projects / deliverable / linux.git / blob
? search:


7b455980e9bfcd520f127c26541f8f1180ad794f
[deliverable/linux.git] / net / netlink / af_netlink.c
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/smp_lock.h>
49 #include <linux/notifier.h>
50 #include <linux/security.h>
51 #include <linux/jhash.h>
52 #include <linux/jiffies.h>
53 #include <linux/random.h>
54 #include <linux/bitops.h>
55 #include <linux/mm.h>
56 #include <linux/types.h>
57 #include <linux/audit.h>
58 #include <linux/selinux.h>
59
60 #include <net/sock.h>
61 #include <net/scm.h>
62 #include <net/netlink.h>
63
64 #define NLGRPSZ(x) (ALIGN(x, sizeof(unsigned long) * 8) / 8)
65
66 struct netlink_sock {
67 /* struct sock has to be the first member of netlink_sock */
68 struct sock sk;
69 u32 pid;
70 u32 dst_pid;
71 u32 dst_group;
72 u32 flags;
73 u32 subscriptions;
74 u32 ngroups;
75 unsigned long *groups;
76 unsigned long state;
77 wait_queue_head_t wait;
78 struct netlink_callback *cb;
79 spinlock_t cb_lock;
80 void (*data_ready)(struct sock *sk, int bytes);
81 struct module *module;
82 };
83
84 #define NETLINK_KERNEL_SOCKET 0x1
85 #define NETLINK_RECV_PKTINFO 0x2
86
87 static inline struct netlink_sock *nlk_sk(struct sock *sk)
88 {
89 return (struct netlink_sock *)sk;
90 }
91
92 struct nl_pid_hash {
93 struct hlist_head *table;
94 unsigned long rehash_time;
95
96 unsigned int mask;
97 unsigned int shift;
98
99 unsigned int entries;
100 unsigned int max_shift;
101
102 u32 rnd;
103 };
104
105 struct netlink_table {
106 struct nl_pid_hash hash;
107 struct hlist_head mc_list;
108 unsigned long *listeners;
109 unsigned int nl_nonroot;
110 unsigned int groups;
111 struct module *module;
112 int registered;
113 };
114
115 static struct netlink_table *nl_table;
116
117 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
118
119 static int netlink_dump(struct sock *sk);
120 static void netlink_destroy_callback(struct netlink_callback *cb);
121
122 static DEFINE_RWLOCK(nl_table_lock);
123 static atomic_t nl_table_users = ATOMIC_INIT(0);
124
125 static ATOMIC_NOTIFIER_HEAD(netlink_chain);
126
127 static u32 netlink_group_mask(u32 group)
128 {
129 return group ? 1 << (group - 1) : 0;
130 }
131
132 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
133 {
134 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
135 }
136
137 static void netlink_sock_destruct(struct sock *sk)
138 {
139 skb_queue_purge(&sk->sk_receive_queue);
140
141 if (!sock_flag(sk, SOCK_DEAD)) {
142 printk("Freeing alive netlink socket %p\n", sk);
143 return;
144 }
145 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
146 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
147 BUG_TRAP(!nlk_sk(sk)->cb);
148 BUG_TRAP(!nlk_sk(sk)->groups);
149 }
150
151 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
152 * Look, when several writers sleep and reader wakes them up, all but one
153 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
154 * this, _but_ remember, it adds useless work on UP machines.
155 */
156
157 static void netlink_table_grab(void)
158 {
159 write_lock_irq(&nl_table_lock);
160
161 if (atomic_read(&nl_table_users)) {
162 DECLARE_WAITQUEUE(wait, current);
163
164 add_wait_queue_exclusive(&nl_table_wait, &wait);
165 for(;;) {
166 set_current_state(TASK_UNINTERRUPTIBLE);
167 if (atomic_read(&nl_table_users) == 0)
168 break;
169 write_unlock_irq(&nl_table_lock);
170 schedule();
171 write_lock_irq(&nl_table_lock);
172 }
173
174 __set_current_state(TASK_RUNNING);
175 remove_wait_queue(&nl_table_wait, &wait);
176 }
177 }
178
179 static __inline__ void netlink_table_ungrab(void)
180 {
181 write_unlock_irq(&nl_table_lock);
182 wake_up(&nl_table_wait);
183 }
184
185 static __inline__ void
186 netlink_lock_table(void)
187 {
188 /* read_lock() synchronizes us to netlink_table_grab */
189
190 read_lock(&nl_table_lock);
191 atomic_inc(&nl_table_users);
192 read_unlock(&nl_table_lock);
193 }
194
195 static __inline__ void
196 netlink_unlock_table(void)
197 {
198 if (atomic_dec_and_test(&nl_table_users))
199 wake_up(&nl_table_wait);
200 }
201
202 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
203 {
204 struct nl_pid_hash *hash = &nl_table[protocol].hash;
205 struct hlist_head *head;
206 struct sock *sk;
207 struct hlist_node *node;
208
209 read_lock(&nl_table_lock);
210 head = nl_pid_hashfn(hash, pid);
211 sk_for_each(sk, node, head) {
212 if (nlk_sk(sk)->pid == pid) {
213 sock_hold(sk);
214 goto found;
215 }
216 }
217 sk = NULL;
218 found:
219 read_unlock(&nl_table_lock);
220 return sk;
221 }
222
223 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
224 {
225 if (size <= PAGE_SIZE)
226 return kmalloc(size, GFP_ATOMIC);
227 else
228 return (struct hlist_head *)
229 __get_free_pages(GFP_ATOMIC, get_order(size));
230 }
231
232 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
233 {
234 if (size <= PAGE_SIZE)
235 kfree(table);
236 else
237 free_pages((unsigned long)table, get_order(size));
238 }
239
240 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
241 {
242 unsigned int omask, mask, shift;
243 size_t osize, size;
244 struct hlist_head *otable, *table;
245 int i;
246
247 omask = mask = hash->mask;
248 osize = size = (mask + 1) * sizeof(*table);
249 shift = hash->shift;
250
251 if (grow) {
252 if (++shift > hash->max_shift)
253 return 0;
254 mask = mask * 2 + 1;
255 size *= 2;
256 }
257
258 table = nl_pid_hash_alloc(size);
259 if (!table)
260 return 0;
261
262 memset(table, 0, size);
263 otable = hash->table;
264 hash->table = table;
265 hash->mask = mask;
266 hash->shift = shift;
267 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
268
269 for (i = 0; i <= omask; i++) {
