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>
66 #include <net/net_namespace.h>
69 #include <net/netlink.h>
71 #include "af_netlink.h"
75 unsigned long masks[0];
79 #define NETLINK_S_CONGESTED 0x0
81 static inline int netlink_is_kernel(struct sock *sk)
83 return nlk_sk(sk)->flags & NETLINK_F_KERNEL_SOCKET;
86 struct netlink_table *nl_table __read_mostly;
87 EXPORT_SYMBOL_GPL(nl_table);
89 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
91 static struct lock_class_key nlk_cb_mutex_keys[MAX_LINKS];
93 static const char *const nlk_cb_mutex_key_strings[MAX_LINKS + 1] = {
96 "nlk_cb_mutex-USERSOCK",
97 "nlk_cb_mutex-FIREWALL",
98 "nlk_cb_mutex-SOCK_DIAG",
101 "nlk_cb_mutex-SELINUX",
102 "nlk_cb_mutex-ISCSI",
103 "nlk_cb_mutex-AUDIT",
104 "nlk_cb_mutex-FIB_LOOKUP",
105 "nlk_cb_mutex-CONNECTOR",
106 "nlk_cb_mutex-NETFILTER",
107 "nlk_cb_mutex-IP6_FW",
108 "nlk_cb_mutex-DNRTMSG",
109 "nlk_cb_mutex-KOBJECT_UEVENT",
110 "nlk_cb_mutex-GENERIC",
112 "nlk_cb_mutex-SCSITRANSPORT",
113 "nlk_cb_mutex-ECRYPTFS",
115 "nlk_cb_mutex-CRYPTO",
126 "nlk_cb_mutex-MAX_LINKS"
129 static int netlink_dump(struct sock *sk);
130 static void netlink_skb_destructor(struct sk_buff *skb);
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 memcpy(skb_put(new, len), 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(atomic_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(atomic_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;
693 if (!atomic_dec_and_test(&sk->sk_refcnt))
696 if (nlk->cb_running && nlk->cb.done) {
697 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
698 schedule_work(&nlk->work);
705 static int netlink_release(struct socket *sock)
707 struct sock *sk = sock->sk;
708 struct netlink_sock *nlk;
718 * OK. Socket is unlinked, any packets that arrive now
722 /* must not acquire netlink_table_lock in any way again before unbind
723 * and notifying genetlink is done as otherwise it might deadlock
725 if (nlk->netlink_unbind) {
728 for (i = 0; i < nlk->ngroups; i++)
729 if (test_bit(i, nlk->groups))
730 nlk->netlink_unbind(sock_net(sk), i + 1);
732 if (sk->sk_protocol == NETLINK_GENERIC &&
733 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
734 wake_up(&genl_sk_destructing_waitq);
737 wake_up_interruptible_all(&nlk->wait);
739 skb_queue_purge(&sk->sk_write_queue);
741 if (nlk->portid && nlk->bound) {
742 struct netlink_notify n = {
744 .protocol = sk->sk_protocol,
745 .portid = nlk->portid,
747 blocking_notifier_call_chain(&netlink_chain,
748 NETLINK_URELEASE, &n);
751 module_put(nlk->module);
753 if (netlink_is_kernel(sk)) {
754 netlink_table_grab();
755 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
756 if (--nl_table[sk->sk_protocol].registered == 0) {
757 struct listeners *old;
759 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
760 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
762 nl_table[sk->sk_protocol].module = NULL;
763 nl_table[sk->sk_protocol].bind = NULL;
764 nl_table[sk->sk_protocol].unbind = NULL;
765 nl_table[sk->sk_protocol].flags = 0;
766 nl_table[sk->sk_protocol].registered = 0;
768 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 if (nlk->netlink_bind && groups) {
988 for (group = 0; group < nlk->ngroups; group++) {
989 if (!test_bit(group, &groups))
991 err = nlk->netlink_bind(net, group + 1);
994 netlink_undo_bind(group, groups, sk);
999 /* No need for barriers here as we return to user-space without
1000 * using any of the bound attributes.
1003 err = nladdr->nl_pid ?
1004 netlink_insert(sk, nladdr->nl_pid) :
1005 netlink_autobind(sock);
1007 netlink_undo_bind(nlk->ngroups, groups, sk);
1012 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1015 netlink_table_grab();
1016 netlink_update_subscriptions(sk, nlk->subscriptions +
1018 hweight32(nlk->groups[0]));
1019 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1020 netlink_update_listeners(sk);
1021 netlink_table_ungrab();
1026 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1027 int alen, int flags)
1030 struct sock *sk = sock->sk;
1031 struct netlink_sock *nlk = nlk_sk(sk);
1032 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1034 if (alen < sizeof(addr->sa_family))
1037 if (addr->sa_family == AF_UNSPEC) {
1038 sk->sk_state = NETLINK_UNCONNECTED;
1039 nlk->dst_portid = 0;
1043 if (addr->sa_family != AF_NETLINK)
1046 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1047 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1050 /* No need for barriers here as we return to user-space without
1051 * using any of the bound attributes.
