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);
131 static void netlink_skb_destructor(struct sk_buff *skb);
133 /* nl_table locking explained:
134 * Lookup and traversal are protected with an RCU read-side lock. Insertion
135 * and removal are protected with per bucket lock while using RCU list
136 * modification primitives and may run in parallel to RCU protected lookups.
137 * Destruction of the Netlink socket may only occur *after* nl_table_lock has
138 * been acquired * either during or after the socket has been removed from
139 * the list and after an RCU grace period.
141 DEFINE_RWLOCK(nl_table_lock);
142 EXPORT_SYMBOL_GPL(nl_table_lock);
143 static atomic_t nl_table_users = ATOMIC_INIT(0);
145 #define nl_deref_protected(X) rcu_dereference_protected(X, lockdep_is_held(&nl_table_lock));
147 static BLOCKING_NOTIFIER_HEAD(netlink_chain);
149 static DEFINE_SPINLOCK(netlink_tap_lock);
150 static struct list_head netlink_tap_all __read_mostly;
152 static const struct rhashtable_params netlink_rhashtable_params;
154 static inline u32 netlink_group_mask(u32 group)
156 return group ? 1 << (group - 1) : 0;
159 static struct sk_buff *netlink_to_full_skb(const struct sk_buff *skb,
162 unsigned int len = skb_end_offset(skb);
165 new = alloc_skb(len, gfp_mask);
169 NETLINK_CB(new).portid = NETLINK_CB(skb).portid;
170 NETLINK_CB(new).dst_group = NETLINK_CB(skb).dst_group;
171 NETLINK_CB(new).creds = NETLINK_CB(skb).creds;
173 skb_put_data(new, skb->data, len);
177 int netlink_add_tap(struct netlink_tap *nt)
179 if (unlikely(nt->dev->type != ARPHRD_NETLINK))
182 spin_lock(&netlink_tap_lock);
183 list_add_rcu(&nt->list, &netlink_tap_all);
184 spin_unlock(&netlink_tap_lock);
186 __module_get(nt->module);
190 EXPORT_SYMBOL_GPL(netlink_add_tap);
192 static int __netlink_remove_tap(struct netlink_tap *nt)
195 struct netlink_tap *tmp;
197 spin_lock(&netlink_tap_lock);
199 list_for_each_entry(tmp, &netlink_tap_all, list) {
201 list_del_rcu(&nt->list);
207 pr_warn("__netlink_remove_tap: %p not found\n", nt);
209 spin_unlock(&netlink_tap_lock);
212 module_put(nt->module);
214 return found ? 0 : -ENODEV;
217 int netlink_remove_tap(struct netlink_tap *nt)
221 ret = __netlink_remove_tap(nt);
226 EXPORT_SYMBOL_GPL(netlink_remove_tap);
228 static bool netlink_filter_tap(const struct sk_buff *skb)
230 struct sock *sk = skb->sk;
232 /* We take the more conservative approach and
233 * whitelist socket protocols that may pass.
235 switch (sk->sk_protocol) {
237 case NETLINK_USERSOCK:
238 case NETLINK_SOCK_DIAG:
241 case NETLINK_FIB_LOOKUP:
242 case NETLINK_NETFILTER:
243 case NETLINK_GENERIC:
250 static int __netlink_deliver_tap_skb(struct sk_buff *skb,
251 struct net_device *dev)
253 struct sk_buff *nskb;
254 struct sock *sk = skb->sk;
259 if (is_vmalloc_addr(skb->head))
260 nskb = netlink_to_full_skb(skb, GFP_ATOMIC);
262 nskb = skb_clone(skb, GFP_ATOMIC);
265 nskb->protocol = htons((u16) sk->sk_protocol);
266 nskb->pkt_type = netlink_is_kernel(sk) ?
267 PACKET_KERNEL : PACKET_USER;
268 skb_reset_network_header(nskb);
269 ret = dev_queue_xmit(nskb);
270 if (unlikely(ret > 0))
271 ret = net_xmit_errno(ret);
278 static void __netlink_deliver_tap(struct sk_buff *skb)
281 struct netlink_tap *tmp;
283 if (!netlink_filter_tap(skb))
286 list_for_each_entry_rcu(tmp, &netlink_tap_all, list) {
287 ret = __netlink_deliver_tap_skb(skb, tmp->dev);
293 static void netlink_deliver_tap(struct sk_buff *skb)
297 if (unlikely(!list_empty(&netlink_tap_all)))
298 __netlink_deliver_tap(skb);
303 static void netlink_deliver_tap_kernel(struct sock *dst, struct sock *src,
306 if (!(netlink_is_kernel(dst) && netlink_is_kernel(src)))
307 netlink_deliver_tap(skb);
310 static void netlink_overrun(struct sock *sk)
312 struct netlink_sock *nlk = nlk_sk(sk);
314 if (!(nlk->flags & NETLINK_F_RECV_NO_ENOBUFS)) {
315 if (!test_and_set_bit(NETLINK_S_CONGESTED,
316 &nlk_sk(sk)->state)) {
317 sk->sk_err = ENOBUFS;
318 sk->sk_error_report(sk);
321 atomic_inc(&sk->sk_drops);
324 static void netlink_rcv_wake(struct sock *sk)
326 struct netlink_sock *nlk = nlk_sk(sk);
328 if (skb_queue_empty(&sk->sk_receive_queue))
329 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
330 if (!test_bit(NETLINK_S_CONGESTED, &nlk->state))
331 wake_up_interruptible(&nlk->wait);
334 static void netlink_skb_destructor(struct sk_buff *skb)
336 if (is_vmalloc_addr(skb->head)) {
338 !atomic_dec_return(&(skb_shinfo(skb)->dataref)))
347 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
349 WARN_ON(skb->sk != NULL);
351 skb->destructor = netlink_skb_destructor;
352 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
353 sk_mem_charge(sk, skb->truesize);
356 static void netlink_sock_destruct(struct sock *sk)
358 struct netlink_sock *nlk = nlk_sk(sk);
360 if (nlk->cb_running) {
362 nlk->cb.done(&nlk->cb);
363 module_put(nlk->cb.module);
364 kfree_skb(nlk->cb.skb);
367 skb_queue_purge(&sk->sk_receive_queue);
369 if (!sock_flag(sk, SOCK_DEAD)) {
370 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
374 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
375 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
376 WARN_ON(nlk_sk(sk)->groups);
379 static void netlink_sock_destruct_work(struct work_struct *work)
381 struct netlink_sock *nlk = container_of(work, struct netlink_sock,
387 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
388 * SMP. Look, when several writers sleep and reader wakes them up, all but one
389 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
390 * this, _but_ remember, it adds useless work on UP machines.
