2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 * Patrick McHardy <kaber@trash.net>
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
13 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
14 * added netlink_proto_exit
15 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
16 * use nlk_sk, as sk->protinfo is on a diet 8)
17 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
18 * - inc module use count of module that owns
19 * the kernel socket in case userspace opens
20 * socket of same protocol
21 * - remove all module support, since netlink is
22 * mandatory if CONFIG_NET=y these days
25 #include <linux/module.h>
27 #include <linux/capability.h>
28 #include <linux/kernel.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/sched.h>
32 #include <linux/errno.h>
33 #include <linux/string.h>
34 #include <linux/stat.h>
35 #include <linux/socket.h>
37 #include <linux/fcntl.h>
38 #include <linux/termios.h>
39 #include <linux/sockios.h>
40 #include <linux/net.h>
42 #include <linux/slab.h>
43 #include <linux/uaccess.h>
44 #include <linux/skbuff.h>
45 #include <linux/netdevice.h>
46 #include <linux/rtnetlink.h>
47 #include <linux/proc_fs.h>
48 #include <linux/seq_file.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/mutex.h>
59 #include <linux/vmalloc.h>
60 #include <linux/if_arp.h>
61 #include <linux/rhashtable.h>
62 #include <asm/cacheflush.h>
63 #include <linux/hash.h>
64 #include <linux/genetlink.h>
65 #include <linux/net_namespace.h>
67 #include <net/net_namespace.h>
70 #include <net/netlink.h>
72 #include "af_netlink.h"
76 unsigned long masks[0];
80 #define NETLINK_S_CONGESTED 0x0
82 static inline int netlink_is_kernel(struct sock *sk)
84 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
87 struct netlink_table *nl_table __read_mostly;
88 EXPORT_SYMBOL_GPL(nl_table);
90 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
92 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
94 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
97 "nlk_cb_mutex-USERSOCK",
98 "nlk_cb_mutex-FIREWALL",
99 "nlk_cb_mutex-SOCK_DIAG",
100 "nlk_cb_mutex-NFLOG",
102 "nlk_cb_mutex-SELINUX",
103 "nlk_cb_mutex-ISCSI",
104 "nlk_cb_mutex-AUDIT",
105 "nlk_cb_mutex-FIB_LOOKUP",
106 "nlk_cb_mutex-CONNECTOR",
107 "nlk_cb_mutex-NETFILTER",
108 "nlk_cb_mutex-IP6_FW",
109 "nlk_cb_mutex-DNRTMSG",
110 "nlk_cb_mutex-KOBJECT_UEVENT",
111 "nlk_cb_mutex-GENERIC",
113 "nlk_cb_mutex-SCSITRANSPORT",
114 "nlk_cb_mutex-ECRYPTFS",
116 "nlk_cb_mutex-CRYPTO",
127 "nlk_cb_mutex-MAX_LINKS"
130 static int netlink_dump(struct sock *sk);
132 /* nl_table locking explained:
133 * Lookup and traversal are protected with an RCU read-side lock. Insertion
134 * and removal are protected with per bucket lock while using RCU list
135 * modification primitives and may run in parallel to RCU protected lookups.
136 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
137 * been acquired * either during or after the socket has been removed from
138 * the list and after an RCU grace period.
140 DEFINE_RWLOCK(nl_table_lock);
141 EXPORT_SYMBOL_GPL(nl_table_lock);
142 static atomic_t nl_table_users = ATOMIC_INIT(0);
144 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
146 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
148 static DEFINE_SPINLOCK(netlink_tap_lock);
149 static struct list_head netlink_tap_all __read_mostly;
151 static const struct rhashtable_params netlink_rhashtable_params;
153 static inline u32 netlink_group_mask(u32 group)
155 return group ? 1 << (group - 1) : 0;
158 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
161 unsigned int len = skb_end_offset(skb);
164 new = alloc_skb(len, gfp_mask);
168 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
169 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
170 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
172 skb_put_data(new, skb->data, len);
176 int netlink_add_tap(struct netlink_tap *nt)
178 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
181 spin_lock(&netlink_tap_lock);
182 list_add_rcu(&nt->list, &netlink_tap_all);
183 spin_unlock(&netlink_tap_lock);
185 __module_get(nt->module);
189 EXPORT_SYMBOL_GPL(netlink_add_tap);
191 static int __netlink_remove_tap(struct netlink_tap *nt)
194 struct netlink_tap *tmp;
196 spin_lock(&netlink_tap_lock);
198 list_for_each_entry(tmp, &netlink_tap_all, list) {
200 list_del_rcu(&nt->list);
206 pr_warn("__netlink_remove_tap: %p not found\n", nt);
208 spin_unlock(&netlink_tap_lock);
211 module_put(nt->module);
213 return found ? 0 : -ENODEV;
216 int netlink_remove_tap(struct netlink_tap *nt)
220 ret = __netlink_remove_tap(nt);
225 EXPORT_SYMBOL_GPL(netlink_remove_tap);
227 static bool netlink_filter_tap(const struct sk_buff *skb)
229 struct sock *sk = skb->sk;
231 /* We take the more conservative approach and
232 * whitelist socket protocols that may pass.
234 switch (sk->sk_protocol) {
236 case NETLINK_USERSOCK:
237 case NETLINK_SOCK_DIAG:
240 case NETLINK_FIB_LOOKUP:
241 case NETLINK_NETFILTER:
242 case NETLINK_GENERIC:
249 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
250 struct net_device *dev)
252 struct sk_buff *nskb;
253 struct sock *sk = skb->sk;
258 if (is_vmalloc_addr(skb->head))
259 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
261 nskb = skb_clone(skb, GFP_ATOMIC);
264 nskb->protocol = htons((u16) sk->sk_protocol);
265 nskb->pkt_type = netlink_is_kernel(sk) ?