1054 err = netlink_autobind(sock);
1057 sk->sk_state = NETLINK_CONNECTED;
1058 nlk->dst_portid = nladdr->nl_pid;
1059 nlk->dst_group = ffs(nladdr->nl_groups);
1065 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1066 int *addr_len, int peer)
1068 struct sock *sk = sock->sk;
1069 struct netlink_sock *nlk = nlk_sk(sk);
1070 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1072 nladdr->nl_family = AF_NETLINK;
1074 *addr_len = sizeof(*nladdr);
1077 nladdr->nl_pid = nlk->dst_portid;
1078 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1080 nladdr->nl_pid = nlk->portid;
1081 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1086 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1089 /* try to hand this ioctl down to the NIC drivers.
1091 return -ENOIOCTLCMD;
1094 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1097 struct netlink_sock *nlk;
1099 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1101 return ERR_PTR(-ECONNREFUSED);
1103 /* Don't bother queuing skb if kernel socket has no input function */
1105 if (sock->sk_state == NETLINK_CONNECTED &&
1106 nlk->dst_portid != nlk_sk(ssk)->portid) {
1108 return ERR_PTR(-ECONNREFUSED);
1113 struct sock *netlink_getsockbyfilp(struct file *filp)
1115 struct inode *inode = file_inode(filp);
1118 if (!S_ISSOCK(inode->i_mode))
1119 return ERR_PTR(-ENOTSOCK);
1121 sock = SOCKET_I(inode)->sk;
1122 if (sock->sk_family != AF_NETLINK)
1123 return ERR_PTR(-EINVAL);
1129 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1132 struct sk_buff *skb;
1135 if (size <= NLMSG_GOODSIZE || broadcast)
1136 return alloc_skb(size, GFP_KERNEL);
1138 size = SKB_DATA_ALIGN(size) +
1139 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1141 data = vmalloc(size);
1145 skb = __build_skb(data, size);
1149 skb->destructor = netlink_skb_destructor;
1155 * Attach a skb to a netlink socket.
1156 * The caller must hold a reference to the destination socket. On error, the
1157 * reference is dropped. The skb is not send to the destination, just all
1158 * all error checks are performed and memory in the queue is reserved.
1160 * < 0: error. skb freed, reference to sock dropped.
1162 * 1: repeat lookup - reference dropped while waiting for socket memory.
1164 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1165 long *timeo, struct sock *ssk)
1167 struct netlink_sock *nlk;
1171 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1172 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1173 DECLARE_WAITQUEUE(wait, current);
1175 if (!ssk || netlink_is_kernel(ssk))
1176 netlink_overrun(sk);
1182 __set_current_state(TASK_INTERRUPTIBLE);
1183 add_wait_queue(&nlk->wait, &wait);
1185 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1186 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1187 !sock_flag(sk, SOCK_DEAD))
1188 *timeo = schedule_timeout(*timeo);
1190 __set_current_state(TASK_RUNNING);
1191 remove_wait_queue(&nlk->wait, &wait);
1194 if (signal_pending(current)) {
1196 return sock_intr_errno(*timeo);
1200 netlink_skb_set_owner_r(skb, sk);
1204 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1208 netlink_deliver_tap(skb);
1210 skb_queue_tail(&sk->sk_receive_queue, skb);
1211 sk->sk_data_ready(sk);
1215 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1217 int len = __netlink_sendskb(sk, skb);
1223 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1229 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1233 WARN_ON(skb->sk != NULL);
1234 delta = skb->end - skb->tail;
1235 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1238 if (skb_shared(skb)) {
1239 struct sk_buff *nskb = skb_clone(skb, allocation);
1246 pskb_expand_head(skb, 0, -delta,
1247 (allocation & ~__GFP_DIRECT_RECLAIM) |
1248 __GFP_NOWARN | __GFP_NORETRY);
1252 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1256 struct netlink_sock *nlk = nlk_sk(sk);
1258 ret = -ECONNREFUSED;
1259 if (nlk->netlink_rcv != NULL) {
1261 netlink_skb_set_owner_r(skb, sk);
1262 NETLINK_CB(skb).sk = ssk;
1263 netlink_deliver_tap_kernel(sk, ssk, skb);
1264 nlk->netlink_rcv(skb);
1273 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1274 u32 portid, int nonblock)
1280 skb = netlink_trim(skb, gfp_any());
1282 timeo = sock_sndtimeo(ssk, nonblock);
1284 sk = netlink_getsockbyportid(ssk, portid);
1289 if (netlink_is_kernel(sk))
1290 return netlink_unicast_kernel(sk, skb, ssk);