270 struct sock *sk;
271 struct hlist_node *node, *tmp;
272
273 sk_for_each_safe(sk, node, tmp, &otable[i])
274 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
275 }
276
277 nl_pid_hash_free(otable, osize);
278 hash->rehash_time = jiffies + 10 * 60 * HZ;
279 return 1;
280 }
281
282 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
283 {
284 int avg = hash->entries >> hash->shift;
285
286 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
287 return 1;
288
289 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
290 nl_pid_hash_rehash(hash, 0);
291 return 1;
292 }
293
294 return 0;
295 }
296
297 static const struct proto_ops netlink_ops;
298
299 static void
300 netlink_update_listeners(struct sock *sk)
301 {
302 struct netlink_table *tbl = &nl_table[sk->sk_protocol];
303 struct hlist_node *node;
304 unsigned long mask;
305 unsigned int i;
306
307 for (i = 0; i < NLGRPSZ(tbl->groups)/sizeof(unsigned long); i++) {
308 mask = 0;
309 sk_for_each_bound(sk, node, &tbl->mc_list)
310 mask |= nlk_sk(sk)->groups[i];
311 tbl->listeners[i] = mask;
312 }
313 /* this function is only called with the netlink table "grabbed", which
314 * makes sure updates are visible before bind or setsockopt return. */
315 }
316
317 static int netlink_insert(struct sock *sk, u32 pid)
318 {
319 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
320 struct hlist_head *head;
321 int err = -EADDRINUSE;
322 struct sock *osk;
323 struct hlist_node *node;
324 int len;
325
326 netlink_table_grab();
327 head = nl_pid_hashfn(hash, pid);
328 len = 0;
329 sk_for_each(osk, node, head) {
330 if (nlk_sk(osk)->pid == pid)
331 break;
332 len++;
333 }
334 if (node)
335 goto err;
336
337 err = -EBUSY;
338 if (nlk_sk(sk)->pid)
339 goto err;
340
341 err = -ENOMEM;
342 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
343 goto err;
344
345 if (len && nl_pid_hash_dilute(hash, len))
346 head = nl_pid_hashfn(hash, pid);
347 hash->entries++;
348 nlk_sk(sk)->pid = pid;
349 sk_add_node(sk, head);
350 err = 0;
351
352 err:
353 netlink_table_ungrab();
354 return err;
355 }
356
357 static void netlink_remove(struct sock *sk)
358 {
359 netlink_table_grab();
360 if (sk_del_node_init(sk))
361 nl_table[sk->sk_protocol].hash.entries--;
362 if (nlk_sk(sk)->subscriptions)
363 __sk_del_bind_node(sk);
364 netlink_table_ungrab();
365 }
366
367 static struct proto netlink_proto = {
368 .name = "NETLINK",
369 .owner = THIS_MODULE,
370 .obj_size = sizeof(struct netlink_sock),
371 };
372
373 static int __netlink_create(struct socket *sock, int protocol)
374 {
375 struct sock *sk;
376 struct netlink_sock *nlk;
377
378 sock->ops = &netlink_ops;
379
380 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
381 if (!sk)
382 return -ENOMEM;
383
384 sock_init_data(sock, sk);
385
386 nlk = nlk_sk(sk);
387 spin_lock_init(&nlk->cb_lock);
388 init_waitqueue_head(&nlk->wait);
389
390 sk->sk_destruct = netlink_sock_destruct;
391 sk->sk_protocol = protocol;
392 return 0;
393 }
394
395 static int netlink_create(struct socket *sock, int protocol)
396 {
397 struct module *module = NULL;
398 struct netlink_sock *nlk;
399 int err = 0;
400
401 sock->state = SS_UNCONNECTED;
402
403 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
404 return -ESOCKTNOSUPPORT;
405
406 if (protocol<0 || protocol >= MAX_LINKS)
407 return -EPROTONOSUPPORT;
408
409 netlink_lock_table();
410 #ifdef CONFIG_KMOD
411 if (!nl_table[protocol].registered) {
412 netlink_unlock_table();
413 request_module("net-pf-%d-proto-%d", PF_NETLINK, protocol);
414 netlink_lock_table();
415 }
416 #endif
417 if (nl_table[protocol].registered &&
418 try_module_get(nl_table[protocol].module))
419 module = nl_table[protocol].module;
420 netlink_unlock_table();
421
422 if ((err = __netlink_create(sock, protocol)) < 0)
423 goto out_module;
424
425 nlk = nlk_sk(sock->sk);
426 nlk->module = module;
427 out:
428 return err;
429
430 out_module:
431 module_put(module);
432 goto out;
433 }
434
435 static int netlink_release(struct socket *sock)
436 {
437 struct sock *sk = sock->sk;
438 struct netlink_sock *nlk;
439
440 if (!sk)
441 return 0;
442
443 netlink_remove(sk);
444 sock_orphan(sk);
445 nlk = nlk_sk(sk);
446
447 spin_lock(&nlk->cb_lock);
448 if (nlk->cb) {
449 if (nlk->cb->done)
450 nlk->cb->done(nlk->cb);
451 netlink_destroy_callback(nlk->cb);
452 nlk->cb = NULL;
453 }
454 spin_unlock(&nlk->cb_lock);
455
456 /* OK. Socket is unlinked, and, therefore,
457 no new packets will arrive */
458
459 sock->sk = NULL;
460 wake_up_interruptible_all(&nlk->wait);
461
462 skb_queue_purge(&sk->sk_write_queue);
463
464 if (nlk->pid && !nlk->subscriptions) {
465 struct netlink_notify n = {
466 .protocol = sk->sk_protocol,
467 .pid = nlk->pid,
468 };
469 atomic_notifier_call_chain(&netlink_chain,
470 NETLINK_URELEASE, &n);
471 }
472
473 module_put(nlk->module);
474
475 netlink_table_grab();
476 if (nlk->flags & NETLINK_KERNEL_SOCKET) {
477 kfree(nl_table[sk->sk_protocol].listeners);
478 nl_table[sk->sk_protocol].module = NULL;
479 nl_table[sk->sk_protocol].registered = 0;
480 } else if (nlk->subscriptions)
481 netlink_update_listeners(sk);
482 netlink_table_ungrab();
483
484 kfree(nlk->groups);
485 nlk->groups = NULL;
486
487 sock_put(sk);
488 return 0;
489 }
490
491 static int netlink_autobind(struct socket *sock)
492 {
493 struct sock *sk = sock->sk;
494 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
495 struct hlist_head *head;
496 struct sock *osk;
497 struct hlist_node *node;
498 s32 pid = current->tgid;
499 int err;
500 static s32 rover = -4097;
501
502 retry:
503 cond_resched();
504 netlink_table_grab();
505 head = nl_pid_hashfn(hash, pid);
506 sk_for_each(osk, node, head) {
