMBOL_GPL(nl_table);
88 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
90 static int netlink_dump(struct sock *sk);
92 DEFINE_RWLOCK(nl_table_lock);
93 EXPORT_SYMBOL_GPL(nl_table_lock);
94 static atomic_t nl_table_users = ATOMIC_INIT(0);
96 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
98 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
100 static inline u32 netlink_group_mask(u32 group)
102 return group ? 1 << (group - 1) : 0;
105 static inline struct hlist_head *nl_portid_hashfn(struct nl_portid_hash *hash, u32 portid)
107 return &hash->table[jhash_1word(portid, hash->rnd) & hash->mask];
110 static void netlink_destroy_callback(struct netlink_callback *cb)
116 static void netlink_consume_callback(struct netlink_callback *cb)
118 consume_skb(cb->skb);
122 static void netlink_sock_destruct(struct sock *sk)
124 struct netlink_sock *nlk = nlk_sk(sk);
128 nlk->cb->done(nlk->cb);
130 module_put(nlk->cb->module);
131 netlink_destroy_callback(nlk->cb);
134 skb_queue_purge(&sk->sk_receive_queue);
136 if (!sock_flag(sk, SOCK_DEAD)) {
137 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
141 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
142 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
143 WARN_ON(nlk_sk(sk)->groups);
146 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
147 * SMP. Look, when several writers sleep and reader wakes them up, all but one
148 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
149 * this, _but_ remember, it adds useless work on UP machines.
152 void netlink_table_grab(void)
153 __acquires(nl_table_lock)
157 write_lock_irq(&nl_table_lock);
159 if (atomic_read(&nl_table_users)) {
160 DECLARE_WAITQUEUE(wait, current);
162 add_wait_queue_exclusive(&nl_table_wait, &wait);
164 set_current_state(TASK_UNINTERRUPTIBLE);
165 if (atomic_read(&nl_table_users) == 0)
167 write_unlock_irq(&nl_table_lock);
169 write_lock_irq(&nl_table_lock);
172 __set_current_state(TASK_RUNNING);
173 remove_wait_queue(&nl_table_wait, &wait);
177 void netlink_table_ungrab(void)
178 __releases(nl_table_lock)
180 write_unlock_irq(&nl_table_lock);
181 wake_up(&nl_table_wait);
185 netlink_lock_table(void)
187 /* read_lock() synchronizes us to netlink_table_grab */
189 read_lock(&nl_table_lock);
190 atomic_inc(&nl_table_users);
191 read_unlock(&nl_table_lock);
195 netlink_unlock_table(void)
197 if (atomic_dec_and_test(&nl_table_users))
198 wake_up(&nl_table_wait);
201 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
203 struct nl_portid_hash *hash = &nl_table[protocol].hash;
204 struct hlist_head *head;
207 read_lock(&nl_table_lock);
208 head = nl_portid_hashfn(hash, portid);
209 sk_for_each(sk, head) {
210 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->portid == portid)) {
217 read_unlock(&nl_table_lock);
221 static struct hlist_head *nl_portid_hash_zalloc(size_t size)
223 if (size <= PAGE_SIZE)
224 return kzalloc(size, GFP_ATOMIC);
226 return (struct hlist_head *)
227 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
231 static void nl_portid_hash_free(struct hlist_head *table, size_t size)
233 if (size <= PAGE_SIZE)
236 free_pages((unsigned long)table, get_order(size));
239 static int nl_portid_hash_rehash(struct nl_portid_hash *hash, int grow)
241 unsigned int omask, mask, shift;
243 struct hlist_head *otable, *table;
246 omask = mask = hash->mask;
247 osize = size = (mask + 1) * sizeof(*table);
251 if (++shift > hash->max_shift)
257 table = nl_portid_hash_zalloc(size);
261 otable = hash->table;
265 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
267 for (i = 0; i <= omask; i++) {
269 struct hlist_node *tmp;
271 sk_for_each_safe(sk, tmp, &otable[i])
272 __sk_add_node(sk, nl_portid_hashfn(hash, nlk_sk(sk)->portid));
275 nl_portid_hash_free(otable, osize);
276 hash->rehash_time = jiffies + 10 * 60 * HZ;
280 static inline int nl_portid_hash_dilute(struct nl_portid_hash *hash, int len)
282 int avg = hash->entries >> hash->shift;
284 if (unlikely(avg > 1) && nl_portid_hash_rehash(hash, 1))
287 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
288 nl_portid_hash_rehash(hash, 0);
295 static const struct proto_ops netlink_ops;
298 netlink_update_listeners(struct sock *sk)
300 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
303 struct listeners *listeners;
305 listeners = nl_deref_protected(tbl->listeners);
309 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
311 sk_for_each_bound(sk, &tbl->mc_list) {
312 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
313 mask |= nlk_sk(sk)->groups[i];
315 listeners->masks[i] = mask;
317 /* this function is only called with the netlink table "grabbed", which
318 * makes sure updates are visible before bind or setsockopt return. */
321 static int netlink_insert(struct sock *sk, struct net *net, u32 portid)
323 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash;
324 struct hlist_head *head;
325 int err = -EADDRINUSE;
329 netlink_table_grab();
330 head = nl_portid_hashfn(hash, portid);
332 sk_for_each(osk, head) {
333 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->portid == portid))
341 if (nlk_sk(sk)->portid)
345 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
348 if (len && nl_portid_hash_dilute(hash, len))
349 head = nl_portid_hashfn(hash, portid);
351 nlk_sk(sk)->portid = portid;
352 sk_add_node(sk, head);