393 void netlink_table_grab(void)
394 __acquires(nl_table_lock)
398 write_lock_irq(&nl_table_lock);
400 if (atomic_read(&nl_table_users)) {
401 DECLARE_WAITQUEUE(wait, current);
403 add_wait_queue_exclusive(&nl_table_wait, &wait);
405 set_current_state(TASK_UNINTERRUPTIBLE);
406 if (atomic_read(&nl_table_users) == 0)
408 write_unlock_irq(&nl_table_lock);
410 write_lock_irq(&nl_table_lock);
413 __set_current_state(TASK_RUNNING);
414 remove_wait_queue(&nl_table_wait, &wait);
418 void netlink_table_ungrab(void)
419 __releases(nl_table_lock)
421 write_unlock_irq(&nl_table_lock);
422 wake_up(&nl_table_wait);
426 netlink_lock_table(void)
428 /* read_lock() synchronizes us to netlink_table_grab */
430 read_lock(&nl_table_lock);
431 atomic_inc(&nl_table_users);
432 read_unlock(&nl_table_lock);
436 netlink_unlock_table(void)
438 if (atomic_dec_and_test(&nl_table_users))
439 wake_up(&nl_table_wait);
442 struct netlink_compare_arg
448 /* Doing sizeof directly may yield 4 extra bytes on 64-bit. */
449 #define netlink_compare_arg_len \
450 (offsetof(struct netlink_compare_arg, portid) + sizeof(u32))
452 static inline int netlink_compare(struct rhashtable_compare_arg *arg,
455 const struct netlink_compare_arg *x = arg->key;
456 const struct netlink_sock *nlk = ptr;
458 return nlk->portid != x->portid ||
459 !net_eq(sock_net(&nlk->sk), read_pnet(&x->pnet));
462 static void netlink_compare_arg_init(struct netlink_compare_arg *arg,
463 struct net *net, u32 portid)
465 memset(arg, 0, sizeof(*arg));
466 write_pnet(&arg->pnet, net);
467 arg->portid = portid;
470 static struct sock *__netlink_lookup(struct netlink_table *table, u32 portid,
473 struct netlink_compare_arg arg;
475 netlink_compare_arg_init(&arg, net, portid);
476 return rhashtable_lookup_fast(&table->hash, &arg,
477 netlink_rhashtable_params);
480 static int __netlink_insert(struct netlink_table *table, struct sock *sk)
482 struct netlink_compare_arg arg;
484 netlink_compare_arg_init(&arg, sock_net(sk), nlk_sk(sk)->portid);
485 return rhashtable_lookup_insert_key(&table->hash, &arg,
487 netlink_rhashtable_params);
490 static struct sock *netlink_lookup(struct net *net, int protocol, u32 portid)
492 struct netlink_table *table = &nl_table[protocol];
496 sk = __netlink_lookup(table, portid, net);
504 static const struct proto_ops netlink_ops;
507 netlink_update_listeners(struct sock *sk)
509 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
512 struct listeners *listeners;
514 listeners = nl_deref_protected(tbl->listeners);
518 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
520 sk_for_each_bound(sk, &tbl->mc_list) {
521 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
522 mask |= nlk_sk(sk)->groups[i];
524 listeners->masks[i] = mask;
526 /* this function is only called with the netlink table "grabbed", which
527 * makes sure updates are visible before bind or setsockopt return. */
530 static int netlink_insert(struct sock *sk, u32 portid)
532 struct netlink_table *table = &nl_table[sk->sk_protocol];
537 err = nlk_sk(sk)->portid == portid ? 0 : -EBUSY;
538 if (nlk_sk(sk)->bound)
542 if (BITS_PER_LONG > 32 &&
543 unlikely(atomic_read(&table->hash.nelems) >= UINT_MAX))
546 nlk_sk(sk)->portid = portid;
549 err = __netlink_insert(table, sk);
551 /* In case the hashtable backend returns with -EBUSY
552 * from here, it must not escape to the caller.
554 if (unlikely(err == -EBUSY))
562 /* We need to ensure that the socket is hashed and visible. */
564 nlk_sk(sk)->bound = portid;
571 static void netlink_remove(struct sock *sk)
573 struct netlink_table *table;
575 table = &nl_table[sk->sk_protocol];
576 if (!rhashtable_remove_fast(&table->hash, &nlk_sk(sk)->node,
577 netlink_rhashtable_params)) {
578 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
582 netlink_table_grab();
583 if (nlk_sk(sk)->subscriptions) {
584 __sk_del_bind_node(sk);
585 netlink_update_listeners(sk);
587 if (sk->sk_protocol == NETLINK_GENERIC)
588 atomic_inc(&genl_sk_destructing_cnt);
589 netlink_table_ungrab();
592 static struct proto netlink_proto = {
594 .owner = THIS_MODULE,
595 .obj_size = sizeof(struct netlink_sock),
598 static int __netlink_create(struct net *net, struct socket *sock,
599 struct mutex *cb_mutex, int protocol,
603 struct netlink_sock *nlk;
605 sock->ops = &netlink_ops;
607 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, kern);
611 sock_init_data(sock, sk);
615 nlk->cb_mutex = cb_mutex;
617 nlk->cb_mutex = &nlk->cb_def_mutex;
618 mutex_init(nlk->cb_mutex);
619 lockdep_set_class_and_name(nlk->cb_mutex,
620 nlk_cb_mutex_keys + protocol,
621 nlk_cb_mutex_key_strings[protocol]);
623 init_waitqueue_head(&nlk->wait);
625 sk->sk_destruct = netlink_sock_destruct;
626 sk->sk_protocol = protocol;
630 static int netlink_create(struct net *net, struct socket *sock, int protocol,
633 struct module *module = NULL;
634 struct mutex *cb_mutex;
635 struct netlink_sock *nlk;
636 int (*bind)(struct net *net, int group);
637 void (*unbind)(struct net *net, int group);
640 sock->state = SS_UNCONNECTED;
642 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
643 return -ESOCKTNOSUPPORT;
645 if (protocol < 0 || protocol >= MAX_LINKS)
646 return -EPROTONOSUPPORT;
648 netlink_lock_table();
649 #ifdef CONFIG_MODULES
650 if (!nl_table[protocol].registered) {
651 netlink_unlock_table();
652 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
653 netlink_lock_table();
656 if (nl_table[protocol].registered &&
657 try_module_get(nl_table[protocol].module))
658 module = nl_table[protocol].module;
660 err = -EPROTONOSUPPORT;
661 cb_mutex = nl_table[protocol].cb_mutex;
662 bind = nl_table[protocol].bind;
663 unbind = nl_table[protocol].unbind;
664 netlink_unlock_table();
669 err = __netlink_create(net, sock, cb_mutex, protocol, kern);
674 sock_prot_inuse_add(net, &netlink_proto, 1);
677 nlk = nlk_sk(sock->sk);
678 nlk->module = module;
679 nlk->netlink_bind = bind;
680 nlk->netlink_unbind = unbind;
689 static void deferred_put_nlk_sk(struct rcu_head *head)
691 struct netlink_sock *nlk = container_of(head, struct netlink_sock, rcu);
692 struct sock *sk = &nlk->sk;
697 if (!refcount_dec_and_test(&sk->sk_refcnt))
700 if (nlk->cb_running && nlk->cb.done) {
701 INIT_WORK(&nlk->work, netlink_sock_destruct_work);
702 schedule_work(&nlk->work);
709 static int netlink_release(struct socket *sock)
711 struct sock *sk = sock->sk;
712 struct netlink_sock *nlk;
722 * OK. Socket is unlinked, any packets that arrive now
726 /* must not acquire netlink_table_lock in any way again before unbind
727 * and notifying genetlink is done as otherwise it might deadlock
729 if (nlk->netlink_unbind) {
732 for (i = 0; i < nlk->ngroups; i++)
733 if (test_bit(i, nlk->groups))
734 nlk->netlink_unbind(sock_net(sk), i + 1);
736 if (sk->sk_protocol == NETLINK_GENERIC &&
737 atomic_dec_return(&genl_sk_destructing_cnt) == 0)
738 wake_up(&genl_sk_destructing_waitq);
741 wake_up_interruptible_all(&nlk->wait);
743 skb_queue_purge(&sk->sk_write_queue);
745 if (nlk->portid && nlk->bound) {
746 struct netlink_notify n = {
748 .protocol = sk->sk_protocol,
749 .portid = nlk->portid,
751 blocking_notifier_call_chain(&netlink_chain,
752 NETLINK_URELEASE, &n);
755 module_put(nlk->module);
757 if (netlink_is_kernel(sk)) {
758 netlink_table_grab();
759 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
760 if (--nl_table[sk->sk_protocol].registered == 0) {
761 struct listeners *old;
763 old = nl_deref_protected(nl_table[sk->sk_protocol].listeners);
764 RCU_INIT_POINTER(nl_table[sk->sk_protocol].listeners, NULL);
766 nl_table[sk->sk_protocol].module = NULL;
767 nl_table[sk->sk_protocol].bind = NULL;
768 nl_table[sk->sk_protocol].unbind = NULL;
769 nl_table[sk->sk_protocol].flags = 0;
770 nl_table[sk->sk_protocol].registered = 0;
772 netlink_table_ungrab();
776 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
778 call_rcu(&nlk->rcu, deferred_put_nlk_sk);
782 static int netlink_autobind(struct socket *sock)
784 struct sock *sk = sock->sk;
785 struct net *net = sock_net(sk);
786 struct netlink_table *table = &nl_table[sk->sk_protocol];
787 s32 portid = task_tgid_vnr(current);
795 ok = !__netlink_lookup(table, portid, net);
798 /* Bind collision, search negative portid values. */
800 /* rover will be in range [S32_MIN, -4097] */
801 rover = S32_MIN + prandom_u32_max(-4096 - S32_MIN);
802 else if (rover >= -4096)
808 err = netlink_insert(sk, portid);
809 if (err == -EADDRINUSE)
812 /* If 2 threads race to autobind, that is fine. */
820 * __netlink_ns_capable - General netlink message capability test
821 * @nsp: NETLINK_CB of the socket buffer holding a netlink command from userspace.