507 if (nlk_sk(osk)->pid == pid) {
508 /* Bind collision, search negative pid values. */
509 pid = rover--;
510 if (rover > -4097)
511 rover = -4097;
512 netlink_table_ungrab();
513 goto retry;
514 }
515 }
516 netlink_table_ungrab();
517
518 err = netlink_insert(sk, pid);
519 if (err == -EADDRINUSE)
520 goto retry;
521
522 /* If 2 threads race to autobind, that is fine. */
523 if (err == -EBUSY)
524 err = 0;
525
526 return err;
527 }
528
529 static inline int netlink_capable(struct socket *sock, unsigned int flag)
530 {
531 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
532 capable(CAP_NET_ADMIN);
533 }
534
535 static void
536 netlink_update_subscriptions(struct sock *sk, unsigned int subscriptions)
537 {
538 struct netlink_sock *nlk = nlk_sk(sk);
539
540 if (nlk->subscriptions && !subscriptions)
541 __sk_del_bind_node(sk);
542 else if (!nlk->subscriptions && subscriptions)
543 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
544 nlk->subscriptions = subscriptions;
545 }
546
547 static int netlink_alloc_groups(struct sock *sk)
548 {
549 struct netlink_sock *nlk = nlk_sk(sk);
550 unsigned int groups;
551 int err = 0;
552
553 netlink_lock_table();
554 groups = nl_table[sk->sk_protocol].groups;
555 if (!nl_table[sk->sk_protocol].registered)
556 err = -ENOENT;
557 netlink_unlock_table();
558
559 if (err)
560 return err;
561
562 nlk->groups = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
563 if (nlk->groups == NULL)
564 return -ENOMEM;
565 nlk->ngroups = groups;
566 return 0;
567 }
568
569 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
570 {
571 struct sock *sk = sock->sk;
572 struct netlink_sock *nlk = nlk_sk(sk);
573 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
574 int err;
575
576 if (nladdr->nl_family != AF_NETLINK)
577 return -EINVAL;
578
579 /* Only superuser is allowed to listen multicasts */
580 if (nladdr->nl_groups) {
581 if (!netlink_capable(sock, NL_NONROOT_RECV))
582 return -EPERM;
583 if (nlk->groups == NULL) {
584 err = netlink_alloc_groups(sk);
585 if (err)
586 return err;
587 }
588 }
589
590 if (nlk->pid) {
591 if (nladdr->nl_pid != nlk->pid)
592 return -EINVAL;
593 } else {
594 err = nladdr->nl_pid ?
595 netlink_insert(sk, nladdr->nl_pid) :
596 netlink_autobind(sock);
597 if (err)
598 return err;
599 }
600
601 if (!nladdr->nl_groups && (nlk->groups == NULL || !(u32)nlk->groups[0]))
602 return 0;
603
604 netlink_table_grab();
605 netlink_update_subscriptions(sk, nlk->subscriptions +
606 hweight32(nladdr->nl_groups) -
607 hweight32(nlk->groups[0]));
608 nlk->groups[0] = (nlk->groups[0] & ~0xffffffffUL) | nladdr->nl_groups;
609 netlink_update_listeners(sk);
610 netlink_table_ungrab();
611
612 return 0;
613 }
614
615 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
616 int alen, int flags)
617 {
618 int err = 0;
619 struct sock *sk = sock->sk;
620 struct netlink_sock *nlk = nlk_sk(sk);
621 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
622
623 if (addr->sa_family == AF_UNSPEC) {
624 sk->sk_state = NETLINK_UNCONNECTED;
625 nlk->dst_pid = 0;
626 nlk->dst_group = 0;
627 return 0;
628 }
629 if (addr->sa_family != AF_NETLINK)
630 return -EINVAL;
631
632 /* Only superuser is allowed to send multicasts */
633 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
634 return -EPERM;
635
636 if (!nlk->pid)
637 err = netlink_autobind(sock);
638
639 if (err == 0) {
640 sk->sk_state = NETLINK_CONNECTED;
641 nlk->dst_pid = nladdr->nl_pid;
642 nlk->dst_group = ffs(nladdr->nl_groups);
643 }
644
645 return err;
646 }
647
648 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
649 {
650 struct sock *sk = sock->sk;
651 struct netlink_sock *nlk = nlk_sk(sk);
652 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
653
654 nladdr->nl_family = AF_NETLINK;
655 nladdr->nl_pad = 0;
656 *addr_len = sizeof(*nladdr);
657
658 if (peer) {
659 nladdr->nl_pid = nlk->dst_pid;
660 nladdr->nl_groups = netlink_group_mask(nlk->dst_group);
661 } else {
662 nladdr->nl_pid = nlk->pid;
663 nladdr->nl_groups = nlk->groups ? nlk->groups[0] : 0;
664 }
665 return 0;
666 }
667
668 static void netlink_overrun(struct sock *sk)
669 {
670 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
671 sk->sk_err = ENOBUFS;
672 sk->sk_error_report(sk);
673 }
674 }
675
676 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
677 {
678 int protocol = ssk->sk_protocol;
679 struct sock *sock;
680 struct netlink_sock *nlk;
681
682 sock = netlink_lookup(protocol, pid);
683 if (!sock)
684 return ERR_PTR(-ECONNREFUSED);
685
686 /* Don't bother queuing skb if kernel socket has no input function */
687 nlk = nlk_sk(sock);
688 if ((nlk->pid == 0 && !nlk->data_ready) ||
689 (sock->sk_state == NETLINK_CONNECTED &&
690 nlk->dst_pid != nlk_sk(ssk)->pid)) {
691 sock_put(sock);
692 return ERR_PTR(-ECONNREFUSED);
693 }
694 return sock;
695 }
696
697 struct sock *netlink_getsockbyfilp(struct file *filp)
698 {
699 struct inode *inode = filp->f_path.dentry->d_inode;
700 struct sock *sock;
701
702 if (!S_ISSOCK(inode->i_mode))
703 return ERR_PTR(-ENOTSOCK);
704
705 sock = SOCKET_I(inode)->sk;
706 if (sock->sk_family != AF_NETLINK)
707 return ERR_PTR(-EINVAL);
708
709 sock_hold(sock);
710 return sock;
711 }
712
713 /*
714 * Attach a skb to a netlink socket.
715 * The caller must hold a reference to the destination socket. On error, the
716 * reference is dropped. The skb is not send to the destination, just all
717 * all error checks are performed and memory in the queue is reserved.
718 * Return values:
719 * < 0: error. skb freed, reference to sock dropped.
720 * 0: continue
721 * 1: repeat lookup - reference dropped while waiting for socket memory.
