?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[NET]: Fix race when opening a proc file while a network namespace is exiting.
[linux-2.6/cjktty.git] / net / netlink / af_netlink.c
blobdc9f8c2ab1d5ac51dbaed64027959ad9094612bc
1 /*
2 * NETLINK Kernel-user communication protocol.
3 *
4 * Authors: Alan Cox <alan@redhat.com>
5 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
6 *
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.
11 *
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
22 */
23
24 #include <linux/module.h>
25
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/selinux.h>
58 #include <linux/mutex.h>
59
60 #include <net/net_namespace.h>
61 #include <net/sock.h>
62 #include <net/scm.h>
63 #include <net/netlink.h>
64
65 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
66 #define NLGRPLONGS(x) (NLGRPSZ(x)/sizeof(unsigned long))
67
68 struct netlink_sock {
69 /* struct sock has to be the first member of netlink_sock */
70 struct sock sk;
71 u32 pid;
72 u32 dst_pid;
73 u32 dst_group;
74 u32 flags;
75 u32 subscriptions;
76 u32 ngroups;
77 unsigned long *groups;
78 unsigned long state;
79 wait_queue_head_t wait;
80 struct netlink_callback *cb;
81 struct mutex *cb_mutex;
82 struct mutex cb_def_mutex;
83 void (*data_ready)(struct sock *sk, int bytes);
84 struct module *module;
85 };
86
87 #define NETLINK_KERNEL_SOCKET 0x1
88 #define NETLINK_RECV_PKTINFO 0x2
89
90 static inline struct netlink_sock *nlk_sk(struct sock *sk)
91 {
92 return container_of(sk, struct netlink_sock, sk);
93 }
94
95 struct nl_pid_hash {
96 struct hlist_head *table;
97 unsigned long rehash_time;
98
99 unsigned int mask;
100 unsigned int shift;
101
102 unsigned int entries;
103 unsigned int max_shift;
104
105 u32 rnd;
106 };
107
108 struct netlink_table {
109 struct nl_pid_hash hash;
110 struct hlist_head mc_list;
111 unsigned long *listeners;
112 unsigned int nl_nonroot;
113 unsigned int groups;
114 struct mutex *cb_mutex;
115 struct module *module;
116 int registered;
117 };
118
119 static struct netlink_table *nl_table;
120
121 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
122
123 static int netlink_dump(struct sock *sk);
124 static void netlink_destroy_callback(struct netlink_callback *cb);
125 static void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb);
126
127 static DEFINE_RWLOCK(nl_table_lock);
128 static atomic_t nl_table_users = ATOMIC_INIT(0);
129
130 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
131
132 static u32 netlink_group_mask(u32 group)
133 {
134 return group ? 1 << (group - 1) : 0;
135 }
136
137 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
138 {
139 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
140 }
141
142 static void netlink_sock_destruct(struct sock *sk)
143 {
144 struct netlink_sock *nlk = nlk_sk(sk);
145
146 if (nlk->cb) {
147 if (nlk->cb->done)
148 nlk->cb->done(nlk->cb);
149 netlink_destroy_callback(nlk->cb);
150 }
151
152 skb_queue_purge(&sk->sk_receive_queue);
153
154 if (!sock_flag(sk, SOCK_DEAD)) {
155 printk("Freeing alive netlink socket %p\n", sk);
156 return;
157 }
158 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
159 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
160 BUG_TRAP(!nlk_sk(sk)->groups);
161 }
162
163 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
164 * Look, when several writers sleep and reader wakes them up, all but one
165 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
166 * this, _but_ remember, it adds useless work on UP machines.
167 */
168
169 static void netlink_table_grab(void)
170 {
171 write_lock_irq(&nl_table_lock);
172
173 if (atomic_read(&nl_table_users)) {
174 DECLARE_WAITQUEUE(wait, current);
175
176 add_wait_queue_exclusive(&nl_table_wait, &wait);
177 for(;;) {
178 set_current_state(TASK_UNINTERRUPTIBLE);
179 if (atomic_read(&nl_table_users) == 0)
180 break;
181 write_unlock_irq(&nl_table_lock);
182 schedule();
183 write_lock_irq(&nl_table_lock);
184 }
185
186 __set_current_state(TASK_RUNNING);
187 remove_wait_queue(&nl_table_wait, &wait);
188 }
189 }
190
191 static __inline__ void netlink_table_ungrab(void)
192 {
193 write_unlock_irq(&nl_table_lock);
194 wake_up(&nl_table_wait);
195 }
196
197 static __inline__ void
198 netlink_lock_table(void)
199 {
200 /* read_lock() synchronizes us to netlink_table_grab */
201
202 read_lock(&nl_table_lock);
203 atomic_inc(&nl_table_users);
204 read_unlock(&nl_table_lock);
205 }
206
207 static __inline__ void
208 netlink_unlock_table(void)
209 {
210 if (atomic_dec_and_test(&nl_table_users))
211 wake_up(&nl_table_wait);
212 }
213
214 static __inline__ struct sock *netlink_lookup(struct net *net, int protocol, u32 pid)
215 {
216 struct nl_pid_hash *hash = &nl_table[protocol].hash;
217 struct hlist_head *head;
218 struct sock *sk;
219 struct hlist_node *node;
220
221 read_lock(&nl_table_lock);
222 head = nl_pid_hashfn(hash, pid);
223 sk_for_each(sk, node, head) {
224 if ((sk->sk_net == net) && (nlk_sk(sk)->pid == pid)) {
225 sock_hold(sk);
226 goto found;
227 }
228 }
229 sk = NULL;
230 found:
231 read_unlock(&nl_table_lock);
232 return sk;
233 }
234
235 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
236 {
237 if (size <= PAGE_SIZE)
238 return kmalloc(size, GFP_ATOMIC);
239 else
240 return (struct hlist_head *)
241 __get_free_pages(GFP_ATOMIC, get_order(size));
242 }
243
244 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
245 {
246 if (size <= PAGE_SIZE)
247 kfree(table);
248 else
249 free_pages((unsigned long)table, get_order(size));
250 }
251
252 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
253 {
254 unsigned int omask, mask, shift;
255 size_t osize, size;
256 struct hlist_head *otable, *table;
257 int i;
258
259 omask = mask = hash->mask;
260 osize = size = (mask + 1) * sizeof(*table);
261 shift = hash->shift;
262
263 if (grow) {
264 if (++shift > hash->max_shift)
265 return 0;
266 mask = mask * 2 + 1;
267 size *= 2;
268 }
269
270 table = nl_pid_hash_alloc(size);
271 if (!table)
272 return 0;
273
274 memset(table, 0, size);
275 otable = hash->table;
276 hash->table = table;
277 hash->mask = mask;
278 hash->shift = shift;
279 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
280
281 for (i = 0; i <= omask; i++) {
282 struct sock *sk;
283 struct hlist_node *node, *tmp;
284
285 sk_for_each_safe(sk, node, tmp, &otable[i])
286 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
287 }
288
289 nl_pid_hash_free(otable, osize);
290 hash->rehash_time = jiffies + 10 * 60 * HZ;
291 return 1;
292 }
293
294 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
295 {
296 int avg = hash->entries >> hash->shift;
297
298 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
299 return 1;
300
301 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
302 nl_pid_hash_rehash(hash, 0);
303 return 1;
304 }
305
306 return 0;
307 }
308
309 static const struct proto_ops netlink_ops;
310
311 static void
312 netlink_update_listeners(struct sock *sk)
313 {
314 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
315 struct hlist_node *node;
316 unsigned long mask;
317 unsigned int i;
318
319 for (i = 0; i < NLGRPLONGS(tbl->groups); i++) {
320 mask = 0;
321 sk_for_each_bound(sk, node, &tbl->mc_list) {
322 if (i < NLGRPLONGS(nlk_sk(sk)->ngroups))
323 mask |= nlk_sk(sk)->groups[i];
324 }