822 * @user_ns: The user namespace of the capability to use
823 * @cap: The capability to use
825 * Test to see if the opener of the socket we received the message
826 * from had when the netlink socket was created and the sender of the
827 * message has has the capability @cap in the user namespace @user_ns.
829 bool __netlink_ns_capable(const struct netlink_skb_parms *nsp,
830 struct user_namespace *user_ns, int cap)
832 return ((nsp->flags & NETLINK_SKB_DST) ||
833 file_ns_capable(nsp->sk->sk_socket->file, user_ns, cap)) &&
834 ns_capable(user_ns, cap);
836 EXPORT_SYMBOL(__netlink_ns_capable);
839 * netlink_ns_capable - General netlink message capability test
840 * @skb: socket buffer holding a netlink command from userspace
841 * @user_ns: The user namespace of the capability to use
842 * @cap: The capability to use
844 * Test to see if the opener of the socket we received the message
845 * from had when the netlink socket was created and the sender of the
846 * message has has the capability @cap in the user namespace @user_ns.
848 bool netlink_ns_capable(const struct sk_buff *skb,
849 struct user_namespace *user_ns, int cap)
851 return __netlink_ns_capable(&NETLINK_CB(skb), user_ns, cap);
853 EXPORT_SYMBOL(netlink_ns_capable);
856 * netlink_capable - Netlink global message capability test
857 * @skb: socket buffer holding a netlink command from userspace
858 * @cap: The capability to use
860 * Test to see if the opener of the socket we received the message
861 * from had when the netlink socket was created and the sender of the
862 * message has has the capability @cap in all user namespaces.
864 bool netlink_capable(const struct sk_buff *skb, int cap)
866 return netlink_ns_capable(skb, &init_user_ns, cap);
868 EXPORT_SYMBOL(netlink_capable);
871 * netlink_net_capable - Netlink network namespace message capability test
872 * @skb: socket buffer holding a netlink command from userspace
873 * @cap: The capability to use
875 * Test to see if the opener of the socket we received the message
876 * from had when the netlink socket was created and the sender of the
877 * message has has the capability @cap over the network namespace of
878 * the socket we received the message from.
880 bool netlink_net_capable(const struct sk_buff *skb, int cap)
882 return netlink_ns_capable(skb, sock_net(skb->sk)->user_ns, cap);
884 EXPORT_SYMBOL(netlink_net_capable);
886 static inline int netlink_allowed(const struct socket *sock, unsigned int flag)
888 return (nl_table[sock->sk->sk_protocol].flags & flag) ||
889 ns_capable(sock_net(sock->sk)->user_ns, CAP_NET_ADMIN);
893 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
895 struct netlink_sock *nlk = nlk_sk(sk);
897 if (nlk->subscriptions && !subscriptions)
898 __sk_del_bind_node(sk);
899 else if (!nlk->subscriptions && subscriptions)
900 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
901 nlk->subscriptions = subscriptions;
904 static int netlink_realloc_groups(struct sock *sk)
906 struct netlink_sock *nlk = nlk_sk(sk);
908 unsigned long *new_groups;
911 netlink_table_grab();
913 groups = nl_table[sk->sk_protocol].groups;
914 if (!nl_table[sk->sk_protocol].registered) {
919 if (nlk->ngroups >= groups)
922 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
923 if (new_groups == NULL) {
927 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
928 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
930 nlk->groups = new_groups;
931 nlk->ngroups = groups;
933 netlink_table_ungrab();
937 static void netlink_undo_bind(int group, long unsigned int groups,
940 struct netlink_sock *nlk = nlk_sk(sk);
943 if (!nlk->netlink_unbind)
946 for (undo = 0; undo < group; undo++)
947 if (test_bit(undo, &groups))
948 nlk->netlink_unbind(sock_net(sk), undo + 1);
951 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
954 struct sock *sk = sock->sk;
955 struct net *net = sock_net(sk);
956 struct netlink_sock *nlk = nlk_sk(sk);
957 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
959 long unsigned int groups = nladdr->nl_groups;
962 if (addr_len < sizeof(struct sockaddr_nl))
965 if (nladdr->nl_family != AF_NETLINK)
968 /* Only superuser is allowed to listen multicasts */
970 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
972 err = netlink_realloc_groups(sk);
979 /* Ensure nlk->portid is up-to-date. */
982 if (nladdr->nl_pid != nlk->portid)
986 netlink_lock_table();
987 if (nlk->netlink_bind && groups) {
990 for (group = 0; group < nlk->ngroups; group++) {
991 if (!test_bit(group, &groups))
993 err = nlk->netlink_bind(net, group + 1);
996 netlink_undo_bind(group, groups, sk);
1001 /* No need for barriers here as we return to user-space without
1002 * using any of the bound attributes.
1005 err = nladdr->nl_pid ?
1006 netlink_insert(sk, nladdr->nl_pid) :
1007 netlink_autobind(sock);
1009 netlink_undo_bind(nlk->ngroups, groups, sk);
1014 if (!groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
1016 netlink_unlock_table();
1018 netlink_table_grab();
1019 netlink_update_subscriptions(sk, nlk->subscriptions +
1021 hweight32(nlk->groups[0]));
1022 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | groups;
1023 netlink_update_listeners(sk);
1024 netlink_table_ungrab();
1029 netlink_unlock_table();
1033 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
1034 int alen, int flags)
1037 struct sock *sk = sock->sk;
1038 struct netlink_sock *nlk = nlk_sk(sk);
1039 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
1041 if (alen < sizeof(addr->sa_family))
1044 if (addr->sa_family == AF_UNSPEC) {
1045 sk->sk_state = NETLINK_UNCONNECTED;
1046 nlk->dst_portid = 0;
1050 if (addr->sa_family != AF_NETLINK)
1053 if ((nladdr->nl_groups || nladdr->nl_pid) &&
1054 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1057 /* No need for barriers here as we return to user-space without
1058 * using any of the bound attributes.
1061 err = netlink_autobind(sock);
1064 sk->sk_state = NETLINK_CONNECTED;
1065 nlk->dst_portid = nladdr->nl_pid;
1066 nlk->dst_group = ffs(nladdr->nl_groups);
1072 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
1073 int *addr_len, int peer)
1075 struct sock *sk = sock->sk;
1076 struct netlink_sock *nlk = nlk_sk(sk);
1077 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
1079 nladdr->nl_family = AF_NETLINK;
1081 *addr_len = sizeof(*nladdr);
1084 nladdr->nl_pid = nlk->dst_portid;
1085 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
1087 nladdr->nl_pid = nlk->portid;
1088 netlink_lock_table();
1089 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
1090 netlink_unlock_table();
1095 static int netlink_ioctl(struct socket *sock, unsigned int cmd,
1098 /* try to hand this ioctl down to the NIC drivers.
