Xilinx: ARM: Update Zynq defconfig to enable USB errata.
[linux-2.6-xlnx.git] / net / netlink / af_netlink.c
blobb3025a603d56b21f1529262025cc2d6f00993cc9
1 /*
2 * NETLINK Kernel-user communication protocol.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
12 * Tue Jun 26 14:36:48 MEST 2001 Herbert "herp" Rosmanith
13 * added netlink_proto_exit
14 * Tue Jan 22 18:32:44 BRST 2002 Arnaldo C. de Melo <acme@conectiva.com.br>
15 * use nlk_sk, as sk->protinfo is on a diet 8)
16 * Fri Jul 22 19:51:12 MEST 2005 Harald Welte <laforge@gnumonks.org>
17 * - inc module use count of module that owns
18 * the kernel socket in case userspace opens
19 * socket of same protocol
20 * - remove all module support, since netlink is
21 * mandatory if CONFIG_NET=y these days
24 #include <linux/module.h>
26 #include <linux/capability.h>
27 #include <linux/kernel.h>
28 #include <linux/init.h>
29 #include <linux/signal.h>
30 #include <linux/sched.h>
31 #include <linux/errno.h>
32 #include <linux/string.h>
33 #include <linux/stat.h>
34 #include <linux/socket.h>
35 #include <linux/un.h>
36 #include <linux/fcntl.h>
37 #include <linux/termios.h>
38 #include <linux/sockios.h>
39 #include <linux/net.h>
40 #include <linux/fs.h>
41 #include <linux/slab.h>
42 #include <asm/uaccess.h>
43 #include <linux/skbuff.h>
44 #include <linux/netdevice.h>
45 #include <linux/rtnetlink.h>
46 #include <linux/proc_fs.h>
47 #include <linux/seq_file.h>
48 #include <linux/notifier.h>
49 #include <linux/security.h>
50 #include <linux/jhash.h>
51 #include <linux/jiffies.h>
52 #include <linux/random.h>
53 #include <linux/bitops.h>
54 #include <linux/mm.h>
55 #include <linux/types.h>
56 #include <linux/audit.h>
57 #include <linux/mutex.h>
59 #include <net/net_namespace.h>
60 #include <net/sock.h>
61 #include <net/scm.h>
62 #include <net/netlink.h>
64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
65 #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
67 struct netlink_sock {
68 /* struct sock has to be the first member of netlink_sock */
69 struct sock sk;
70 u32 pid;
71 u32 dst_pid;
72 u32 dst_group;
73 u32 flags;
74 u32 subscriptions;
75 u32 ngroups;
76 unsigned long *groups;
77 unsigned long state;
78 wait_queue_head_t wait;
79 struct netlink_callback *cb;
80 struct mutex *cb_mutex;
81 struct mutex cb_def_mutex;
82 void (*netlink_rcv)(struct sk_buff *skb);
83 struct module *module;
86 struct listeners {
87 struct rcu_head rcu;
88 unsigned long masks[0];
91 #define NETLINK_KERNEL_SOCKET 0x1
92 #define NETLINK_RECV_PKTINFO 0x2
93 #define NETLINK_BROADCAST_SEND_ERROR 0x4
94 #define NETLINK_RECV_NO_ENOBUFS 0x8
96 static inline struct netlink_sock *nlk_sk(struct sock *sk)
98 return container_of(sk, struct netlink_sock, sk);
101 static inline int netlink_is_kernel(struct sock *sk)
103 return nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET;
106 struct nl_pid_hash {
107 struct hlist_head *table;
108 unsigned long rehash_time;
110 unsigned int mask;
111 unsigned int shift;
113 unsigned int entries;
114 unsigned int max_shift;
116 u32 rnd;
119 struct netlink_table {
120 struct nl_pid_hash hash;
121 struct hlist_head mc_list;
122 struct listeners __rcu *listeners;
123 unsigned int nl_nonroot;
124 unsigned int groups;
125 struct mutex *cb_mutex;
126 struct module *module;
127 int registered;
130 static struct netlink_table *nl_table;
132 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
134 static int netlink_dump(struct sock *sk);
136 static DEFINE_RWLOCK(nl_table_lock);
137 static atomic_t nl_table_users = ATOMIC_INIT(0);
139 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
141 static inline u32 netlink_group_mask(u32 group)
143 return group ? 1 << (group - 1) : 0;
146 static inline struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
148 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
151 static void netlink_destroy_callback(struct netlink_callback *cb)
153 kfree_skb(cb->skb);
154 kfree(cb);
157 static void netlink_consume_callback(struct netlink_callback *cb)
159 consume_skb(cb->skb);
160 kfree(cb);
163 static void netlink_sock_destruct(struct sock *sk)
165 struct netlink_sock *nlk = nlk_sk(sk);
167 if (nlk->cb) {
168 if (nlk->cb->done)
169 nlk->cb->done(nlk->cb);
170 netlink_destroy_callback(nlk->cb);
173 skb_queue_purge(&sk->sk_receive_queue);
175 if (!sock_flag(sk, SOCK_DEAD)) {
176 printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
177 return;
180 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
181 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
182 WARN_ON(nlk_sk(sk)->groups);
185 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on
186 * SMP. Look, when several writers sleep and reader wakes them up, all but one
187 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
188 * this, _but_ remember, it adds useless work on UP machines.
191 void netlink_table_grab(void)
192 __acquires(nl_table_lock)
194 might_sleep();
196 write_lock_irq(&nl_table_lock);
198 if (atomic_read(&nl_table_users)) {
199 DECLARE_WAITQUEUE(wait, current);
201 add_wait_queue_exclusive(&nl_table_wait, &wait);
202 for (;;) {
203 set_current_state(TASK_UNINTERRUPTIBLE);
204 if (atomic_read(&nl_table_users) == 0)
205 break;
206 write_unlock_irq(&nl_table_lock);
207 schedule();
208 write_lock_irq(&nl_table_lock);
211 __set_current_state(TASK_RUNNING);
212 remove_wait_queue(&nl_table_wait, &wait);
216 void netlink_table_ungrab(void)
217 __releases(nl_table_lock)
219 write_unlock_irq(&nl_table_lock);
220 wake_up(&nl_table_wait);
223 static inline void
224 netlink_lock_table(void)
226 /* read_lock() synchronizes us to netlink_table_grab */
228 read_lock(&nl_table_lock);
229 atomic_inc(&nl_table_users);
230 read_unlock(&nl_table_lock);
233 static inline void
234 netlink_unlock_table(void)
236 if (atomic_dec_and_test(&nl_table_users))
237 wake_up(&nl_table_wait);
240 static struct sock *netlink_lookup(struct net *net, int protocol, u32 pid)
242 struct nl_pid_hash *hash = &nl_table[protocol].hash;