722 */
723 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock,
724 long timeo, struct sock *ssk)
725 {
726 struct netlink_sock *nlk;
727
728 nlk = nlk_sk(sk);
729
730 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
731 test_bit(0, &nlk->state)) {
732 DECLARE_WAITQUEUE(wait, current);
733 if (!timeo) {
734 if (!ssk || nlk_sk(ssk)->pid == 0)
735 netlink_overrun(sk);
736 sock_put(sk);
737 kfree_skb(skb);
738 return -EAGAIN;
739 }
740
741 __set_current_state(TASK_INTERRUPTIBLE);
742 add_wait_queue(&nlk->wait, &wait);
743
744 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
745 test_bit(0, &nlk->state)) &&
746 !sock_flag(sk, SOCK_DEAD))
747 timeo = schedule_timeout(timeo);
748
749 __set_current_state(TASK_RUNNING);
750 remove_wait_queue(&nlk->wait, &wait);
751 sock_put(sk);
752
753 if (signal_pending(current)) {
754 kfree_skb(skb);
755 return sock_intr_errno(timeo);
756 }
757 return 1;
758 }
759 skb_set_owner_r(skb, sk);
760 return 0;
761 }
762
763 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
764 {
765 int len = skb->len;
766
767 skb_queue_tail(&sk->sk_receive_queue, skb);
768 sk->sk_data_ready(sk, len);
769 sock_put(sk);
770 return len;
771 }
772
773 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
774 {
775 kfree_skb(skb);
776 sock_put(sk);
777 }
778
779 static inline struct sk_buff *netlink_trim(struct sk_buff *skb,
780 gfp_t allocation)
781 {
782 int delta;
783
784 skb_orphan(skb);
785
786 delta = skb->end - skb->tail;
787 if (delta * 2 < skb->truesize)
788 return skb;
789
790 if (skb_shared(skb)) {
791 struct sk_buff *nskb = skb_clone(skb, allocation);
792 if (!nskb)
793 return skb;
794 kfree_skb(skb);
795 skb = nskb;
796 }
797
798 if (!pskb_expand_head(skb, 0, -delta, allocation))
799 skb->truesize -= delta;
800
801 return skb;
802 }
803
804 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
805 {
806 struct sock *sk;
807 int err;
808 long timeo;
809
810 skb = netlink_trim(skb, gfp_any());
811
812 timeo = sock_sndtimeo(ssk, nonblock);
813 retry:
814 sk = netlink_getsockbypid(ssk, pid);
815 if (IS_ERR(sk)) {
816 kfree_skb(skb);
817 return PTR_ERR(sk);
818 }
819 err = netlink_attachskb(sk, skb, nonblock, timeo, ssk);
820 if (err == 1)
821 goto retry;
822 if (err)
823 return err;
824
825 return netlink_sendskb(sk, skb, ssk->sk_protocol);
826 }
827
828 int netlink_has_listeners(struct sock *sk, unsigned int group)
829 {
830 int res = 0;
831
832 BUG_ON(!(nlk_sk(sk)->flags & NETLINK_KERNEL_SOCKET));
833 if (group - 1 < nl_table[sk->sk_protocol].groups)
834 res = test_bit(group - 1, nl_table[sk->sk_protocol].listeners);
835 return res;
836 }
837 EXPORT_SYMBOL_GPL(netlink_has_listeners);
838
839 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
840 {
841 struct netlink_sock *nlk = nlk_sk(sk);
842
843 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
844 !test_bit(0, &nlk->state)) {
845 skb_set_owner_r(skb, sk);
846 skb_queue_tail(&sk->sk_receive_queue, skb);
847 sk->sk_data_ready(sk, skb->len);
848 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
849 }
850 return -1;
851 }
852
853 struct netlink_broadcast_data {
854 struct sock *exclude_sk;
855 u32 pid;
856 u32 group;
857 int failure;
858 int congested;
859 int delivered;
860 gfp_t allocation;
861 struct sk_buff *skb, *skb2;
862 };
863
864 static inline int do_one_broadcast(struct sock *sk,
865 struct netlink_broadcast_data *p)
866 {
867 struct netlink_sock *nlk = nlk_sk(sk);
868 int val;
869
870 if (p->exclude_sk == sk)
871 goto out;
872
873 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
874 !test_bit(p->group - 1, nlk->groups))
875 goto out;
876
877 if (p->failure) {
878 netlink_overrun(sk);
879 goto out;
880 }
881
882 sock_hold(sk);
883 if (p->skb2 == NULL) {
884 if (skb_shared(p->skb)) {
885 p->skb2 = skb_clone(p->skb, p->allocation);
886 } else {
887 p->skb2 = skb_get(p->skb);
888 /*
889 * skb ownership may have been set when
890 * delivered to a previous socket.
891 */
892 skb_orphan(p->skb2);
893 }
894 }
895 if (p->skb2 == NULL) {
896 netlink_overrun(sk);
897 /* Clone failed. Notify ALL listeners. */
898 p->failure = 1;
899 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
900 netlink_overrun(sk);
901 } else {
902 p->congested |= val;
903 p->delivered = 1;
904 p->skb2 = NULL;
905 }
906 sock_put(sk);
907
908 out:
909 return 0;
910 }
911
912 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
913 u32 group, gfp_t allocation)
914 {
915 struct netlink_broadcast_data info;
916 struct hlist_node *node;
917 struct sock *sk;
918
919 skb = netlink_trim(skb, allocation);
920
921 info.exclude_sk = ssk;
922 info.pid = pid;
923 info.group = group;
924 info.failure = 0;
925 info.congested = 0;
926 info.delivered = 0;
927 info.allocation = allocation;
928 info.skb = skb;
929 info.skb2 = NULL;
930
931 /* While we sleep in clone, do not allow to change socket list */
932
933 netlink_lock_table();
934
935 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
936 do_one_broadcast(sk, &info);
937
938 kfree_skb(skb);
939
940 netlink_unlock_table();
941
942 if (info.skb2)
943 kfree_skb(info.skb2);
944
945 if (info.delivered) {
946 if (info.congested && (allocation & __GFP_WAIT))
947 yield();
948 return 0;
949 }
950 if (info.failure)
951 return -ENOBUFS;
952 return -ESRCH;
953 }
954
955 struct netlink_set_err_data {
956 struct sock *exclude_sk;
957 u32 pid;
958 u32 group;
959 int code;
960 };
961
962 static inline int do_one_set_err(struct sock *sk,
963 struct netlink_set_err_data *p)
964 {
965 struct netlink_sock *nlk = nlk_sk(sk);
966
967 if (sk == p->exclude_sk)
968 goto out;
969
970 if (nlk->pid == p->pid || p->group - 1 >= nlk->ngroups ||
971 !test_bit(p->group - 1, nlk->groups))
972 goto out;
973
974 sk->sk_err = p->code;
975 sk->sk_error_report(sk);
976 out:
977 return 0;
978 }
979
980 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
981 {
982 struct netlink_set_err_data info;
983 struct hlist_node *node;
984 struct sock *sk;
985
986 info.exclude_sk = ssk;
987 info.pid = pid;
988 info.group = group;
989 info.code = code;
990
991 read_lock(&nl_table_lock);
992
993 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
994 do_one_set_err(sk, &info);
995
996 read_unlock(&nl_table_lock);
997 }
998
999 static int netlink_setsockopt(struct socket *sock, int level, int optname,
1000 char __user *optval, int optlen)
1001 {
1002 struct sock *sk = sock->sk;
1003 struct netlink_sock *nlk = nlk_sk(sk);
1004 int val = 0, err;
1005
1006 if (level != SOL_NETLINK)
1007 return -ENOPROTOOPT;
1008
1009 if (optlen >= sizeof(int) &&