356 netlink_table_ungrab();
360 static void netlink_remove(struct sock *sk)
362 netlink_table_grab();
363 if (sk_del_node_init(sk))
364 nl_table[sk->sk_protocol].hash.entries--;
365 if (nlk_sk(sk)->subscriptions)
366 __sk_del_bind_node(sk);
367 netlink_table_ungrab();
370 static struct proto netlink_proto = {
372 .owner = THIS_MODULE,
373 .obj_size = sizeof(struct netlink_sock),
376 static int __netlink_create(struct net *net, struct socket *sock,
377 struct mutex *cb_mutex, int protocol)
380 struct netlink_sock *nlk;
382 sock->ops = &netlink_ops;
384 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
388 sock_init_data(sock, sk);
392 nlk->cb_mutex = cb_mutex;
394 nlk->cb_mutex = &nlk->cb_def_mutex;
395 mutex_init(nlk->cb_mutex);
397 init_waitqueue_head(&nlk->wait);
399 sk->sk_destruct = netlink_sock_destruct;
400 sk->sk_protocol = protocol;
404 static int netlink_create(struct net *net, struct socket *sock, int protocol,
407 struct module *module = NULL;
408 struct mutex *cb_mutex;
409 struct netlink_sock *nlk;
410 void (*bind)(int group);
413 sock->state = SS_UNCONNECTED;
415 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
416 return -ESOCKTNOSUPPORT;
418 if (protocol < 0 || protocol >= MAX_LINKS)
419 return -EPROTONOSUPPORT;
421 netlink_lock_table();
422 #ifdef CONFIG_MODULES
423 if (!nl_table[protocol].registered) {
424 netlink_unlock_table();
425 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
426 netlink_lock_table();
429 if (nl_table[protocol].registered &&
430 try_module_get(nl_table[protocol].module))
431 module = nl_table[protocol].module;
433 err = -EPROTONOSUPPORT;
434 cb_mutex = nl_table[protocol].cb_mutex;
435 bind = nl_table[protocol].bind;
436 netlink_unlock_table();
441 err = __netlink_create(net, sock, cb_mutex, protocol);
446 sock_prot_inuse_add(net, &netlink_proto, 1);
449 nlk = nlk_sk(sock->sk);
450 nlk->module = module;
451 nlk->netlink_bind = bind;
460 static int netlink_release(struct socket *sock)
462 struct sock *sk = sock->sk;
463 struct netlink_sock *nlk;
473 * OK. Socket is unlinked, any packets that arrive now
478 wake_up_interruptible_all(&nlk->wait);
480 skb_queue_purge(&sk->sk_write_queue);
483 struct netlink_notify n = {
485 .protocol = sk->sk_protocol,
486 .portid = nlk->portid,
488 atomic_notifier_call_chain(&netlink_chain,
489 NETLINK_URELEASE, &n);
492 module_put(nlk->module);
494 netlink_table_grab();
495 if (netlink_is_kernel(sk)) {
496 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
497 if (--nl_table[sk->sk_protocol].registered == 0) {
498 struct listeners *old;
500 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
501 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
503 nl_table[sk->sk_protocol].module = NULL;
504 nl_table[sk->sk_protocol].bind = NULL;
505 nl_table[sk->sk_protocol].flags = 0;
506 nl_table[sk->sk_protocol].registered = 0;
508 } else if (nlk->subscriptions) {
509 netlink_update_listeners(sk);
511 netlink_table_ungrab();
517 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
523 static int netlink_autobind(struct socket *sock)
525 struct sock *sk = sock->sk;
526 struct net *net = sock_net(sk);
527 struct nl_portid_hash *hash = &nl_table[sk->sk_protocol].hash;
528 struct hlist_head *head;
530 s32 portid = task_tgid_vnr(current);
532 static s32 rover = -4097;
536 netlink_table_grab();
537 head = nl_portid_hashfn(hash, portid);
538 sk_for_each(osk, head) {
539 if (!net_eq(sock_net(osk), net))
541 if (nlk_sk(osk)->portid == portid) {
542 /* Bind collision, search negative portid values. */
546 netlink_table_ungrab();
550 netlink_table_ungrab();
552 err = netlink_insert(sk, net, portid);
553 if (err == -EADDRINUSE)
556 /* If 2 threads race to autobind, that is fine. */
563 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
565 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
566 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
570 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
572 struct netlink_sock *nlk = nlk_sk(sk);
574 if (nlk->subscriptions && !subscriptions)
575 __sk_del_bind_node(sk);
576 else if (!nlk->subscriptions && subscriptions)
577 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
578 nlk->subscriptions = subscriptions;
581 static int netlink_realloc_groups(struct sock *sk)
583 struct netlink_sock *nlk = nlk_sk(sk);
585 unsigned long *new_groups;
588 netlink_table_grab();
590 groups = nl_table[sk->sk_protocol].groups;
591 if (!nl_table[sk->sk_protocol].registered) {
596 if (nlk->ngroups >= groups)
599 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
600 if (new_groups == NULL) {
604 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
605 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
607 nlk->groups = new_groups;
608 nlk->ngroups = groups;
610 netlink_table_ungrab();
614 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
617 struct sock *sk = sock->sk;
618 struct net *net = sock_net(sk);
619 struct netlink_sock *nlk = nlk_sk(sk);
620 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
623 if (addr_len < sizeof(struct sockaddr_nl))
626 if (nladdr->nl_family != AF_NETLINK)
629 /* Only superuser is allowed to listen multicasts */
630 if (nladdr->nl_groups) {
631 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
633 err = netlink_realloc_groups(sk);
639 if (nladdr->nl_pid != nlk->portid)
642 err = nladdr->nl_pid ?