325 tbl->listeners[i] = mask;
326 }
327 /* this function is only called with the netlink table "grabbed", which
328 * makes sure updates are visible before bind or setsockopt return. */
329 }
330
331 static int netlink_insert(struct sock *sk, struct net *net, u32 pid)
332 {
333 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
334 struct hlist_head *head;
335 int err = -EADDRINUSE;
336 struct sock *osk;
337 struct hlist_node *node;
338 int len;
339
340 netlink_table_grab();
341 head = nl_pid_hashfn(hash, pid);
342 len = 0;
343 sk_for_each(osk, node, head) {
344 if ((osk->sk_net == net) && (nlk_sk(osk)->pid == pid))
345 break;
346 len++;
347 }
348 if (node)
349 goto err;
350
351 err = -EBUSY;
352 if (nlk_sk(sk)->pid)
353 goto err;
354
355 err = -ENOMEM;
356 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
357 goto err;
358
359 if (len && nl_pid_hash_dilute(hash, len))
360 head = nl_pid_hashfn(hash, pid);
361 hash->entries++;
362 nlk_sk(sk)->pid = pid;
363 sk_add_node(sk, head);
364 err = 0;
365
366 err:
367 netlink_table_ungrab();
368 return err;
369 }
370
371 static void netlink_remove(struct sock *sk)
372 {
373 netlink_table_grab();
374 if (sk_del_node_init(sk))
375 nl_table[sk->sk_protocol].hash.entries--;
376 if (nlk_sk(sk)->subscriptions)
377 __sk_del_bind_node(sk);
378 netlink_table_ungrab();
379 }
380
381 static struct proto netlink_proto = {
382 .name = "NETLINK",
383 .owner = THIS_MODULE,
384 .obj_size = sizeof(struct netlink_sock),
385 };
386
387 static int __netlink_create(struct net *net, struct socket *sock,
388 struct mutex *cb_mutex, int protocol)
389 {
390 struct sock *sk;
391 struct netlink_sock *nlk;
392
393 sock->ops = &netlink_ops;
394
395 sk = sk_alloc(net, PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
396 if (!sk)
397 return -ENOMEM;
398
399 sock_init_data(sock, sk);
400
401 nlk = nlk_sk(sk);
402 if (cb_mutex)
403 nlk->cb_mutex = cb_mutex;
404 else {
405 nlk->cb_mutex = &nlk->cb_def_mutex;
406 mutex_init(nlk->cb_mutex);
407 }
408 init_waitqueue_head(&nlk->wait);
409
410 sk->sk_destruct = netlink_sock_destruct;
411 sk->sk_protocol = protocol;
412 return 0;
413 }
414
415 static int netlink_create(struct net *net, struct socket *sock, int protocol)
416 {
417 struct module *module = NULL;
418 struct mutex *cb_mutex;
419 struct netlink_sock *nlk;
420 int err = 0;
421
422 sock->state = SS_UNCONNECTED;
423
424 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
425 return -ESOCKTNOSUPPORT;
426
427 if (protocol<0 || protocol >= MAX_LINKS)
428 return -EPROTONOSUPPORT;
429
430 netlink_lock_table();
431 #ifdef CONFIG_KMOD
432 if (!nl_table[protocol].registered) {
433 netlink_unlock_table();
434 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
435 netlink_lock_table();
436 }
437 #endif
438 if (nl_table[protocol].registered &&
439 try_module_get(nl_table[protocol].module))
440 module = nl_table[protocol].module;
441 cb_mutex = nl_table[protocol].cb_mutex;
442 netlink_unlock_table();
443
444 if ((err = __netlink_create(net, sock, cb_mutex, protocol)) < 0)
445 goto out_module;
446
447 nlk = nlk_sk(sock->sk);
448 nlk->module = module;
449 out:
450 return err;
451
452 out_module:
453 module_put(module);
454 goto out;
455 }
456
457 static int netlink_release(struct socket *sock)
458 {
459 struct sock *sk = sock->sk;
460 struct netlink_sock *nlk;
461
462 if (!sk)
463 return 0;
464
465 netlink_remove(sk);
466 sock_orphan(sk);
467 nlk = nlk_sk(sk);
468
469 /*
470 * OK. Socket is unlinked, any packets that arrive now
471 * will be purged.
472 */
473
474 sock->sk = NULL;
475 wake_up_interruptible_all(&nlk->wait);
476
477 skb_queue_purge(&sk->sk_write_queue);
478
479 if (nlk->pid && !nlk->subscriptions) {
480 struct netlink_notify n = {
481 .net = sk->sk_net,
482 .protocol = sk->sk_protocol,
483 .pid = nlk->pid,
484 };
485 atomic_notifier_call_chain(&netlink_chain,
486 NETLINK_URELEASE, &n);
487 }
488
489 module_put(nlk->module);
490
491 netlink_table_grab();
492 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
493 kfree(nl_table[sk->sk_protocol].listeners);
494 nl_table[sk->sk_protocol].module = NULL;
495 nl_table[sk->sk_protocol].registered = 0;
496 } else if (nlk->subscriptions)
497 netlink_update_listeners(sk);
498 netlink_table_ungrab();
499
500 kfree(nlk->groups);
501 nlk->groups = NULL;
502
503 sock_put(sk);
504 return 0;
505 }
506
507 static int netlink_autobind(struct socket *sock)
508 {
509 struct sock *sk = sock->sk;
510 struct net *net = sk->sk_net;
511 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
512 struct hlist_head *head;
513 struct sock *osk;
514 struct hlist_node *node;
515 s32 pid = current->tgid;
516 int err;
517 static s32 rover = -4097;
518
519 retry:
520 cond_resched();
521 netlink_table_grab();
522 head = nl_pid_hashfn(hash, pid);
523 sk_for_each(osk, node, head) {
524 if ((osk->sk_net != net))
525 continue;
526 if (nlk_sk(osk)->pid == pid) {
527 /* Bind collision, search negative pid values. */
528 pid = rover--;
529 if (rover > -4097)
530 rover = -4097;
531 netlink_table_ungrab();
532 goto retry;
533 }
534 }
535 netlink_table_ungrab();
536
537 err = netlink_insert(sk, net, pid);
538 if (err == -EADDRINUSE)
539 goto retry;
540
541 /* If 2 threads race to autobind, that is fine. */
542 if (err == -EBUSY)
543 err = 0;
544
545 return err;
546 }
547
548 static inline int netlink_capable(struct socket *sock, unsigned int flag)
549 {
550 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
551 capable(CAP_NET_ADMIN);
552 }
553
554 static void
555 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
556 {
557 struct netlink_sock *nlk = nlk_sk(sk);
558
559 if (nlk->subscriptions && !subscriptions)
560 __sk_del_bind_node(sk);
561 else if (!nlk->subscriptions && subscriptions)
562 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
563 nlk->subscriptions = subscriptions;
564 }
565
566 static int netlink_realloc_groups(struct sock *sk)
567 {
568 struct netlink_sock *nlk = nlk_sk(sk);
569 unsigned int groups;
570 unsigned long *new_groups;
571 int err = 0;
572
573 netlink_table_grab();
574
575 groups = nl_table[sk->sk_protocol].groups;
576 if (!nl_table[sk->sk_protocol].registered) {
577 err = -ENOENT;
578 goto out_unlock;
579 }
580
581 if (nlk->ngroups >= groups)
582 goto out_unlock;
583
584 new_groups = krealloc(nlk->groups, NLGRPSZ(groups), GFP_ATOMIC);
585 if (new_groups == NULL) {
586 err = -ENOMEM;
587 goto out_unlock;
588 }
589 memset((char*)new_groups + NLGRPSZ(nlk->ngroups), 0,
590 NLGRPSZ(groups) - NLGRPSZ(nlk->ngroups));
591
592 nlk->groups = new_groups;
593 nlk->ngroups = groups;
594 out_unlock:
595 netlink_table_ungrab();
596 return err;
597 }
598
599 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
600 {
601 struct sock *sk = sock->sk;
602 struct net *net = sk->sk_net;
603 struct netlink_sock *nlk = nlk_sk(sk);
604 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
605 int err;
606
607 if (nladdr->nl_family != AF_NETLINK)
608 return -EINVAL;
609
610 /* Only superuser is allowed to listen multicasts */
611 if (nladdr->nl_groups) {
612 if (!netlink_capable(sock, NL_NONROOT_RECV))
613 return -EPERM;
614 err = netlink_realloc_groups(sk);
615 if (err)
616 return err;
617 }
618
619 if (nlk->pid) {
620 if (nladdr->nl_pid != nlk->pid)
621 return -EINVAL;
622 } else {
623 err = nladdr->nl_pid ?