1010 get_user(val, (int __user *)optval))
1011 return -EFAULT;
1012
1013 switch (optname) {
1014 case NETLINK_PKTINFO:
1015 if (val)
1016 nlk->flags |= NETLINK_RECV_PKTINFO;
1017 else
1018 nlk->flags &= ~NETLINK_RECV_PKTINFO;
1019 err = 0;
1020 break;
1021 case NETLINK_ADD_MEMBERSHIP:
1022 case NETLINK_DROP_MEMBERSHIP: {
1023 unsigned int subscriptions;
1024 int old, new = optname == NETLINK_ADD_MEMBERSHIP ? 1 : 0;
1025
1026 if (!netlink_capable(sock, NL_NONROOT_RECV))
1027 return -EPERM;
1028 if (nlk->groups == NULL) {
1029 err = netlink_alloc_groups(sk);
1030 if (err)
1031 return err;
1032 }
1033 if (!val || val - 1 >= nlk->ngroups)
1034 return -EINVAL;
1035 netlink_table_grab();
1036 old = test_bit(val - 1, nlk->groups);
1037 subscriptions = nlk->subscriptions - old + new;
1038 if (new)
1039 __set_bit(val - 1, nlk->groups);
1040 else
1041 __clear_bit(val - 1, nlk->groups);
1042 netlink_update_subscriptions(sk, subscriptions);
1043 netlink_update_listeners(sk);
1044 netlink_table_ungrab();
1045 err = 0;
1046 break;
1047 }
1048 default:
1049 err = -ENOPROTOOPT;
1050 }
1051 return err;
1052 }
1053
1054 static int netlink_getsockopt(struct socket *sock, int level, int optname,
1055 char __user *optval, int __user *optlen)
1056 {
1057 struct sock *sk = sock->sk;
1058 struct netlink_sock *nlk = nlk_sk(sk);
1059 int len, val, err;
1060
1061 if (level != SOL_NETLINK)
1062 return -ENOPROTOOPT;
1063
1064 if (get_user(len, optlen))
1065 return -EFAULT;
1066 if (len < 0)
1067 return -EINVAL;
1068
1069 switch (optname) {
1070 case NETLINK_PKTINFO:
1071 if (len < sizeof(int))
1072 return -EINVAL;
1073 len = sizeof(int);
1074 val = nlk->flags & NETLINK_RECV_PKTINFO ? 1 : 0;
1075 if (put_user(len, optlen) ||
1076 put_user(val, optval))
1077 return -EFAULT;
1078 err = 0;
1079 break;
1080 default:
1081 err = -ENOPROTOOPT;
1082 }
1083 return err;
1084 }
1085
1086 static void netlink_cmsg_recv_pktinfo(struct msghdr *msg, struct sk_buff *skb)
1087 {
1088 struct nl_pktinfo info;
1089
1090 info.group = NETLINK_CB(skb).dst_group;
1091 put_cmsg(msg, SOL_NETLINK, NETLINK_PKTINFO, sizeof(info), &info);
1092 }
1093
1094 static inline void netlink_rcv_wake(struct sock *sk)
1095 {
1096 struct netlink_sock *nlk = nlk_sk(sk);
1097
1098 if (skb_queue_empty(&sk->sk_receive_queue))
1099 clear_bit(0, &nlk->state);
1100 if (!test_bit(0, &nlk->state))
1101 wake_up_interruptible(&nlk->wait);
1102 }
1103
1104 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
1105 struct msghdr *msg, size_t len)
1106 {
1107 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1108 struct sock *sk = sock->sk;
1109 struct netlink_sock *nlk = nlk_sk(sk);
1110 struct sockaddr_nl *addr=msg->msg_name;
1111 u32 dst_pid;
1112 u32 dst_group;
1113 struct sk_buff *skb;
1114 int err;
1115 struct scm_cookie scm;
1116
1117 if (msg->msg_flags&MSG_OOB)
1118 return -EOPNOTSUPP;
1119
1120 if (NULL == siocb->scm)
1121 siocb->scm = &scm;
1122 err = scm_send(sock, msg, siocb->scm);
1123 if (err < 0)
1124 return err;
1125
1126 if (msg->msg_namelen) {
1127 if (addr->nl_family != AF_NETLINK)
1128 return -EINVAL;
1129 dst_pid = addr->nl_pid;
1130 dst_group = ffs(addr->nl_groups);
1131 if (dst_group && !netlink_capable(sock, NL_NONROOT_SEND))
1132 return -EPERM;
1133 } else {
1134 dst_pid = nlk->dst_pid;
1135 dst_group = nlk->dst_group;
1136 }
1137
1138 if (!nlk->pid) {
1139 err = netlink_autobind(sock);
1140 if (err)
1141 goto out;
1142 }
1143
1144 err = -EMSGSIZE;
1145 if (len > sk->sk_sndbuf - 32)
1146 goto out;
1147 err = -ENOBUFS;
1148 skb = alloc_skb(len, GFP_KERNEL);
1149 if (skb==NULL)
1150 goto out;
1151
1152 NETLINK_CB(skb).pid = nlk->pid;
1153 NETLINK_CB(skb).dst_group = dst_group;
1154 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
1155 selinux_get_task_sid(current, &(NETLINK_CB(skb).sid));
1156 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
1157
1158 /* What can I do? Netlink is asynchronous, so that
1159 we will have to save current capabilities to
1160 check them, when this message will be delivered
1161 to corresponding kernel module. --ANK (980802)
1162 */
1163
1164 err = -EFAULT;
1165 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
1166 kfree_skb(skb);
1167 goto out;
1168 }
1169
1170 err = security_netlink_send(sk, skb);
1171 if (err) {
1172 kfree_skb(skb);
1173 goto out;
1174 }
1175
1176 if (dst_group) {
1177 atomic_inc(&skb->users);
1178 netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
1179 }
1180 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
1181
1182 out:
1183 return err;
1184 }
1185
1186 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
1187 struct msghdr *msg, size_t len,
1188 int flags)
1189 {
1190 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
1191 struct scm_cookie scm;
1192 struct sock *sk = sock->sk;
1193 struct netlink_sock *nlk = nlk_sk(sk);
1194 int noblock = flags&MSG_DONTWAIT;
1195 size_t copied;
1196 struct sk_buff *skb;
1197 int err;
1198
1199 if (flags&MSG_OOB)
1200 return -EOPNOTSUPP;
1201
1202 copied = 0;
1203
1204 skb = skb_recv_datagram(sk,flags,noblock,&err);
1205 if (skb==NULL)
1206 goto out;
1207
1208 msg->msg_namelen = 0;
1209
1210 copied = skb->len;
1211 if (len < copied) {
1212 msg->msg_flags |= MSG_TRUNC;
1213 copied = len;
1214 }
1215
1216 skb_reset_transport_header(skb);
1217 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1218
1219 if (msg->msg_name) {
1220 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
1221 addr->nl_family = AF_NETLINK;
1222 addr->nl_pad = 0;
1223 addr->nl_pid = NETLINK_CB(skb).pid;
1224 addr->nl_groups = netlink_group_mask(NETLINK_CB(skb).dst_group);
1225 msg->msg_namelen = sizeof(*addr);
1226 }
1227
1228 if (nlk->flags & NETLINK_RECV_PKTINFO)
1229 netlink_cmsg_recv_pktinfo(msg, skb);
1230
1231 if (NULL == siocb->scm) {
1232 memset(&scm, 0, sizeof(scm));
1233 siocb->scm = &scm;
1234 }
1235 siocb->scm->creds = *NETLINK_CREDS(skb);
1236 skb_free_datagram(sk, skb);
1237
1238 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
1239 netlink_dump(sk);
1240
1241 scm_recv(sock, msg, siocb->scm, flags);
1242
1243 if (flags & MSG_TRUNC)
1244 copied = skb->len;
1245
1246 out:
1247 netlink_rcv_wake(sk);
1248 return err ? : copied;
1249 }
1250
1251 static void netlink_data_ready(struct sock *sk, int len)
1252 {
1253 struct netlink_sock *nlk = nlk_sk(sk);
1254
1255 if (nlk->data_ready)
1256 nlk->data_ready(sk, len);
1257 netlink_rcv_wake(sk);
1258 }
1259
1260 /*
1261 * We export these functions to other modules. They provide a
1262 * complete set of kernel non-blocking support for message
1263 * queueing.