266 PACKET_KERNEL : PACKET_USER;
267 skb_reset_network_header(nskb);
268 ret = dev_queue_xmit(nskb);
269 if (unlikely(ret > 0))
270 ret = net_xmit_errno(ret);
277 static void __netlink_deliver_tap(struct sk_buff *skb)
280 struct netlink_tap *tmp;
282 if (!netlink_filter_tap(skb))
285 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
286 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
292 static void netlink_deliver_tap(struct sk_buff *skb)
296 if (unlikely(!list_empty(&netlink_tap_all)))
297 __netlink_deliver_tap(skb);
302 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
305 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
306 netlink_deliver_tap(skb);
309 static void netlink_overrun(struct sock *sk)
311 struct netlink_sock *nlk = nlk_sk(sk);
313 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
314 if (!test_and_set_bit(NETLINK_S_CONGESTED,
315 &nlk_sk(sk)->state)) {
316 sk->sk_err = ENOBUFS;
317 sk->sk_error_report(sk);
320 atomic_inc(&sk->sk_drops);
323 static void netlink_rcv_wake(struct sock *sk)
325 struct netlink_sock *nlk = nlk_sk(sk);
327 if (skb_queue_empty(&sk->sk_receive_queue))
328 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
329 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
330 wake_up_interruptible(&nlk->wait);
333 static void netlink_skb_destructor(struct sk_buff *skb)
335 if (is_vmalloc_addr(skb->head)) {
337 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
346 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
348 WARN_ON(skb->sk != NULL);
350 skb->destructor = netlink_skb_destructor;
351 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
352 sk_mem_charge(sk, skb->truesize);
355 static void netlink_sock_destruct(struct sock *sk)
357 struct netlink_sock *nlk = nlk_sk(sk);
359 if (nlk->cb_running) {
361 nlk->cb.done(&nlk->cb);
362 module_put(nlk->cb.module);
363 kfree_skb(nlk->cb.skb);
366 skb_queue_purge(&sk->sk_receive_queue);
368 if (!sock_flag(sk, SOCK_DEAD)) {
369 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
373 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
374 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
375 WARN_ON(nlk_sk(sk)->groups);
378 static void netlink_sock_destruct_work(struct work_struct *work)
380 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
386 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
387 * SMP. Look, when several writers sleep and reader wakes them up, all but one
388 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
389 * this, _but_ remember, it adds useless work on UP machines.
392 void netlink_table_grab(void)
393 __acquires(nl_table_lock)
397 write_lock_irq(&nl_table_lock);
399 if (atomic_read(&nl_table_users)) {
400 DECLARE_WAITQUEUE(wait, current);
402 add_wait_queue_exclusive(&nl_table_wait, &wait);
404 set_current_state(TASK_UNINTERRUPTIBLE);
405 if (atomic_read(&nl_table_users) == 0)
407 write_unlock_irq(&nl_table_lock);
409 write_lock_irq(&nl_table_lock);
412 __set_current_state(TASK_RUNNING);
413 remove_wait_queue(&nl_table_wait, &wait);
417 void netlink_table_ungrab(void)
418 __releases(nl_table_lock)
420 write_unlock_irq(&nl_table_lock);
421 wake_up(&nl_table_wait);
425 netlink_lock_table(void)
427 /* read_lock() synchronizes us to netlink_table_grab */
429 read_lock(&nl_table_lock);
430 atomic_inc(&nl_table_users);
431 read_unlock(&nl_table_lock);
435 netlink_unlock_table(void)
437 if (atomic_dec_and_test(&nl_table_users))
438 wake_up(&nl_table_wait);
441 struct netlink_compare_arg
447 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
448 #define netlink_compare_arg_len \
449 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
451 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
454 const struct netlink_compare_arg *x = arg->key;
455 const struct netlink_sock *nlk = ptr;
457 return nlk->portid != x->portid ||
458 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
461 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
462 struct net *net, u32 portid)
464 memset(arg, 0, sizeof(*arg));
465 write_pnet(&arg->pnet, net);
466 arg->portid = portid;
469 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
472 struct netlink_compare_arg arg;
474 netlink_compare_arg_init(&arg, net, portid);
475 return rhashtable_lookup_fast(&table->hash, &arg,
476 netlink_rhashtable_params);
479 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
481 struct netlink_compare_arg arg;
483 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
484 return rhashtable_lookup_insert_key(&table->hash, &arg,
486 netlink_rhashtable_params);
489 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
491 struct netlink_table *table = &nl_table[protocol];
495 sk = __netlink_lookup(table, portid, net);
503 static const struct proto_ops netlink_ops;
506 netlink_update_listeners(struct sock *sk)
508 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
511 struct listeners *listeners;
513 listeners = nl_deref_protected(tbl->listeners);
517 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
519 sk_for_each_bound(sk, &tbl->mc_list) {
520 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
521 mask |= nlk_sk(sk)->groups[i];
523 listeners->masks[i] = mask;
525 /* this function is only called with the netlink table "grabbed", which
526 * makes sure updates are visible before bind or setsockopt return. */
529 static int netlink_insert(struct sock *sk, u32 portid)
531 struct netlink_table *table = &nl_table[sk->sk_protocol];
536 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
537 if (nlk_sk(sk)->bound)
541 if (BITS_PER_LONG > 32 &&
542 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
545 nlk_sk(sk)->portid = portid;
548 err = __netlink_insert(table, sk);
550 /* In case the hashtable backend returns with -EBUSY
551 * from here, it must not escape to the caller.
553 if (unlikely(err == -EBUSY))
561 /* We need to ensure that the socket is hashed and visible. */
563 nlk_sk(sk)->bound = portid;
570 static void netlink_remove(struct sock *sk)
572 struct netlink_table *table;
574 table = &nl_table[sk->sk_protocol];
575 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
576 netlink_rhashtable_params)) {
577 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
581 netlink_table_grab();
582 if (nlk_sk(sk)->subscriptions) {
583 __sk_del_bind_node(sk);
584 netlink_update_listeners(sk);
586 if (sk->sk_protocol == NETLINK_GENERIC)
587 atomic_inc(&genl_sk_destructing_cnt);
588 netlink_table_ungrab();
591 static struct proto netlink_proto = {
593 .owner = THIS_MODULE,
594 .obj_size = sizeof(struct netlink_sock),
597 static int __netlink_create(struct net *net, struct socket *sock,
598 struct mutex *cb_mutex, int protocol,
602 struct netlink_sock *nlk;
604 sock->ops = &netlink_ops;
606 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
610 sock_init_data(sock, sk);
614 nlk->cb_mutex = cb_mutex;
616 nlk->cb_mutex = &nlk->cb_def_mutex;
617 mutex_init(nlk->cb_mutex);
618 lockdep_set_class_and_name(nlk->cb_mutex,
619 nlk_cb_mutex_keys + protocol,
620 nlk_cb_mutex_key_strings[protocol]);
622 init_waitqueue_head(&nlk->wait);
624 sk->sk_destruct = netlink_sock_destruct;
625 sk->sk_protocol = protocol;
629 static int netlink_create(struct net *net, struct socket *sock, int protocol,
632 struct module *module = NULL;
633 struct mutex *cb_mutex;
634 struct netlink_sock *nlk;
635 int (*bind)(struct net *net, int group);
636 void (*unbind)(struct net *net, int group);
639 sock->state = SS_UNCONNECTED;
641 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
642 return -ESOCKTNOSUPPORT;
644 if (protocol < 0 || protocol >= MAX_LINKS)
645 return -EPROTONOSUPPORT;
647 netlink_lock_table();
648 #ifdef CONFIG_MODULES
649 if (!nl_table[protocol].registered) {
650 netlink_unlock_table();
651 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
652 netlink_lock_table();