643 netlink_insert(sk, net, nladdr->nl_pid) :
644 netlink_autobind(sock);
649 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
652 netlink_table_grab();
653 netlink_update_subscriptions(sk, nlk->subscriptions +
654 hweight32(nladdr->nl_groups) -
655 hweight32(nlk->groups[0]));
656 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
657 netlink_update_listeners(sk);
658 netlink_table_ungrab();
660 if (nlk->netlink_bind && nlk->groups[0]) {
663 for (i=0; i<nlk->ngroups; i++) {
664 if (test_bit(i, nlk->groups))
665 nlk->netlink_bind(i);
672 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
676 struct sock *sk = sock->sk;
677 struct netlink_sock *nlk = nlk_sk(sk);
678 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
680 if (alen < sizeof(addr->sa_family))
683 if (addr->sa_family == AF_UNSPEC) {
684 sk->sk_state = NETLINK_UNCONNECTED;
689 if (addr->sa_family != AF_NETLINK)
692 /* Only superuser is allowed to send multicasts */
693 if (nladdr->nl_groups && !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
697 err = netlink_autobind(sock);
700 sk->sk_state = NETLINK_CONNECTED;
701 nlk->dst_portid = nladdr->nl_pid;
702 nlk->dst_group = ffs(nladdr->nl_groups);
708 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
709 int *addr_len, int peer)
711 struct sock *sk = sock->sk;
712 struct netlink_sock *nlk = nlk_sk(sk);
713 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
715 nladdr->nl_family = AF_NETLINK;
717 *addr_len = sizeof(*nladdr);
720 nladdr->nl_pid = nlk->dst_portid;
721 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
723 nladdr->nl_pid = nlk->portid;
724 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
729 static void netlink_overrun(struct sock *sk)
731 struct netlink_sock *nlk = nlk_sk(sk);
733 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
734 if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
735 sk->sk_err = ENOBUFS;
736 sk->sk_error_report(sk);
739 atomic_inc(&sk->sk_drops);
742 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
745 struct netlink_sock *nlk;
747 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
749 return ERR_PTR(-ECONNREFUSED);
751 /* Don't bother queuing skb if kernel socket has no input function */
753 if (sock->sk_state == NETLINK_CONNECTED &&
754 nlk->dst_portid != nlk_sk(ssk)->portid) {
756 return ERR_PTR(-ECONNREFUSED);
761 struct sock *netlink_getsockbyfilp(struct file *filp)
763 struct inode *inode = file_inode(filp);
766 if (!S_ISSOCK(inode->i_mode))
767 return ERR_PTR(-ENOTSOCK);
769 sock = SOCKET_I(inode)->sk;
770 if (sock->sk_family != AF_NETLINK)
771 return ERR_PTR(-EINVAL);
778 * Attach a skb to a netlink socket.
779 * The caller must hold a reference to the destination socket. On error, the
780 * reference is dropped. The skb is not send to the destination, just all
781 * all error checks are performed and memory in the queue is reserved.
783 * < 0: error. skb freed, reference to sock dropped.
785 * 1: repeat lookup - reference dropped while waiting for socket memory.