243 struct hlist_head *head;
244 struct sock *sk;
245 struct hlist_node *node;
247 read_lock(&nl_table_lock);
248 head = nl_pid_hashfn(hash, pid);
249 sk_for_each(sk, node, head) {
250 if (net_eq(sock_net(sk), net) && (nlk_sk(sk)->pid == pid)) {
251 sock_hold(sk);
252 goto found;
255 sk = NULL;
256 found:
257 read_unlock(&nl_table_lock);
258 return sk;
261 static struct hlist_head *nl_pid_hash_zalloc(size_t size)
263 if (size <= PAGE_SIZE)
264 return kzalloc(size, GFP_ATOMIC);
265 else
266 return (struct hlist_head *)
267 __get_free_pages(GFP_ATOMIC | __GFP_ZERO,
268 get_order(size));
271 static void nl_pid_hash_free(struct hlist_head *table, size_t size)
273 if (size <= PAGE_SIZE)
274 kfree(table);
275 else
276 free_pages((unsigned long)table, get_order(size));
279 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
281 unsigned int omask, mask, shift;
282 size_t osize, size;
283 struct hlist_head *otable, *table;
284 int i;
286 omask = mask = hash->mask;
287 osize = size = (mask + 1) * sizeof(*table);
288 shift = hash->shift;
290 if (grow) {
291 if (++shift > hash->max_shift)
292 return 0;
293 mask = mask * 2 + 1;
294 size *= 2;
297 table = nl_pid_hash_zalloc(size);
298 if (!table)
299 return 0;
301 otable = hash->table;
302 hash->table = table;
303 hash->mask = mask;
304 hash->shift = shift;
305 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
307 for (i = 0; i <= omask; i++) {
308 struct sock *sk;
309 struct hlist_node *node, *tmp;
311 sk_for_each_safe(sk, node, tmp, &otable[i])
312 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
315 nl_pid_hash_free(otable, osize);
316 hash->rehash_time = jiffies + 10 * 60 * HZ;
317 return 1;
320 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
322 int avg = hash->entries >> hash->shift;
324 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
325 return 1;
327 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
328 nl_pid_hash_rehash(hash, 0);
329 return 1;
332 return 0;
335 static const struct proto_ops netlink_ops;
337 static void
338 netlink_update_listeners(struct sock *sk)
340 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
341 struct hlist_node *node;
342 unsigned long mask;
343 unsigned int i;
345 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
346 mask = 0;
347 sk_for_each_bound(sk, node, &tbl->mc_list) {
348 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
349 mask |= nlk_sk(sk)->groups[i];
351 tbl->listeners->masks[i] = mask;
353 /* this function is only called with the netlink table "grabbed", which
354 * makes sure updates are visible before bind or setsockopt return. */
357 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
359 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
360 struct hlist_head *head;
361 int err = -EADDRINUSE;
362 struct sock *osk;
363 struct hlist_node *node;
364 int len;
366 netlink_table_grab();
367 head = nl_pid_hashfn(hash, pid);
368 len = 0;
369 sk_for_each(osk, node, head) {
370 if (net_eq(sock_net(osk), net) && (nlk_sk(osk)->pid == pid))
371 break;
372 len++;
374 if (node)
375 goto err;
377 err = -EBUSY;
378 if (nlk_sk(sk)->pid)
379 goto err;
381 err = -ENOMEM;
382 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
383 goto err;
385 if (len && nl_pid_hash_dilute(hash, len))
386 head = nl_pid_hashfn(hash, pid);
387 hash->entries++;
388 nlk_sk(sk)->pid = pid;
389 sk_add_node(sk, head);
390 err = 0;
392 err:
393 netlink_table_ungrab();
394 return err;
397 static void netlink_remove(struct sock *sk)
399 netlink_table_grab();
400 if (sk_del_node_init(sk))
401 nl_table[sk->sk_protocol].hash.entries--;
402 if (nlk_sk(sk)->subscriptions)
403 __sk_del_bind_node(sk);
404 netlink_table_ungrab();
407 static struct proto netlink_proto = {
408 .name = "NETLINK",
409 .owner = THIS_MODULE,
410 .obj_size = sizeof(struct netlink_sock),
413 static int __netlink_create(struct net *net, struct socket *sock,
414 struct mutex *cb_mutex, int protocol)
416 struct sock *sk;
417 struct netlink_sock *nlk;
419 sock->ops = &netlink_ops;
421 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto);
422 if (!sk)
423 return -ENOMEM;
425 sock_init_data(sock, sk);
427 nlk = nlk_sk(sk);
428 if (cb_mutex) {
429 nlk->cb_mutex = cb_mutex;
430 } else {
431 nlk->cb_mutex = &nlk->cb_def_mutex;
432 mutex_init(nlk->cb_mutex);
434 init_waitqueue_head(&nlk->wait);
436 sk->sk_destruct = netlink_sock_destruct;
437 sk->sk_protocol = protocol;
438 return 0;
441 static int netlink_create(struct net *net, struct socket *sock, int protocol,
442 int kern)
444 struct module *module = NULL;
445 struct mutex *cb_mutex;
446 struct netlink_sock *nlk;
447 int err = 0;
449 sock->state = SS_UNCONNECTED;
451 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
452 return -ESOCKTNOSUPPORT;
454 if (protocol < 0 || protocol >= MAX_LINKS)
455 return -EPROTONOSUPPORT;
457 netlink_lock_table();
458 #ifdef CONFIG_MODULES
459 if (!nl_table[protocol].registered) {
460 netlink_unlock_table();
461 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
462 netlink_lock_table();
464 #endif
465 if (nl_table[protocol].registered &&
466 try_module_get(nl_table[protocol].module))
467 module = nl_table[protocol].module;
468 else
469 err = -EPROTONOSUPPORT;
470 cb_mutex = nl_table[protocol].cb_mutex;
471 netlink_unlock_table();
473 if (err < 0)
474 goto out;
476 err = __netlink_create(net, sock, cb_mutex, protocol);
477 if (err < 0)
478 goto out_module;
480 local_bh_disable();
481 sock_prot_inuse_add(net, &netlink_proto, 1);
482 local_bh_enable();
484 nlk = nlk_sk(sock->sk);
485 nlk->module = module;
486 out:
487 return err;
489 out_module:
490 module_put(module);
491 goto out;
494 static int netlink_release(struct socket *sock)
496 struct sock *sk = sock->sk;
497 struct netlink_sock *nlk;
499 if (!sk)
500 return 0;
502 netlink_remove(sk);
503 sock_orphan(sk);
504 nlk = nlk_sk(sk);
507 * OK. Socket is unlinked, any packets that arrive now
508 * will be purged.