624 netlink_insert(sk, net, nladdr->nl_pid) :
625 netlink_autobind(sock);
626 if (err)
627 return err;
628 }
629
630 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
631 return 0;
632
633 netlink_table_grab();
634 netlink_update_subscriptions(sk, nlk->subscriptions +
635 hweight32(nladdr->nl_groups) -
636 hweight32(nlk->groups[0]));
637 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
638 netlink_update_listeners(sk);
639 netlink_table_ungrab();
640
641 return 0;
642 }
643
644 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
645 int alen, int flags)
646 {
647 int err = 0;
648 struct sock *sk = sock->sk;
649 struct netlink_sock *nlk = nlk_sk(sk);
650 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
651
652 if (addr->sa_family == AF_UNSPEC) {
653 sk->sk_state = NETLINK_UNCONNECTED;
654 nlk->dst_pid = 0;
655 nlk->dst_group = 0;
656 return 0;
657 }
658 if (addr->sa_family != AF_NETLINK)
659 return -EINVAL;
660
661 /* Only superuser is allowed to send multicasts */
662 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
663 return -EPERM;
664
665 if (!nlk->pid)
666 err = netlink_autobind(sock);
667
668 if (err == 0) {
669 sk->sk_state = NETLINK_CONNECTED;
670 nlk->dst_pid = nladdr->nl_pid;
671 nlk->dst_group = ffs(nladdr->nl_groups);
672 }
673
674 return err;
675 }
676
677 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
678 {
679 struct sock *sk = sock->sk;
680 struct netlink_sock *nlk = nlk_sk(sk);
681 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
682
683 nladdr->nl_family = AF_NETLINK;
684 nladdr->nl_pad = 0;
685 *addr_len = sizeof(*nladdr);
686
687 if (peer) {
688 nladdr->nl_pid = nlk->dst_pid;
689 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
690 } else {
691 nladdr->nl_pid = nlk->pid;
692 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
693 }
694 return 0;
695 }
696
697 static void netlink_overrun(struct sock *sk)
698 {
699 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
700 sk->sk_err = ENOBUFS;
701 sk->sk_error_report(sk);
702 }
703 }
704
705 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
706 {
707 int protocol = ssk->sk_protocol;
708 struct net *net;
709 struct sock *sock;
710 struct netlink_sock *nlk;
711
712 net = ssk->sk_net;
713 sock = netlink_lookup(net, protocol, pid);
714 if (!sock)
715 return ERR_PTR(-ECONNREFUSED);
716
717 /* Don't bother queuing skb if kernel socket has no input function */
718 nlk = nlk_sk(sock);
719 if ((nlk->pid == 0 && !nlk->data_ready) ||
720 (sock->sk_state == NETLINK_CONNECTED &&
721 nlk->dst_pid != nlk_sk(ssk)->pid)) {
722 sock_put(sock);
723 return ERR_PTR(-ECONNREFUSED);
724 }
725 return sock;
726 }
727
728 struct sock *netlink_getsockbyfilp(struct file *filp)
729 {
730 struct inode *inode = filp->f_path.dentry->d_inode;
731 struct sock *sock;
732
733 if (!S_ISSOCK(inode->i_mode))
734 return ERR_PTR(-ENOTSOCK);
735
736 sock = SOCKET_I(inode)->sk;
737 if (sock->sk_family != AF_NETLINK)
738 return ERR_PTR(-EINVAL);
739
740 sock_hold(sock);
741 return sock;
742 }
743
744 /*
745 * Attach a skb to a netlink socket.
746 * The caller must hold a reference to the destination socket. On error, the
747 * reference is dropped. The skb is not send to the destination, just all
748 * all error checks are performed and memory in the queue is reserved.
749 * Return values:
750 * < 0: error. skb freed, reference to sock dropped.
751 * 0: continue
752 * 1: repeat lookup - reference dropped while waiting for socket memory.