787 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
788 long *timeo, struct sock *ssk)
790 struct netlink_sock *nlk;
794 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
795 test_bit(NETLINK_CONGESTED, &nlk->state)) {
796 DECLARE_WAITQUEUE(wait, current);
798 if (!ssk || netlink_is_kernel(ssk))
805 __set_current_state(TASK_INTERRUPTIBLE);
806 add_wait_queue(&nlk->wait, &wait);
808 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
809 test_bit(NETLINK_CONGESTED, &nlk->state)) &&
810 !sock_flag(sk, SOCK_DEAD))
811 *timeo = schedule_timeout(*timeo);
813 __set_current_state(TASK_RUNNING);
814 remove_wait_queue(&nlk->wait, &wait);
817 if (signal_pending(current)) {
819 return sock_intr_errno(*timeo);
823 skb_set_owner_r(skb, sk);
827 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
831 skb_queue_tail(&sk->sk_receive_queue, skb);
832 sk->sk_data_ready(sk, len);
836 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
838 int len = __netlink_sendskb(sk, skb);
844 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
850 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
854 WARN_ON(skb->sk != NULL);
856 delta = skb->end - skb->tail;
857 if (delta * 2 < skb->truesize)
860 if (skb_shared(skb)) {
861 struct sk_buff *nskb = skb_clone(skb, allocation);
868 if (!pskb_expand_head(skb, 0, -delta, allocation))
869 skb->truesize -= delta;
874 static void netlink_rcv_wake(struct sock *sk)
876 struct netlink_sock *nlk = nlk_sk(sk);
878 if (skb_queue_empty(&sk->sk_receive_queue))
879 clear_bit(NETLINK_CONGESTED, &nlk->state);
880 if (!test_bit(NETLINK_CONGESTED, &nlk->state))
881 wake_up_interruptible(&nlk->wait);
884 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
888 struct netlink_sock *nlk = nlk_sk(sk);
891 if (nlk->netlink_rcv != NULL) {
893 skb_set_owner_r(skb, sk);
894 NETLINK_CB(skb).sk = ssk;
895 nlk->netlink_rcv(skb);
904 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
905 u32 portid, int nonblock)
911 skb = netlink_trim(skb, gfp_any());
913 timeo = sock_sndtimeo(ssk, nonblock);
915 sk = netlink_getsockbyportid(ssk, portid);
920 if (netlink_is_kernel(sk))
921 return netlink_unicast_kernel(sk, skb, ssk);
923 if (sk_filter(sk, skb)) {
930 err = netlink_attachskb(sk, skb, &timeo, ssk);
936 return netlink_sendskb(sk, skb);
938 EXPORT_SYMBOL(netlink_unicast);
940 int netlink_has_listeners(struct sock *sk, unsigned int group)
943 struct listeners *listeners;
945 BUG_ON(!netlink_is_kernel(sk));
948 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
950 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
951 res = test_bit(group - 1, listeners->masks);
957 EXPORT_SYMBOL_GPL(netlink_has_listeners);
959 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
961 struct netlink_sock *nlk = nlk_sk(sk);
963 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
964 !test_bit(NETLINK_CONGESTED, &nlk->state)) {
965 skb_set_owner_r(skb, sk);
966 __netlink_sendskb(sk, skb);
967 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
972 struct netlink_broadcast_data {
973 struct sock *exclude_sk;
978 int delivery_failure;
982 struct sk_buff *skb, *skb2;
983 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
987 static int do_one_broadcast(struct sock *sk,
988 struct netlink_broadcast_data *p)
990 struct netlink_sock *nlk = nlk_sk(sk);
993 if (p->exclude_sk == sk)
996 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
997 !test_bit(p->group - 1, nlk->groups))
1000 if (!net_eq(sock_net(sk), p->net))
1004 netlink_overrun(sk);
1009 if (p->skb2 == NULL) {
1010 if (skb_shared(p->skb)) {
1011 p->skb2 = skb_clone(p->skb, p->allocation);
1013 p->skb2 = skb_get(p->skb);
1015 * skb ownership may have been set when
1016 * delivered to a previous socket.
1018 skb_orphan(p->skb2);
1021 if (p->skb2 == NULL) {
1022 netlink_overrun(sk);
1023 /* Clone failed. Notify ALL listeners. */
1025 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1026 p->delivery_failure = 1;
1027 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1030 } else if (sk_filter(sk, p->skb2)) {
1033 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1034 netlink_overrun(sk);
1035 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1036 p->delivery_failure = 1;
1038 p->congested |= val;
1048 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1049 u32 group, gfp_t allocation,
1050 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1053 struct net *net = sock_net(ssk);
1054 struct netlink_broadcast_data info;
1057 skb = netlink_trim(skb, allocation);
1059 info.exclude_sk = ssk;
1061 info.portid = portid;
1064 info.delivery_failure = 0;
1067 info.allocation = allocation;
1070 info.tx_filter = filter;
1071 info.tx_data = filter_data;
1073 /* While we sleep in clone, do not allow to change socket list */
1075 netlink_lock_table();
1077 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1078 do_one_broadcast(sk, &info);
1082 netlink_unlock_table();
1084 if (info.delivery_failure) {
1085 kfree_skb(info.skb2);
1088 consume_skb(info.skb2);
1090 if (info.delivered) {
1091 if (info.congested && (allocation & __GFP_WAIT))
1097 EXPORT_SYMBOL(netlink_broadcast_filtered);
1099 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1100 u32 group, gfp_t allocation)
1102 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1105 EXPORT_SYMBOL(netlink_broadcast);
1107 struct netlink_set_err_data {
1108 struct sock *exclude_sk;
1114 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1116 struct netlink_sock *nlk = nlk_sk(sk);
1119 if (sk == p->exclude_sk)
1122 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1125 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1126 !test_bit(p->group - 1, nlk->groups))
1129 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1134 sk->sk_err = p->code;
1135 sk->sk_error_report(sk);
1141 * netlink_set_err - report error to broadcast listeners
1142 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1143 * @portid: the PORTID of a process that we want to skip (if any)
1144 * @groups: the broadcast group that will notice the error
1145 * @code: error code, must be negative (as usual in kernelspace)
1147 * This function returns the number of broadcast listeners that have set the
1148 * NETLINK_RECV_NO_ENOBUFS socket option.