655 if (nl_table[protocol].registered &&
656 try_module_get(nl_table[protocol].module))
657 module = nl_table[protocol].module;
659 err = -EPROTONOSUPPORT;
660 cb_mutex = nl_table[protocol].cb_mutex;
661 bind = nl_table[protocol].bind;
662 unbind = nl_table[protocol].unbind;
663 netlink_unlock_table();
668 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
673 sock_prot_inuse_add(net, &netlink_proto, 1);
676 nlk = nlk_sk(sock->sk);
677 nlk->module = module;
678 nlk->netlink_bind = bind;
679 nlk->netlink_unbind = unbind;
688 static void deferred_put_nlk_sk(struct rcu_head *head)
690 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
691 struct sock *sk = &nlk->sk;
696 if (!refcount_dec_and_test(&sk->sk_refcnt))
699 if (nlk->cb_running && nlk->cb.done) {
700 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
701 schedule_work(&nlk->work);
708 static int netlink_release(struct socket *sock)
710 struct sock *sk = sock->sk;
711 struct netlink_sock *nlk;
721 * OK. Socket is unlinked, any packets that arrive now
725 /* must not acquire netlink_table_lock in any way again before unbind
726 * and notifying genetlink is done as otherwise it might deadlock
728 if (nlk->netlink_unbind) {
731 for (i = 0; i < nlk->ngroups; i++)
732 if (test_bit(i, nlk->groups))
733 nlk->netlink_unbind(sock_net(sk), i + 1);
735 if (sk->sk_protocol == NETLINK_GENERIC &&
736 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
737 wake_up(&genl_sk_destructing_waitq);
740 wake_up_interruptible_all(&nlk->wait);
742 skb_queue_purge(&sk->sk_write_queue);
744 if (nlk->portid && nlk->bound) {
745 struct netlink_notify n = {
747 .protocol = sk->sk_protocol,
748 .portid = nlk->portid,
750 blocking_notifier_call_chain(&netlink_chain,
751 NETLINK_URELEASE, &n);
754 module_put(nlk->module);
756 if (netlink_is_kernel(sk)) {
757 netlink_table_grab();
758 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
759 if (--nl_table[sk->sk_protocol].registered == 0) {
760 struct listeners *old;
762 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
763 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
765 nl_table[sk->sk_protocol].module = NULL;
766 nl_table[sk->sk_protocol].bind = NULL;
767 nl_table[sk->sk_protocol].unbind = NULL;
768 nl_table[sk->sk_protocol].flags = 0;
769 nl_table[sk->sk_protocol].registered = 0;
771 netlink_table_ungrab();
775 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
777 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
781 static int netlink_autobind(struct socket *sock)
783 struct sock *sk = sock->sk;
784 struct net *net = sock_net(sk);
785 struct netlink_table *table = &nl_table[sk->sk_protocol];
786 s32 portid = task_tgid_vnr(current);
794 ok = !__netlink_lookup(table, portid, net);
797 /* Bind collision, search negative portid values. */
799 /* rover will be in range [S32_MIN, -4097] */
800 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
801 else if (rover >= -4096)
807 err = netlink_insert(sk, portid);
808 if (err == -EADDRINUSE)
811 /* If 2 threads race to autobind, that is fine. */
819 * __netlink_ns_capable - General netlink message capability test
820 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
821 * @user_ns: The user namespace of the capability to use
822 * @cap: The capability to use
824 * Test to see if the opener of the socket we received the message
825 * from had when the netlink socket was created and the sender of the
826 * message has has the capability @cap in the user namespace @user_ns.
828 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
829 struct user_namespace *user_ns, int cap)
831 return ((nsp->flags & NETLINK_SKB_DST) ||
832 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
833 ns_capable(user_ns, cap);
835 EXPORT_SYMBOL(__netlink_ns_capable);
838 * netlink_ns_capable - General netlink message capability test
839 * @skb: socket buffer holding a netlink command from userspace
840 * @user_ns: The user namespace of the capability to use
841 * @cap: The capability to use
843 * Test to see if the opener of the socket we received the message
844 * from had when the netlink socket was created and the sender of the
845 * message has has the capability @cap in the user namespace @user_ns.
847 bool netlink_ns_capable(const struct sk_buff *skb,
848 struct user_namespace *user_ns, int cap)
850 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
852 EXPORT_SYMBOL(netlink_ns_capable);
855 * netlink_capable - Netlink global message capability test
856 * @skb: socket buffer holding a netlink command from userspace
857 * @cap: The capability to use
859 * Test to see if the opener of the socket we received the message
860 * from had when the netlink socket was created and the sender of the
861 * message has has the capability @cap in all user namespaces.
863 bool netlink_capable(const struct sk_buff *skb, int cap)
865 return netlink_ns_capable(skb, &init_user_ns, cap);
867 EXPORT_SYMBOL(netlink_capable);
870 * netlink_net_capable - Netlink network namespace message capability test
871 * @skb: socket buffer holding a netlink command from userspace
872 * @cap: The capability to use
874 * Test to see if the opener of the socket we received the message
875 * from had when the netlink socket was created and the sender of the
876 * message has has the capability @cap over the network namespace of
877 * the socket we received the message from.
879 bool netlink_net_capable(const struct sk_buff *skb, int cap)
881 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
883 EXPORT_SYMBOL(netlink_net_capable);
885 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
887 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
888 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
892 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
894 struct netlink_sock *nlk = nlk_sk(sk);
896 if (nlk->subscriptions && !subscriptions)
897 __sk_del_bind_node(sk);
898 else if (!nlk->subscriptions && subscriptions)
899 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
900 nlk->subscriptions = subscriptions;
903 static int netlink_realloc_groups(struct sock *sk)
905 struct netlink_sock *nlk = nlk_sk(sk);
907 unsigned long *new_groups;
910 netlink_table_grab();
912 groups = nl_table[sk->sk_protocol].groups;
913 if (!nl_table[sk->sk_protocol].registered) {
918 if (nlk->ngroups >= groups)
921 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
922 if (new_groups == NULL) {
926 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
927 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
929 nlk->groups = new_groups;
930 nlk->ngroups = groups;
932 netlink_table_ungrab();
936 static void netlink_undo_bind(int group, long unsigned int groups,
939 struct netlink_sock *nlk = nlk_sk(sk);
942 if (!nlk->netlink_unbind)
945 for (undo = 0; undo < group; undo++)
946 if (test_bit(undo, &groups))
947 nlk->netlink_unbind(sock_net(sk), undo + 1);
950 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
953 struct sock *sk = sock->sk;
954 struct net *net = sock_net(sk);
955 struct netlink_sock *nlk = nlk_sk(sk);
956 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
958 long unsigned int groups = nladdr->nl_groups;
961 if (addr_len < sizeof(struct sockaddr_nl))
964 if (nladdr->nl_family != AF_NETLINK)
967 /* Only superuser is allowed to listen multicasts */
969 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
971 err = netlink_realloc_groups(sk);
978 /* Ensure nlk->portid is up-to-date. */
981 if (nladdr->nl_pid != nlk->portid)
985 netlink_lock_table();
986 if (nlk->netlink_bind && groups) {
989 for (group = 0; group < nlk->ngroups; group++) {
990 if (!test_bit(group, &groups))
992 err = nlk->netlink_bind(net, group + 1);
995 netlink_undo_bind(group, groups, sk);
1000 /* No need for barriers here as we return to user-space without
1001 * using any of the bound attributes.