511 sock->sk = NULL;
512 wake_up_interruptible_all(&nlk->wait);
514 skb_queue_purge(&sk->sk_write_queue);
516 if (nlk->pid) {
517 struct netlink_notify n = {
518 .net = sock_net(sk),
519 .protocol = sk->sk_protocol,
520 .pid = nlk->pid,
522 atomic_notifier_call_chain(&netlink_chain,
523 NETLINK_URELEASE, &n);
526 module_put(nlk->module);
528 netlink_table_grab();
529 if (netlink_is_kernel(sk)) {
530 BUG_ON(nl_table[sk->sk_protocol].registered == 0);
531 if (--nl_table[sk->sk_protocol].registered == 0) {
532 kfree(nl_table[sk->sk_protocol].listeners);
533 nl_table[sk->sk_protocol].module = NULL;
534 nl_table[sk->sk_protocol].registered = 0;
536 } else if (nlk->subscriptions) {
537 netlink_update_listeners(sk);
539 netlink_table_ungrab();
541 kfree(nlk->groups);
542 nlk->groups = NULL;
544 local_bh_disable();
545 sock_prot_inuse_add(sock_net(sk), &netlink_proto, -1);
546 local_bh_enable();
547 sock_put(sk);
548 return 0;
551 static int netlink_autobind(struct socket *sock)
553 struct sock *sk = sock->sk;
554 struct net *net = sock_net(sk);
555 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
556 struct hlist_head *head;
557 struct sock *osk;
558 struct hlist_node *node;
559 s32 pid = task_tgid_vnr(current);
560 int err;
561 static s32 rover = -4097;
563 retry:
564 cond_resched();
565 netlink_table_grab();
566 head = nl_pid_hashfn(hash, pid);
567 sk_for_each(osk, node, head) {
568 if (!net_eq(sock_net(osk), net))
569 continue;
570 if (nlk_sk(osk)->pid == pid) {
571 /* Bind collision, search negative pid values. */
572 pid = rover--;
573 if (rover > -4097)
574 rover = -4097;
575 netlink_table_ungrab();
576 goto retry;
579 netlink_table_ungrab();
581 err = netlink_insert(sk, net, pid);
582 if (err == -EADDRINUSE)
583 goto retry;
585 /* If 2 threads race to autobind, that is fine. */
586 if (err == -EBUSY)
587 err = 0;
589 return err;
592 static inline int netlink_capable(const struct socket *sock, unsigned int flag)
594 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
595 capable(CAP_NET_ADMIN);
598 static void
599 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
601 struct netlink_sock *nlk = nlk_sk(sk);
603 if (nlk->subscriptions && !subscriptions)
604 __sk_del_bind_node(sk);
605 else if (!nlk->subscriptions && subscriptions)
606 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
607 nlk->subscriptions = subscriptions;
610 static int netlink_realloc_groups(struct sock *sk)
612 struct netlink_sock *nlk = nlk_sk(sk);
613 unsigned int groups;
614 unsigned long *new_groups;
615 int err = 0;
617 netlink_table_grab();
619 groups = nl_table[sk->sk_protocol].groups;
620 if (!nl_table[sk->sk_protocol].registered) {
621 err = -ENOENT;
622 goto out_unlock;
625 if (nlk->ngroups >= groups)
626 goto out_unlock;
628 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
629 if (new_groups == NULL) {
630 err = -ENOMEM;
631 goto out_unlock;
633 memset((char *)new_groups + NLGRPSZ(nlk->ngroups), 0,
634 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
636 nlk->groups = new_groups;
637 nlk->ngroups = groups;
638 out_unlock:
639 netlink_table_ungrab();
640 return err;
643 static int netlink_bind(struct socket *sock, struct sockaddr *addr,
644 int addr_len)
646 struct sock *sk = sock->sk;
647 struct net *net = sock_net(sk);
648 struct netlink_sock *nlk = nlk_sk(sk);
649 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
650 int err;
652 if (nladdr->nl_family != AF_NETLINK)
653 return -EINVAL;
655 /* Only superuser is allowed to listen multicasts */
656 if (nladdr->nl_groups) {
657 if (!netlink_capable(sock, NL_NONROOT_RECV))
658 return -EPERM;
659 err = netlink_realloc_groups(sk);
660 if (err)
661 return err;
664 if (nlk->pid) {
665 if (nladdr->nl_pid != nlk->pid)
666 return -EINVAL;
667 } else {
668 err = nladdr->nl_pid ?
669 netlink_insert(sk, net, nladdr->nl_pid) :
670 netlink_autobind(sock);
671 if (err)
672 return err;
675 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
676 return 0;
678 netlink_table_grab();
679 netlink_update_subscriptions(sk, nlk->subscriptions +
680 hweight32(nladdr->nl_groups) -
681 hweight32(nlk->groups[0]));
682 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
683 netlink_update_listeners(sk);
684 netlink_table_ungrab();
686 return 0;
689 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
690 int alen, int flags)
692 int err = 0;
693 struct sock *sk = sock->sk;
694 struct netlink_sock *nlk = nlk_sk(sk);
695 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
697 if (alen < sizeof(addr->sa_family))
698 return -EINVAL;
700 if (addr->sa_family == AF_UNSPEC) {
701 sk->sk_state = NETLINK_UNCONNECTED;
702 nlk->dst_pid = 0;
703 nlk->dst_group = 0;
704 return 0;
706 if (addr->sa_family != AF_NETLINK)
707 return -EINVAL;
709 /* Only superuser is allowed to send multicasts */
710 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
711 return -EPERM;
713 if (!nlk->pid)
714 err = netlink_autobind(sock);
716 if (err == 0) {
717 sk->sk_state = NETLINK_CONNECTED;
718 nlk->dst_pid = nladdr->nl_pid;
719 nlk->dst_group = ffs(nladdr->nl_groups);
722 return err;
725 static int netlink_getname(struct socket *sock, struct sockaddr *addr,
726 int *addr_len, int peer)
728 struct sock *sk = sock->sk;
729 struct netlink_sock *nlk = nlk_sk(sk);
730 DECLARE_SOCKADDR(struct sockaddr_nl *, nladdr, addr);
732 nladdr->nl_family = AF_NETLINK;
733 nladdr->nl_pad = 0;
734 *addr_len = sizeof(*nladdr);
736 if (peer) {
737 nladdr->nl_pid = nlk->dst_pid;
738 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
739 } else {
740 nladdr->nl_pid = nlk->pid;
741 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
743 return 0;
746 static void netlink_overrun(struct sock *sk)
748 struct netlink_sock *nlk = nlk_sk(sk);
750 if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
751 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
752 sk->sk_err = ENOBUFS;
753 sk->sk_error_report(sk);
756 atomic_inc(&sk->sk_drops);
759 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
761 struct sock *sock;
762 struct netlink_sock *nlk;
764 sock = netlink_lookup(sock_net(ssk), ssk->sk_protocol, pid);
765 if (!sock)
766 return ERR_PTR(-ECONNREFUSED);
768 /* Don't bother queuing skb if kernel socket has no input function */
769 nlk = nlk_sk(sock);
770 if (sock->sk_state == NETLINK_CONNECTED &&
771 nlk->dst_pid != nlk_sk(ssk)->pid) {
772 sock_put(sock);
773 return ERR_PTR(-ECONNREFUSED);
775 return sock;
778 struct sock *netlink_getsockbyfilp(struct file *filp)
780 struct inode *inode = filp->f_path.dentry->d_inode;
781 struct sock *sock;
783 if (!S_ISSOCK(inode->i_mode))
784 return ERR_PTR(-ENOTSOCK);
786 sock = SOCKET_I(inode)->sk;
787 if (sock->sk_family != AF_NETLINK)
788 return ERR_PTR(-EINVAL);
790 sock_hold(sock);
791 return sock;
795 * Attach a skb to a netlink socket.
796 * The caller must hold a reference to the destination socket. On error, the
797 * reference is dropped. The skb is not send to the destination, just all
798 * all error checks are performed and memory in the queue is reserved.
799 * Return values:
800 * < 0: error. skb freed, reference to sock dropped.
801 * 0: continue
802 * 1: repeat lookup - reference dropped while waiting for socket memory.