1292 if (sk_filter(sk, skb)) {
1299 err = netlink_attachskb(sk, skb, &timeo, ssk);
1305 return netlink_sendskb(sk, skb);
1307 EXPORT_SYMBOL(netlink_unicast);
1309 int netlink_has_listeners(struct sock *sk, unsigned int group)
1312 struct listeners *listeners;
1314 BUG_ON(!netlink_is_kernel(sk));
1317 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1319 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1320 res = test_bit(group - 1, listeners->masks);
1326 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1328 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1330 struct netlink_sock *nlk = nlk_sk(sk);
1332 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1333 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1334 netlink_skb_set_owner_r(skb, sk);
1335 __netlink_sendskb(sk, skb);
1336 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1341 struct netlink_broadcast_data {
1342 struct sock *exclude_sk;
1347 int delivery_failure;
1351 struct sk_buff *skb, *skb2;
1352 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1356 static void do_one_broadcast(struct sock *sk,
1357 struct netlink_broadcast_data *p)
1359 struct netlink_sock *nlk = nlk_sk(sk);
1362 if (p->exclude_sk == sk)
1365 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1366 !test_bit(p->group - 1, nlk->groups))
1369 if (!net_eq(sock_net(sk), p->net)) {
1370 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1373 if (!peernet_has_id(sock_net(sk), p->net))
1376 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1382 netlink_overrun(sk);
1387 if (p->skb2 == NULL) {
1388 if (skb_shared(p->skb)) {
1389 p->skb2 = skb_clone(p->skb, p->allocation);
1391 p->skb2 = skb_get(p->skb);
1393 * skb ownership may have been set when
1394 * delivered to a previous socket.
1396 skb_orphan(p->skb2);
1399 if (p->skb2 == NULL) {
1400 netlink_overrun(sk);
1401 /* Clone failed. Notify ALL listeners. */
1403 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1404 p->delivery_failure = 1;
1407 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1412 if (sk_filter(sk, p->skb2)) {
1417 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1418 NETLINK_CB(p->skb2).nsid_is_set = true;
1419 val = netlink_broadcast_deliver(sk, p->skb2);
1421 netlink_overrun(sk);
1422 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1423 p->delivery_failure = 1;
1425 p->congested |= val;
1433 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1434 u32 group, gfp_t allocation,
1435 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1438 struct net *net = sock_net(ssk);
1439 struct netlink_broadcast_data info;
1442 skb = netlink_trim(skb, allocation);
1444 info.exclude_sk = ssk;
1446 info.portid = portid;
1449 info.delivery_failure = 0;
1452 info.allocation = allocation;
1455 info.tx_filter = filter;
1456 info.tx_data = filter_data;
1458 /* While we sleep in clone, do not allow to change socket list */
1460 netlink_lock_table();
1462 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1463 do_one_broadcast(sk, &info);
1467 netlink_unlock_table();
1469 if (info.delivery_failure) {
1470 kfree_skb(info.skb2);
1473 consume_skb(info.skb2);
1475 if (info.delivered) {
1476 if (info.congested && gfpflags_allow_blocking(allocation))
1482 EXPORT_SYMBOL(netlink_broadcast_filtered);
1484 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1485 u32 group, gfp_t allocation)
1487 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1490 EXPORT_SYMBOL(netlink_broadcast);
1492 struct netlink_set_err_data {
1493 struct sock *exclude_sk;
1499 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1501 struct netlink_sock *nlk = nlk_sk(sk);
1504 if (sk == p->exclude_sk)
1507 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1510 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1511 !test_bit(p->group - 1, nlk->groups))
1514 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1519 sk->sk_err = p->code;
1520 sk->sk_error_report(sk);
1526 * netlink_set_err - report error to broadcast listeners
1527 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1528 * @portid: the PORTID of a process that we want to skip (if any)
1529 * @group: the broadcast group that will notice the error
1530 * @code: error code, must be negative (as usual in kernelspace)
1532 * This function returns the number of broadcast listeners that have set the
1533 * NETLINK_NO_ENOBUFS socket option.