1004 err = nladdr->nl_pid ?
1005 netlink_insert(sk, nladdr->nl_pid) :
1006 netlink_autobind(sock);
1008 netlink_undo_bind(nlk->ngroups, groups, sk);
1013 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1015 netlink_unlock_table();
1017 netlink_table_grab();
1018 netlink_update_subscriptions(sk, nlk->subscriptions +
1020 hweight32(nlk->groups[0]));
1021 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1022 netlink_update_listeners(sk);
1023 netlink_table_ungrab();
1028 netlink_unlock_table();
1032 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1033 int alen, int flags)
1036 struct sock *sk = sock->sk;
1037 struct netlink_sock *nlk = nlk_sk(sk);
1038 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1040 if (alen < sizeof(addr->sa_family))
1043 if (addr->sa_family == AF_UNSPEC) {
1044 sk->sk_state = NETLINK_UNCONNECTED;
1045 nlk->dst_portid = 0;
1049 if (addr->sa_family != AF_NETLINK)
1052 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1053 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1056 /* No need for barriers here as we return to user-space without
1057 * using any of the bound attributes.
1060 err = netlink_autobind(sock);
1063 sk->sk_state = NETLINK_CONNECTED;
1064 nlk->dst_portid = nladdr->nl_pid;
1065 nlk->dst_group = ffs(nladdr->nl_groups);
1071 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1072 int *addr_len, int peer)
1074 struct sock *sk = sock->sk;
1075 struct netlink_sock *nlk = nlk_sk(sk);
1076 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1078 nladdr->nl_family = AF_NETLINK;
1080 *addr_len = sizeof(*nladdr);
1083 nladdr->nl_pid = nlk->dst_portid;
1084 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1086 nladdr->nl_pid = nlk->portid;
1087 netlink_lock_table();
1088 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1089 netlink_unlock_table();
1094 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1097 /* try to hand this ioctl down to the NIC drivers.
1099 return -ENOIOCTLCMD;
1102 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1105 struct netlink_sock *nlk;
1107 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1109 return ERR_PTR(-ECONNREFUSED);
1111 /* Don't bother queuing skb if kernel socket has no input function */
1113 if (sock->sk_state == NETLINK_CONNECTED &&
1114 nlk->dst_portid != nlk_sk(ssk)->portid) {
1116 return ERR_PTR(-ECONNREFUSED);
1121 struct sock *netlink_getsockbyfilp(struct file *filp)
1123 struct inode *inode = file_inode(filp);
1126 if (!S_ISSOCK(inode->i_mode))
1127 return ERR_PTR(-ENOTSOCK);
1129 sock = SOCKET_I(inode)->sk;
1130 if (sock->sk_family != AF_NETLINK)
1131 return ERR_PTR(-EINVAL);
1137 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1140 struct sk_buff *skb;
1143 if (size <= NLMSG_GOODSIZE || broadcast)
1144 return alloc_skb(size, GFP_KERNEL);
1146 size = SKB_DATA_ALIGN(size) +
1147 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1149 data = vmalloc(size);
1153 skb = __build_skb(data, size);
1157 skb->destructor = netlink_skb_destructor;
1163 * Attach a skb to a netlink socket.
1164 * The caller must hold a reference to the destination socket. On error, the
1165 * reference is dropped. The skb is not send to the destination, just all
1166 * all error checks are performed and memory in the queue is reserved.
1168 * < 0: error. skb freed, reference to sock dropped.
1170 * 1: repeat lookup - reference dropped while waiting for socket memory.
1172 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1173 long *timeo, struct sock *ssk)
1175 struct netlink_sock *nlk;
1179 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1180 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1181 DECLARE_WAITQUEUE(wait, current);
1183 if (!ssk || netlink_is_kernel(ssk))
1184 netlink_overrun(sk);
1190 __set_current_state(TASK_INTERRUPTIBLE);
1191 add_wait_queue(&nlk->wait, &wait);
1193 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1194 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1195 !sock_flag(sk, SOCK_DEAD))
1196 *timeo = schedule_timeout(*timeo);
1198 __set_current_state(TASK_RUNNING);
1199 remove_wait_queue(&nlk->wait, &wait);
1202 if (signal_pending(current)) {
1204 return sock_intr_errno(*timeo);
1208 netlink_skb_set_owner_r(skb, sk);
1212 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1216 netlink_deliver_tap(skb);
1218 skb_queue_tail(&sk->sk_receive_queue, skb);
1219 sk->sk_data_ready(sk);
1223 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1225 int len = __netlink_sendskb(sk, skb);
1231 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1237 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1241 WARN_ON(skb->sk != NULL);
1242 delta = skb->end - skb->tail;
1243 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1246 if (skb_shared(skb)) {
1247 struct sk_buff *nskb = skb_clone(skb, allocation);
1254 pskb_expand_head(skb, 0, -delta,
1255 (allocation & ~__GFP_DIRECT_RECLAIM) |
1256 __GFP_NOWARN | __GFP_NORETRY);
1260 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1264 struct netlink_sock *nlk = nlk_sk(sk);
1266 ret = -ECONNREFUSED;
1267 if (nlk->netlink_rcv != NULL) {
1269 netlink_skb_set_owner_r(skb, sk);
1270 NETLINK_CB(skb).sk = ssk;
1271 netlink_deliver_tap_kernel(sk, ssk, skb);
1272 nlk->netlink_rcv(skb);
1281 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1282 u32 portid, int nonblock)
1288 skb = netlink_trim(skb, gfp_any());
1290 timeo = sock_sndtimeo(ssk, nonblock);
1292 sk = netlink_getsockbyportid(ssk, portid);
1297 if (netlink_is_kernel(sk))
1298 return netlink_unicast_kernel(sk, skb, ssk);
1300 if (sk_filter(sk, skb)) {
1307 err = netlink_attachskb(sk, skb, &timeo, ssk);
1313 return netlink_sendskb(sk, skb);
1315 EXPORT_SYMBOL(netlink_unicast);
1317 int netlink_has_listeners(struct sock *sk, unsigned int group)
1320 struct listeners *listeners;
1322 BUG_ON(!netlink_is_kernel(sk));
1325 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1327 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1328 res = test_bit(group - 1, listeners->masks);
1334 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1336 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1338 struct netlink_sock *nlk = nlk_sk(sk);
1340 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1341 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1342 netlink_skb_set_owner_r(skb, sk);
1343 __netlink_sendskb(sk, skb);
1344 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1349 struct netlink_broadcast_data {
1350 struct sock *exclude_sk;
1355 int delivery_failure;
1359 struct sk_buff *skb, *skb2;
1360 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1364 static void do_one_broadcast(struct sock *sk,
1365 struct netlink_broadcast_data *p)
1367 struct netlink_sock *nlk = nlk_sk(sk);
1370 if (p->exclude_sk == sk)
1373 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1374 !test_bit(p->group - 1, nlk->groups))
1377 if (!net_eq(sock_net(sk), p->net)) {
1378 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1381 if (!peernet_has_id(sock_net(sk), p->net))
1384 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1390 netlink_overrun(sk);
1395 if (p->skb2 == NULL) {
1396 if (skb_shared(p->skb)) {
1397 p->skb2 = skb_clone(p->skb, p->allocation);
1399 p->skb2 = skb_get(p->skb);
1401 * skb ownership may have been set when
1402 * delivered to a previous socket.