753 */
754 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
755 long timeo, struct sock *ssk)
756 {
757 struct netlink_sock *nlk;
758
759 nlk = nlk_sk(sk);
760
761 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
762 test_bit(0, &nlk->state)) {
763 DECLARE_WAITQUEUE(wait, current);
764 if (!timeo) {
765 if (!ssk || nlk_sk(ssk)->pid == 0)
766 netlink_overrun(sk);
767 sock_put(sk);
768 kfree_skb(skb);
769 return -EAGAIN;
770 }
771
772 __set_current_state(TASK_INTERRUPTIBLE);
773 add_wait_queue(&nlk->wait, &wait);
774
775 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
776 test_bit(0, &nlk->state)) &&
777 !sock_flag(sk, SOCK_DEAD))
778 timeo = schedule_timeout(timeo);
779
780 __set_current_state(TASK_RUNNING);
781 remove_wait_queue(&nlk->wait, &wait);
782 sock_put(sk);
783
784 if (signal_pending(current)) {
785 kfree_skb(skb);
786 return sock_intr_errno(timeo);
787 }
788 return 1;
789 }
790 skb_set_owner_r(skb, sk);
791 return 0;
792 }
793
794 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
795 {
796 int len = skb->len;
797
798 skb_queue_tail(&sk->sk_receive_queue, skb);
799 sk->sk_data_ready(sk, len);
800 sock_put(sk);
801 return len;
802 }
803
804 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
805 {
806 kfree_skb(skb);
807 sock_put(sk);
808 }
809
810 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
811 gfp_t allocation)
812 {
813 int delta;
814
815 skb_orphan(skb);
816
817 delta = skb->end - skb->tail;
818 if (delta * 2 < skb->truesize)
819 return skb;
820
821 if (skb_shared(skb)) {
822 struct sk_buff *nskb = skb_clone(skb, allocation);
823 if (!nskb)
824 return skb;
825 kfree_skb(skb);
826 skb = nskb;
827 }
828
829 if (!pskb_expand_head(skb, 0, -delta, allocation))
830 skb->truesize -= delta;
831
832 return skb;
833 }
834
835 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
836 {
837 struct sock *sk;
838 int err;
839 long timeo;
840
841 skb = netlink_trim(skb, gfp_any());
842
843 timeo = sock_sndtimeo(ssk, nonblock);
844 retry:
845 sk = netlink_getsockbypid(ssk, pid);
846 if (IS_ERR(sk)) {
847 kfree_skb(skb);
848 return PTR_ERR(sk);
849 }
850 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
851 if (err == 1)
852 goto retry;
853 if (err)
854 return err;
855
856 return netlink_sendskb(sk, skb, ssk->sk_protocol);
857 }
858
859 int netlink_has_listeners(struct sock *sk, unsigned int group)
860 {
861 int res = 0;
862 unsigned long *listeners;
863
864 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
865
866 rcu_read_lock();
867 listeners = rcu_dereference(nl_table[sk->sk_protocol].listeners);
868
869 if (group - 1 < nl_table[sk->sk_protocol].groups)
870 res = test_bit(group - 1, listeners);
871
872 rcu_read_unlock();
873
874 return res;
875 }
876 EXPORT_SYMBOL_GPL(netlink_has_listeners);
877
878 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
879 {
880 struct netlink_sock *nlk = nlk_sk(sk);
881
882 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
883 !test_bit(0, &nlk->state)) {
884 skb_set_owner_r(skb, sk);
885 skb_queue_tail(&sk->sk_receive_queue, skb);
886 sk->sk_data_ready(sk, skb->len);
887 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
888 }
889 return -1;
890 }
891
892 struct netlink_broadcast_data {
893 struct sock *exclude_sk;
894 struct net *net;
895 u32 pid;
896 u32 group;
897 int failure;
898 int congested;
899 int delivered;
900 gfp_t allocation;
901 struct sk_buff *skb, *skb2;
902 };
903
904 static inline int do_one_broadcast(struct sock *sk,
905 struct netlink_broadcast_data *p)
906 {
907 struct netlink_sock *nlk = nlk_sk(sk);
908 int val;
909
910 if (p->exclude_sk == sk)
911 goto out;
912
913 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
914 !test_bit(p->group - 1, nlk->groups))
915 goto out;
916
917 if ((sk->sk_net != p->net))
918 goto out;
919
920 if (p->failure) {
921 netlink_overrun(sk);
922 goto out;
923 }
924
925 sock_hold(sk);
926 if (p->skb2 == NULL) {
927 if (skb_shared(p->skb)) {
928 p->skb2 = skb_clone(p->skb, p->allocation);
929 } else {
930 p->skb2 = skb_get(p->skb);
931 /*
932 * skb ownership may have been set when
933 * delivered to a previous socket.
934 */
935 skb_orphan(p->skb2);
936 }
937 }
938 if (p->skb2 == NULL) {
939 netlink_overrun(sk);
940 /* Clone failed. Notify ALL listeners. */
941 p->failure = 1;
942 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
943 netlink_overrun(sk);
944 } else {
945 p->congested |= val;
946 p->delivered = 1;
947 p->skb2 = NULL;
948 }
949 sock_put(sk);
950
951 out:
952 return 0;
953 }
954
955 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
956 u32 group, gfp_t allocation)
957 {
958 struct net *net = ssk->sk_net;
959 struct netlink_broadcast_data info;
960 struct hlist_node *node;
961 struct sock *sk;
962
963 skb = netlink_trim(skb, allocation);
964
965 info.exclude_sk = ssk;
966 info.net = net;
967 info.pid = pid;
968 info.group = group;
969 info.failure = 0;
970 info.congested = 0;
971 info.delivered = 0;
972 info.allocation = allocation;
973 info.skb = skb;
974 info.skb2 = NULL;
975
976 /* While we sleep in clone, do not allow to change socket list */
977
978 netlink_lock_table();
979
980 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
981 do_one_broadcast(sk, &info);
982
983 kfree_skb(skb);
984
985 netlink_unlock_table();
986
987 if (info.skb2)
988 kfree_skb(info.skb2);
989
990 if (info.delivered) {
991 if (info.congested && (allocation & __GFP_WAIT))
992 yield();
993 return 0;
994 }
995 if (info.failure)
996 return -ENOBUFS;
997 return -ESRCH;
998 }
999
1000 struct netlink_set_err_data {
1001 struct sock *exclude_sk;
1002 u32 pid;
1003 u32 group;
1004 int code;
1005 };
1006
1007 static inline int do_one_set_err(struct sock *sk,
1008 struct netlink_set_err_data *p)
1009 {
1010 struct netlink_sock *nlk = nlk_sk(sk);
1011
1012 if (sk == p->exclude_sk)
1013 goto out;
1014
1015 if (sk->sk_net != p->exclude_sk->sk_net)
1016 goto out;
1017
1018 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
1019 !test_bit(p->group - 1, nlk->groups))
1020 goto out;
1021
1022 sk->sk_err = p->code;
1023 sk->sk_error_report(sk);
1024 out:
1025 return 0;
1026 }
1027
1028 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
1029 {
1030 struct netlink_set_err_data info;
1031 struct hlist_node *node;
1032 struct sock *sk;
1033
1034 info.exclude_sk = ssk;
1035 info.pid = pid;
1036 info.group = group;
1037 info.code = code;
1038
1039 read_lock(&nl_table_lock);
1040
1041 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
1042 do_one_set_err(sk, &info);
1043
1044 read_unlock(&nl_table_lock);
1045 }
1046
1047 /* must be called with netlink table grabbed */
1048 static void netlink_update_socket_mc(struct netlink_sock *nlk,
1049 unsigned int group,
1050 int is_new)
1051 {
1052 int old, new = !!is_new, subscriptions;
1053