1150 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1152 struct netlink_set_err_data info;
1156 info.exclude_sk = ssk;
1157 info.portid = portid;
1159 /* sk->sk_err wants a positive error value */
1162 read_lock(&nl_table_lock);
1164 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1165 ret += do_one_set_err(sk, &info);
1167 read_unlock(&nl_table_lock);
1170 EXPORT_SYMBOL(netlink_set_err);
1172 /* must be called with netlink table grabbed */
1173 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1177 int old, new = !!is_new, subscriptions;
1179 old = test_bit(group - 1, nlk->groups);
1180 subscriptions = nlk->subscriptions - old + new;
1182 __set_bit(group - 1, nlk->groups);
1184 __clear_bit(group - 1, nlk->groups);
1185 netlink_update_subscriptions(&nlk->sk, subscriptions);
1186 netlink_update_listeners(&nlk->sk);
1189 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1190 char __user *optval, unsigned int optlen)
1192 struct sock *sk = sock->sk;
1193 struct netlink_sock *nlk = nlk_sk(sk);
1194 unsigned int val = 0;
1197 if (level != SOL_NETLINK)
1198 return -ENOPROTOOPT;
1200 if (optlen >= sizeof(int) &&
1201 get_user(val, (unsigned int __user *)optval))
1205 case NETLINK_PKTINFO:
1207 nlk->flags |= NETLINK_RECV_PKTINFO;
1209 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1212 case NETLINK_ADD_MEMBERSHIP:
1213 case NETLINK_DROP_MEMBERSHIP: {
1214 if (!netlink_capable(sock, NL_CFG_F_NONROOT_RECV))
1216 err = netlink_realloc_groups(sk);
1219 if (!val || val - 1 >= nlk->ngroups)
1221 netlink_table_grab();
1222 netlink_update_socket_mc(nlk, val,
1223 optname == NETLINK_ADD_MEMBERSHIP);
1224 netlink_table_ungrab();
1226 if (nlk->netlink_bind)
1227 nlk->netlink_bind(val);
1232 case NETLINK_BROADCAST_ERROR:
1234 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1236 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1239 case NETLINK_NO_ENOBUFS:
1241 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1242 clear_bit(NETLINK_CONGESTED, &nlk->state);
1243 wake_up_interruptible(&nlk->wait);
1245 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1255 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1256 char __user *optval, int __user *optlen)
1258 struct sock *sk = sock->sk;
1259 struct netlink_sock *nlk = nlk_sk(sk);
1262 if (level != SOL_NETLINK)
1263 return -ENOPROTOOPT;
1265 if (get_user(len, optlen))
1271 case NETLINK_PKTINFO:
1272 if (len < sizeof(int))
1275 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1276 if (put_user(len, optlen) ||
1277 put_user(val, optval))
1281 case NETLINK_BROADCAST_ERROR:
1282 if (len < sizeof(int))
1285 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1286 if (put_user(len, optlen) ||
1287 put_user(val, optval))
1291 case NETLINK_NO_ENOBUFS:
1292 if (len < sizeof(int))
1295 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1296 if (put_user(len, optlen) ||
1297 put_user(val, optval))
1307 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1309 struct nl_pktinfo info;
1311 info.group = NETLINK_CB(skb).dst_group;
1312 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1315 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1316 struct msghdr *msg, size_t len)
1318 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1319 struct sock *sk = sock->sk;
1320 struct netlink_sock *nlk = nlk_sk(sk);
1321 struct sockaddr_nl *addr = msg->msg_name;
1324 struct sk_buff *skb;
1326 struct scm_cookie scm;
1328 if (msg->msg_flags&MSG_OOB)
1331 if (NULL == siocb->scm)
1334 err = scm_send(sock, msg, siocb->scm, true);
1338 if (msg->msg_namelen) {
1340 if (addr->nl_family != AF_NETLINK)
1342 dst_portid = addr->nl_pid;
1343 dst_group = ffs(addr->nl_groups);
1345 if ((dst_group || dst_portid) &&
1346 !netlink_capable(sock, NL_CFG_F_NONROOT_SEND))
1349 dst_portid = nlk->dst_portid;
1350 dst_group = nlk->dst_group;
1354 err = netlink_autobind(sock);
1360 if (len > sk->sk_sndbuf - 32)
1363 skb = alloc_skb(len, GFP_KERNEL);
1367 NETLINK_CB(skb).portid = nlk->portid;
1368 NETLINK_CB(skb).dst_group = dst_group;
1369 NETLINK_CB(skb).creds = siocb->scm->creds;
1372 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1377 err = security_netlink_send(sk, skb);
1384 atomic_inc(&skb->users);
1385 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1387 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1390 scm_destroy(siocb->scm);
1394 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1395 struct msghdr *msg, size_t len,
1398 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1399 struct scm_cookie scm;
1400 struct sock *sk = sock->sk;
1401 struct netlink_sock *nlk = nlk_sk(sk);
1402 int noblock = flags&MSG_DONTWAIT;
1404 struct sk_buff *skb, *data_skb;
1412 skb = skb_recv_datagram(sk, flags, noblock, &err);
1418 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1419 if (unlikely(skb_shinfo(skb)->frag_list)) {
1421 * If this skb has a frag_list, then here that means that we
1422 * will have to use the frag_list skb's data for compat tasks
1423 * and the regular skb's data for normal (non-compat) tasks.