1535 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1537 struct netlink_set_err_data info;
1541 info.exclude_sk = ssk;
1542 info.portid = portid;
1544 /* sk->sk_err wants a positive error value */
1547 read_lock(&nl_table_lock);
1549 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1550 ret += do_one_set_err(sk, &info);
1552 read_unlock(&nl_table_lock);
1555 EXPORT_SYMBOL(netlink_set_err);
1557 /* must be called with netlink table grabbed */
1558 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1562 int old, new = !!is_new, subscriptions;
1564 old = test_bit(group - 1, nlk->groups);
1565 subscriptions = nlk->subscriptions - old + new;
1567 __set_bit(group - 1, nlk->groups);
1569 __clear_bit(group - 1, nlk->groups);
1570 netlink_update_subscriptions(&nlk->sk, subscriptions);
1571 netlink_update_listeners(&nlk->sk);
1574 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1575 char __user *optval, unsigned int optlen)
1577 struct sock *sk = sock->sk;
1578 struct netlink_sock *nlk = nlk_sk(sk);
1579 unsigned int val = 0;
1582 if (level != SOL_NETLINK)
1583 return -ENOPROTOOPT;
1585 if (optlen >= sizeof(int) &&
1586 get_user(val, (unsigned int __user *)optval))
1590 case NETLINK_PKTINFO:
1592 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1594 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1597 case NETLINK_ADD_MEMBERSHIP:
1598 case NETLINK_DROP_MEMBERSHIP: {
1599 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1601 err = netlink_realloc_groups(sk);
1604 if (!val || val - 1 >= nlk->ngroups)
1606 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1607 err = nlk->netlink_bind(sock_net(sk), val);
1611 netlink_table_grab();
1612 netlink_update_socket_mc(nlk, val,
1613 optname == NETLINK_ADD_MEMBERSHIP);
1614 netlink_table_ungrab();
1615 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1616 nlk->netlink_unbind(sock_net(sk), val);
1621 case NETLINK_BROADCAST_ERROR:
1623 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1625 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1628 case NETLINK_NO_ENOBUFS:
1630 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1631 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1632 wake_up_interruptible(&nlk->wait);
1634 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1638 case NETLINK_LISTEN_ALL_NSID:
1639 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1643 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1645 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1648 case NETLINK_CAP_ACK:
1650 nlk->flags |= NETLINK_F_CAP_ACK;
1652 nlk->flags &= ~NETLINK_F_CAP_ACK;
1655 case NETLINK_EXT_ACK:
1657 nlk->flags |= NETLINK_F_EXT_ACK;
1659 nlk->flags &= ~NETLINK_F_EXT_ACK;
1668 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1669 char __user *optval, int __user *optlen)
1671 struct sock *sk = sock->sk;
1672 struct netlink_sock *nlk = nlk_sk(sk);
1675 if (level != SOL_NETLINK)
1676 return -ENOPROTOOPT;
1678 if (get_user(len, optlen))
1684 case NETLINK_PKTINFO:
1685 if (len < sizeof(int))
1688 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1689 if (put_user(len, optlen) ||
1690 put_user(val, optval))
1694 case NETLINK_BROADCAST_ERROR:
1695 if (len < sizeof(int))
1698 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1699 if (put_user(len, optlen) ||
1700 put_user(val, optval))
1704 case NETLINK_NO_ENOBUFS:
1705 if (len < sizeof(int))
1708 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1709 if (put_user(len, optlen) ||
1710 put_user(val, optval))
1714 case NETLINK_LIST_MEMBERSHIPS: {
1715 int pos, idx, shift;
1718 netlink_lock_table();
1719 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1720 if (len - pos < sizeof(u32))
1723 idx = pos / sizeof(unsigned long);
1724 shift = (pos % sizeof(unsigned long)) * 8;
1725 if (put_user((u32)(nlk->groups[idx] >> shift),
1726 (u32 __user *)(optval + pos))) {
1731 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1733 netlink_unlock_table();
1736 case NETLINK_CAP_ACK:
1737 if (len < sizeof(int))
1740 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1741 if (put_user(len, optlen) ||
1742 put_user(val, optval))
1746 case NETLINK_EXT_ACK:
1747 if (len < sizeof(int))
1750 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1751 if (put_user(len, optlen) || put_user(val, optval))
1761 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1763 struct nl_pktinfo info;
1765 info.group = NETLINK_CB(skb).dst_group;
1766 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1769 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1770 struct sk_buff *skb)
1772 if (!NETLINK_CB(skb).nsid_is_set)
1775 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1776 &NETLINK_CB(skb).nsid);
1779 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1781 struct sock *sk = sock->sk;
1782 struct netlink_sock *nlk = nlk_sk(sk);
1783 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1786 struct sk_buff *skb;
1788 struct scm_cookie scm;
1789 u32 netlink_skb_flags = 0;
1791 if (msg->msg_flags&MSG_OOB)
1794 err = scm_send(sock, msg, &scm, true);
1798 if (msg->msg_namelen) {
1800 if (addr->nl_family != AF_NETLINK)
1802 dst_portid = addr->nl_pid;
1803 dst_group = ffs(addr->nl_groups);
1805 if ((dst_group || dst_portid) &&
1806 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1808 netlink_skb_flags |= NETLINK_SKB_DST;
1810 dst_portid = nlk->dst_portid;
1811 dst_group = nlk->dst_group;
1815 err = netlink_autobind(sock);
1819 /* Ensure nlk is hashed and visible. */
1824 if (len > sk->sk_sndbuf - 32)
1827 skb = netlink_alloc_large_skb(len, dst_group);
1831 NETLINK_CB(skb).portid = nlk->portid;
1832 NETLINK_CB(skb).dst_group = dst_group;
1833 NETLINK_CB(skb).creds = scm.creds;
1834 NETLINK_CB(skb).flags = netlink_skb_flags;
1837 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1842 err = security_netlink_send(sk, skb);
1849 atomic_inc(&skb->users);
1850 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1852 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1859 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1862 struct scm_cookie scm;
1863 struct sock *sk = sock->sk;
1864 struct netlink_sock *nlk = nlk_sk(sk);
1865 int noblock = flags&MSG_DONTWAIT;
1867 struct sk_buff *skb, *data_skb;
1875 skb = skb_recv_datagram(sk, flags, noblock, &err);
1881 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1882 if (unlikely(skb_shinfo(skb)->frag_list)) {
1884 * If this skb has a frag_list, then here that means that we
1885 * will have to use the frag_list skb's data for compat tasks
1886 * and the regular skb's data for normal (non-compat) tasks.