1404 skb_orphan(p->skb2);
1407 if (p->skb2 == NULL) {
1408 netlink_overrun(sk);
1409 /* Clone failed. Notify ALL listeners. */
1411 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1412 p->delivery_failure = 1;
1415 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1420 if (sk_filter(sk, p->skb2)) {
1425 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1426 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1427 NETLINK_CB(p->skb2).nsid_is_set = true;
1428 val = netlink_broadcast_deliver(sk, p->skb2);
1430 netlink_overrun(sk);
1431 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1432 p->delivery_failure = 1;
1434 p->congested |= val;
1442 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1443 u32 group, gfp_t allocation,
1444 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1447 struct net *net = sock_net(ssk);
1448 struct netlink_broadcast_data info;
1451 skb = netlink_trim(skb, allocation);
1453 info.exclude_sk = ssk;
1455 info.portid = portid;
1458 info.delivery_failure = 0;
1461 info.allocation = allocation;
1464 info.tx_filter = filter;
1465 info.tx_data = filter_data;
1467 /* While we sleep in clone, do not allow to change socket list */
1469 netlink_lock_table();
1471 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1472 do_one_broadcast(sk, &info);
1476 netlink_unlock_table();
1478 if (info.delivery_failure) {
1479 kfree_skb(info.skb2);
1482 consume_skb(info.skb2);
1484 if (info.delivered) {
1485 if (info.congested && gfpflags_allow_blocking(allocation))
1491 EXPORT_SYMBOL(netlink_broadcast_filtered);
1493 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1494 u32 group, gfp_t allocation)
1496 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1499 EXPORT_SYMBOL(netlink_broadcast);
1501 struct netlink_set_err_data {
1502 struct sock *exclude_sk;
1508 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1510 struct netlink_sock *nlk = nlk_sk(sk);
1513 if (sk == p->exclude_sk)
1516 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1519 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1520 !test_bit(p->group - 1, nlk->groups))
1523 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1528 sk->sk_err = p->code;
1529 sk->sk_error_report(sk);
1535 * netlink_set_err - report error to broadcast listeners
1536 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1537 * @portid: the PORTID of a process that we want to skip (if any)
1538 * @group: the broadcast group that will notice the error
1539 * @code: error code, must be negative (as usual in kernelspace)
1541 * This function returns the number of broadcast listeners that have set the
1542 * NETLINK_NO_ENOBUFS socket option.
1544 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1546 struct netlink_set_err_data info;
1550 info.exclude_sk = ssk;
1551 info.portid = portid;
1553 /* sk->sk_err wants a positive error value */
1556 read_lock(&nl_table_lock);
1558 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1559 ret += do_one_set_err(sk, &info);
1561 read_unlock(&nl_table_lock);
1564 EXPORT_SYMBOL(netlink_set_err);
1566 /* must be called with netlink table grabbed */
1567 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1571 int old, new = !!is_new, subscriptions;
1573 old = test_bit(group - 1, nlk->groups);
1574 subscriptions = nlk->subscriptions - old + new;
1576 __set_bit(group - 1, nlk->groups);
1578 __clear_bit(group - 1, nlk->groups);
1579 netlink_update_subscriptions(&nlk->sk, subscriptions);
1580 netlink_update_listeners(&nlk->sk);
1583 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1584 char __user *optval, unsigned int optlen)
1586 struct sock *sk = sock->sk;
1587 struct netlink_sock *nlk = nlk_sk(sk);
1588 unsigned int val = 0;
1591 if (level != SOL_NETLINK)
1592 return -ENOPROTOOPT;
1594 if (optlen >= sizeof(int) &&
1595 get_user(val, (unsigned int __user *)optval))
1599 case NETLINK_PKTINFO:
1601 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1603 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1606 case NETLINK_ADD_MEMBERSHIP:
1607 case NETLINK_DROP_MEMBERSHIP: {
1608 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1610 err = netlink_realloc_groups(sk);
1613 if (!val || val - 1 >= nlk->ngroups)
1615 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1616 err = nlk->netlink_bind(sock_net(sk), val);
1620 netlink_table_grab();
1621 netlink_update_socket_mc(nlk, val,
1622 optname == NETLINK_ADD_MEMBERSHIP);
1623 netlink_table_ungrab();
1624 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1625 nlk->netlink_unbind(sock_net(sk), val);
1630 case NETLINK_BROADCAST_ERROR:
1632 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1634 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1637 case NETLINK_NO_ENOBUFS:
1639 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1640 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1641 wake_up_interruptible(&nlk->wait);
1643 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1647 case NETLINK_LISTEN_ALL_NSID:
1648 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1652 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1654 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1657 case NETLINK_CAP_ACK:
1659 nlk->flags |= NETLINK_F_CAP_ACK;
1661 nlk->flags &= ~NETLINK_F_CAP_ACK;
1664 case NETLINK_EXT_ACK:
1666 nlk->flags |= NETLINK_F_EXT_ACK;
1668 nlk->flags &= ~NETLINK_F_EXT_ACK;
1677 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1678 char __user *optval, int __user *optlen)
1680 struct sock *sk = sock->sk;
1681 struct netlink_sock *nlk = nlk_sk(sk);
1684 if (level != SOL_NETLINK)
1685 return -ENOPROTOOPT;
1687 if (get_user(len, optlen))
1693 case NETLINK_PKTINFO:
1694 if (len < sizeof(int))
1697 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1698 if (put_user(len, optlen) ||
1699 put_user(val, optval))
1703 case NETLINK_BROADCAST_ERROR:
1704 if (len < sizeof(int))
1707 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1708 if (put_user(len, optlen) ||
1709 put_user(val, optval))
1713 case NETLINK_NO_ENOBUFS:
1714 if (len < sizeof(int))
1717 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1718 if (put_user(len, optlen) ||
1719 put_user(val, optval))
1723 case NETLINK_LIST_MEMBERSHIPS: {
1724 int pos, idx, shift;
1727 netlink_lock_table();
1728 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1729 if (len - pos < sizeof(u32))
1732 idx = pos / sizeof(unsigned long);
1733 shift = (pos % sizeof(unsigned long)) * 8;
1734 if (put_user((u32)(nlk->groups[idx] >> shift),
1735 (u32 __user *)(optval + pos))) {
1740 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1742 netlink_unlock_table();
1745 case NETLINK_CAP_ACK:
1746 if (len < sizeof(int))
1749 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1750 if (put_user(len, optlen) ||
1751 put_user(val, optval))