804 int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
805 long *timeo, struct sock *ssk)
807 struct netlink_sock *nlk;
809 nlk = nlk_sk(sk);
811 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
812 test_bit(0, &nlk->state)) {
813 DECLARE_WAITQUEUE(wait, current);
814 if (!*timeo) {
815 if (!ssk || netlink_is_kernel(ssk))
816 netlink_overrun(sk);
817 sock_put(sk);
818 kfree_skb(skb);
819 return -EAGAIN;
822 __set_current_state(TASK_INTERRUPTIBLE);
823 add_wait_queue(&nlk->wait, &wait);
825 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
826 test_bit(0, &nlk->state)) &&
827 !sock_flag(sk, SOCK_DEAD))
828 *timeo = schedule_timeout(*timeo);
830 __set_current_state(TASK_RUNNING);
831 remove_wait_queue(&nlk->wait, &wait);
832 sock_put(sk);
834 if (signal_pending(current)) {
835 kfree_skb(skb);
836 return sock_intr_errno(*timeo);
838 return 1;
840 skb_set_owner_r(skb, sk);
841 return 0;
844 static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
846 int len = skb->len;
848 skb_queue_tail(&sk->sk_receive_queue, skb);
849 sk->sk_data_ready(sk, len);
850 return len;
853 int netlink_sendskb(struct sock *sk, struct sk_buff *skb)
855 int len = __netlink_sendskb(sk, skb);
857 sock_put(sk);
858 return len;
861 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
863 kfree_skb(skb);
864 sock_put(sk);
867 static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
869 int delta;
871 skb_orphan(skb);
873 delta = skb->end - skb->tail;
874 if (delta * 2 < skb->truesize)
875 return skb;
877 if (skb_shared(skb)) {
878 struct sk_buff *nskb = skb_clone(skb, allocation);
879 if (!nskb)
880 return skb;
881 consume_skb(skb);
882 skb = nskb;
885 if (!pskb_expand_head(skb, 0, -delta, allocation))
886 skb->truesize -= delta;
888 return skb;
891 static void netlink_rcv_wake(struct sock *sk)
893 struct netlink_sock *nlk = nlk_sk(sk);
895 if (skb_queue_empty(&sk->sk_receive_queue))
896 clear_bit(0, &nlk->state);
897 if (!test_bit(0, &nlk->state))
898 wake_up_interruptible(&nlk->wait);
901 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
903 int ret;
904 struct netlink_sock *nlk = nlk_sk(sk);
906 ret = -ECONNREFUSED;
907 if (nlk->netlink_rcv != NULL) {
908 ret = skb->len;
909 skb_set_owner_r(skb, sk);
910 nlk->netlink_rcv(skb);
911 consume_skb(skb);
912 } else {
913 kfree_skb(skb);
915 sock_put(sk);
916 return ret;
919 int netlink_unicast(struct sock *ssk, struct sk_buff *skb,
920 u32 pid, int nonblock)
922 struct sock *sk;
923 int err;
924 long timeo;
926 skb = netlink_trim(skb, gfp_any());
928 timeo = sock_sndtimeo(ssk, nonblock);
929 retry:
930 sk = netlink_getsockbypid(ssk, pid);
931 if (IS_ERR(sk)) {
932 kfree_skb(skb);
933 return PTR_ERR(sk);
935 if (netlink_is_kernel(sk))
936 return netlink_unicast_kernel(sk, skb);
938 if (sk_filter(sk, skb)) {
939 err = skb->len;
940 kfree_skb(skb);
941 sock_put(sk);
942 return err;
945 err = netlink_attachskb(sk, skb, &timeo, ssk);
946 if (err == 1)
947 goto retry;
948 if (err)
949 return err;
951 return netlink_sendskb(sk, skb);
953 EXPORT_SYMBOL(netlink_unicast);
955 int netlink_has_listeners(struct sock *sk, unsigned int group)
957 int res = 0;
958 struct listeners *listeners;
960 BUG_ON(!netlink_is_kernel(sk));
962 rcu_read_lock();
963 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
965 if (group - 1 < nl_table[sk->sk_protocol].groups)
966 res = test_bit(group - 1, listeners->masks);
968 rcu_read_unlock();
970 return res;
972 EXPORT_SYMBOL_GPL(netlink_has_listeners);
974 static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
976 struct netlink_sock *nlk = nlk_sk(sk);
978 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
979 !test_bit(0, &nlk->state)) {
980 skb_set_owner_r(skb, sk);
981 __netlink_sendskb(sk, skb);
982 return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
984 return -1;
987 struct netlink_broadcast_data {
988 struct sock *exclude_sk;
989 struct net *net;
990 u32 pid;
991 u32 group;
992 int failure;
993 int delivery_failure;
994 int congested;
995 int delivered;
996 gfp_t allocation;
997 struct sk_buff *skb, *skb2;
998 int (*tx_filter)(struct sock *dsk, struct sk_buff *skb, void *data);
999 void *tx_data;
1002 static int do_one_broadcast(struct sock *sk,
1003 struct netlink_broadcast_data *p)
1005 struct netlink_sock *nlk = nlk_sk(sk);
1006 int val;
1008 if (p->exclude_sk == sk)
1009 goto out;
1011 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1012 !test_bit(p->group - 1, nlk->groups))
1013 goto out;
1015 if (!net_eq(sock_net(sk), p->net))
1016 goto out;
1018 if (p->failure) {
1019 netlink_overrun(sk);
1020 goto out;
1023 sock_hold(sk);
1024 if (p->skb2 == NULL) {
1025 if (skb_shared(p->skb)) {
1026 p->skb2 = skb_clone(p->skb, p->allocation);
1027 } else {
1028 p->skb2 = skb_get(p->skb);
1030 * skb ownership may have been set when
1031 * delivered to a previous socket.
1033 skb_orphan(p->skb2);
1036 if (p->skb2 == NULL) {
1037 netlink_overrun(sk);
1038 /* Clone failed. Notify ALL listeners. */
1039 p->failure = 1;
1040 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1041 p->delivery_failure = 1;
1042 } else if (p->tx_filter && p->tx_filter(sk, p->skb2, p->tx_data)) {
1043 kfree_skb(p->skb2);
1044 p->skb2 = NULL;
1045 } else if (sk_filter(sk, p->skb2)) {
1046 kfree_skb(p->skb2);
1047 p->skb2 = NULL;
1048 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
1049 netlink_overrun(sk);
1050 if (nlk->flags & NETLINK_BROADCAST_SEND_ERROR)
1051 p->delivery_failure = 1;
1052 } else {
1053 p->congested |= val;
1054 p->delivered = 1;
1055 p->skb2 = NULL;
1057 sock_put(sk);
1059 out:
1060 return 0;
1063 int netlink_broadcast_filtered(struct sock *ssk, struct sk_buff *skb, u32 pid,
1064 u32 group, gfp_t allocation,
1065 int (*filter)(struct sock *dsk, struct sk_buff *skb, void *data),
1066 void *filter_data)
1068 struct net *net = sock_net(ssk);
1069 struct netlink_broadcast_data info;
1070 struct hlist_node *node;
1071 struct sock *sk;
1073 skb = netlink_trim(skb, allocation);
1075 info.exclude_sk = ssk;
1076 info.net = net;
1077 info.pid = pid;
1078 info.group = group;
1079 info.failure = 0;
1080 info.delivery_failure = 0;
1081 info.congested = 0;
1082 info.delivered = 0;
1083 info.allocation = allocation;
1084 info.skb = skb;
1085 info.skb2 = NULL;
1086 info.tx_filter = filter;
1087 info.tx_data = filter_data;
1089 /* While we sleep in clone, do not allow to change socket list */
1091 netlink_lock_table();
1093 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1094 do_one_broadcast(sk, &info);
1096 consume_skb(skb);
1098 netlink_unlock_table();
1100 if (info.delivery_failure) {
1101 kfree_skb(info.skb2);
1102 return -ENOBUFS;
1104 consume_skb(info.skb2);
1106 if (info.delivered) {
1107 if (info.congested && (allocation & __GFP_WAIT))
1108 yield();
1109 return 0;
1111 return -ESRCH;
1113 EXPORT_SYMBOL(netlink_broadcast_filtered);
1115 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
1116 u32 group, gfp_t allocation)
1118 return netlink_broadcast_filtered(ssk, skb, pid, group, allocation,
1119 NULL, NULL);
1121 EXPORT_SYMBOL(netlink_broadcast);
1123 struct netlink_set_err_data {
1124 struct sock *exclude_sk;
1125 u32 pid;
1126 u32 group;
1127 int code;
1130 static int do_one_set_err(struct sock *sk, struct netlink_set_err_data *p)
1132 struct netlink_sock *nlk = nlk_sk(sk);
1133 int ret = 0;
1135 if (sk == p->exclude_sk)
1136 goto out;
1138 if (!net_eq(sock_net(sk), sock_net(p->exclude_sk)))
1139 goto out;
1141 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1142 !test_bit(p->group - 1, nlk->groups))
1143 goto out;
1145 if (p->code == ENOBUFS && nlk->flags & NETLINK_RECV_NO_ENOBUFS) {
1146 ret = 1;
1147 goto out;
1150 sk->sk_err = p->code;
1151 sk->sk_error_report(sk);
1152 out:
1153 return ret;
1157 * netlink_set_err - report error to broadcast listeners
1158 * @ssk: the kernel netlink socket, as returned by netlink_kernel_create()
1159 * @pid: the PID of a process that we want to skip (if any)
1160 * @groups: the broadcast group that will notice the error
1161 * @code: error code, must be negative (as usual in kernelspace)
1163 * This function returns the number of broadcast listeners that have set the
1164 * NETLINK_RECV_NO_ENOBUFS socket option.