1425 * If we need to send the compat skb, assign it to the
1426 * 'data_skb' variable so that it will be used below for data
1427 * copying. We keep 'skb' for everything else, including
1428 * freeing both later.
1430 if (flags & MSG_CMSG_COMPAT)
1431 data_skb = skb_shinfo(skb)->frag_list;
1435 msg->msg_namelen = 0;
1437 copied = data_skb->len;
1439 msg->msg_flags |= MSG_TRUNC;
1443 skb_reset_transport_header(data_skb);
1444 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1446 if (msg->msg_name) {
1447 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1448 addr->nl_family = AF_NETLINK;
1450 addr->nl_pid = NETLINK_CB(skb).portid;
1451 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1452 msg->msg_namelen = sizeof(*addr);
1455 if (nlk->flags & NETLINK_RECV_PKTINFO)
1456 netlink_cmsg_recv_pktinfo(msg, skb);
1458 if (NULL == siocb->scm) {
1459 memset(&scm, 0, sizeof(scm));
1462 siocb->scm->creds = *NETLINK_CREDS(skb);
1463 if (flags & MSG_TRUNC)
1464 copied = data_skb->len;
1466 skb_free_datagram(sk, skb);
1468 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1469 ret = netlink_dump(sk);
1472 sk->sk_error_report(sk);
1476 scm_recv(sock, msg, siocb->scm, flags);
1478 netlink_rcv_wake(sk);
1479 return err ? : copied;
1482 static void netlink_data_ready(struct sock *sk, int len)
1488 * We export these functions to other modules. They provide a
1489 * complete set of kernel non-blocking support for message
1494 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1495 struct netlink_kernel_cfg *cfg)
1497 struct socket *sock;
1499 struct netlink_sock *nlk;
1500 struct listeners *listeners = NULL;
1501 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1502 unsigned int groups;
1506 if (unit < 0 || unit >= MAX_LINKS)
1509 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1513 * We have to just have a reference on the net from sk, but don't
1514 * get_net it. Besides, we cannot get and then put the net here.
1515 * So we create one inside init_net and the move it to net.
1518 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1519 goto out_sock_release_nosk;
1522 sk_change_net(sk, net);
1524 if (!cfg || cfg->groups < 32)
1527 groups = cfg->groups;
1529 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1531 goto out_sock_release;
1533 sk->sk_data_ready = netlink_data_ready;
1534 if (cfg && cfg->input)
1535 nlk_sk(sk)->netlink_rcv = cfg->input;
1537 if (netlink_insert(sk, net, 0))
1538 goto out_sock_release;
1541 nlk->flags |= NETLINK_KERNEL_SOCKET;
1543 netlink_table_grab();
1544 if (!nl_table[unit].registered) {
1545 nl_table[unit].groups = groups;
1546 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1547 nl_table[unit].cb_mutex = cb_mutex;
1548 nl_table[unit].module = module;
1550 nl_table[unit].bind = cfg->bind;
1551 nl_table[unit].flags = cfg->flags;
1553 nl_table[unit].registered = 1;
1556 nl_table[unit].registered++;
1558 netlink_table_ungrab();
1563 netlink_kernel_release(sk);
1566 out_sock_release_nosk:
1570 EXPORT_SYMBOL(__netlink_kernel_create);
1573 netlink_kernel_release(struct sock *sk)
1575 sk_release_kernel(sk);
1577 EXPORT_SYMBOL(netlink_kernel_release);
1579 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1581 struct listeners *new, *old;
1582 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1587 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1588 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1591 old = nl_deref_protected(tbl->listeners);
1592 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1593 rcu_assign_pointer(tbl->listeners, new);
1595 kfree_rcu(old, rcu);
1597 tbl->groups = groups;
1603 * netlink_change_ngroups - change number of multicast groups
1605 * This changes the number of multicast groups that are available
1606 * on a certain netlink family. Note that it is not possible to
1607 * change the number of groups to below 32. Also note that it does
1608 * not implicitly call netlink_clear_multicast_users() when the
1609 * number of groups is reduced.
1611 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1612 * @groups: The new number of groups.
1614 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1618 netlink_table_grab();
1619 err = __netlink_change_ngroups(sk, groups);
1620 netlink_table_ungrab();
1625 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1628 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1630 sk_for_each_bound(sk, &tbl->mc_list)
1631 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1635 * netlink_clear_multicast_users - kick off multicast listeners
1637 * This function removes all listeners from the given group.
1638 * @ksk: The kernel netlink socket, as returned by
1639 * netlink_kernel_create().
1640 * @group: The multicast group to clear.
1642 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1644 netlink_table_grab();
1645 __netlink_clear_multicast_users(ksk, group);
1646 netlink_table_ungrab();
1650 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
1652 struct nlmsghdr *nlh;
1653 int size = nlmsg_msg_size(len);
1655 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
1656 nlh->nlmsg_type = type;
1657 nlh->nlmsg_len = size;
1658 nlh->nlmsg_flags = flags;
1659 nlh->nlmsg_pid = portid;
1660 nlh->nlmsg_seq = seq;
1661 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
1662 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
1665 EXPORT_SYMBOL(__nlmsg_put);
1668 * It looks a bit ugly.