1755 case NETLINK_EXT_ACK:
1756 if (len < sizeof(int))
1759 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1760 if (put_user(len, optlen) || put_user(val, optval))
1770 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1772 struct nl_pktinfo info;
1774 info.group = NETLINK_CB(skb).dst_group;
1775 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1778 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1779 struct sk_buff *skb)
1781 if (!NETLINK_CB(skb).nsid_is_set)
1784 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1785 &NETLINK_CB(skb).nsid);
1788 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1790 struct sock *sk = sock->sk;
1791 struct netlink_sock *nlk = nlk_sk(sk);
1792 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1795 struct sk_buff *skb;
1797 struct scm_cookie scm;
1798 u32 netlink_skb_flags = 0;
1800 if (msg->msg_flags&MSG_OOB)
1803 err = scm_send(sock, msg, &scm, true);
1807 if (msg->msg_namelen) {
1809 if (addr->nl_family != AF_NETLINK)
1811 dst_portid = addr->nl_pid;
1812 dst_group = ffs(addr->nl_groups);
1814 if ((dst_group || dst_portid) &&
1815 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1817 netlink_skb_flags |= NETLINK_SKB_DST;
1819 dst_portid = nlk->dst_portid;
1820 dst_group = nlk->dst_group;
1824 err = netlink_autobind(sock);
1828 /* Ensure nlk is hashed and visible. */
1833 if (len > sk->sk_sndbuf - 32)
1836 skb = netlink_alloc_large_skb(len, dst_group);
1840 NETLINK_CB(skb).portid = nlk->portid;
1841 NETLINK_CB(skb).dst_group = dst_group;
1842 NETLINK_CB(skb).creds = scm.creds;
1843 NETLINK_CB(skb).flags = netlink_skb_flags;
1846 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1851 err = security_netlink_send(sk, skb);
1858 refcount_inc(&skb->users);
1859 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1861 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1868 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1871 struct scm_cookie scm;
1872 struct sock *sk = sock->sk;
1873 struct netlink_sock *nlk = nlk_sk(sk);
1874 int noblock = flags&MSG_DONTWAIT;
1876 struct sk_buff *skb, *data_skb;
1884 skb = skb_recv_datagram(sk, flags, noblock, &err);
1890 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1891 if (unlikely(skb_shinfo(skb)->frag_list)) {
1893 * If this skb has a frag_list, then here that means that we
1894 * will have to use the frag_list skb's data for compat tasks
1895 * and the regular skb's data for normal (non-compat) tasks.
1897 * If we need to send the compat skb, assign it to the
1898 * 'data_skb' variable so that it will be used below for data
1899 * copying. We keep 'skb' for everything else, including
1900 * freeing both later.
1902 if (flags & MSG_CMSG_COMPAT)
1903 data_skb = skb_shinfo(skb)->frag_list;
1907 /* Record the max length of recvmsg() calls for future allocations */
1908 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1909 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1910 SKB_WITH_OVERHEAD(32768));
1912 copied = data_skb->len;
1914 msg->msg_flags |= MSG_TRUNC;
1918 skb_reset_transport_header(data_skb);
1919 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1921 if (msg->msg_name) {
1922 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1923 addr->nl_family = AF_NETLINK;
1925 addr->nl_pid = NETLINK_CB(skb).portid;
1926 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1927 msg->msg_namelen = sizeof(*addr);
1930 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1931 netlink_cmsg_recv_pktinfo(msg, skb);
1932 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1933 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1935 memset(&scm, 0, sizeof(scm));
1936 scm.creds = *NETLINK_CREDS(skb);
1937 if (flags & MSG_TRUNC)
1938 copied = data_skb->len;
1940 skb_free_datagram(sk, skb);
1942 if (nlk->cb_running &&
1943 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1944 ret = netlink_dump(sk);
1947 sk->sk_error_report(sk);
1951 scm_recv(sock, msg, &scm, flags);
1953 netlink_rcv_wake(sk);
1954 return err ? : copied;
1957 static void netlink_data_ready(struct sock *sk)
1963 * We export these functions to other modules. They provide a
1964 * complete set of kernel non-blocking support for message
1969 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1970 struct netlink_kernel_cfg *cfg)
1972 struct socket *sock;
1974 struct netlink_sock *nlk;
1975 struct listeners *listeners = NULL;
1976 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1977 unsigned int groups;
1981 if (unit < 0 || unit >= MAX_LINKS)
1984 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1987 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1988 goto out_sock_release_nosk;
1992 if (!cfg || cfg->groups < 32)
1995 groups = cfg->groups;
1997 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1999 goto out_sock_release;
2001 sk->sk_data_ready = netlink_data_ready;
2002 if (cfg && cfg->input)
2003 nlk_sk(sk)->netlink_rcv = cfg->input;
2005 if (netlink_insert(sk, 0))
2006 goto out_sock_release;
2009 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2011 netlink_table_grab();
2012 if (!nl_table[unit].registered) {
2013 nl_table[unit].groups = groups;
2014 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2015 nl_table[unit].cb_mutex = cb_mutex;
2016 nl_table[unit].module = module;
2018 nl_table[unit].bind = cfg->bind;
2019 nl_table[unit].unbind = cfg->unbind;
2020 nl_table[unit].flags = cfg->flags;
2022 nl_table[unit].compare = cfg->compare;
2024 nl_table[unit].registered = 1;
2027 nl_table[unit].registered++;
2029 netlink_table_ungrab();
2034 netlink_kernel_release(sk);
2037 out_sock_release_nosk:
2041 EXPORT_SYMBOL(__netlink_kernel_create);
2044 netlink_kernel_release(struct sock *sk)
2046 if (sk == NULL || sk->sk_socket == NULL)
2049 sock_release(sk->sk_socket);
2051 EXPORT_SYMBOL(netlink_kernel_release);
2053 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2055 struct listeners *new, *old;
2056 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2061 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2062 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2065 old = nl_deref_protected(tbl->listeners);
2066 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2067 rcu_assign_pointer(tbl->listeners, new);
2069 kfree_rcu(old, rcu);
2071 tbl->groups = groups;
2077 * netlink_change_ngroups - change number of multicast groups
2079 * This changes the number of multicast groups that are available
2080 * on a certain netlink family. Note that it is not possible to
2081 * change the number of groups to below 32. Also note that it does
2082 * not implicitly call netlink_clear_multicast_users() when the
2083 * number of groups is reduced.
2085 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2086 * @groups: The new number of groups.