1100 return -ENOIOCTLCMD;
1103 static struct sock *netlink_getsockbyportid(struct sock *ssk, u32 portid)
1106 struct netlink_sock *nlk;
1108 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, portid);
1110 return ERR_PTR(-ECONNREFUSED);
1112 /* Don't bother queuing skb if kernel socket has no input function */
1114 if (sock->sk_state == NETLINK_CONNECTED &&
1115 nlk->dst_portid != nlk_sk(ssk)->portid) {
1117 return ERR_PTR(-ECONNREFUSED);
1122 struct sock *netlink_getsockbyfilp(struct file *filp)
1124 struct inode *inode = file_inode(filp);
1127 if (!S_ISSOCK(inode->i_mode))
1128 return ERR_PTR(-ENOTSOCK);
1130 sock = SOCKET_I(inode)->sk;
1131 if (sock->sk_family != AF_NETLINK)
1132 return ERR_PTR(-EINVAL);
1138 static struct sk_buff *netlink_alloc_large_skb(unsigned int size,
1141 struct sk_buff *skb;
1144 if (size <= NLMSG_GOODSIZE || broadcast)
1145 return alloc_skb(size, GFP_KERNEL);
1147 size = SKB_DATA_ALIGN(size) +
1148 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1150 data = vmalloc(size);
1154 skb = __build_skb(data, size);
1158 skb->destructor = netlink_skb_destructor;
1164 * Attach a skb to a netlink socket.
1165 * The caller must hold a reference to the destination socket. On error, the
1166 * reference is dropped. The skb is not send to the destination, just all
1167 * all error checks are performed and memory in the queue is reserved.
1169 * < 0: error. skb freed, reference to sock dropped.
1171 * 1: repeat lookup - reference dropped while waiting for socket memory.
1173 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
1174 long *timeo, struct sock *ssk)
1176 struct netlink_sock *nlk;
1180 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1181 test_bit(NETLINK_S_CONGESTED, &nlk->state))) {
1182 DECLARE_WAITQUEUE(wait, current);
1184 if (!ssk || netlink_is_kernel(ssk))
1185 netlink_overrun(sk);
1191 __set_current_state(TASK_INTERRUPTIBLE);
1192 add_wait_queue(&nlk->wait, &wait);
1194 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
1195 test_bit(NETLINK_S_CONGESTED, &nlk->state)) &&
1196 !sock_flag(sk, SOCK_DEAD))
1197 *timeo = schedule_timeout(*timeo);
1199 __set_current_state(TASK_RUNNING);
1200 remove_wait_queue(&nlk->wait, &wait);
1203 if (signal_pending(current)) {
1205 return sock_intr_errno(*timeo);
1209 netlink_skb_set_owner_r(skb, sk);
1213 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1217 netlink_deliver_tap(skb);
1219 skb_queue_tail(&sk->sk_receive_queue, skb);
1220 sk->sk_data_ready(sk);
1224 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
1226 int len = __netlink_sendskb(sk, skb);
1232 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
1238 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
1242 WARN_ON(skb->sk != NULL);
1243 delta = skb->end - skb->tail;
1244 if (is_vmalloc_addr(skb->head) || delta * 2 < skb->truesize)
1247 if (skb_shared(skb)) {
1248 struct sk_buff *nskb = skb_clone(skb, allocation);
1255 pskb_expand_head(skb, 0, -delta,
1256 (allocation & ~__GFP_DIRECT_RECLAIM) |
1257 __GFP_NOWARN | __GFP_NORETRY);
1261 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb,
1265 struct netlink_sock *nlk = nlk_sk(sk);
1267 ret = -ECONNREFUSED;
1268 if (nlk->netlink_rcv != NULL) {
1270 netlink_skb_set_owner_r(skb, sk);
1271 NETLINK_CB(skb).sk = ssk;
1272 netlink_deliver_tap_kernel(sk, ssk, skb);
1273 nlk->netlink_rcv(skb);
1282 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
1283 u32 portid, int nonblock)
1289 skb = netlink_trim(skb, gfp_any());
1291 timeo = sock_sndtimeo(ssk, nonblock);
1293 sk = netlink_getsockbyportid(ssk, portid);
1298 if (netlink_is_kernel(sk))
1299 return netlink_unicast_kernel(sk, skb, ssk);
1301 if (sk_filter(sk, skb)) {
1308 err = netlink_attachskb(sk, skb, &timeo, ssk);
1314 return netlink_sendskb(sk, skb);
1316 EXPORT_SYMBOL(netlink_unicast);
1318 int netlink_has_listeners(struct sock *sk, unsigned int group)
1321 struct listeners *listeners;
1323 BUG_ON(!netlink_is_kernel(sk));
1326 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
1328 if (listeners && group - 1 < nl_table[sk->sk_protocol].groups)
1329 res = test_bit(group - 1, listeners->masks);
1335 EXPORT_SYMBOL_GPL(netlink_has_listeners);
1337 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
1339 struct netlink_sock *nlk = nlk_sk(sk);
1341 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
1342 !test_bit(NETLINK_S_CONGESTED, &nlk->state)) {
1343 netlink_skb_set_owner_r(skb, sk);
1344 __netlink_sendskb(sk, skb);
1345 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
1350 struct netlink_broadcast_data {
1351 struct sock *exclude_sk;
1356 int delivery_failure;
1360 struct sk_buff *skb, *skb2;
1361 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
1365 static void do_one_broadcast(struct sock *sk,
1366 struct netlink_broadcast_data *p)
1368 struct netlink_sock *nlk = nlk_sk(sk);
1371 if (p->exclude_sk == sk)
1374 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1375 !test_bit(p->group - 1, nlk->groups))
1378 if (!net_eq(sock_net(sk), p->net)) {
1379 if (!(nlk->flags & NETLINK_F_LISTEN_ALL_NSID))
1382 if (!peernet_has_id(sock_net(sk), p->net))
1385 if (!file_ns_capable(sk->sk_socket->file, p->net->user_ns,
1391 netlink_overrun(sk);
1396 if (p->skb2 == NULL) {
1397 if (skb_shared(p->skb)) {
1398 p->skb2 = skb_clone(p->skb, p->allocation);
1400 p->skb2 = skb_get(p->skb);
1402 * skb ownership may have been set when
1403 * delivered to a previous socket.
1405 skb_orphan(p->skb2);
1408 if (p->skb2 == NULL) {
1409 netlink_overrun(sk);
1410 /* Clone failed. Notify ALL listeners. */
1412 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1413 p->delivery_failure = 1;
1416 if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1421 if (sk_filter(sk, p->skb2)) {
1426 NETLINK_CB(p->skb2).nsid = peernet2id(sock_net(sk), p->net);
1427 if (NETLINK_CB(p->skb2).nsid != NETNSA_NSID_NOT_ASSIGNED)
1428 NETLINK_CB(p->skb2).nsid_is_set = true;
1429 val = netlink_broadcast_deliver(sk, p->skb2);
1431 netlink_overrun(sk);
1432 if (nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR)
1433 p->delivery_failure = 1;
1435 p->congested |= val;
1443 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 portid,
1444 u32 group, gfp_t allocation,
1445 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1448 struct net *net = sock_net(ssk);
1449 struct netlink_broadcast_data info;
1452 skb = netlink_trim(skb, allocation);
1454 info.exclude_sk = ssk;
1456 info.portid = portid;
1459 info.delivery_failure = 0;
1462 info.allocation = allocation;
1465 info.tx_filter = filter;
1466 info.tx_data = filter_data;
1468 /* While we sleep in clone, do not allow to change socket list */
1470 netlink_lock_table();
1472 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1473 do_one_broadcast(sk, &info);
1477 netlink_unlock_table();
1479 if (info.delivery_failure) {
1480 kfree_skb(info.skb2);
1483 consume_skb(info.skb2);
1485 if (info.delivered) {
1486 if (info.congested && gfpflags_allow_blocking(allocation))
1492 EXPORT_SYMBOL(netlink_broadcast_filtered);
1494 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 portid,
1495 u32 group, gfp_t allocation)
1497 return netlink_broadcast_filtered(ssk, skb, portid, group, allocation,
1500 EXPORT_SYMBOL(netlink_broadcast);
1502 struct netlink_set_err_data {
1503 struct sock *exclude_sk;
1509 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1511 struct netlink_sock *nlk = nlk_sk(sk);
1514 if (sk == p->exclude_sk)
1517 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1520 if (nlk->portid == p->portid || p->group - 1 >= nlk->ngroups ||
1521 !test_bit(p->group - 1, nlk->groups))
1524 if (p->code == ENOBUFS && nlk->flags & NETLINK_F_RECV_NO_ENOBUFS) {
1529 sk->sk_err = p->code;
1530 sk->sk_error_report(sk);
1536 * netlink_set_err - report error to broadcast listeners
1537 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1538 * @portid: the PORTID of a process that we want to skip (if any)
1539 * @group: the broadcast group that will notice the error
1540 * @code: error code, must be negative (as usual in kernelspace)
1542 * This function returns the number of broadcast listeners that have set the
1543 * NETLINK_NO_ENOBUFS socket option.