1054 old = test_bit(group - 1, nlk->groups);
1055 subscriptions = nlk->subscriptions - old + new;
1056 if (new)
1057 __set_bit(group - 1, nlk->groups);
1058 else
1059 __clear_bit(group - 1, nlk->groups);
1060 netlink_update_subscriptions(&nlk->sk, subscriptions);
1061 netlink_update_listeners(&nlk->sk);
1062 }
1063
1064 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1065 char __user *optval, int optlen)
1066 {
1067 struct sock *sk = sock->sk;
1068 struct netlink_sock *nlk = nlk_sk(sk);
1069 unsigned int val = 0;
1070 int err;
1071
1072 if (level != SOL_NETLINK)
1073 return -ENOPROTOOPT;
1074
1075 if (optlen >= sizeof(int) &&
1076 get_user(val, (unsigned int __user *)optval))
1077 return -EFAULT;
1078
1079 switch (optname) {
1080 case NETLINK_PKTINFO:
1081 if (val)
1082 nlk->flags |= NETLINK_RECV_PKTINFO;
1083 else
1084 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1085 err = 0;
1086 break;
1087 case NETLINK_ADD_MEMBERSHIP:
1088 case NETLINK_DROP_MEMBERSHIP: {
1089 if (!netlink_capable(sock, NL_NONROOT_RECV))
1090 return -EPERM;
1091 err = netlink_realloc_groups(sk);
1092 if (err)
1093 return err;
1094 if (!val || val - 1 >= nlk->ngroups)
1095 return -EINVAL;
1096 netlink_table_grab();
1097 netlink_update_socket_mc(nlk, val,
1098 optname == NETLINK_ADD_MEMBERSHIP);
1099 netlink_table_ungrab();
1100 err = 0;
1101 break;
1102 }
1103 default:
1104 err = -ENOPROTOOPT;
1105 }
1106 return err;
1107 }
1108
1109 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1110 char __user *optval, int __user *optlen)
1111 {
1112 struct sock *sk = sock->sk;
1113 struct netlink_sock *nlk = nlk_sk(sk);
1114 int len, val, err;
1115
1116 if (level != SOL_NETLINK)
1117 return -ENOPROTOOPT;
1118
1119 if (get_user(len, optlen))
1120 return -EFAULT;
1121 if (len < 0)
1122 return -EINVAL;
1123
1124 switch (optname) {
1125 case NETLINK_PKTINFO:
1126 if (len < sizeof(int))
1127 return -EINVAL;
1128 len = sizeof(int);
1129 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1130 if (put_user(len, optlen) ||
1131 put_user(val, optval))
1132 return -EFAULT;
1133 err = 0;
1134 break;
1135 default:
1136 err = -ENOPROTOOPT;
1137 }
1138 return err;
1139 }
1140
1141 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1142 {
1143 struct nl_pktinfo info;
1144
1145 info.group = NETLINK_CB(skb).dst_group;
1146 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1147 }
1148
1149 static inline void netlink_rcv_wake(struct sock *sk)
1150 {
1151 struct netlink_sock *nlk = nlk_sk(sk);
1152
1153 if (skb_queue_empty(&sk->sk_receive_queue))
1154 clear_bit(0, &nlk->state);
1155 if (!test_bit(0, &nlk->state))
1156 wake_up_interruptible(&nlk->wait);
1157 }
1158
1159 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1160 struct msghdr *msg, size_t len)
1161 {
1162 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1163 struct sock *sk = sock->sk;
1164 struct netlink_sock *nlk = nlk_sk(sk);
1165 struct sockaddr_nl *addr=msg->msg_name;
1166 u32 dst_pid;
1167 u32 dst_group;
1168 struct sk_buff *skb;
1169 int err;
1170 struct scm_cookie scm;
1171
1172 if (msg->msg_flags&MSG_OOB)
1173 return -EOPNOTSUPP;
1174
1175 if (NULL == siocb->scm)
1176 siocb->scm = &scm;
1177 err = scm_send(sock, msg, siocb->scm);
1178 if (err < 0)
1179 return err;
1180
1181 if (msg->msg_namelen) {
1182 if (addr->nl_family != AF_NETLINK)
1183 return -EINVAL;
1184 dst_pid = addr->nl_pid;
1185 dst_group = ffs(addr->nl_groups);
1186 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1187 return -EPERM;
1188 } else {
1189 dst_pid = nlk->dst_pid;
1190 dst_group = nlk->dst_group;
1191 }
1192
1193 if (!nlk->pid) {
1194 err = netlink_autobind(sock);
1195 if (err)
1196 goto out;
1197 }
1198
1199 err = -EMSGSIZE;
1200 if (len > sk->sk_sndbuf - 32)
1201 goto out;
1202 err = -ENOBUFS;
1203 skb = alloc_skb(len, GFP_KERNEL);
1204 if (skb==NULL)
1205 goto out;
1206
1207 NETLINK_CB(skb).pid = nlk->pid;
1208 NETLINK_CB(skb).dst_group = dst_group;
1209 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1210 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1211 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1212
1213 /* What can I do? Netlink is asynchronous, so that
1214 we will have to save current capabilities to
1215 check them, when this message will be delivered
1216 to corresponding kernel module. --ANK (980802)
1217 */
1218
1219 err = -EFAULT;
1220 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1221 kfree_skb(skb);
1222 goto out;
1223 }
1224
1225 err = security_netlink_send(sk, skb);
1226 if (err) {
1227 kfree_skb(skb);
1228 goto out;
1229 }
1230
1231 if (dst_group) {
1232 atomic_inc(&skb->users);
1233 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1234 }
1235 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1236
1237 out:
1238 return err;
1239 }
1240
1241 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1242 struct msghdr *msg, size_t len,
1243 int flags)
1244 {
1245 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1246 struct scm_cookie scm;
1247 struct sock *sk = sock->sk;
1248 struct netlink_sock *nlk = nlk_sk(sk);
1249 int noblock = flags&MSG_DONTWAIT;
1250 size_t copied;
1251 struct sk_buff *skb;
1252 int err;
1253
1254 if (flags&MSG_OOB)
1255 return -EOPNOTSUPP;
1256
1257 copied = 0;
1258
1259 skb = skb_recv_datagram(sk,flags,noblock,&err);
1260 if (skb==NULL)
1261 goto out;
1262
1263 msg->msg_namelen = 0;
1264
1265 copied = skb->len;
1266 if (len < copied) {
1267 msg->msg_flags |= MSG_TRUNC;
1268 copied = len;
1269 }
1270
1271 skb_reset_transport_header(skb);
1272 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1273
1274 if (msg->msg_name) {
1275 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1276 addr->nl_family = AF_NETLINK;
1277 addr->nl_pad = 0;
1278 addr->nl_pid = NETLINK_CB(skb).pid;
1279 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1280 msg->msg_namelen = sizeof(*addr);
1281 }
1282
1283 if (nlk->flags & NETLINK_RECV_PKTINFO)
1284 netlink_cmsg_recv_pktinfo(msg, skb);
1285
1286 if (NULL == siocb->scm) {
1287 memset(&scm, 0, sizeof(scm));
1288 siocb->scm = &scm;
1289 }
1290 siocb->scm->creds = *NETLINK_CREDS(skb);
1291 if (flags & MSG_TRUNC)
1292 copied = skb->len;
1293 skb_free_datagram(sk, skb);
1294
1295 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1296 netlink_dump(sk);
1297
1298 scm_recv(sock, msg, siocb->scm, flags);
1299 out:
1300 netlink_rcv_wake(sk);
1301 return err ? : copied;
1302 }
1303
1304 static void netlink_data_ready(struct sock *sk, int len)
1305 {
1306 struct netlink_sock *nlk = nlk_sk(sk);
1307
1308 if (nlk->data_ready)
1309 nlk->data_ready(sk, len);
1310 netlink_rcv_wake(sk);
1311 }
1312
1313 /*
1314 * We export these functions to other modules. They provide a
1315 * complete set of kernel non-blocking support for message
1316 * queueing.