1166 int netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1168 struct netlink_set_err_data info;
1169 struct hlist_node *node;
1170 struct sock *sk;
1171 int ret = 0;
1173 info.exclude_sk = ssk;
1174 info.pid = pid;
1175 info.group = group;
1176 /* sk->sk_err wants a positive error value */
1177 info.code = -code;
1179 read_lock(&nl_table_lock);
1181 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1182 ret += do_one_set_err(sk, &info);
1184 read_unlock(&nl_table_lock);
1185 return ret;
1187 EXPORT_SYMBOL(netlink_set_err);
1189 /* must be called with netlink table grabbed */
1190 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1191 unsigned int group,
1192 int is_new)
1194 int old, new = !!is_new, subscriptions;
1196 old = test_bit(group - 1, nlk->groups);
1197 subscriptions = nlk->subscriptions - old + new;
1198 if (new)
1199 __set_bit(group - 1, nlk->groups);
1200 else
1201 __clear_bit(group - 1, nlk->groups);
1202 netlink_update_subscriptions(&nlk->sk, subscriptions);
1203 netlink_update_listeners(&nlk->sk);
1206 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1207 char __user *optval, unsigned int optlen)
1209 struct sock *sk = sock->sk;
1210 struct netlink_sock *nlk = nlk_sk(sk);
1211 unsigned int val = 0;
1212 int err;
1214 if (level != SOL_NETLINK)
1215 return -ENOPROTOOPT;
1217 if (optlen >= sizeof(int) &&
1218 get_user(val, (unsigned int __user *)optval))
1219 return -EFAULT;
1221 switch (optname) {
1222 case NETLINK_PKTINFO:
1223 if (val)
1224 nlk->flags |= NETLINK_RECV_PKTINFO;
1225 else
1226 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1227 err = 0;
1228 break;
1229 case NETLINK_ADD_MEMBERSHIP:
1230 case NETLINK_DROP_MEMBERSHIP: {
1231 if (!netlink_capable(sock, NL_NONROOT_RECV))
1232 return -EPERM;
1233 err = netlink_realloc_groups(sk);
1234 if (err)
1235 return err;
1236 if (!val || val - 1 >= nlk->ngroups)
1237 return -EINVAL;
1238 netlink_table_grab();
1239 netlink_update_socket_mc(nlk, val,
1240 optname == NETLINK_ADD_MEMBERSHIP);
1241 netlink_table_ungrab();
1242 err = 0;
1243 break;
1245 case NETLINK_BROADCAST_ERROR:
1246 if (val)
1247 nlk->flags |= NETLINK_BROADCAST_SEND_ERROR;
1248 else
1249 nlk->flags &= ~NETLINK_BROADCAST_SEND_ERROR;
1250 err = 0;
1251 break;
1252 case NETLINK_NO_ENOBUFS:
1253 if (val) {
1254 nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
1255 clear_bit(0, &nlk->state);
1256 wake_up_interruptible(&nlk->wait);
1257 } else {
1258 nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
1260 err = 0;
1261 break;
1262 default:
1263 err = -ENOPROTOOPT;
1265 return err;
1268 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1269 char __user *optval, int __user *optlen)
1271 struct sock *sk = sock->sk;
1272 struct netlink_sock *nlk = nlk_sk(sk);
1273 int len, val, err;
1275 if (level != SOL_NETLINK)
1276 return -ENOPROTOOPT;
1278 if (get_user(len, optlen))
1279 return -EFAULT;
1280 if (len < 0)
1281 return -EINVAL;
1283 switch (optname) {
1284 case NETLINK_PKTINFO:
1285 if (len < sizeof(int))
1286 return -EINVAL;
1287 len = sizeof(int);
1288 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1289 if (put_user(len, optlen) ||
1290 put_user(val, optval))
1291 return -EFAULT;
1292 err = 0;
1293 break;
1294 case NETLINK_BROADCAST_ERROR:
1295 if (len < sizeof(int))
1296 return -EINVAL;
1297 len = sizeof(int);
1298 val = nlk->flags & NETLINK_BROADCAST_SEND_ERROR ? 1 : 0;
1299 if (put_user(len, optlen) ||
1300 put_user(val, optval))
1301 return -EFAULT;
1302 err = 0;
1303 break;
1304 case NETLINK_NO_ENOBUFS:
1305 if (len < sizeof(int))
1306 return -EINVAL;
1307 len = sizeof(int);
1308 val = nlk->flags & NETLINK_RECV_NO_ENOBUFS ? 1 : 0;
1309 if (put_user(len, optlen) ||
1310 put_user(val, optval))
1311 return -EFAULT;
1312 err = 0;
1313 break;
1314 default:
1315 err = -ENOPROTOOPT;
1317 return err;
1320 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1322 struct nl_pktinfo info;
1324 info.group = NETLINK_CB(skb).dst_group;
1325 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1328 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1329 struct msghdr *msg, size_t len)
1331 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1332 struct sock *sk = sock->sk;
1333 struct netlink_sock *nlk = nlk_sk(sk);
1334 struct sockaddr_nl *addr = msg->msg_name;
1335 u32 dst_pid;
1336 u32 dst_group;
1337 struct sk_buff *skb;
1338 int err;
1339 struct scm_cookie scm;
1341 if (msg->msg_flags&MSG_OOB)
1342 return -EOPNOTSUPP;
1344 if (NULL == siocb->scm)
1345 siocb->scm = &scm;
1347 err = scm_send(sock, msg, siocb->scm);
1348 if (err < 0)
1349 return err;
1351 if (msg->msg_namelen) {
1352 err = -EINVAL;
1353 if (addr->nl_family != AF_NETLINK)
1354 goto out;
1355 dst_pid = addr->nl_pid;
1356 dst_group = ffs(addr->nl_groups);
1357 err = -EPERM;
1358 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1359 goto out;
1360 } else {
1361 dst_pid = nlk->dst_pid;
1362 dst_group = nlk->dst_group;
1365 if (!nlk->pid) {
1366 err = netlink_autobind(sock);
1367 if (err)
1368 goto out;
1371 err = -EMSGSIZE;
1372 if (len > sk->sk_sndbuf - 32)
1373 goto out;
1374 err = -ENOBUFS;
1375 skb = alloc_skb(len, GFP_KERNEL);
1376 if (skb == NULL)
1377 goto out;
1379 NETLINK_CB(skb).pid = nlk->pid;
1380 NETLINK_CB(skb).dst_group = dst_group;
1381 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1383 err = -EFAULT;
1384 if (memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len)) {
1385 kfree_skb(skb);
1386 goto out;
1389 err = security_netlink_send(sk, skb);
1390 if (err) {
1391 kfree_skb(skb);
1392 goto out;
1395 if (dst_group) {
1396 atomic_inc(&skb->users);
1397 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1399 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1401 out:
1402 scm_destroy(siocb->scm);
1403 return err;
1406 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1407 struct msghdr *msg, size_t len,
1408 int flags)
1410 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1411 struct scm_cookie scm;
1412 struct sock *sk = sock->sk;
1413 struct netlink_sock *nlk = nlk_sk(sk);
1414 int noblock = flags&MSG_DONTWAIT;
1415 size_t copied;
1416 struct sk_buff *skb, *data_skb;
1417 int err, ret;
1419 if (flags&MSG_OOB)
1420 return -EOPNOTSUPP;
1422 copied = 0;
1424 skb = skb_recv_datagram(sk, flags, noblock, &err);
1425 if (skb == NULL)
1426 goto out;
1428 data_skb = skb;
1430 #ifdef CONFIG_COMPAT_NETLINK_MESSAGES
1431 if (unlikely(skb_shinfo(skb)->frag_list)) {
1433 * If this skb has a frag_list, then here that means that we
1434 * will have to use the frag_list skb's data for compat tasks
1435 * and the regular skb's data for normal (non-compat) tasks.