1888 * If we need to send the compat skb, assign it to the
1889 * 'data_skb' variable so that it will be used below for data
1890 * copying. We keep 'skb' for everything else, including
1891 * freeing both later.
1893 if (flags & MSG_CMSG_COMPAT)
1894 data_skb = skb_shinfo(skb)->frag_list;
1898 /* Record the max length of recvmsg() calls for future allocations */
1899 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1900 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1901 SKB_WITH_OVERHEAD(32768));
1903 copied = data_skb->len;
1905 msg->msg_flags |= MSG_TRUNC;
1909 skb_reset_transport_header(data_skb);
1910 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1912 if (msg->msg_name) {
1913 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1914 addr->nl_family = AF_NETLINK;
1916 addr->nl_pid = NETLINK_CB(skb).portid;
1917 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1918 msg->msg_namelen = sizeof(*addr);
1921 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1922 netlink_cmsg_recv_pktinfo(msg, skb);
1923 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1924 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1926 memset(&scm, 0, sizeof(scm));
1927 scm.creds = *NETLINK_CREDS(skb);
1928 if (flags & MSG_TRUNC)
1929 copied = data_skb->len;
1931 skb_free_datagram(sk, skb);
1933 if (nlk->cb_running &&
1934 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1935 ret = netlink_dump(sk);
1938 sk->sk_error_report(sk);
1942 scm_recv(sock, msg, &scm, flags);
1944 netlink_rcv_wake(sk);
1945 return err ? : copied;
1948 static void netlink_data_ready(struct sock *sk)
1954 * We export these functions to other modules. They provide a
1955 * complete set of kernel non-blocking support for message
1960 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1961 struct netlink_kernel_cfg *cfg)
1963 struct socket *sock;
1965 struct netlink_sock *nlk;
1966 struct listeners *listeners = NULL;
1967 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1968 unsigned int groups;
1972 if (unit < 0 || unit >= MAX_LINKS)
1975 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1978 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1979 goto out_sock_release_nosk;
1983 if (!cfg || cfg->groups < 32)
1986 groups = cfg->groups;
1988 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1990 goto out_sock_release;
1992 sk->sk_data_ready = netlink_data_ready;
1993 if (cfg && cfg->input)
1994 nlk_sk(sk)->netlink_rcv = cfg->input;
1996 if (netlink_insert(sk, 0))
1997 goto out_sock_release;
2000 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2002 netlink_table_grab();
2003 if (!nl_table[unit].registered) {
2004 nl_table[unit].groups = groups;
2005 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2006 nl_table[unit].cb_mutex = cb_mutex;
2007 nl_table[unit].module = module;
2009 nl_table[unit].bind = cfg->bind;
2010 nl_table[unit].unbind = cfg->unbind;
2011 nl_table[unit].flags = cfg->flags;
2013 nl_table[unit].compare = cfg->compare;
2015 nl_table[unit].registered = 1;
2018 nl_table[unit].registered++;
2020 netlink_table_ungrab();
2025 netlink_kernel_release(sk);
2028 out_sock_release_nosk:
2032 EXPORT_SYMBOL(__netlink_kernel_create);
2035 netlink_kernel_release(struct sock *sk)
2037 if (sk == NULL || sk->sk_socket == NULL)
2040 sock_release(sk->sk_socket);
2042 EXPORT_SYMBOL(netlink_kernel_release);
2044 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2046 struct listeners *new, *old;
2047 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2052 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2053 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2056 old = nl_deref_protected(tbl->listeners);
2057 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2058 rcu_assign_pointer(tbl->listeners, new);
2060 kfree_rcu(old, rcu);
2062 tbl->groups = groups;
2068 * netlink_change_ngroups - change number of multicast groups
2070 * This changes the number of multicast groups that are available
2071 * on a certain netlink family. Note that it is not possible to
2072 * change the number of groups to below 32. Also note that it does
2073 * not implicitly call netlink_clear_multicast_users() when the
2074 * number of groups is reduced.