1545 int netlink_set_err(struct sock *ssk, u32 portid, u32 group, int code)
1547 struct netlink_set_err_data info;
1551 info.exclude_sk = ssk;
1552 info.portid = portid;
1554 /* sk->sk_err wants a positive error value */
1557 read_lock(&nl_table_lock);
1559 sk_for_each_bound(sk, &nl_table[ssk->sk_protocol].mc_list)
1560 ret += do_one_set_err(sk, &info);
1562 read_unlock(&nl_table_lock);
1565 EXPORT_SYMBOL(netlink_set_err);
1567 /* must be called with netlink table grabbed */
1568 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1572 int old, new = !!is_new, subscriptions;
1574 old = test_bit(group - 1, nlk->groups);
1575 subscriptions = nlk->subscriptions - old + new;
1577 __set_bit(group - 1, nlk->groups);
1579 __clear_bit(group - 1, nlk->groups);
1580 netlink_update_subscriptions(&nlk->sk, subscriptions);
1581 netlink_update_listeners(&nlk->sk);
1584 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1585 char __user *optval, unsigned int optlen)
1587 struct sock *sk = sock->sk;
1588 struct netlink_sock *nlk = nlk_sk(sk);
1589 unsigned int val = 0;
1592 if (level != SOL_NETLINK)
1593 return -ENOPROTOOPT;
1595 if (optlen >= sizeof(int) &&
1596 get_user(val, (unsigned int __user *)optval))
1600 case NETLINK_PKTINFO:
1602 nlk->flags |= NETLINK_F_RECV_PKTINFO;
1604 nlk->flags &= ~NETLINK_F_RECV_PKTINFO;
1607 case NETLINK_ADD_MEMBERSHIP:
1608 case NETLINK_DROP_MEMBERSHIP: {
1609 if (!netlink_allowed(sock, NL_CFG_F_NONROOT_RECV))
1611 err = netlink_realloc_groups(sk);
1614 if (!val || val - 1 >= nlk->ngroups)
1616 if (optname == NETLINK_ADD_MEMBERSHIP && nlk->netlink_bind) {
1617 err = nlk->netlink_bind(sock_net(sk), val);
1621 netlink_table_grab();
1622 netlink_update_socket_mc(nlk, val,
1623 optname == NETLINK_ADD_MEMBERSHIP);
1624 netlink_table_ungrab();
1625 if (optname == NETLINK_DROP_MEMBERSHIP && nlk->netlink_unbind)
1626 nlk->netlink_unbind(sock_net(sk), val);
1631 case NETLINK_BROADCAST_ERROR:
1633 nlk->flags |= NETLINK_F_BROADCAST_SEND_ERROR;
1635 nlk->flags &= ~NETLINK_F_BROADCAST_SEND_ERROR;
1638 case NETLINK_NO_ENOBUFS:
1640 nlk->flags |= NETLINK_F_RECV_NO_ENOBUFS;
1641 clear_bit(NETLINK_S_CONGESTED, &nlk->state);
1642 wake_up_interruptible(&nlk->wait);
1644 nlk->flags &= ~NETLINK_F_RECV_NO_ENOBUFS;
1648 case NETLINK_LISTEN_ALL_NSID:
1649 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_BROADCAST))
1653 nlk->flags |= NETLINK_F_LISTEN_ALL_NSID;
1655 nlk->flags &= ~NETLINK_F_LISTEN_ALL_NSID;
1658 case NETLINK_CAP_ACK:
1660 nlk->flags |= NETLINK_F_CAP_ACK;
1662 nlk->flags &= ~NETLINK_F_CAP_ACK;
1665 case NETLINK_EXT_ACK:
1667 nlk->flags |= NETLINK_F_EXT_ACK;
1669 nlk->flags &= ~NETLINK_F_EXT_ACK;
1678 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1679 char __user *optval, int __user *optlen)
1681 struct sock *sk = sock->sk;
1682 struct netlink_sock *nlk = nlk_sk(sk);
1685 if (level != SOL_NETLINK)
1686 return -ENOPROTOOPT;
1688 if (get_user(len, optlen))
1694 case NETLINK_PKTINFO:
1695 if (len < sizeof(int))
1698 val = nlk->flags & NETLINK_F_RECV_PKTINFO ? 1 : 0;
1699 if (put_user(len, optlen) ||
1700 put_user(val, optval))
1704 case NETLINK_BROADCAST_ERROR:
1705 if (len < sizeof(int))
1708 val = nlk->flags & NETLINK_F_BROADCAST_SEND_ERROR ? 1 : 0;
1709 if (put_user(len, optlen) ||
1710 put_user(val, optval))
1714 case NETLINK_NO_ENOBUFS:
1715 if (len < sizeof(int))
1718 val = nlk->flags & NETLINK_F_RECV_NO_ENOBUFS ? 1 : 0;
1719 if (put_user(len, optlen) ||
1720 put_user(val, optval))
1724 case NETLINK_LIST_MEMBERSHIPS: {
1725 int pos, idx, shift;
1728 netlink_lock_table();
1729 for (pos = 0; pos * 8 < nlk->ngroups; pos += sizeof(u32)) {
1730 if (len - pos < sizeof(u32))
1733 idx = pos / sizeof(unsigned long);
1734 shift = (pos % sizeof(unsigned long)) * 8;
1735 if (put_user((u32)(nlk->groups[idx] >> shift),
1736 (u32 __user *)(optval + pos))) {
1741 if (put_user(ALIGN(nlk->ngroups / 8, sizeof(u32)), optlen))
1743 netlink_unlock_table();
1746 case NETLINK_CAP_ACK:
1747 if (len < sizeof(int))
1750 val = nlk->flags & NETLINK_F_CAP_ACK ? 1 : 0;
1751 if (put_user(len, optlen) ||
1752 put_user(val, optval))
1756 case NETLINK_EXT_ACK:
1757 if (len < sizeof(int))
1760 val = nlk->flags & NETLINK_F_EXT_ACK ? 1 : 0;
1761 if (put_user(len, optlen) || put_user(val, optval))
1771 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1773 struct nl_pktinfo info;
1775 info.group = NETLINK_CB(skb).dst_group;
1776 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1779 static void netlink_cmsg_listen_all_nsid(struct sock *sk, struct msghdr *msg,
1780 struct sk_buff *skb)
1782 if (!NETLINK_CB(skb).nsid_is_set)
1785 put_cmsg(msg, SOL_NETLINK, NETLINK_LISTEN_ALL_NSID, sizeof(int),
1786 &NETLINK_CB(skb).nsid);
1789 static int netlink_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1791 struct sock *sk = sock->sk;
1792 struct netlink_sock *nlk = nlk_sk(sk);
1793 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1796 struct sk_buff *skb;
1798 struct scm_cookie scm;
1799 u32 netlink_skb_flags = 0;
1801 if (msg->msg_flags&MSG_OOB)
1804 err = scm_send(sock, msg, &scm, true);
1808 if (msg->msg_namelen) {
1810 if (addr->nl_family != AF_NETLINK)
1812 dst_portid = addr->nl_pid;
1813 dst_group = ffs(addr->nl_groups);
1815 if ((dst_group || dst_portid) &&
1816 !netlink_allowed(sock, NL_CFG_F_NONROOT_SEND))
1818 netlink_skb_flags |= NETLINK_SKB_DST;
1820 dst_portid = nlk->dst_portid;
1821 dst_group = nlk->dst_group;
1825 err = netlink_autobind(sock);
1829 /* Ensure nlk is hashed and visible. */
1834 if (len > sk->sk_sndbuf - 32)
1837 skb = netlink_alloc_large_skb(len, dst_group);
1841 NETLINK_CB(skb).portid = nlk->portid;
1842 NETLINK_CB(skb).dst_group = dst_group;
1843 NETLINK_CB(skb).creds = scm.creds;
1844 NETLINK_CB(skb).flags = netlink_skb_flags;
1847 if (memcpy_from_msg(skb_put(skb, len), msg, len)) {
1852 err = security_netlink_send(sk, skb);
1859 refcount_inc(&skb->users);
1860 netlink_broadcast(sk, skb, dst_portid, dst_group, GFP_KERNEL);