1264 */
1265
1266 struct sock *
1267 netlink_kernel_create(int unit, unsigned int groups,
1268 void (*input)(struct sock *sk, int len),
1269 struct module *module)
1270 {
1271 struct socket *sock;
1272 struct sock *sk;
1273 struct netlink_sock *nlk;
1274 unsigned long *listeners = NULL;
1275
1276 BUG_ON(!nl_table);
1277
1278 if (unit<0 || unit>=MAX_LINKS)
1279 return NULL;
1280
1281 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1282 return NULL;
1283
1284 if (__netlink_create(sock, unit) < 0)
1285 goto out_sock_release;
1286
1287 if (groups < 32)
1288 groups = 32;
1289
1290 listeners = kzalloc(NLGRPSZ(groups), GFP_KERNEL);
1291 if (!listeners)
1292 goto out_sock_release;
1293
1294 sk = sock->sk;
1295 sk->sk_data_ready = netlink_data_ready;
1296 if (input)
1297 nlk_sk(sk)->data_ready = input;
1298
1299 if (netlink_insert(sk, 0))
1300 goto out_sock_release;
1301
1302 nlk = nlk_sk(sk);
1303 nlk->flags |= NETLINK_KERNEL_SOCKET;
1304
1305 netlink_table_grab();
1306 nl_table[unit].groups = groups;
1307 nl_table[unit].listeners = listeners;
1308 nl_table[unit].module = module;
1309 nl_table[unit].registered = 1;
1310 netlink_table_ungrab();
1311
1312 return sk;
1313
1314 out_sock_release:
1315 kfree(listeners);
1316 sock_release(sock);
1317 return NULL;
1318 }
1319
1320 void netlink_set_nonroot(int protocol, unsigned int flags)
1321 {
1322 if ((unsigned int)protocol < MAX_LINKS)
1323 nl_table[protocol].nl_nonroot = flags;
1324 }
1325
1326 static void netlink_destroy_callback(struct netlink_callback *cb)
1327 {
1328 if (cb->skb)
1329 kfree_skb(cb->skb);
1330 kfree(cb);
1331 }
1332
1333 /*
1334 * It looks a bit ugly.
1335 * It would be better to create kernel thread.
1336 */
1337
1338 static int netlink_dump(struct sock *sk)
1339 {
1340 struct netlink_sock *nlk = nlk_sk(sk);
1341 struct netlink_callback *cb;
1342 struct sk_buff *skb;
1343 struct nlmsghdr *nlh;
1344 int len, err = -ENOBUFS;
1345
1346 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1347 if (!skb)
1348 goto errout;
1349
1350 spin_lock(&nlk->cb_lock);
1351
1352 cb = nlk->cb;
1353 if (cb == NULL) {
1354 err = -EINVAL;
1355 goto errout_skb;
1356 }
1357
1358 len = cb->dump(skb, cb);
1359
1360 if (len > 0) {
1361 spin_unlock(&nlk->cb_lock);
1362 skb_queue_tail(&sk->sk_receive_queue, skb);
1363 sk->sk_data_ready(sk, len);
1364 return 0;
1365 }
1366
1367 nlh = nlmsg_put_answer(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1368 if (!nlh)
1369 goto errout_skb;
1370
1371 memcpy(nlmsg_data(nlh), &len, sizeof(len));
1372
1373 skb_queue_tail(&sk->sk_receive_queue, skb);
1374 sk->sk_data_ready(sk, skb->len);
1375
1376 if (cb->done)
1377 cb->done(cb);
1378 nlk->cb = NULL;
1379 spin_unlock(&nlk->cb_lock);
1380
1381 netlink_destroy_callback(cb);
1382 return 0;
1383
1384 errout_skb:
1385 spin_unlock(&nlk->cb_lock);
1386 kfree_skb(skb);
1387 errout:
1388 return err;
1389 }
1390
1391 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1392 struct nlmsghdr *nlh,
1393 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1394 int (*done)(struct netlink_callback*))
1395 {
1396 struct netlink_callback *cb;
1397 struct sock *sk;
1398 struct netlink_sock *nlk;
1399
1400 cb = kzalloc(sizeof(*cb), GFP_KERNEL);
1401 if (cb == NULL)
1402 return -ENOBUFS;
1403
1404 cb->dump = dump;
1405 cb->done = done;
1406 cb->nlh = nlh;
1407 atomic_inc(&skb->users);
1408 cb->skb = skb;
1409
1410 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1411 if (sk == NULL) {
1412 netlink_destroy_callback(cb);
1413 return -ECONNREFUSED;
1414 }
1415 nlk = nlk_sk(sk);
1416 /* A dump or destruction is in progress... */
1417 spin_lock(&nlk->cb_lock);
1418 if (nlk->cb || sock_flag(sk, SOCK_DEAD)) {
1419 spin_unlock(&nlk->cb_lock);
1420 netlink_destroy_callback(cb);
1421 sock_put(sk);
1422 return -EBUSY;
1423 }
1424 nlk->cb = cb;
1425 spin_unlock(&nlk->cb_lock);
1426
1427 netlink_dump(sk);
1428 sock_put(sk);
1429 return 0;
1430 }
1431
1432 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1433 {
1434 struct sk_buff *skb;
1435 struct nlmsghdr *rep;
1436 struct nlmsgerr *errmsg;
1437 size_t payload = sizeof(*errmsg);
1438
1439 /* error messages get the original request appened */
1440 if (err)
1441 payload += nlmsg_len(nlh);
1442
1443 skb = nlmsg_new(payload, GFP_KERNEL);
1444 if (!skb) {
1445 struct sock *sk;
1446
1447 sk = netlink_lookup(in_skb->sk->sk_protocol,
1448 NETLINK_CB(in_skb).pid);
1449 if (sk) {
1450 sk->sk_err = ENOBUFS;
1451 sk->sk_error_report(sk);
1452 sock_put(sk);
1453 }
1454 return;
1455 }
1456
1457 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1458 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1459 errmsg = nlmsg_data(rep);
1460 errmsg->error = err;
1461 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(*nlh));
1462 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1463 }
1464
1465 static int netlink_rcv_skb(struct sk_buff *skb, int (*cb)(struct sk_buff *,