2076 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2077 * @groups: The new number of groups.
2079 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2083 netlink_table_grab();
2084 err = __netlink_change_ngroups(sk, groups);
2085 netlink_table_ungrab();
2090 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2093 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2095 sk_for_each_bound(sk, &tbl->mc_list)
2096 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2100 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2102 struct nlmsghdr *nlh;
2103 int size = nlmsg_msg_size(len);
2105 nlh = (struct nlmsghdr *)skb_put(skb, NLMSG_ALIGN(size));
2106 nlh->nlmsg_type = type;
2107 nlh->nlmsg_len = size;
2108 nlh->nlmsg_flags = flags;
2109 nlh->nlmsg_pid = portid;
2110 nlh->nlmsg_seq = seq;
2111 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2112 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2115 EXPORT_SYMBOL(__nlmsg_put);
2118 * It looks a bit ugly.
2119 * It would be better to create kernel thread.
2122 static int netlink_dump(struct sock *sk)
2124 struct netlink_sock *nlk = nlk_sk(sk);
2125 struct netlink_callback *cb;
2126 struct sk_buff *skb = NULL;
2127 struct nlmsghdr *nlh;
2128 struct module *module;
2129 int len, err = -ENOBUFS;
2133 mutex_lock(nlk->cb_mutex);
2134 if (!nlk->cb_running) {
2139 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2142 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2143 * required, but it makes sense to _attempt_ a 16K bytes allocation
2144 * to reduce number of system calls on dump operations, if user
2145 * ever provided a big enough buffer.
2148 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2150 if (alloc_min_size < nlk->max_recvmsg_len) {
2151 alloc_size = nlk->max_recvmsg_len;
2152 skb = alloc_skb(alloc_size,
2153 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2154 __GFP_NOWARN | __GFP_NORETRY);
2157 alloc_size = alloc_min_size;
2158 skb = alloc_skb(alloc_size, GFP_KERNEL);
2163 /* Trim skb to allocated size. User is expected to provide buffer as
2164 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2165 * netlink_recvmsg())). dump will pack as many smaller messages as
2166 * could fit within the allocated skb. skb is typically allocated
2167 * with larger space than required (could be as much as near 2x the
2168 * requested size with align to next power of 2 approach). Allowing
2169 * dump to use the excess space makes it difficult for a user to have a
2170 * reasonable static buffer based on the expected largest dump of a
2171 * single netdev. The outcome is MSG_TRUNC error.
2173 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2174 netlink_skb_set_owner_r(skb, sk);
2176 len = cb->dump(skb, cb);
2179 mutex_unlock(nlk->cb_mutex);
2181 if (sk_filter(sk, skb))
2184 __netlink_sendskb(sk, skb);
2188 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2192 nl_dump_check_consistent(cb, nlh);
2194 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2196 if (sk_filter(sk, skb))
2199 __netlink_sendskb(sk, skb);
2204 nlk->cb_running = false;
2205 module = cb->module;
2207 mutex_unlock(nlk->cb_mutex);
2213 mutex_unlock(nlk->cb_mutex);
2218 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2219 const struct nlmsghdr *nlh,
2220 struct netlink_dump_control *control)
2222 struct netlink_callback *cb;
2224 struct netlink_sock *nlk;
2227 atomic_inc(&skb->users);
2229 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2231 ret = -ECONNREFUSED;
2236 mutex_lock(nlk->cb_mutex);
2237 /* A dump is in progress... */
2238 if (nlk->cb_running) {
2242 /* add reference of module which cb->dump belongs to */
2243 if (!try_module_get(control->module)) {
2244 ret = -EPROTONOSUPPORT;
2249 memset(cb, 0, sizeof(*cb));
2250 cb->start = control->start;
2251 cb->dump = control->dump;
2252 cb->done = control->done;
2254 cb->data = control->data;
2255 cb->module = control->module;
2256 cb->min_dump_alloc = control->min_dump_alloc;
2259 nlk->cb_running = true;
2261 mutex_unlock(nlk->cb_mutex);
2266 ret = netlink_dump(sk);
2272 /* We successfully started a dump, by returning -EINTR we
2273 * signal not to send ACK even if it was requested.