1669 * It would be better to create kernel thread.
1672 static int netlink_dump(struct sock *sk)
1674 struct netlink_sock *nlk = nlk_sk(sk);
1675 struct netlink_callback *cb;
1676 struct sk_buff *skb = NULL;
1677 struct nlmsghdr *nlh;
1678 int len, err = -ENOBUFS;
1681 mutex_lock(nlk->cb_mutex);
1689 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
1691 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
1695 len = cb->dump(skb, cb);
1698 mutex_unlock(nlk->cb_mutex);
1700 if (sk_filter(sk, skb))
1703 __netlink_sendskb(sk, skb);
1707 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1711 nl_dump_check_consistent(cb, nlh);
1713 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1715 if (sk_filter(sk, skb))
1718 __netlink_sendskb(sk, skb);
1723 mutex_unlock(nlk->cb_mutex);
1725 module_put(cb->module);
1726 netlink_consume_callback(cb);
1730 mutex_unlock(nlk->cb_mutex);
1735 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1736 const struct nlmsghdr *nlh,
1737 struct netlink_dump_control *control)
1739 struct netlink_callback *cb;
1741 struct netlink_sock *nlk;
1744 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1748 cb->dump = control->dump;
1749 cb->done = control->done;
1751 cb->data = control->data;
1752 cb->module = control->module;
1753 cb->min_dump_alloc = control->min_dump_alloc;
1754 atomic_inc(&skb->users);
1757 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
1759 netlink_destroy_callback(cb);
1760 return -ECONNREFUSED;
1764 mutex_lock(nlk->cb_mutex);
1765 /* A dump is in progress... */
1767 mutex_unlock(nlk->cb_mutex);
1768 netlink_destroy_callback(cb);
1772 /* add reference of module which cb->dump belongs to */
1773 if (!try_module_get(cb->module)) {
1774 mutex_unlock(nlk->cb_mutex);
1775 netlink_destroy_callback(cb);
1776 ret = -EPROTONOSUPPORT;
1781 mutex_unlock(nlk->cb_mutex);
1783 ret = netlink_dump(sk);
1790 /* We successfully started a dump, by returning -EINTR we
1791 * signal not to send ACK even if it was requested.
1795 EXPORT_SYMBOL(__netlink_dump_start);
1797 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1799 struct sk_buff *skb;
1800 struct nlmsghdr *rep;
1801 struct nlmsgerr *errmsg;
1802 size_t payload = sizeof(*errmsg);
1804 /* error messages get the original request appened */
1806 payload += nlmsg_len(nlh);
1808 skb = nlmsg_new(payload, GFP_KERNEL);
1812 sk = netlink_lookup(sock_net(in_skb->sk),
1813 in_skb->sk->sk_protocol,
1814 NETLINK_CB(in_skb).portid);
1816 sk->sk_err = ENOBUFS;
1817 sk->sk_error_report(sk);
1823 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
1824 NLMSG_ERROR, payload, 0);
1825 errmsg = nlmsg_data(rep);
1826 errmsg->error = err;
1827 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1828 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
1830 EXPORT_SYMBOL(netlink_ack);
1832 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1835 struct nlmsghdr *nlh;
1838 while (skb->len >= nlmsg_total_size(0)) {
1841 nlh = nlmsg_hdr(skb);
1844 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1847 /* Only requests are handled by the kernel */
1848 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1851 /* Skip control messages */
1852 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1860 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1861 netlink_ack(skb, nlh, err);
1864 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1865 if (msglen > skb->len)
1867 skb_pull(skb, msglen);
1872 EXPORT_SYMBOL(netlink_rcv_skb);
1875 * nlmsg_notify - send a notification netlink message
1876 * @sk: netlink socket to use
1877 * @skb: notification message
1878 * @portid: destination netlink portid for reports or 0
1879 * @group: destination multicast group or 0
1880 * @report: 1 to report back, 0 to disable
1881 * @flags: allocation flags
1883 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
1884 unsigned int group, int report, gfp_t flags)
1889 int exclude_portid = 0;
1892 atomic_inc(&skb->users);
1893 exclude_portid = portid;
1896 /* errors reported via destination sk->sk_err, but propagate
1897 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1898 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
1904 err2 = nlmsg_unicast(sk, skb, portid);
1905 if (!err || err == -ESRCH)
1911 EXPORT_SYMBOL(nlmsg_notify);
1913 #ifdef CONFIG_PROC_FS
1914 struct nl_seq_iter {
1915 struct seq_net_private p;
1920 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1922 struct nl_seq_iter *iter = seq->private;
1927 for (i = 0; i < MAX_LINKS; i++) {
1928 struct nl_portid_hash *hash = &nl_table[i].hash;
1930 for (j = 0; j <= hash->mask; j++) {
1931 sk_for_each(s, &hash->table[j]) {
1932 if (sock_net(s) != seq_file_net(seq))
1946 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1947 __acquires(nl_table_lock)
1949 read_lock(&nl_table_lock);