1466 struct nlmsghdr *, int *))
1467 {
1468 struct nlmsghdr *nlh;
1469 int err;
1470
1471 while (skb->len >= nlmsg_total_size(0)) {
1472 nlh = nlmsg_hdr(skb);
1473 err = 0;
1474
1475 if (nlh->nlmsg_len < NLMSG_HDRLEN || skb->len < nlh->nlmsg_len)
1476 return 0;
1477
1478 /* Only requests are handled by the kernel */
1479 if (!(nlh->nlmsg_flags & NLM_F_REQUEST))
1480 goto skip;
1481
1482 if (cb(skb, nlh, &err) < 0) {
1483 /* Not an error, but we have to interrupt processing
1484 * here. Note: that in this case we do not pull
1485 * message from skb, it will be processed later.
1486 */
1487 if (err == 0)
1488 return -1;
1489 }
1490 skip:
1491 if (nlh->nlmsg_flags & NLM_F_ACK || err)
1492 netlink_ack(skb, nlh, err);
1493
1494 netlink_queue_skip(nlh, skb);
1495 }
1496
1497 return 0;
1498 }
1499
1500 /**
1501 * nelink_run_queue - Process netlink receive queue.
1502 * @sk: Netlink socket containing the queue
1503 * @qlen: Place to store queue length upon entry
1504 * @cb: Callback function invoked for each netlink message found
1505 *
1506 * Processes as much as there was in the queue upon entry and invokes
1507 * a callback function for each netlink message found. The callback
1508 * function may refuse a message by returning a negative error code
1509 * but setting the error pointer to 0 in which case this function
1510 * returns with a qlen != 0.
1511 *
1512 * qlen must be initialized to 0 before the initial entry, afterwards
1513 * the function may be called repeatedly until qlen reaches 0.
1514 */
1515 void netlink_run_queue(struct sock *sk, unsigned int *qlen,
1516 int (*cb)(struct sk_buff *, struct nlmsghdr *, int *))
1517 {
1518 struct sk_buff *skb;
1519
1520 if (!*qlen || *qlen > skb_queue_len(&sk->sk_receive_queue))
1521 *qlen = skb_queue_len(&sk->sk_receive_queue);
1522
1523 for (; *qlen; (*qlen)--) {
1524 skb = skb_dequeue(&sk->sk_receive_queue);
1525 if (netlink_rcv_skb(skb, cb)) {
1526 if (skb->len)
1527 skb_queue_head(&sk->sk_receive_queue, skb);
1528 else {
1529 kfree_skb(skb);
1530 (*qlen)--;
1531 }
1532 break;
1533 }
1534
1535 kfree_skb(skb);
1536 }
1537 }
1538
1539 /**
1540 * netlink_queue_skip - Skip netlink message while processing queue.
1541 * @nlh: Netlink message to be skipped
1542 * @skb: Socket buffer containing the netlink messages.
1543 *
1544 * Pulls the given netlink message off the socket buffer so the next
1545 * call to netlink_queue_run() will not reconsider the message.
1546 */
1547 void netlink_queue_skip(struct nlmsghdr *nlh, struct sk_buff *skb)
1548 {
1549 int msglen = NLMSG_ALIGN(nlh->nlmsg_len);
1550
1551 if (msglen > skb->len)
1552 msglen = skb->len;
1553
1554 skb_pull(skb, msglen);
1555 }
1556
1557 /**
1558 * nlmsg_notify - send a notification netlink message
1559 * @sk: netlink socket to use
1560 * @skb: notification message
1561 * @pid: destination netlink pid for reports or 0
1562 * @group: destination multicast group or 0
1563 * @report: 1 to report back, 0 to disable
1564 * @flags: allocation flags
1565 */
1566 int nlmsg_notify(struct sock *sk, struct sk_buff *skb, u32 pid,
1567 unsigned int group, int report, gfp_t flags)
1568 {
1569 int err = 0;
1570
1571 if (group) {
1572 int exclude_pid = 0;
1573
1574 if (report) {
1575 atomic_inc(&skb->users);
1576 exclude_pid = pid;
1577 }
1578
1579 /* errors reported via destination sk->sk_err */
1580 nlmsg_multicast(sk, skb, exclude_pid, group, flags);
1581 }
1582
1583 if (report)
1584 err = nlmsg_unicast(sk, skb, pid);
1585
1586 return err;
1587 }
1588
1589 #ifdef CONFIG_PROC_FS
1590 struct nl_seq_iter {
1591 int link;
1592 int hash_idx;
1593 };
1594
1595 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1596 {
1597 struct nl_seq_iter *iter = seq->private;
1598 int i, j;
1599 struct sock *s;
1600 struct hlist_node *node;
1601 loff_t off = 0;
1602
1603 for (i=0; i<MAX_LINKS; i++) {
1604 struct nl_pid_hash *hash = &nl_table[i].hash;
1605
1606 for (j = 0; j <= hash->mask; j++) {
1607 sk_for_each(s, node, &hash->table[j]) {
1608 if (off == pos) {
1609 iter->link = i;
1610 iter->hash_idx = j;
1611 return s;
1612 }
1613 ++off;
1614 }
1615 }
1616 }
1617 return NULL;
1618 }
1619
1620 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1621 {
1622 read_lock(&nl_table_lock);
1623 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1624 }
1625
1626 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1627 {
1628 struct sock *s;
1629 struct nl_seq_iter *iter;
1630 int i, j;
1631
1632 ++*pos;
1633
1634 if (v == SEQ_START_TOKEN)
1635 return netlink_seq_socket_idx(seq, 0);
1636
1637 s = sk_next(v);
1638 if (s)
1639 return s;
1640
1641 iter = seq->private;
1642 i = iter->link;
1643 j = iter->hash_idx + 1;
1644
1645 do {
1646 struct nl_pid_hash *hash = &nl_table[i].hash;
1647
1648 for (; j <= hash->mask; j++) {
1649 s = sk_head(&hash->table[j]);
1650 if (s) {
1651 iter->link = i;
1652 iter->hash_idx = j;
1653 return s;
1654 }
1655 }
1656
1657 j = 0;