1317 */
1318
1319 struct sock *
1320 netlink_kernel_create(struct net *net, int unit, unsigned int groups,
1321 void (*input)(struct sock *sk, int len),
1322 struct mutex *cb_mutex, struct module *module)
1323 {
1324 struct socket *sock;
1325 struct sock *sk;
1326 struct netlink_sock *nlk;
1327 unsigned long *listeners = NULL;
1328
1329 BUG_ON(!nl_table);
1330
1331 if (unit<0 || unit>=MAX_LINKS)
1332 return NULL;
1333
1334 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1335 return NULL;
1336
1337 if (__netlink_create(net, sock, cb_mutex, unit) < 0)
1338 goto out_sock_release;
1339
1340 if (groups < 32)
1341 groups = 32;
1342
1343 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1344 if (!listeners)
1345 goto out_sock_release;
1346
1347 sk = sock->sk;
1348 sk->sk_data_ready = netlink_data_ready;
1349 if (input)
1350 nlk_sk(sk)->data_ready = input;
1351
1352 if (netlink_insert(sk, net, 0))
1353 goto out_sock_release;
1354
1355 nlk = nlk_sk(sk);
1356 nlk->flags |= NETLINK_KERNEL_SOCKET;
1357
1358 netlink_table_grab();
1359 if (!nl_table[unit].registered) {
1360 nl_table[unit].groups = groups;
1361 nl_table[unit].listeners = listeners;
1362 nl_table[unit].cb_mutex = cb_mutex;
1363 nl_table[unit].module = module;
1364 nl_table[unit].registered = 1;
1365 }
1366 netlink_table_ungrab();
1367
1368 return sk;
1369
1370 out_sock_release:
1371 kfree(listeners);
1372 sock_release(sock);
1373 return NULL;
1374 }
1375
1376 /**
1377 * netlink_change_ngroups - change number of multicast groups
1378 *
1379 * This changes the number of multicast groups that are available
1380 * on a certain netlink family. Note that it is not possible to
1381 * change the number of groups to below 32. Also note that it does
1382 * not implicitly call netlink_clear_multicast_users() when the
1383 * number of groups is reduced.
1384 *
1385 * @sk: The kernel netlink socket, as returned by netlink_kernel_create().
1386 * @groups: The new number of groups.
1387 */
1388 int netlink_change_ngroups(struct sock *sk, unsigned int groups)
1389 {
1390 unsigned long *listeners, *old = NULL;
1391 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
1392 int err = 0;
1393
1394 if (groups < 32)
1395 groups = 32;
1396
1397 netlink_table_grab();
1398 if (NLGRPSZ(tbl->groups) < NLGRPSZ(groups)) {
1399 listeners = kzalloc(NLGRPSZ(groups), GFP_ATOMIC);
1400 if (!listeners) {
1401 err = -ENOMEM;
1402 goto out_ungrab;
1403 }
1404 old = tbl->listeners;
1405 memcpy(listeners, old, NLGRPSZ(tbl->groups));
1406 rcu_assign_pointer(tbl->listeners, listeners);
1407 }
1408 tbl->groups = groups;
1409
1410 out_ungrab:
1411 netlink_table_ungrab();
1412 synchronize_rcu();
1413 kfree(old);
1414 return err;
1415 }
1416 EXPORT_SYMBOL(netlink_change_ngroups);
1417
1418 /**
1419 * netlink_clear_multicast_users - kick off multicast listeners
1420 *
1421 * This function removes all listeners from the given group.
1422 * @ksk: The kernel netlink socket, as returned by
1423 * netlink_kernel_create().
1424 * @group: The multicast group to clear.
1425 */
1426 void netlink_clear_multicast_users(struct sock *ksk, unsigned int group)
1427 {
1428 struct sock *sk;
1429 struct hlist_node *node;
1430 struct netlink_table *tbl = &nl_table[ksk->sk_protocol];
1431
1432 netlink_table_grab();
1433
1434 sk_for_each_bound(sk, node, &tbl->mc_list)
1435 netlink_update_socket_mc(nlk_sk(sk), group, 0);
1436
1437 netlink_table_ungrab();
1438 }
1439 EXPORT_SYMBOL(netlink_clear_multicast_users);
1440
1441 void netlink_set_nonroot(int protocol, unsigned int flags)
1442 {
1443 if ((unsigned int)protocol < MAX_LINKS)
1444 nl_table[protocol].nl_nonroot = flags;
1445 }
1446
1447 static void netlink_destroy_callback(struct netlink_callback *cb)
1448 {
1449 if (cb->skb)
1450 kfree_skb(cb->skb);
1451 kfree(cb);
1452 }
1453
1454 /*
1455 * It looks a bit ugly.
1456 * It would be better to create kernel thread.
1457 */
1458
1459 static int netlink_dump(struct sock *sk)
1460 {
1461 struct netlink_sock *nlk = nlk_sk(sk);
1462 struct netlink_callback *cb;
1463 struct sk_buff *skb;
1464 struct nlmsghdr *nlh;
1465 int len, err = -ENOBUFS;
1466
1467 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1468 if (!skb)
1469 goto errout;
1470
1471 mutex_lock(nlk->cb_mutex);
1472
1473 cb = nlk->cb;
1474 if (cb == NULL) {
1475 err = -EINVAL;
1476 goto errout_skb;
1477 }
1478
1479 len = cb->dump(skb, cb);
1480
1481 if (len > 0) {
1482 mutex_unlock(nlk->cb_mutex);
1483 skb_queue_tail(&sk->sk_receive_queue, skb);
1484 sk->sk_data_ready(sk, len);
1485 return 0;
1486 }
1487
1488 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1489 if (!nlh)
1490 goto errout_skb;
1491
1492 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1493
1494 skb_queue_tail(&sk->sk_receive_queue, skb);
1495 sk->sk_data_ready(sk, skb->len);
1496
1497 if (cb->done)
1498 cb->done(cb);
1499 nlk->cb = NULL;
1500 mutex_unlock(nlk->cb_mutex);
1501
1502 netlink_destroy_callback(cb);
1503 return 0;
1504
1505 errout_skb:
1506 mutex_unlock(nlk->cb_mutex);
1507 kfree_skb(skb);
1508 errout:
1509 return err;
1510 }
1511
1512 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1513 struct nlmsghdr *nlh,
1514 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1515 int (*done)(struct netlink_callback*))
1516 {
1517 struct netlink_callback *cb;
1518 struct sock *sk;
1519 struct netlink_sock *nlk;
1520
1521 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1522 if (cb == NULL)
1523 return -ENOBUFS;
1524
1525 cb->dump = dump;
1526 cb->done = done;
1527 cb->nlh = nlh;
1528 atomic_inc(&skb->users);
1529 cb->skb = skb;
1530
1531 sk = netlink_lookup(ssk->sk_net, ssk->sk_protocol, NETLINK_CB(skb).pid);
1532 if (sk == NULL) {
1533 netlink_destroy_callback(cb);
1534 return -ECONNREFUSED;
1535 }
1536 nlk = nlk_sk(sk);
1537 /* A dump is in progress... */
1538 mutex_lock(nlk->cb_mutex);
1539 if (nlk->cb) {
1540 mutex_unlock(nlk->cb_mutex);
1541 netlink_destroy_callback(cb);
1542 sock_put(sk);
1543 return -EBUSY;
1544 }
1545 nlk->cb = cb;
1546 mutex_unlock(nlk->cb_mutex);
1547
1548 netlink_dump(sk);
1549 sock_put(sk);
1550