1862 err = netlink_unicast(sk, skb, dst_portid, msg->msg_flags&MSG_DONTWAIT);
1869 static int netlink_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1872 struct scm_cookie scm;
1873 struct sock *sk = sock->sk;
1874 struct netlink_sock *nlk = nlk_sk(sk);
1875 int noblock = flags&MSG_DONTWAIT;
1877 struct sk_buff *skb, *data_skb;
1885 skb = skb_recv_datagram(sk, flags, noblock, &err);
1891 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1892 if (unlikely(skb_shinfo(skb)->frag_list)) {
1894 * If this skb has a frag_list, then here that means that we
1895 * will have to use the frag_list skb's data for compat tasks
1896 * and the regular skb's data for normal (non-compat) tasks.
1898 * If we need to send the compat skb, assign it to the
1899 * 'data_skb' variable so that it will be used below for data
1900 * copying. We keep 'skb' for everything else, including
1901 * freeing both later.
1903 if (flags & MSG_CMSG_COMPAT)
1904 data_skb = skb_shinfo(skb)->frag_list;
1908 /* Record the max length of recvmsg() calls for future allocations */
1909 nlk->max_recvmsg_len = max(nlk->max_recvmsg_len, len);
1910 nlk->max_recvmsg_len = min_t(size_t, nlk->max_recvmsg_len,
1911 SKB_WITH_OVERHEAD(32768));
1913 copied = data_skb->len;
1915 msg->msg_flags |= MSG_TRUNC;
1919 skb_reset_transport_header(data_skb);
1920 err = skb_copy_datagram_msg(data_skb, 0, msg, copied);
1922 if (msg->msg_name) {
1923 DECLARE_SOCKADDR(struct sockaddr_nl *, addr, msg->msg_name);
1924 addr->nl_family = AF_NETLINK;
1926 addr->nl_pid = NETLINK_CB(skb).portid;
1927 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1928 msg->msg_namelen = sizeof(*addr);
1931 if (nlk->flags & NETLINK_F_RECV_PKTINFO)
1932 netlink_cmsg_recv_pktinfo(msg, skb);
1933 if (nlk->flags & NETLINK_F_LISTEN_ALL_NSID)
1934 netlink_cmsg_listen_all_nsid(sk, msg, skb);
1936 memset(&scm, 0, sizeof(scm));
1937 scm.creds = *NETLINK_CREDS(skb);
1938 if (flags & MSG_TRUNC)
1939 copied = data_skb->len;
1941 skb_free_datagram(sk, skb);
1943 if (nlk->cb_running &&
1944 atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1945 ret = netlink_dump(sk);
1948 sk->sk_error_report(sk);
1952 scm_recv(sock, msg, &scm, flags);
1954 netlink_rcv_wake(sk);
1955 return err ? : copied;
1958 static void netlink_data_ready(struct sock *sk)
1964 * We export these functions to other modules. They provide a
1965 * complete set of kernel non-blocking support for message
1970 __netlink_kernel_create(struct net *net, int unit, struct module *module,
1971 struct netlink_kernel_cfg *cfg)
1973 struct socket *sock;
1975 struct netlink_sock *nlk;
1976 struct listeners *listeners = NULL;
1977 struct mutex *cb_mutex = cfg ? cfg->cb_mutex : NULL;
1978 unsigned int groups;
1982 if (unit < 0 || unit >= MAX_LINKS)
1985 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1988 if (__netlink_create(net, sock, cb_mutex, unit, 1) < 0)
1989 goto out_sock_release_nosk;
1993 if (!cfg || cfg->groups < 32)
1996 groups = cfg->groups;
1998 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2000 goto out_sock_release;
2002 sk->sk_data_ready = netlink_data_ready;
2003 if (cfg && cfg->input)
2004 nlk_sk(sk)->netlink_rcv = cfg->input;
2006 if (netlink_insert(sk, 0))
2007 goto out_sock_release;
2010 nlk->flags |= NETLINK_F_KERNEL_SOCKET;
2012 netlink_table_grab();
2013 if (!nl_table[unit].registered) {
2014 nl_table[unit].groups = groups;
2015 rcu_assign_pointer(nl_table[unit].listeners, listeners);
2016 nl_table[unit].cb_mutex = cb_mutex;
2017 nl_table[unit].module = module;
2019 nl_table[unit].bind = cfg->bind;
2020 nl_table[unit].unbind = cfg->unbind;
2021 nl_table[unit].flags = cfg->flags;
2023 nl_table[unit].compare = cfg->compare;
2025 nl_table[unit].registered = 1;
2028 nl_table[unit].registered++;
2030 netlink_table_ungrab();
2035 netlink_kernel_release(sk);
2038 out_sock_release_nosk:
2042 EXPORT_SYMBOL(__netlink_kernel_create);
2045 netlink_kernel_release(struct sock *sk)
2047 if (sk == NULL || sk->sk_socket == NULL)
2050 sock_release(sk->sk_socket);
2052 EXPORT_SYMBOL(netlink_kernel_release);
2054 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
2056 struct listeners *new, *old;
2057 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
2062 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
2063 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
2066 old = nl_deref_protected(tbl->listeners);
2067 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
2068 rcu_assign_pointer(tbl->listeners, new);
2070 kfree_rcu(old, rcu);
2072 tbl->groups = groups;
2078 * netlink_change_ngroups - change number of multicast groups
2080 * This changes the number of multicast groups that are available
2081 * on a certain netlink family. Note that it is not possible to
2082 * change the number of groups to below 32. Also note that it does
2083 * not implicitly call netlink_clear_multicast_users() when the
2084 * number of groups is reduced.
2086 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
2087 * @groups: The new number of groups.
2089 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
2093 netlink_table_grab();
2094 err = __netlink_change_ngroups(sk, groups);
2095 netlink_table_ungrab();
2100 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
2103 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
2105 sk_for_each_bound(sk, &tbl->mc_list)
2106 netlink_update_socket_mc(nlk_sk(sk), group, 0);
2110 __nlmsg_put(struct sk_buff *skb, u32 portid, u32 seq, int type, int len, int flags)
2112 struct nlmsghdr *nlh;
2113 int size = nlmsg_msg_size(len);
2115 nlh = skb_put(skb, NLMSG_ALIGN(size));
2116 nlh->nlmsg_type = type;
2117 nlh->nlmsg_len = size;
2118 nlh->nlmsg_flags = flags;
2119 nlh->nlmsg_pid = portid;
2120 nlh->nlmsg_seq = seq;
2121 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
2122 memset(nlmsg_data(nlh) + len, 0, NLMSG_ALIGN(size) - size);
2125 EXPORT_SYMBOL(__nlmsg_put);
2128 * It looks a bit ugly.