2279 mutex_unlock(nlk->cb_mutex);
2284 EXPORT_SYMBOL(__netlink_dump_start);
2286 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2287 const struct netlink_ext_ack *extack)
2289 struct sk_buff *skb;
2290 struct nlmsghdr *rep;
2291 struct nlmsgerr *errmsg;
2292 size_t payload = sizeof(*errmsg);
2294 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2295 unsigned int flags = 0;
2297 /* Error messages get the original request appened, unless the user
2298 * requests to cap the error message, and get extra error data if
2302 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2303 payload += nlmsg_len(nlh);
2305 flags |= NLM_F_CAPPED;
2306 if (nlk->flags & NETLINK_F_EXT_ACK && extack) {
2308 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2309 if (extack->bad_attr)
2310 tlvlen += nla_total_size(sizeof(u32));
2313 flags |= NLM_F_CAPPED;
2315 if (nlk->flags & NETLINK_F_EXT_ACK &&
2316 extack && extack->cookie_len)
2317 tlvlen += nla_total_size(extack->cookie_len);
2321 flags |= NLM_F_ACK_TLVS;
2323 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2327 sk = netlink_lookup(sock_net(in_skb->sk),
2328 in_skb->sk->sk_protocol,
2329 NETLINK_CB(in_skb).portid);
2331 sk->sk_err = ENOBUFS;
2332 sk->sk_error_report(sk);
2338 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2339 NLMSG_ERROR, payload, flags);
2340 errmsg = nlmsg_data(rep);
2341 errmsg->error = err;
2342 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2344 if (nlk->flags & NETLINK_F_EXT_ACK && extack) {
2347 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2349 if (extack->bad_attr &&
2350 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2351 (u8 *)extack->bad_attr >= in_skb->data +
2353 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2354 (u8 *)extack->bad_attr -
2357 if (extack->cookie_len)
2358 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2364 nlmsg_end(skb, rep);
2366 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2368 EXPORT_SYMBOL(netlink_ack);
2370 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2372 struct netlink_ext_ack *))
2374 struct netlink_ext_ack extack = {};
2375 struct nlmsghdr *nlh;
2378 while (skb->len >= nlmsg_total_size(0)) {
2381 nlh = nlmsg_hdr(skb);
2384 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2387 /* Only requests are handled by the kernel */
2388 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2391 /* Skip control messages */
2392 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2395 err = cb(skb, nlh, &extack);
2400 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2401 netlink_ack(skb, nlh, err, &extack);
2404 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2405 if (msglen > skb->len)
2407 skb_pull(skb, msglen);
2412 EXPORT_SYMBOL(netlink_rcv_skb);
2415 * nlmsg_notify - send a notification netlink message
2416 * @sk: netlink socket to use
2417 * @skb: notification message
2418 * @portid: destination netlink portid for reports or 0
2419 * @group: destination multicast group or 0
2420 * @report: 1 to report back, 0 to disable
2421 * @flags: allocation flags
2423 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2424 unsigned int group, int report, gfp_t flags)
2429 int exclude_portid = 0;
2432 atomic_inc(&skb->users);
2433 exclude_portid = portid;
2436 /* errors reported via destination sk->sk_err, but propagate
2437 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2438 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2444 err2 = nlmsg_unicast(sk, skb, portid);
2445 if (!err || err == -ESRCH)
2451 EXPORT_SYMBOL(nlmsg_notify);
2453 #ifdef CONFIG_PROC_FS
2454 struct nl_seq_iter {
2455 struct seq_net_private p;
2456 struct rhashtable_iter hti;
2460 static int netlink_walk_start(struct nl_seq_iter *iter)
2464 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2467 iter->link = MAX_LINKS;
2471 err = rhashtable_walk_start(&iter->hti);
2472 return err == -EAGAIN ? 0 : err;
2475 static void netlink_walk_stop(struct nl_seq_iter *iter)
2477 rhashtable_walk_stop(&iter->hti);
2478 rhashtable_walk_exit(&iter->hti);
2481 static void *__netlink_seq_next(struct seq_file *seq)
2483 struct nl_seq_iter *iter = seq->private;
2484 struct netlink_sock *nlk;
2490 nlk = rhashtable_walk_next(&iter->hti);
2493 if (PTR_ERR(nlk) == -EAGAIN)
2502 netlink_walk_stop(iter);
2503 if (++iter->link >= MAX_LINKS)
2506 err = netlink_walk_start(iter);
2508 return ERR_PTR(err);
2510 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2515 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2517 struct nl_seq_iter *iter = seq->private;
2518 void *obj = SEQ_START_TOKEN;
2524 err = netlink_walk_start(iter);
2526 return ERR_PTR(err);