2088 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2092 netlink_table_grab();
2093 err = __netlink_change_ngroups(sk, groups);
2094 netlink_table_ungrab();
2099 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2102 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2104 sk_for_each_bound(sk, &tbl->mc_list)
2105 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2109 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2111 struct nlmsghdr *nlh;
2112 int size = nlmsg_msg_size(len);
2114 nlh = skb_put(skb, NLMSG_ALIGN(size));
2115 nlh->nlmsg_type = type;
2116 nlh->nlmsg_len = size;
2117 nlh->nlmsg_flags = flags;
2118 nlh->nlmsg_pid = portid;
2119 nlh->nlmsg_seq = seq;
2120 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2121 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2124 EXPORT_SYMBOL(__nlmsg_put);
2127 * It looks a bit ugly.
2128 * It would be better to create kernel thread.
2131 static int netlink_dump(struct sock *sk)
2133 struct netlink_sock *nlk = nlk_sk(sk);
2134 struct netlink_callback *cb;
2135 struct sk_buff *skb = NULL;
2136 struct nlmsghdr *nlh;
2137 struct module *module;
2142 mutex_lock(nlk->cb_mutex);
2143 if (!nlk->cb_running) {
2148 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2151 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2152 * required, but it makes sense to _attempt_ a 16K bytes allocation
2153 * to reduce number of system calls on dump operations, if user
2154 * ever provided a big enough buffer.
2157 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2159 if (alloc_min_size < nlk->max_recvmsg_len) {
2160 alloc_size = nlk->max_recvmsg_len;
2161 skb = alloc_skb(alloc_size,
2162 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2163 __GFP_NOWARN | __GFP_NORETRY);
2166 alloc_size = alloc_min_size;
2167 skb = alloc_skb(alloc_size, GFP_KERNEL);
2172 /* Trim skb to allocated size. User is expected to provide buffer as
2173 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2174 * netlink_recvmsg())). dump will pack as many smaller messages as
2175 * could fit within the allocated skb. skb is typically allocated
2176 * with larger space than required (could be as much as near 2x the
2177 * requested size with align to next power of 2 approach). Allowing
2178 * dump to use the excess space makes it difficult for a user to have a
2179 * reasonable static buffer based on the expected largest dump of a
2180 * single netdev. The outcome is MSG_TRUNC error.
2182 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2183 netlink_skb_set_owner_r(skb, sk);
2185 if (nlk->dump_done_errno > 0)
2186 nlk->dump_done_errno = cb->dump(skb, cb);
2188 if (nlk->dump_done_errno > 0 ||
2189 skb_tailroom(skb) < nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
2190 mutex_unlock(nlk->cb_mutex);
2192 if (sk_filter(sk, skb))
2195 __netlink_sendskb(sk, skb);
2199 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE,
2200 sizeof(nlk->dump_done_errno), NLM_F_MULTI);
2204 nl_dump_check_consistent(cb, nlh);
2206 memcpy(nlmsg_data(nlh), &nlk->dump_done_errno,
2207 sizeof(nlk->dump_done_errno));
2209 if (sk_filter(sk, skb))
2212 __netlink_sendskb(sk, skb);
2217 nlk->cb_running = false;
2218 module = cb->module;
2220 mutex_unlock(nlk->cb_mutex);
2226 mutex_unlock(nlk->cb_mutex);
2231 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2232 const struct nlmsghdr *nlh,
2233 struct netlink_dump_control *control)
2235 struct netlink_callback *cb;
2237 struct netlink_sock *nlk;
2240 refcount_inc(&skb->users);
2242 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2244 ret = -ECONNREFUSED;
2249 mutex_lock(nlk->cb_mutex);
2250 /* A dump is in progress... */
2251 if (nlk->cb_running) {
2255 /* add reference of module which cb->dump belongs to */
2256 if (!try_module_get(control->module)) {
2257 ret = -EPROTONOSUPPORT;
2262 memset(cb, 0, sizeof(*cb));
2263 cb->start = control->start;
2264 cb->dump = control->dump;
2265 cb->done = control->done;
2267 cb->data = control->data;
2268 cb->module = control->module;
2269 cb->min_dump_alloc = control->min_dump_alloc;
2273 ret = cb->start(cb);
2278 nlk->cb_running = true;
2279 nlk->dump_done_errno = INT_MAX;
2281 mutex_unlock(nlk->cb_mutex);
2283 ret = netlink_dump(sk);
2290 /* We successfully started a dump, by returning -EINTR we
2291 * signal not to send ACK even if it was requested.
2297 mutex_unlock(nlk->cb_mutex);
2302 EXPORT_SYMBOL(__netlink_dump_start);
2304 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2305 const struct netlink_ext_ack *extack)
2307 struct sk_buff *skb;
2308 struct nlmsghdr *rep;
2309 struct nlmsgerr *errmsg;
2310 size_t payload = sizeof(*errmsg);
2312 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2313 unsigned int flags = 0;
2314 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2316 /* Error messages get the original request appened, unless the user
2317 * requests to cap the error message, and get extra error data if
2320 if (nlk_has_extack && extack && extack->_msg)
2321 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2324 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2325 payload += nlmsg_len(nlh);
2327 flags |= NLM_F_CAPPED;
2328 if (nlk_has_extack && extack && extack->bad_attr)
2329 tlvlen += nla_total_size(sizeof(u32));
2331 flags |= NLM_F_CAPPED;
2333 if (nlk_has_extack && extack && extack->cookie_len)
2334 tlvlen += nla_total_size(extack->cookie_len);
2338 flags |= NLM_F_ACK_TLVS;
2340 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2342 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2343 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2347 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2348 NLMSG_ERROR, payload, flags);
2349 errmsg = nlmsg_data(rep);
2350 errmsg->error = err;
2351 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2353 if (nlk_has_extack && extack) {
2355 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2359 if (extack->bad_attr &&
2360 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2361 (u8 *)extack->bad_attr >= in_skb->data +
2363 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2364 (u8 *)extack->bad_attr -
2367 if (extack->cookie_len)
2368 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2374 nlmsg_end(skb, rep);
2376 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2378 EXPORT_SYMBOL(netlink_ack);
2380 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2382 struct netlink_ext_ack *))
2384 struct netlink_ext_ack extack = {};
2385 struct nlmsghdr *nlh;
2388 while (skb->len >= nlmsg_total_size(0)) {
2391 nlh = nlmsg_hdr(skb);
2394 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2397 /* Only requests are handled by the kernel */
2398 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2401 /* Skip control messages */
2402 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2405 err = cb(skb, nlh, &extack);