1950 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1953 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1956 struct nl_seq_iter *iter;
1961 if (v == SEQ_START_TOKEN)
1962 return netlink_seq_socket_idx(seq, 0);
1964 iter = seq->private;
1968 } while (s && sock_net(s) != seq_file_net(seq));
1973 j = iter->hash_idx + 1;
1976 struct nl_portid_hash *hash = &nl_table[i].hash;
1978 for (; j <= hash->mask; j++) {
1979 s = sk_head(&hash->table[j]);
1980 while (s && sock_net(s) != seq_file_net(seq))
1990 } while (++i < MAX_LINKS);
1995 static void netlink_seq_stop(struct seq_file *seq, void *v)
1996 __releases(nl_table_lock)
1998 read_unlock(&nl_table_lock);
2002 static int netlink_seq_show(struct seq_file *seq, void *v)
2004 if (v == SEQ_START_TOKEN) {
2006 "sk Eth Pid Groups "
2007 "Rmem Wmem Dump Locks Drops Inode\n");
2010 struct netlink_sock *nlk = nlk_sk(s);
2012 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2016 nlk->groups ? (u32)nlk->groups[0] : 0,
2017 sk_rmem_alloc_get(s),
2018 sk_wmem_alloc_get(s),
2020 atomic_read(&s->sk_refcnt),
2021 atomic_read(&s->sk_drops),
2029 static const struct seq_operations netlink_seq_ops = {
2030 .start = netlink_seq_start,
2031 .next = netlink_seq_next,
2032 .stop = netlink_seq_stop,
2033 .show = netlink_seq_show,
2037 static int netlink_seq_open(struct inode *inode, struct file *file)
2039 return seq_open_net(inode, file, &netlink_seq_ops,
2040 sizeof(struct nl_seq_iter));
2043 static const struct file_operations netlink_seq_fops = {
2044 .owner = THIS_MODULE,
2045 .open = netlink_seq_open,
2047 .llseek = seq_lseek,
2048 .release = seq_release_net,
2053 int netlink_register_notifier(struct notifier_block *nb)
2055 return atomic_notifier_chain_register(&netlink_chain, nb);
2057 EXPORT_SYMBOL(netlink_register_notifier);
2059 int netlink_unregister_notifier(struct notifier_block *nb)
2061 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2063 EXPORT_SYMBOL(netlink_unregister_notifier);
2065 static const struct proto_ops netlink_ops = {
2066 .family = PF_NETLINK,
2067 .owner = THIS_MODULE,
2068 .release = netlink_release,
2069 .bind = netlink_bind,
2070 .connect = netlink_connect,
2071 .socketpair = sock_no_socketpair,
2072 .accept = sock_no_accept,
2073 .getname = netlink_getname,
2074 .poll = datagram_poll,
2075 .ioctl = sock_no_ioctl,
2076 .listen = sock_no_listen,
2077 .shutdown = sock_no_shutdown,
2078 .setsockopt = netlink_setsockopt,
2079 .getsockopt = netlink_getsockopt,
2080 .sendmsg = netlink_sendmsg,
2081 .recvmsg = netlink_recvmsg,
2082 .mmap = sock_no_mmap,
2083 .sendpage = sock_no_sendpage,
2086 static const struct net_proto_family netlink_family_ops = {
2087 .family = PF_NETLINK,
2088 .create = netlink_create,
2089 .owner = THIS_MODULE, /* for consistency 8) */
2092 static int __net_init netlink_net_init(struct net *net)
2094 #ifdef CONFIG_PROC_FS
2095 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2101 static void __net_exit netlink_net_exit(struct net *net)
2103 #ifdef CONFIG_PROC_FS
2104 remove_proc_entry("netlink", net->proc_net);
2108 static void __init netlink_add_usersock_entry(void)
2110 struct listeners *listeners;
2113 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2115 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2117 netlink_table_grab();
2119 nl_table[NETLINK_USERSOCK].groups = groups;
2120 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2121 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2122 nl_table[NETLINK_USERSOCK].registered = 1;
2123 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2125 netlink_table_ungrab();
2128 static struct pernet_operations __net_initdata netlink_net_ops = {
2129 .init = netlink_net_init,
2130 .exit = netlink_net_exit,
2133 static int __init netlink_proto_init(void)
2136 unsigned long limit;
2138 int err = proto_register(&netlink_proto, 0);
2143 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2145 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2149 if (totalram_pages >= (128 * 1024))
2150 limit = totalram_pages >> (21 - PAGE_SHIFT);
2152 limit = totalram_pages >> (23 - PAGE_SHIFT);
2154 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2155 limit = (1UL << order) / sizeof(struct hlist_head);
2156 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2158 for (i = 0; i < MAX_LINKS; i++) {
2159 struct nl_portid_hash *hash = &nl_table[i].hash;
2161 hash->table = nl_portid_hash_zalloc(1 * sizeof(*hash->table));
2164 nl_portid_hash_free(nl_table[i].hash.table,
2165 1 * sizeof(*hash->table));
2169 hash->max_shift = order;
2172 hash->rehash_time = jiffies;
2175 netlink_add_usersock_entry();
2177 sock_register(&netlink_family_ops);
2178 register_pernet_subsys(&netlink_net_ops);
2179 /* The netlink device handler may be needed early. */
2184 panic("netlink_init: Cannot allocate nl_table\n");
2187 core_initcall(netlink_proto_init);