1437 * If we need to send the compat skb, assign it to the
1438 * 'data_skb' variable so that it will be used below for data
1439 * copying. We keep 'skb' for everything else, including
1440 * freeing both later.
1442 if (flags & MSG_CMSG_COMPAT)
1443 data_skb = skb_shinfo(skb)->frag_list;
1445 #endif
1447 msg->msg_namelen = 0;
1449 copied = data_skb->len;
1450 if (len < copied) {
1451 msg->msg_flags |= MSG_TRUNC;
1452 copied = len;
1455 skb_reset_transport_header(data_skb);
1456 err = skb_copy_datagram_iovec(data_skb, 0, msg->msg_iov, copied);
1458 if (msg->msg_name) {
1459 struct sockaddr_nl *addr = (struct sockaddr_nl *)msg->msg_name;
1460 addr->nl_family = AF_NETLINK;
1461 addr->nl_pad = 0;
1462 addr->nl_pid = NETLINK_CB(skb).pid;
1463 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1464 msg->msg_namelen = sizeof(*addr);
1467 if (nlk->flags & NETLINK_RECV_PKTINFO)
1468 netlink_cmsg_recv_pktinfo(msg, skb);
1470 if (NULL == siocb->scm) {
1471 memset(&scm, 0, sizeof(scm));
1472 siocb->scm = &scm;
1474 siocb->scm->creds = *NETLINK_CREDS(skb);
1475 if (flags & MSG_TRUNC)
1476 copied = data_skb->len;
1478 skb_free_datagram(sk, skb);
1480 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2) {
1481 ret = netlink_dump(sk);
1482 if (ret) {
1483 sk->sk_err = ret;
1484 sk->sk_error_report(sk);
1488 scm_recv(sock, msg, siocb->scm, flags);
1489 out:
1490 netlink_rcv_wake(sk);
1491 return err ? : copied;
1494 static void netlink_data_ready(struct sock *sk, int len)
1496 BUG();
1500 * We export these functions to other modules. They provide a
1501 * complete set of kernel non-blocking support for message
1502 * queueing.
1505 struct sock *
1506 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1507 void (*input)(struct sk_buff *skb),
1508 struct mutex *cb_mutex, struct module *module)
1510 struct socket *sock;
1511 struct sock *sk;
1512 struct netlink_sock *nlk;
1513 struct listeners *listeners = NULL;
1515 BUG_ON(!nl_table);
1517 if (unit < 0 || unit >= MAX_LINKS)
1518 return NULL;
1520 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1521 return NULL;
1524 * We have to just have a reference on the net from sk, but don't
1525 * get_net it. Besides, we cannot get and then put the net here.
1526 * So we create one inside init_net and the move it to net.
1529 if (__netlink_create(&init_net, sock, cb_mutex, unit) < 0)
1530 goto out_sock_release_nosk;
1532 sk = sock->sk;
1533 sk_change_net(sk, net);
1535 if (groups < 32)
1536 groups = 32;
1538 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
1539 if (!listeners)
1540 goto out_sock_release;
1542 sk->sk_data_ready = netlink_data_ready;
1543 if (input)
1544 nlk_sk(sk)->netlink_rcv = input;
1546 if (netlink_insert(sk, net, 0))
1547 goto out_sock_release;
1549 nlk = nlk_sk(sk);
1550 nlk->flags |= NETLINK_KERNEL_SOCKET;
1552 netlink_table_grab();
1553 if (!nl_table[unit].registered) {
1554 nl_table[unit].groups = groups;
1555 rcu_assign_pointer(nl_table[unit].listeners, listeners);
1556 nl_table[unit].cb_mutex = cb_mutex;
1557 nl_table[unit].module = module;
1558 nl_table[unit].registered = 1;
1559 } else {
1560 kfree(listeners);
1561 nl_table[unit].registered++;
1563 netlink_table_ungrab();
1564 return sk;
1566 out_sock_release:
1567 kfree(listeners);
1568 netlink_kernel_release(sk);
1569 return NULL;
1571 out_sock_release_nosk:
1572 sock_release(sock);
1573 return NULL;
1575 EXPORT_SYMBOL(netlink_kernel_create);
1578 void
1579 netlink_kernel_release(struct sock *sk)
1581 sk_release_kernel(sk);
1583 EXPORT_SYMBOL(netlink_kernel_release);
1585 int __netlink_change_ngroups(struct sock *sk, unsigned int groups)
1587 struct listeners *new, *old;
1588 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1590 if (groups < 32)
1591 groups = 32;
1593 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1594 new = kzalloc(sizeof(*new) + NLGRPSZ(groups), GFP_ATOMIC);
1595 if (!new)
1596 return -ENOMEM;
1597 old = rcu_dereference_protected(tbl->listeners, 1);
1598 memcpy(new->masks, old->masks, NLGRPSZ(tbl->groups));
1599 rcu_assign_pointer(tbl->listeners, new);
1601 kfree_rcu(old, rcu);
1603 tbl->groups = groups;
1605 return 0;
1609 * netlink_change_ngroups - change number of multicast groups
1611 * This changes the number of multicast groups that are available
1612 * on a certain netlink family. Note that it is not possible to
1613 * change the number of groups to below 32. Also note that it does
1614 * not implicitly call netlink_clear_multicast_users() when the
1615 * number of groups is reduced.
1617 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1618 * @groups: The new number of groups.
1620 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1622 int err;
1624 netlink_table_grab();
1625 err = __netlink_change_ngroups(sk, groups);
1626 netlink_table_ungrab();
1628 return err;
1631 void __netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1633 struct sock *sk;
1634 struct hlist_node *node;
1635 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1637 sk_for_each_bound(sk, node, &tbl->mc_list)
1638 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1642 * netlink_clear_multicast_users - kick off multicast listeners
1644 * This function removes all listeners from the given group.
1645 * @ksk: The kernel netlink socket, as returned by
1646 * netlink_kernel_create().
1647 * @group: The multicast group to clear.