1658 } while (++i < MAX_LINKS);
1659
1660 return NULL;
1661 }
1662
1663 static void netlink_seq_stop(struct seq_file *seq, void *v)
1664 {
1665 read_unlock(&nl_table_lock);
1666 }
1667
1668
1669 static int netlink_seq_show(struct seq_file *seq, void *v)
1670 {
1671 if (v == SEQ_START_TOKEN)
1672 seq_puts(seq,
1673 "sk Eth Pid Groups "
1674 "Rmem Wmem Dump Locks\n");
1675 else {
1676 struct sock *s = v;
1677 struct netlink_sock *nlk = nlk_sk(s);
1678
1679 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1680 s,
1681 s->sk_protocol,
1682 nlk->pid,
1683 nlk->groups ? (u32)nlk->groups[0] : 0,
1684 atomic_read(&s->sk_rmem_alloc),
1685 atomic_read(&s->sk_wmem_alloc),
1686 nlk->cb,
1687 atomic_read(&s->sk_refcnt)
1688 );
1689
1690 }
1691 return 0;
1692 }
1693
1694 static struct seq_operations netlink_seq_ops = {
1695 .start = netlink_seq_start,
1696 .next = netlink_seq_next,
1697 .stop = netlink_seq_stop,
1698 .show = netlink_seq_show,
1699 };
1700
1701
1702 static int netlink_seq_open(struct inode *inode, struct file *file)
1703 {
1704 struct seq_file *seq;
1705 struct nl_seq_iter *iter;
1706 int err;
1707
1708 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1709 if (!iter)
1710 return -ENOMEM;
1711
1712 err = seq_open(file, &netlink_seq_ops);
1713 if (err) {
1714 kfree(iter);
1715 return err;
1716 }
1717
1718 seq = file->private_data;
1719 seq->private = iter;
1720 return 0;
1721 }
1722
1723 static const struct file_operations netlink_seq_fops = {
1724 .owner = THIS_MODULE,
1725 .open = netlink_seq_open,
1726 .read = seq_read,
1727 .llseek = seq_lseek,
1728 .release = seq_release_private,
1729 };
1730
1731 #endif
1732
1733 int netlink_register_notifier(struct notifier_block *nb)
1734 {
1735 return atomic_notifier_chain_register(&netlink_chain, nb);
1736 }
1737
1738 int netlink_unregister_notifier(struct notifier_block *nb)
1739 {
1740 return atomic_notifier_chain_unregister(&netlink_chain, nb);
1741 }
1742
1743 static const struct proto_ops netlink_ops = {
1744 .family = PF_NETLINK,
1745 .owner = THIS_MODULE,
1746 .release = netlink_release,
1747 .bind = netlink_bind,
1748 .connect = netlink_connect,
1749 .socketpair = sock_no_socketpair,
1750 .accept = sock_no_accept,
1751 .getname = netlink_getname,
1752 .poll = datagram_poll,
1753 .ioctl = sock_no_ioctl,
1754 .listen = sock_no_listen,
1755 .shutdown = sock_no_shutdown,
1756 .setsockopt = netlink_setsockopt,
1757 .getsockopt = netlink_getsockopt,
1758 .sendmsg = netlink_sendmsg,
1759 .recvmsg = netlink_recvmsg,
1760 .mmap = sock_no_mmap,
1761 .sendpage = sock_no_sendpage,
1762 };
1763
1764 static struct net_proto_family netlink_family_ops = {
1765 .family = PF_NETLINK,
1766 .create = netlink_create,
1767 .owner = THIS_MODULE, /* for consistency 8) */
1768 };
1769
1770 static int __init netlink_proto_init(void)
1771 {
1772 struct sk_buff *dummy_skb;
1773 int i;
1774 unsigned long max;
1775 unsigned int order;
1776 int err = proto_register(&netlink_proto, 0);
1777
1778 if (err != 0)
1779 goto out;
1780
1781 BUILD_BUG_ON(sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb));
1782
1783 nl_table = kcalloc(MAX_LINKS, sizeof(*nl_table), GFP_KERNEL);
1784 if (!nl_table)
1785 goto panic;
1786
1787 if (num_physpages >= (128 * 1024))
1788 max = num_physpages >> (21 - PAGE_SHIFT);
1789 else
1790 max = num_physpages >> (23 - PAGE_SHIFT);
1791
1792 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1793 max = (1UL << order) / sizeof(struct hlist_head);
1794 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1795
1796 for (i = 0; i < MAX_LINKS; i++) {
1797 struct nl_pid_hash *hash = &nl_table[i].hash;
1798
1799 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1800 if (!hash->table) {
1801 while (i-- > 0)
1802 nl_pid_hash_free(nl_table[i].hash.table,
1803 1 * sizeof(*hash->table));
1804 kfree(nl_table);
1805 goto panic;
1806 }
1807 memset(hash->table, 0, 1 * sizeof(*hash->table));
1808 hash->max_shift = order;
1809 hash->shift = 0;
1810 hash->mask = 0;
1811 hash->rehash_time = jiffies;
1812 }
1813
1814 sock_register(&netlink_family_ops);
1815 #ifdef CONFIG_PROC_FS
1816 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1817 #endif
1818 /* The netlink device handler may be needed early. */
1819 rtnetlink_init();
1820 out:
1821 return err;
1822 panic:
1823 panic("netlink_init: Cannot allocate nl_table\n");
1824 }
1825
1826 core_initcall(netlink_proto_init);
1827
1828 EXPORT_SYMBOL(netlink_ack);
1829 EXPORT_SYMBOL(netlink_run_queue);
1830 EXPORT_SYMBOL(netlink_queue_skip);
1831 EXPORT_SYMBOL(netlink_broadcast);
1832 EXPORT_SYMBOL(netlink_dump_start);
1833 EXPORT_SYMBOL(netlink_kernel_create);
1834 EXPORT_SYMBOL(netlink_register_notifier);
1835 EXPORT_SYMBOL(netlink_set_nonroot);
1836 EXPORT_SYMBOL(netlink_unicast);
1837 EXPORT_SYMBOL(netlink_unregister_notifier);
1838 EXPORT_SYMBOL(nlmsg_notify);
Linux kernel - Deliverables
This page took 0.086977 seconds and 4 git commands to generate.