2410 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2411 netlink_ack(skb, nlh, err, &extack);
2414 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2415 if (msglen > skb->len)
2417 skb_pull(skb, msglen);
2422 EXPORT_SYMBOL(netlink_rcv_skb);
2425 * nlmsg_notify - send a notification netlink message
2426 * @sk: netlink socket to use
2427 * @skb: notification message
2428 * @portid: destination netlink portid for reports or 0
2429 * @group: destination multicast group or 0
2430 * @report: 1 to report back, 0 to disable
2431 * @flags: allocation flags
2433 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2434 unsigned int group, int report, gfp_t flags)
2439 int exclude_portid = 0;
2442 refcount_inc(&skb->users);
2443 exclude_portid = portid;
2446 /* errors reported via destination sk->sk_err, but propagate
2447 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2448 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2454 err2 = nlmsg_unicast(sk, skb, portid);
2455 if (!err || err == -ESRCH)
2461 EXPORT_SYMBOL(nlmsg_notify);
2463 #ifdef CONFIG_PROC_FS
2464 struct nl_seq_iter {
2465 struct seq_net_private p;
2466 struct rhashtable_iter hti;
2470 static int netlink_walk_start(struct nl_seq_iter *iter)
2474 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2477 iter->link = MAX_LINKS;
2481 err = rhashtable_walk_start(&iter->hti);
2482 return err == -EAGAIN ? 0 : err;
2485 static void netlink_walk_stop(struct nl_seq_iter *iter)
2487 rhashtable_walk_stop(&iter->hti);
2488 rhashtable_walk_exit(&iter->hti);
2491 static void *__netlink_seq_next(struct seq_file *seq)
2493 struct nl_seq_iter *iter = seq->private;
2494 struct netlink_sock *nlk;
2500 nlk = rhashtable_walk_next(&iter->hti);
2503 if (PTR_ERR(nlk) == -EAGAIN)
2512 netlink_walk_stop(iter);
2513 if (++iter->link >= MAX_LINKS)
2516 err = netlink_walk_start(iter);
2518 return ERR_PTR(err);
2520 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2525 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2527 struct nl_seq_iter *iter = seq->private;
2528 void *obj = SEQ_START_TOKEN;
2534 err = netlink_walk_start(iter);
2536 return ERR_PTR(err);
2538 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2539 obj = __netlink_seq_next(seq);
2544 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2547 return __netlink_seq_next(seq);
2550 static void netlink_seq_stop(struct seq_file *seq, void *v)
2552 struct nl_seq_iter *iter = seq->private;
2554 if (iter->link >= MAX_LINKS)
2557 netlink_walk_stop(iter);
2561 static int netlink_seq_show(struct seq_file *seq, void *v)
2563 if (v == SEQ_START_TOKEN) {
2565 "sk Eth Pid Groups "
2566 "Rmem Wmem Dump Locks Drops Inode\n");
2569 struct netlink_sock *nlk = nlk_sk(s);
2571 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2575 nlk->groups ? (u32)nlk->groups[0] : 0,
2576 sk_rmem_alloc_get(s),
2577 sk_wmem_alloc_get(s),
2579 refcount_read(&s->sk_refcnt),
2580 atomic_read(&s->sk_drops),
2588 static const struct seq_operations netlink_seq_ops = {
2589 .start = netlink_seq_start,
2590 .next = netlink_seq_next,
2591 .stop = netlink_seq_stop,
2592 .show = netlink_seq_show,
2596 static int netlink_seq_open(struct inode *inode, struct file *file)
2598 return seq_open_net(inode, file, &netlink_seq_ops,
2599 sizeof(struct nl_seq_iter));
2602 static const struct file_operations netlink_seq_fops = {
2603 .owner = THIS_MODULE,
2604 .open = netlink_seq_open,
2606 .llseek = seq_lseek,
2607 .release = seq_release_net,
2612 int netlink_register_notifier(struct notifier_block *nb)
2614 return blocking_notifier_chain_register(&netlink_chain, nb);
2616 EXPORT_SYMBOL(netlink_register_notifier);
2618 int netlink_unregister_notifier(struct notifier_block *nb)
2620 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2622 EXPORT_SYMBOL(netlink_unregister_notifier);
2624 static const struct proto_ops netlink_ops = {
2625 .family = PF_NETLINK,
2626 .owner = THIS_MODULE,
2627 .release = netlink_release,
2628 .bind = netlink_bind,
2629 .connect = netlink_connect,
2630 .socketpair = sock_no_socketpair,
2631 .accept = sock_no_accept,
2632 .getname = netlink_getname,
2633 .poll = datagram_poll,
2634 .ioctl = netlink_ioctl,
2635 .listen = sock_no_listen,
2636 .shutdown = sock_no_shutdown,
2637 .setsockopt = netlink_setsockopt,
2638 .getsockopt = netlink_getsockopt,
2639 .sendmsg = netlink_sendmsg,
2640 .recvmsg = netlink_recvmsg,
2641 .mmap = sock_no_mmap,
2642 .sendpage = sock_no_sendpage,
2645 static const struct net_proto_family netlink_family_ops = {
2646 .family = PF_NETLINK,
2647 .create = netlink_create,
2648 .owner = THIS_MODULE, /* for consistency 8) */
2651 static int __net_init netlink_net_init(struct net *net)
2653 #ifdef CONFIG_PROC_FS
2654 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2660 static void __net_exit netlink_net_exit(struct net *net)
2662 #ifdef CONFIG_PROC_FS
2663 remove_proc_entry("netlink", net->proc_net);
2667 static void __init netlink_add_usersock_entry(void)
2669 struct listeners *listeners;
2672 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2674 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2676 netlink_table_grab();
2678 nl_table[NETLINK_USERSOCK].groups = groups;
2679 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2680 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2681 nl_table[NETLINK_USERSOCK].registered = 1;
2682 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2684 netlink_table_ungrab();
2687 static struct pernet_operations __net_initdata netlink_net_ops = {
2688 .init = netlink_net_init,
2689 .exit = netlink_net_exit,
2692 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2694 const struct netlink_sock *nlk = data;
2695 struct netlink_compare_arg arg;
2697 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2698 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2701 static const struct rhashtable_params netlink_rhashtable_params = {
2702 .head_offset = offsetof(struct netlink_sock, node),
2703 .key_len = netlink_compare_arg_len,
2704 .obj_hashfn = netlink_hash,
2705 .obj_cmpfn = netlink_compare,
2706 .automatic_shrinking = true,
2709 static int __init netlink_proto_init(void)
2712 int err = proto_register(&netlink_proto, 0);
2717 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2719 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2723 for (i = 0; i < MAX_LINKS; i++) {
2724 if (rhashtable_init(&nl_table[i].hash,
2725 &netlink_rhashtable_params) < 0) {
2727 rhashtable_destroy(&nl_table[i].hash);
2733 INIT_LIST_HEAD(&netlink_tap_all);
2735 netlink_add_usersock_entry();
2737 sock_register(&netlink_family_ops);
2738 register_pernet_subsys(&netlink_net_ops);
2739 /* The netlink device handler may be needed early. */
2744 panic("netlink_init: Cannot allocate nl_table\n");
2747 core_initcall(netlink_proto_init);