1649 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1651 netlink_table_grab();
1652 __netlink_clear_multicast_users(ksk, group);
1653 netlink_table_ungrab();
1656 void netlink_set_nonroot(int protocol, unsigned int flags)
1658 if ((unsigned int)protocol < MAX_LINKS)
1659 nl_table[protocol].nl_nonroot = flags;
1661 EXPORT_SYMBOL(netlink_set_nonroot);
1663 struct nlmsghdr *
1664 __nlmsg_put(struct sk_buff *skb, u32 pid, u32 seq, int type, int len, int flags)
1666 struct nlmsghdr *nlh;
1667 int size = NLMSG_LENGTH(len);
1669 nlh = (struct nlmsghdr*)skb_put(skb, NLMSG_ALIGN(size));
1670 nlh->nlmsg_type = type;
1671 nlh->nlmsg_len = size;
1672 nlh->nlmsg_flags = flags;
1673 nlh->nlmsg_pid = pid;
1674 nlh->nlmsg_seq = seq;
1675 if (!__builtin_constant_p(size) || NLMSG_ALIGN(size) - size != 0)
1676 memset(NLMSG_DATA(nlh) + len, 0, NLMSG_ALIGN(size) - size);
1677 return nlh;
1679 EXPORT_SYMBOL(__nlmsg_put);
1682 * It looks a bit ugly.
1683 * It would be better to create kernel thread.
1686 static int netlink_dump(struct sock *sk)
1688 struct netlink_sock *nlk = nlk_sk(sk);
1689 struct netlink_callback *cb;
1690 struct sk_buff *skb = NULL;
1691 struct nlmsghdr *nlh;
1692 int len, err = -ENOBUFS;
1693 int alloc_size;
1695 mutex_lock(nlk->cb_mutex);
1697 cb = nlk->cb;
1698 if (cb == NULL) {
1699 err = -EINVAL;
1700 goto errout_skb;
1703 alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
1705 skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
1706 if (!skb)
1707 goto errout_skb;
1709 len = cb->dump(skb, cb);
1711 if (len > 0) {
1712 mutex_unlock(nlk->cb_mutex);
1714 if (sk_filter(sk, skb))
1715 kfree_skb(skb);
1716 else
1717 __netlink_sendskb(sk, skb);
1718 return 0;
1721 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1722 if (!nlh)
1723 goto errout_skb;
1725 nl_dump_check_consistent(cb, nlh);
1727 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1729 if (sk_filter(sk, skb))
1730 kfree_skb(skb);
1731 else
1732 __netlink_sendskb(sk, skb);
1734 if (cb->done)
1735 cb->done(cb);
1736 nlk->cb = NULL;
1737 mutex_unlock(nlk->cb_mutex);
1739 netlink_consume_callback(cb);
1740 return 0;
1742 errout_skb:
1743 mutex_unlock(nlk->cb_mutex);
1744 kfree_skb(skb);
1745 return err;
1748 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1749 const struct nlmsghdr *nlh,
1750 struct netlink_dump_control *control)
1752 struct netlink_callback *cb;
1753 struct sock *sk;
1754 struct netlink_sock *nlk;
1755 int ret;
1757 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1758 if (cb == NULL)
1759 return -ENOBUFS;
1761 cb->dump = control->dump;
1762 cb->done = control->done;
1763 cb->nlh = nlh;
1764 cb->data = control->data;
1765 cb->min_dump_alloc = control->min_dump_alloc;
1766 atomic_inc(&skb->users);
1767 cb->skb = skb;
1769 sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
1770 if (sk == NULL) {
1771 netlink_destroy_callback(cb);
1772 return -ECONNREFUSED;
1774 nlk = nlk_sk(sk);
1775 /* A dump is in progress... */
1776 mutex_lock(nlk->cb_mutex);
1777 if (nlk->cb) {
1778 mutex_unlock(nlk->cb_mutex);
1779 netlink_destroy_callback(cb);
1780 sock_put(sk);
1781 return -EBUSY;
1783 nlk->cb = cb;
1784 mutex_unlock(nlk->cb_mutex);
1786 ret = netlink_dump(sk);
1788 sock_put(sk);
1790 if (ret)
1791 return ret;
1793 /* We successfully started a dump, by returning -EINTR we
1794 * signal not to send ACK even if it was requested.
1796 return -EINTR;
1798 EXPORT_SYMBOL(netlink_dump_start);
1800 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1802 struct sk_buff *skb;
1803 struct nlmsghdr *rep;
1804 struct nlmsgerr *errmsg;
1805 size_t payload = sizeof(*errmsg);
1807 /* error messages get the original request appened */
1808 if (err)
1809 payload += nlmsg_len(nlh);
1811 skb = nlmsg_new(payload, GFP_KERNEL);
1812 if (!skb) {
1813 struct sock *sk;
1815 sk = netlink_lookup(sock_net(in_skb->sk),
1816 in_skb->sk->sk_protocol,
1817 NETLINK_CB(in_skb).pid);
1818 if (sk) {
1819 sk->sk_err = ENOBUFS;
1820 sk->sk_error_report(sk);
1821 sock_put(sk);
1823 return;
1826 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1827 NLMSG_ERROR, payload, 0);
1828 errmsg = nlmsg_data(rep);
1829 errmsg->error = err;
1830 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1831 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1833 EXPORT_SYMBOL(netlink_ack);
1835 int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1836 struct nlmsghdr *))
1838 struct nlmsghdr *nlh;
1839 int err;
1841 while (skb->len >= nlmsg_total_size(0)) {
1842 int msglen;
1844 nlh = nlmsg_hdr(skb);
1845 err = 0;
1847 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1848 return 0;
1850 /* Only requests are handled by the kernel */
1851 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1852 goto ack;
1854 /* Skip control messages */
1855 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1856 goto ack;
1858 err = cb(skb, nlh);
1859 if (err == -EINTR)
1860 goto skip;
1862 ack:
1863 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1864 netlink_ack(skb, nlh, err);
1866 skip:
1867 msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1868 if (msglen > skb->len)
1869 msglen = skb->len;
1870 skb_pull(skb, msglen);
1873 return 0;
1875 EXPORT_SYMBOL(netlink_rcv_skb);
1878 * nlmsg_notify - send a notification netlink message
1879 * @sk: netlink socket to use
1880 * @skb: notification message
1881 * @pid: destination netlink pid for reports or 0
1882 * @group: destination multicast group or 0
1883 * @report: 1 to report back, 0 to disable
1884 * @flags: allocation flags
1886 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1887 unsigned int group, int report, gfp_t flags)
1889 int err = 0;
1891 if (group) {
1892 int exclude_pid = 0;
1894 if (report) {
1895 atomic_inc(&skb->users);
1896 exclude_pid = pid;
1899 /* errors reported via destination sk->sk_err, but propagate
1900 * delivery errors if NETLINK_BROADCAST_ERROR flag is set */
1901 err = nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1904 if (report) {
1905 int err2;
1907 err2 = nlmsg_unicast(sk, skb, pid);
1908 if (!err || err == -ESRCH)
1909 err = err2;
1912 return err;
1914 EXPORT_SYMBOL(nlmsg_notify);
1916 #ifdef CONFIG_PROC_FS
1917 struct nl_seq_iter {
1918 struct seq_net_private p;
1919 int link;
1920 int hash_idx;
1923 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1925 struct nl_seq_iter *iter = seq->private;