1551 /* We successfully started a dump, by returning -EINTR we
1552 * signal the queue mangement to interrupt processing of
1553 * any netlink messages so userspace gets a chance to read
1554 * the results. */
1555 return -EINTR;
1556 }
1557
1558 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1559 {
1560 struct sk_buff *skb;
1561 struct nlmsghdr *rep;
1562 struct nlmsgerr *errmsg;
1563 size_t payload = sizeof(*errmsg);
1564
1565 /* error messages get the original request appened */
1566 if (err)
1567 payload += nlmsg_len(nlh);
1568
1569 skb = nlmsg_new(payload, GFP_KERNEL);
1570 if (!skb) {
1571 struct sock *sk;
1572
1573 sk = netlink_lookup(in_skb->sk->sk_net,
1574 in_skb->sk->sk_protocol,
1575 NETLINK_CB(in_skb).pid);
1576 if (sk) {
1577 sk->sk_err = ENOBUFS;
1578 sk->sk_error_report(sk);
1579 sock_put(sk);
1580 }
1581 return;
1582 }
1583
1584 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1585 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1586 errmsg = nlmsg_data(rep);
1587 errmsg->error = err;
1588 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1589 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1590 }
1591
1592 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1593 struct nlmsghdr *))
1594 {
1595 struct nlmsghdr *nlh;
1596 int err;
1597
1598 while (skb->len >= nlmsg_total_size(0)) {
1599 nlh = nlmsg_hdr(skb);
1600 err = 0;
1601
1602 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1603 return 0;
1604
1605 /* Only requests are handled by the kernel */
1606 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1607 goto skip;
1608
1609 /* Skip control messages */
1610 if (nlh->nlmsg_type < NLMSG_MIN_TYPE)
1611 goto skip;
1612
1613 err = cb(skb, nlh);
1614 if (err == -EINTR) {
1615 /* Not an error, but we interrupt processing */
1616 netlink_queue_skip(nlh, skb);
1617 return err;
1618 }
1619 skip:
1620 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1621 netlink_ack(skb, nlh, err);
1622
1623 netlink_queue_skip(nlh, skb);
1624 }
1625
1626 return 0;
1627 }
1628
1629 /**
1630 * nelink_run_queue - Process netlink receive queue.
1631 * @sk: Netlink socket containing the queue
1632 * @qlen: Place to store queue length upon entry
1633 * @cb: Callback function invoked for each netlink message found
1634 *
1635 * Processes as much as there was in the queue upon entry and invokes
1636 * a callback function for each netlink message found. The callback
1637 * function may refuse a message by returning a negative error code
1638 * but setting the error pointer to 0 in which case this function
1639 * returns with a qlen != 0.
1640 *
1641 * qlen must be initialized to 0 before the initial entry, afterwards
1642 * the function may be called repeatedly until qlen reaches 0.
1643 *
1644 * The callback function may return -EINTR to signal that processing
1645 * of netlink messages shall be interrupted. In this case the message
1646 * currently being processed will NOT be requeued onto the receive
1647 * queue.
1648 */
1649 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1650 int (*cb)(struct sk_buff *, struct nlmsghdr *))
1651 {
1652 struct sk_buff *skb;
1653
1654 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1655 *qlen = skb_queue_len(&sk->sk_receive_queue);
1656
1657 for (; *qlen; (*qlen)--) {
1658 skb = skb_dequeue(&sk->sk_receive_queue);
1659 if (netlink_rcv_skb(skb, cb)) {
1660 if (skb->len)
1661 skb_queue_head(&sk->sk_receive_queue, skb);
1662 else {
1663 kfree_skb(skb);
1664 (*qlen)--;
1665 }
1666 break;
1667 }
1668
1669 kfree_skb(skb);
1670 }
1671 }
1672
1673 /**
1674 * netlink_queue_skip - Skip netlink message while processing queue.
1675 * @nlh: Netlink message to be skipped
1676 * @skb: Socket buffer containing the netlink messages.
1677 *
1678 * Pulls the given netlink message off the socket buffer so the next
1679 * call to netlink_queue_run() will not reconsider the message.
1680 */
1681 static void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1682 {
1683 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1684
1685 if (msglen > skb->len)
1686 msglen = skb->len;
1687
1688 skb_pull(skb, msglen);
1689 }
1690
1691 /**
1692 * nlmsg_notify - send a notification netlink message
1693 * @sk: netlink socket to use
1694 * @skb: notification message
1695 * @pid: destination netlink pid for reports or 0
1696 * @group: destination multicast group or 0
1697 * @report: 1 to report back, 0 to disable
1698 * @flags: allocation flags
1699 */
1700 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1701 unsigned int group, int report, gfp_t flags)
1702 {
1703 int err = 0;
1704
1705 if (group) {
1706 int exclude_pid = 0;
1707
1708 if (report) {
1709 atomic_inc(&skb->users);
1710 exclude_pid = pid;
1711 }
1712
1713 /* errors reported via destination sk->sk_err */
1714 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1715 }
1716
1717 if (report)
1718 err = nlmsg_unicast(sk, skb, pid);
1719
1720 return err;
1721 }
1722
1723 #ifdef CONFIG_PROC_FS
1724 struct nl_seq_iter {
1725 struct net *net;
1726 int link;
1727 int hash_idx;
1728 };
1729
1730 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1731 {
1732 struct nl_seq_iter *iter = seq->private;
1733 int i, j;
1734 struct sock *s;
1735 struct hlist_node *node;
1736 loff_t off = 0;
1737
1738 for (i=0; i<MAX_LINKS; i++) {
1739 struct nl_pid_hash *hash = &nl_table[i].hash;
1740
1741 for (j = 0; j <= hash->mask; j++) {
1742 sk_for_each(s, node, &hash->table[j]) {
1743 if (iter->net != s->sk_net)
1744 continue;
1745 if (off == pos) {
1746 iter->link = i;
1747 iter->hash_idx = j;
1748 return s;
1749 }
1750 ++off;
1751 }
1752 }
1753 }
1754 return NULL;
1755 }
1756
1757 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1758 {
1759 read_lock(&nl_table_lock);
1760 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1761 }
1762
1763 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1764 {
1765 struct sock *s;
1766 struct nl_seq_iter *iter;
1767 int i, j;
1768
1769 ++*pos;
1770
1771 if (v == SEQ_START_TOKEN)
1772 return netlink_seq_socket_idx(seq, 0);
1773
1774 iter = seq->private;
1775 s = v;
1776 do {