2129 * It would be better to create kernel thread.
2132 static int netlink_dump(struct sock *sk)
2134 struct netlink_sock *nlk = nlk_sk(sk);
2135 struct netlink_callback *cb;
2136 struct sk_buff *skb = NULL;
2137 struct nlmsghdr *nlh;
2138 struct module *module;
2139 int len, err = -ENOBUFS;
2143 mutex_lock(nlk->cb_mutex);
2144 if (!nlk->cb_running) {
2149 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2152 /* NLMSG_GOODSIZE is small to avoid high order allocations being
2153 * required, but it makes sense to _attempt_ a 16K bytes allocation
2154 * to reduce number of system calls on dump operations, if user
2155 * ever provided a big enough buffer.
2158 alloc_min_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
2160 if (alloc_min_size < nlk->max_recvmsg_len) {
2161 alloc_size = nlk->max_recvmsg_len;
2162 skb = alloc_skb(alloc_size,
2163 (GFP_KERNEL & ~__GFP_DIRECT_RECLAIM) |
2164 __GFP_NOWARN | __GFP_NORETRY);
2167 alloc_size = alloc_min_size;
2168 skb = alloc_skb(alloc_size, GFP_KERNEL);
2173 /* Trim skb to allocated size. User is expected to provide buffer as
2174 * large as max(min_dump_alloc, 16KiB (mac_recvmsg_len capped at
2175 * netlink_recvmsg())). dump will pack as many smaller messages as
2176 * could fit within the allocated skb. skb is typically allocated
2177 * with larger space than required (could be as much as near 2x the
2178 * requested size with align to next power of 2 approach). Allowing
2179 * dump to use the excess space makes it difficult for a user to have a
2180 * reasonable static buffer based on the expected largest dump of a
2181 * single netdev. The outcome is MSG_TRUNC error.
2183 skb_reserve(skb, skb_tailroom(skb) - alloc_size);
2184 netlink_skb_set_owner_r(skb, sk);
2186 len = cb->dump(skb, cb);
2189 mutex_unlock(nlk->cb_mutex);
2191 if (sk_filter(sk, skb))
2194 __netlink_sendskb(sk, skb);
2198 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
2202 nl_dump_check_consistent(cb, nlh);
2204 memcpy(nlmsg_data(nlh), &len, sizeof(len));
2206 if (sk_filter(sk, skb))
2209 __netlink_sendskb(sk, skb);
2214 nlk->cb_running = false;
2215 module = cb->module;
2217 mutex_unlock(nlk->cb_mutex);
2223 mutex_unlock(nlk->cb_mutex);
2228 int __netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
2229 const struct nlmsghdr *nlh,
2230 struct netlink_dump_control *control)
2232 struct netlink_callback *cb;
2234 struct netlink_sock *nlk;
2237 refcount_inc(&skb->users);
2239 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).portid);
2241 ret = -ECONNREFUSED;
2246 mutex_lock(nlk->cb_mutex);
2247 /* A dump is in progress... */
2248 if (nlk->cb_running) {
2252 /* add reference of module which cb->dump belongs to */
2253 if (!try_module_get(control->module)) {
2254 ret = -EPROTONOSUPPORT;
2259 memset(cb, 0, sizeof(*cb));
2260 cb->start = control->start;
2261 cb->dump = control->dump;
2262 cb->done = control->done;
2264 cb->data = control->data;
2265 cb->module = control->module;
2266 cb->min_dump_alloc = control->min_dump_alloc;
2270 ret = cb->start(cb);
2275 nlk->cb_running = true;
2277 mutex_unlock(nlk->cb_mutex);
2279 ret = netlink_dump(sk);
2286 /* We successfully started a dump, by returning -EINTR we
2287 * signal not to send ACK even if it was requested.
2293 mutex_unlock(nlk->cb_mutex);
2298 EXPORT_SYMBOL(__netlink_dump_start);
2300 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err,
2301 const struct netlink_ext_ack *extack)
2303 struct sk_buff *skb;
2304 struct nlmsghdr *rep;
2305 struct nlmsgerr *errmsg;
2306 size_t payload = sizeof(*errmsg);
2308 struct netlink_sock *nlk = nlk_sk(NETLINK_CB(in_skb).sk);
2309 unsigned int flags = 0;
2310 bool nlk_has_extack = nlk->flags & NETLINK_F_EXT_ACK;
2312 /* Error messages get the original request appened, unless the user
2313 * requests to cap the error message, and get extra error data if
2317 if (!(nlk->flags & NETLINK_F_CAP_ACK))
2318 payload += nlmsg_len(nlh);
2320 flags |= NLM_F_CAPPED;
2321 if (nlk_has_extack && extack) {
2323 tlvlen += nla_total_size(strlen(extack->_msg) + 1);
2324 if (extack->bad_attr)
2325 tlvlen += nla_total_size(sizeof(u32));
2328 flags |= NLM_F_CAPPED;
2330 if (nlk_has_extack && extack && extack->cookie_len)
2331 tlvlen += nla_total_size(extack->cookie_len);
2335 flags |= NLM_F_ACK_TLVS;
2337 skb = nlmsg_new(payload + tlvlen, GFP_KERNEL);
2339 NETLINK_CB(in_skb).sk->sk_err = ENOBUFS;
2340 NETLINK_CB(in_skb).sk->sk_error_report(NETLINK_CB(in_skb).sk);
2344 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).portid, nlh->nlmsg_seq,
2345 NLMSG_ERROR, payload, flags);
2346 errmsg = nlmsg_data(rep);
2347 errmsg->error = err;
2348 memcpy(&errmsg->msg, nlh, payload > sizeof(*errmsg) ? nlh->nlmsg_len : sizeof(*nlh));
2350 if (nlk_has_extack && extack) {
2353 WARN_ON(nla_put_string(skb, NLMSGERR_ATTR_MSG,
2355 if (extack->bad_attr &&
2356 !WARN_ON((u8 *)extack->bad_attr < in_skb->data ||
2357 (u8 *)extack->bad_attr >= in_skb->data +
2359 WARN_ON(nla_put_u32(skb, NLMSGERR_ATTR_OFFS,
2360 (u8 *)extack->bad_attr -
2363 if (extack->cookie_len)
2364 WARN_ON(nla_put(skb, NLMSGERR_ATTR_COOKIE,
2370 nlmsg_end(skb, rep);
2372 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).portid, MSG_DONTWAIT);
2374 EXPORT_SYMBOL(netlink_ack);
2376 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
2378 struct netlink_ext_ack *))
2380 struct netlink_ext_ack extack = {};
2381 struct nlmsghdr *nlh;
2384 while (skb->len >= nlmsg_total_size(0)) {
2387 nlh = nlmsg_hdr(skb);
2390 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
2393 /* Only requests are handled by the kernel */
2394 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
2397 /* Skip control messages */
2398 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
2401 err = cb(skb, nlh, &extack);
2406 if (nlh->nlmsg_flags & NLM_F_ACK || err)
2407 netlink_ack(skb, nlh, err, &extack);
2410 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
2411 if (msglen > skb->len)
2413 skb_pull(skb, msglen);
2418 EXPORT_SYMBOL(netlink_rcv_skb);
2421 * nlmsg_notify - send a notification netlink message
2422 * @sk: netlink socket to use
2423 * @skb: notification message
2424 * @portid: destination netlink portid for reports or 0
2425 * @group: destination multicast group or 0
2426 * @report: 1 to report back, 0 to disable
2427 * @flags: allocation flags
2429 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 portid,