1926 int i, j;
1927 struct sock *s;
1928 struct hlist_node *node;
1929 loff_t off = 0;
1931 for (i = 0; i < MAX_LINKS; i++) {
1932 struct nl_pid_hash *hash = &nl_table[i].hash;
1934 for (j = 0; j <= hash->mask; j++) {
1935 sk_for_each(s, node, &hash->table[j]) {
1936 if (sock_net(s) != seq_file_net(seq))
1937 continue;
1938 if (off == pos) {
1939 iter->link = i;
1940 iter->hash_idx = j;
1941 return s;
1943 ++off;
1947 return NULL;
1950 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1951 __acquires(nl_table_lock)
1953 read_lock(&nl_table_lock);
1954 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1957 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1959 struct sock *s;
1960 struct nl_seq_iter *iter;
1961 int i, j;
1963 ++*pos;
1965 if (v == SEQ_START_TOKEN)
1966 return netlink_seq_socket_idx(seq, 0);
1968 iter = seq->private;
1969 s = v;
1970 do {
1971 s = sk_next(s);
1972 } while (s && sock_net(s) != seq_file_net(seq));
1973 if (s)
1974 return s;
1976 i = iter->link;
1977 j = iter->hash_idx + 1;
1979 do {
1980 struct nl_pid_hash *hash = &nl_table[i].hash;
1982 for (; j <= hash->mask; j++) {
1983 s = sk_head(&hash->table[j]);
1984 while (s && sock_net(s) != seq_file_net(seq))
1985 s = sk_next(s);
1986 if (s) {
1987 iter->link = i;
1988 iter->hash_idx = j;
1989 return s;
1993 j = 0;
1994 } while (++i < MAX_LINKS);
1996 return NULL;
1999 static void netlink_seq_stop(struct seq_file *seq, void *v)
2000 __releases(nl_table_lock)
2002 read_unlock(&nl_table_lock);
2006 static int netlink_seq_show(struct seq_file *seq, void *v)
2008 if (v == SEQ_START_TOKEN) {
2009 seq_puts(seq,
2010 "sk Eth Pid Groups "
2011 "Rmem Wmem Dump Locks Drops Inode\n");
2012 } else {
2013 struct sock *s = v;
2014 struct netlink_sock *nlk = nlk_sk(s);
2016 seq_printf(seq, "%pK %-3d %-6d %08x %-8d %-8d %pK %-8d %-8d %-8lu\n",
2018 s->sk_protocol,
2019 nlk->pid,
2020 nlk->groups ? (u32)nlk->groups[0] : 0,
2021 sk_rmem_alloc_get(s),
2022 sk_wmem_alloc_get(s),
2023 nlk->cb,
2024 atomic_read(&s->sk_refcnt),
2025 atomic_read(&s->sk_drops),
2026 sock_i_ino(s)
2030 return 0;
2033 static const struct seq_operations netlink_seq_ops = {
2034 .start = netlink_seq_start,
2035 .next = netlink_seq_next,
2036 .stop = netlink_seq_stop,
2037 .show = netlink_seq_show,
2041 static int netlink_seq_open(struct inode *inode, struct file *file)
2043 return seq_open_net(inode, file, &netlink_seq_ops,
2044 sizeof(struct nl_seq_iter));
2047 static const struct file_operations netlink_seq_fops = {
2048 .owner = THIS_MODULE,
2049 .open = netlink_seq_open,
2050 .read = seq_read,
2051 .llseek = seq_lseek,
2052 .release = seq_release_net,
2055 #endif
2057 int netlink_register_notifier(struct notifier_block *nb)
2059 return atomic_notifier_chain_register(&netlink_chain, nb);
2061 EXPORT_SYMBOL(netlink_register_notifier);
2063 int netlink_unregister_notifier(struct notifier_block *nb)
2065 return atomic_notifier_chain_unregister(&netlink_chain, nb);
2067 EXPORT_SYMBOL(netlink_unregister_notifier);
2069 static const struct proto_ops netlink_ops = {
2070 .family = PF_NETLINK,
2071 .owner = THIS_MODULE,
2072 .release = netlink_release,
2073 .bind = netlink_bind,
2074 .connect = netlink_connect,
2075 .socketpair = sock_no_socketpair,
2076 .accept = sock_no_accept,
2077 .getname = netlink_getname,
2078 .poll = datagram_poll,
2079 .ioctl = sock_no_ioctl,
2080 .listen = sock_no_listen,
2081 .shutdown = sock_no_shutdown,
2082 .setsockopt = netlink_setsockopt,
2083 .getsockopt = netlink_getsockopt,
2084 .sendmsg = netlink_sendmsg,
2085 .recvmsg = netlink_recvmsg,
2086 .mmap = sock_no_mmap,
2087 .sendpage = sock_no_sendpage,
2090 static const struct net_proto_family netlink_family_ops = {
2091 .family = PF_NETLINK,
2092 .create = netlink_create,
2093 .owner = THIS_MODULE, /* for consistency 8) */
2096 static int __net_init netlink_net_init(struct net *net)
2098 #ifdef CONFIG_PROC_FS
2099 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
2100 return -ENOMEM;
2101 #endif
2102 return 0;
2105 static void __net_exit netlink_net_exit(struct net *net)
2107 #ifdef CONFIG_PROC_FS
2108 proc_net_remove(net, "netlink");
2109 #endif
2112 static void __init netlink_add_usersock_entry(void)
2114 struct listeners *listeners;
2115 int groups = 32;
2117 listeners = kzalloc(sizeof(*listeners) + NLGRPSZ(groups), GFP_KERNEL);
2118 if (!listeners)
2119 panic("netlink_add_usersock_entry: Cannot allocate listeners\n");
2121 netlink_table_grab();
2123 nl_table[NETLINK_USERSOCK].groups = groups;
2124 rcu_assign_pointer(nl_table[NETLINK_USERSOCK].listeners, listeners);
2125 nl_table[NETLINK_USERSOCK].module = THIS_MODULE;
2126 nl_table[NETLINK_USERSOCK].registered = 1;
2128 netlink_table_ungrab();
2131 static struct pernet_operations __net_initdata netlink_net_ops = {
2132 .init = netlink_net_init,
2133 .exit = netlink_net_exit,
2136 static int __init netlink_proto_init(void)
2138 struct sk_buff *dummy_skb;
2139 int i;
2140 unsigned long limit;
2141 unsigned int order;
2142 int err = proto_register(&netlink_proto, 0);
2144 if (err != 0)
2145 goto out;
2147 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
2149 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
2150 if (!nl_table)
2151 goto panic;
2153 if (totalram_pages >= (128 * 1024))
2154 limit = totalram_pages >> (21 - PAGE_SHIFT);
2155 else
2156 limit = totalram_pages >> (23 - PAGE_SHIFT);
2158 order = get_bitmask_order(limit) - 1 + PAGE_SHIFT;
2159 limit = (1UL << order) / sizeof(struct hlist_head);
2160 order = get_bitmask_order(min(limit, (unsigned long)UINT_MAX)) - 1;
2162 for (i = 0; i < MAX_LINKS; i++) {
2163 struct nl_pid_hash *hash = &nl_table[i].hash;
2165 hash->table = nl_pid_hash_zalloc(1 * sizeof(*hash->table));
2166 if (!hash->table) {
2167 while (i-- > 0)
2168 nl_pid_hash_free(nl_table[i].hash.table,
2169 1 * sizeof(*hash->table));
2170 kfree(nl_table);
2171 goto panic;
2173 hash->max_shift = order;
2174 hash->shift = 0;
2175 hash->mask = 0;
2176 hash->rehash_time = jiffies;
2179 netlink_add_usersock_entry();
2181 sock_register(&netlink_family_ops);
2182 register_pernet_subsys(&netlink_net_ops);
2183 /* The netlink device handler may be needed early. */
2184 rtnetlink_init();
2185 out:
2186 return err;
2187 panic:
2188 panic("netlink_init: Cannot allocate nl_table\n");
2191 core_initcall(netlink_proto_init);