1777 s = sk_next(s);
1778 } while (s && (iter->net != s->sk_net));
1779 if (s)
1780 return s;
1781
1782 i = iter->link;
1783 j = iter->hash_idx + 1;
1784
1785 do {
1786 struct nl_pid_hash *hash = &nl_table[i].hash;
1787
1788 for (; j <= hash->mask; j++) {
1789 s = sk_head(&hash->table[j]);
1790 while (s && (iter->net != s->sk_net))
1791 s = sk_next(s);
1792 if (s) {
1793 iter->link = i;
1794 iter->hash_idx = j;
1795 return s;
1796 }
1797 }
1798
1799 j = 0;
1800 } while (++i < MAX_LINKS);
1801
1802 return NULL;
1803 }
1804
1805 static void netlink_seq_stop(struct seq_file *seq, void *v)
1806 {
1807 read_unlock(&nl_table_lock);
1808 }
1809
1810
1811 static int netlink_seq_show(struct seq_file *seq, void *v)
1812 {
1813 if (v == SEQ_START_TOKEN)
1814 seq_puts(seq,
1815 "sk Eth Pid Groups "
1816 "Rmem Wmem Dump Locks\n");
1817 else {
1818 struct sock *s = v;
1819 struct netlink_sock *nlk = nlk_sk(s);
1820
1821 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1822 s,
1823 s->sk_protocol,
1824 nlk->pid,
1825 nlk->groups ? (u32)nlk->groups[0] : 0,
1826 atomic_read(&s->sk_rmem_alloc),
1827 atomic_read(&s->sk_wmem_alloc),
1828 nlk->cb,
1829 atomic_read(&s->sk_refcnt)
1830 );
1831
1832 }
1833 return 0;
1834 }
1835
1836 static const struct seq_operations netlink_seq_ops = {
1837 .start = netlink_seq_start,
1838 .next = netlink_seq_next,
1839 .stop = netlink_seq_stop,
1840 .show = netlink_seq_show,
1841 };
1842
1843
1844 static int netlink_seq_open(struct inode *inode, struct file *file)
1845 {
1846 struct seq_file *seq;
1847 struct nl_seq_iter *iter;
1848 int err;
1849
1850 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1851 if (!iter)
1852 return -ENOMEM;
1853
1854 err = seq_open(file, &netlink_seq_ops);
1855 if (err) {
1856 kfree(iter);
1857 return err;
1858 }
1859
1860 seq = file->private_data;
1861 seq->private = iter;
1862 iter->net = get_proc_net(inode);
1863 if (!iter->net) {
1864 seq_release_private(inode, file);
1865 return -ENXIO;
1866 }
1867 return 0;
1868 }
1869
1870 static int netlink_seq_release(struct inode *inode, struct file *file)
1871 {
1872 struct seq_file *seq = file->private_data;
1873 struct nl_seq_iter *iter = seq->private;
1874 put_net(iter->net);
1875 return seq_release_private(inode, file);
1876 }
1877
1878 static const struct file_operations netlink_seq_fops = {
1879 .owner = THIS_MODULE,
1880 .open = netlink_seq_open,
1881 .read = seq_read,
1882 .llseek = seq_lseek,
1883 .release = netlink_seq_release,
1884 };
1885
1886 #endif
1887
1888 int netlink_register_notifier(struct notifier_block *nb)
1889 {
1890 return atomic_notifier_chain_register(&netlink_chain, nb);
1891 }
1892
1893 int netlink_unregister_notifier(struct notifier_block *nb)
1894 {
1895 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1896 }
1897
1898 static const struct proto_ops netlink_ops = {
1899 .family = PF_NETLINK,
1900 .owner = THIS_MODULE,
1901 .release = netlink_release,
1902 .bind = netlink_bind,
1903 .connect = netlink_connect,
1904 .socketpair = sock_no_socketpair,
1905 .accept = sock_no_accept,
1906 .getname = netlink_getname,
1907 .poll = datagram_poll,
1908 .ioctl = sock_no_ioctl,
1909 .listen = sock_no_listen,
1910 .shutdown = sock_no_shutdown,
1911 .setsockopt = netlink_setsockopt,
1912 .getsockopt = netlink_getsockopt,
1913 .sendmsg = netlink_sendmsg,
1914 .recvmsg = netlink_recvmsg,
1915 .mmap = sock_no_mmap,
1916 .sendpage = sock_no_sendpage,
1917 };
1918
1919 static struct net_proto_family netlink_family_ops = {
1920 .family = PF_NETLINK,
1921 .create = netlink_create,
1922 .owner = THIS_MODULE, /* for consistency 8) */
1923 };
1924
1925 static int netlink_net_init(struct net *net)
1926 {
1927 #ifdef CONFIG_PROC_FS
1928 if (!proc_net_fops_create(net, "netlink", 0, &netlink_seq_fops))
1929 return -ENOMEM;
1930 #endif
1931 return 0;
1932 }
1933
1934 static void netlink_net_exit(struct net *net)
1935 {
1936 #ifdef CONFIG_PROC_FS
1937 proc_net_remove(net, "netlink");
1938 #endif
1939 }
1940
1941 static struct pernet_operations netlink_net_ops = {
1942 .init = netlink_net_init,
1943 .exit = netlink_net_exit,
1944 };
1945
1946 static int __init netlink_proto_init(void)
1947 {
1948 struct sk_buff *dummy_skb;
1949 int i;
1950 unsigned long max;
1951 unsigned int order;
1952 int err = proto_register(&netlink_proto, 0);
1953
1954 if (err != 0)
1955 goto out;
1956
1957 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1958
1959 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1960 if (!nl_table)
1961 goto panic;
1962
1963 if (num_physpages >= (128 * 1024))
1964 max = num_physpages >> (21 - PAGE_SHIFT);
1965 else
1966 max = num_physpages >> (23 - PAGE_SHIFT);
1967
1968 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1969 max = (1UL << order) / sizeof(struct hlist_head);
1970 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1971
1972 for (i = 0; i < MAX_LINKS; i++) {
1973 struct nl_pid_hash *hash = &nl_table[i].hash;
1974
1975 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1976 if (!hash->table) {
1977 while (i-- > 0)
1978 nl_pid_hash_free(nl_table[i].hash.table,
1979 1 * sizeof(*hash->table));
1980 kfree(nl_table);
1981 goto panic;
1982 }
1983 memset(hash->table, 0, 1 * sizeof(*hash->table));
1984 hash->max_shift = order;
1985 hash->shift = 0;
1986 hash->mask = 0;
1987 hash->rehash_time = jiffies;
1988 }
1989
1990 sock_register(&netlink_family_ops);
1991 register_pernet_subsys(&netlink_net_ops);
1992 /* The netlink device handler may be needed early. */
1993 rtnetlink_init();
1994 out:
1995 return err;
1996 panic:
1997 panic("netlink_init: Cannot allocate nl_table\n");
1998 }
1999
2000 core_initcall(netlink_proto_init);
2001
2002 EXPORT_SYMBOL(netlink_ack);
2003 EXPORT_SYMBOL(netlink_run_queue);
2004 EXPORT_SYMBOL(netlink_broadcast);
2005 EXPORT_SYMBOL(netlink_dump_start);
2006 EXPORT_SYMBOL(netlink_kernel_create);
2007 EXPORT_SYMBOL(netlink_register_notifier);
2008 EXPORT_SYMBOL(netlink_set_nonroot);
2009 EXPORT_SYMBOL(netlink_unicast);
2010 EXPORT_SYMBOL(netlink_unregister_notifier);
2011 EXPORT_SYMBOL(nlmsg_notify);
CJK support for tty framebuffer vt