2430 unsigned int group, int report, gfp_t flags)
2435 int exclude_portid = 0;
2438 refcount_inc(&skb->users);
2439 exclude_portid = portid;
2442 /* errors reported via destination sk->sk_err, but propagate
2443 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
2444 err = nlmsg_multicast(sk, skb, exclude_portid, group, flags);
2450 err2 = nlmsg_unicast(sk, skb, portid);
2451 if (!err || err == -ESRCH)
2457 EXPORT_SYMBOL(nlmsg_notify);
2459 #ifdef CONFIG_PROC_FS
2460 struct nl_seq_iter {
2461 struct seq_net_private p;
2462 struct rhashtable_iter hti;
2466 static int netlink_walk_start(struct nl_seq_iter *iter)
2470 err = rhashtable_walk_init(&nl_table[iter->link].hash, &iter->hti,
2473 iter->link = MAX_LINKS;
2477 err = rhashtable_walk_start(&iter->hti);
2478 return err == -EAGAIN ? 0 : err;
2481 static void netlink_walk_stop(struct nl_seq_iter *iter)
2483 rhashtable_walk_stop(&iter->hti);
2484 rhashtable_walk_exit(&iter->hti);
2487 static void *__netlink_seq_next(struct seq_file *seq)
2489 struct nl_seq_iter *iter = seq->private;
2490 struct netlink_sock *nlk;
2496 nlk = rhashtable_walk_next(&iter->hti);
2499 if (PTR_ERR(nlk) == -EAGAIN)
2508 netlink_walk_stop(iter);
2509 if (++iter->link >= MAX_LINKS)
2512 err = netlink_walk_start(iter);
2514 return ERR_PTR(err);
2516 } while (sock_net(&nlk->sk) != seq_file_net(seq));
2521 static void *netlink_seq_start(struct seq_file *seq, loff_t *posp)
2523 struct nl_seq_iter *iter = seq->private;
2524 void *obj = SEQ_START_TOKEN;
2530 err = netlink_walk_start(iter);
2532 return ERR_PTR(err);
2534 for (pos = *posp; pos && obj && !IS_ERR(obj); pos--)
2535 obj = __netlink_seq_next(seq);
2540 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2543 return __netlink_seq_next(seq);
2546 static void netlink_seq_stop(struct seq_file *seq, void *v)
2548 struct nl_seq_iter *iter = seq->private;
2550 if (iter->link >= MAX_LINKS)
2553 netlink_walk_stop(iter);
2557 static int netlink_seq_show(struct seq_file *seq, void *v)
2559 if (v == SEQ_START_TOKEN) {
2561 "sk Eth Pid Groups "
2562 "Rmem Wmem Dump Locks Drops Inode\n");
2565 struct netlink_sock *nlk = nlk_sk(s);
2567 seq_printf(seq, "%pK %-3d %-6u %08x %-8d %-8d %d %-8d %-8d %-8lu\n",
2571 nlk->groups ? (u32)nlk->groups[0] : 0,
2572 sk_rmem_alloc_get(s),
2573 sk_wmem_alloc_get(s),
2575 refcount_read(&s->sk_refcnt),
2576 atomic_read(&s->sk_drops),
2584 static const struct seq_operations netlink_seq_ops = {
2585 .start = netlink_seq_start,
2586 .next = netlink_seq_next,
2587 .stop = netlink_seq_stop,
2588 .show = netlink_seq_show,
2592 static int netlink_seq_open(struct inode *inode, struct file *file)
2594 return seq_open_net(inode, file, &netlink_seq_ops,
2595 sizeof(struct nl_seq_iter));
2598 static const struct file_operations netlink_seq_fops = {
2599 .owner = THIS_MODULE,
2600 .open = netlink_seq_open,
2602 .llseek = seq_lseek,
2603 .release = seq_release_net,
2608 int netlink_register_notifier(struct notifier_block *nb)
2610 return blocking_notifier_chain_register(&netlink_chain, nb);
2612 EXPORT_SYMBOL(netlink_register_notifier);
2614 int netlink_unregister_notifier(struct notifier_block *nb)
2616 return blocking_notifier_chain_unregister(&netlink_chain, nb);
2618 EXPORT_SYMBOL(netlink_unregister_notifier);
2620 static const struct proto_ops netlink_ops = {
2621 .family = PF_NETLINK,
2622 .owner = THIS_MODULE,
2623 .release = netlink_release,
2624 .bind = netlink_bind,
2625 .connect = netlink_connect,
2626 .socketpair = sock_no_socketpair,
2627 .accept = sock_no_accept,
2628 .getname = netlink_getname,
2629 .poll = datagram_poll,
2630 .ioctl = netlink_ioctl,
2631 .listen = sock_no_listen,
2632 .shutdown = sock_no_shutdown,
2633 .setsockopt = netlink_setsockopt,
2634 .getsockopt = netlink_getsockopt,
2635 .sendmsg = netlink_sendmsg,
2636 .recvmsg = netlink_recvmsg,
2637 .mmap = sock_no_mmap,
2638 .sendpage = sock_no_sendpage,
2641 static const struct net_proto_family netlink_family_ops = {
2642 .family = PF_NETLINK,
2643 .create = netlink_create,
2644 .owner = THIS_MODULE, /* for consistency 8) */
2647 static int __net_init netlink_net_init(struct net *net)
2649 #ifdef CONFIG_PROC_FS
2650 if (!proc_create("netlink", 0, net->proc_net, &netlink_seq_fops))
2656 static void __net_exit netlink_net_exit(struct net *net)
2658 #ifdef CONFIG_PROC_FS
2659 remove_proc_entry("netlink", net->proc_net);
2663 static void __init netlink_add_usersock_entry(void)
2665 struct listeners *listeners;
2668 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2670 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2672 netlink_table_grab();
2674 nl_table[NETLINK_USERSOCK].groups = groups;
2675 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2676 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2677 nl_table[NETLINK_USERSOCK].registered = 1;
2678 nl_table[NETLINK_USERSOCK].flags = NL_CFG_F_NONROOT_SEND;
2680 netlink_table_ungrab();
2683 static struct pernet_operations __net_initdata netlink_net_ops = {
2684 .init = netlink_net_init,
2685 .exit = netlink_net_exit,
2688 static inline u32 netlink_hash(const void *data, u32 len, u32 seed)
2690 const struct netlink_sock *nlk = data;
2691 struct netlink_compare_arg arg;
2693 netlink_compare_arg_init(&arg, sock_net(&nlk->sk), nlk->portid);
2694 return jhash2((u32 *)&arg, netlink_compare_arg_len / sizeof(u32), seed);
2697 static const struct rhashtable_params netlink_rhashtable_params = {
2698 .head_offset = offsetof(struct netlink_sock, node),
2699 .key_len = netlink_compare_arg_len,
2700 .obj_hashfn = netlink_hash,
2701 .obj_cmpfn = netlink_compare,
2702 .automatic_shrinking = true,
2705 static int __init netlink_proto_init(void)
2708 int err = proto_register(&netlink_proto, 0);
2713 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > FIELD_SIZEOF(struct sk_buff, cb));
2715 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2719 for (i = 0; i < MAX_LINKS; i++) {
2720 if (rhashtable_init(&nl_table[i].hash,
2721 &netlink_rhashtable_params) < 0) {
2723 rhashtable_destroy(&nl_table[i].hash);
2729 INIT_LIST_HEAD(&netlink_tap_all);
2731 netlink_add_usersock_entry();
2733 sock_register(&netlink_family_ops);
2734 register_pernet_subsys(&netlink_net_ops);
2735 /* The netlink device handler may be needed early. */
2740 panic("netlink_init: Cannot allocate nl_table\n");
2743 core_initcall(netlink_proto_init);