2528 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2529 obj = __netlink_seq_next(seq);
2534 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2537 return __netlink_seq_next(seq);
2540 static void netlink_seq_stop(struct seq_file *seq, void *v)
2542 struct nl_seq_iter *iter = seq->private;
2544 if (iter->link >= MAX_LINKS)
2547 netlink_walk_stop(iter);
2551 static int netlink_seq_show(struct seq_file *seq, void *v)
2553 if (v == SEQ_START_TOKEN) {
2555 "sk Eth Pid Groups "
2556 "Rmem Wmem Dump Locks Drops Inode\n");
2559 struct netlink_sock *nlk = nlk_sk(s);
2561 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2565 nlk->groups ? (u32)nlk->groups[0] : 0,
2566 sk_rmem_alloc_get(s),
2567 sk_wmem_alloc_get(s),
2569 atomic_read(&s->sk_refcnt),
2570 atomic_read(&s->sk_drops),
2578 static const struct seq_operations netlink_seq_ops = {
2579 .start = netlink_seq_start,
2580 .next = netlink_seq_next,
2581 .stop = netlink_seq_stop,
2582 .show = netlink_seq_show,
2586 static int netlink_seq_open(struct inode *inode, struct file *file)
2588 return seq_open_net(inode, file, &netlink_seq_ops,
2589 sizeof(struct nl_seq_iter));
2592 static const struct file_operations netlink_seq_fops = {
2593 .owner = THIS_MODULE,
2594 .open = netlink_seq_open,
2596 .llseek = seq_lseek,
2597 .release = seq_release_net,
2602 int netlink_register_notifier(struct notifier_block *nb)
2604 return blocking_notifier_chain_register(&netlink_chain, nb);
2606 EXPORT_SYMBOL(netlink_register_notifier);
2608 int netlink_unregister_notifier(struct notifier_block *nb)
2610 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2612 EXPORT_SYMBOL(netlink_unregister_notifier);
2614 static const struct proto_ops netlink_ops = {
2615 .family = PF_NETLINK,
2616 .owner = THIS_MODULE,
2617 .release = netlink_release,
2618 .bind = netlink_bind,
2619 .connect = netlink_connect,
2620 .socketpair = sock_no_socketpair,
2621 .accept = sock_no_accept,
2622 .getname = netlink_getname,
2623 .poll = datagram_poll,
2624 .ioctl = netlink_ioctl,
2625 .listen = sock_no_listen,
2626 .shutdown = sock_no_shutdown,
2627 .setsockopt = netlink_setsockopt,
2628 .getsockopt = netlink_getsockopt,
2629 .sendmsg = netlink_sendmsg,
2630 .recvmsg = netlink_recvmsg,
2631 .mmap = sock_no_mmap,
2632 .sendpage = sock_no_sendpage,
2635 static const struct net_proto_family netlink_family_ops = {
2636 .family = PF_NETLINK,
2637 .create = netlink_create,
2638 .owner = THIS_MODULE, /* for consistency 8) */
2641 static int __net_init netlink_net_init(struct net *net)
2643 #ifdef CONFIG_PROC_FS
2644 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2650 static void __net_exit netlink_net_exit(struct net *net)
2652 #ifdef CONFIG_PROC_FS
2653 remove_proc_entry("netlink", net->proc_net);
2657 static void __init netlink_add_usersock_entry(void)
2659 struct listeners *listeners;
2662 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2664 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2666 netlink_table_grab();
2668 nl_table[NETLINK_USERSOCK].groups = groups;
2669 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2670 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2671 nl_table[NETLINK_USERSOCK].registered = 1;
2672 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2674 netlink_table_ungrab();
2677 static struct pernet_operations __net_initdata netlink_net_ops = {
2678 .init = netlink_net_init,
2679 .exit = netlink_net_exit,
2682 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2684 const struct netlink_sock *nlk = data;
2685 struct netlink_compare_arg arg;
2687 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2688 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2691 static const struct rhashtable_params netlink_rhashtable_params = {
2692 .head_offset = offsetof(struct netlink_sock, node),
2693 .key_len = netlink_compare_arg_len,
2694 .obj_hashfn = netlink_hash,
2695 .obj_cmpfn = netlink_compare,
2696 .automatic_shrinking = true,
2699 static int __init netlink_proto_init(void)
2702 int err = proto_register(&netlink_proto, 0);
2707 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2709 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2713 for (i = 0; i < MAX_LINKS; i++) {
2714 if (rhashtable_init(&nl_table[i].hash,
2715 &netlink_rhashtable_params) < 0) {
2717 rhashtable_destroy(&nl_table[i].hash);
2723 INIT_LIST_HEAD(&netlink_tap_all);
2725 netlink_add_usersock_entry();
2727 sock_register(&netlink_family_ops);
2728 register_pernet_subsys(&netlink_net_ops);
2729 /* The netlink device handler may be needed early. */
2734 panic("netlink_init: Cannot allocate nl_table\n");
2737 core_initcall(netlink_proto_init);