?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[PATCH] x86_64: Dont use broadcast shortcut to make it cpu hotplug safe.
[linux-2.6.git] / net / netlink / af_netlink.c
blob70bcd4744d9370e44f0389f28754df5aaaa3bee6
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 *
17 */
18
19 #include <linux/config.h>
20 #include <linux/module.h>
21
22 #include <linux/kernel.h>
23 #include <linux/init.h>
24 #include <linux/signal.h>
25 #include <linux/sched.h>
26 #include <linux/errno.h>
27 #include <linux/string.h>
28 #include <linux/stat.h>
29 #include <linux/socket.h>
30 #include <linux/un.h>
31 #include <linux/fcntl.h>
32 #include <linux/termios.h>
33 #include <linux/sockios.h>
34 #include <linux/net.h>
35 #include <linux/fs.h>
36 #include <linux/slab.h>
37 #include <asm/uaccess.h>
38 #include <linux/skbuff.h>
39 #include <linux/netdevice.h>
40 #include <linux/rtnetlink.h>
41 #include <linux/proc_fs.h>
42 #include <linux/seq_file.h>
43 #include <linux/smp_lock.h>
44 #include <linux/notifier.h>
45 #include <linux/security.h>
46 #include <linux/jhash.h>
47 #include <linux/jiffies.h>
48 #include <linux/random.h>
49 #include <linux/bitops.h>
50 #include <linux/mm.h>
51 #include <linux/types.h>
52 #include <linux/audit.h>
53
54 #include <net/sock.h>
55 #include <net/scm.h>
56
57 #define Nprintk(a...)
58
59 struct netlink_sock {
60 /* struct sock has to be the first member of netlink_sock */
61 struct sock sk;
62 u32 pid;
63 unsigned int groups;
64 u32 dst_pid;
65 unsigned int dst_groups;
66 unsigned long state;
67 wait_queue_head_t wait;
68 struct netlink_callback *cb;
69 spinlock_t cb_lock;
70 void (*data_ready)(struct sock *sk, int bytes);
71 };
72
73 static inline struct netlink_sock *nlk_sk(struct sock *sk)
74 {
75 return (struct netlink_sock *)sk;
76 }
77
78 struct nl_pid_hash {
79 struct hlist_head *table;
80 unsigned long rehash_time;
81
82 unsigned int mask;
83 unsigned int shift;
84
85 unsigned int entries;
86 unsigned int max_shift;
87
88 u32 rnd;
89 };
90
91 struct netlink_table {
92 struct nl_pid_hash hash;
93 struct hlist_head mc_list;
94 unsigned int nl_nonroot;
95 };
96
97 static struct netlink_table *nl_table;
98
99 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
100
101 static int netlink_dump(struct sock *sk);
102 static void netlink_destroy_callback(struct netlink_callback *cb);
103
104 static DEFINE_RWLOCK(nl_table_lock);
105 static atomic_t nl_table_users = ATOMIC_INIT(0);
106
107 static struct notifier_block *netlink_chain;
108
109 static struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid)
110 {
111 return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
112 }
113
114 static void netlink_sock_destruct(struct sock *sk)
115 {
116 skb_queue_purge(&sk->sk_receive_queue);
117
118 if (!sock_flag(sk, SOCK_DEAD)) {
119 printk("Freeing alive netlink socket %p\n", sk);
120 return;
121 }
122 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
123 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
124 BUG_TRAP(!nlk_sk(sk)->cb);
125 }
126
127 /* This lock without WQ_FLAG_EXCLUSIVE is good on UP and it is _very_ bad on SMP.
128 * Look, when several writers sleep and reader wakes them up, all but one
129 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
130 * this, _but_ remember, it adds useless work on UP machines.
131 */
132
133 static void netlink_table_grab(void)
134 {
135 write_lock_bh(&nl_table_lock);
136
137 if (atomic_read(&nl_table_users)) {
138 DECLARE_WAITQUEUE(wait, current);
139
140 add_wait_queue_exclusive(&nl_table_wait, &wait);
141 for(;;) {
142 set_current_state(TASK_UNINTERRUPTIBLE);
143 if (atomic_read(&nl_table_users) == 0)
144 break;
145 write_unlock_bh(&nl_table_lock);
146 schedule();
147 write_lock_bh(&nl_table_lock);
148 }
149
150 __set_current_state(TASK_RUNNING);
151 remove_wait_queue(&nl_table_wait, &wait);
152 }
153 }
154
155 static __inline__ void netlink_table_ungrab(void)
156 {
157 write_unlock_bh(&nl_table_lock);
158 wake_up(&nl_table_wait);
159 }
160
161 static __inline__ void
162 netlink_lock_table(void)
163 {
164 /* read_lock() synchronizes us to netlink_table_grab */
165
166 read_lock(&nl_table_lock);
167 atomic_inc(&nl_table_users);
168 read_unlock(&nl_table_lock);
169 }
170
171 static __inline__ void
172 netlink_unlock_table(void)
173 {
174 if (atomic_dec_and_test(&nl_table_users))
175 wake_up(&nl_table_wait);
176 }
177
178 static __inline__ struct sock *netlink_lookup(int protocol, u32 pid)
179 {
180 struct nl_pid_hash *hash = &nl_table[protocol].hash;
181 struct hlist_head *head;
182 struct sock *sk;
183 struct hlist_node *node;
184
185 read_lock(&nl_table_lock);
186 head = nl_pid_hashfn(hash, pid);
187 sk_for_each(sk, node, head) {
188 if (nlk_sk(sk)->pid == pid) {
189 sock_hold(sk);
190 goto found;
191 }
192 }
193 sk = NULL;
194 found:
195 read_unlock(&nl_table_lock);
196 return sk;
197 }
198
199 static inline struct hlist_head *nl_pid_hash_alloc(size_t size)
200 {
201 if (size <= PAGE_SIZE)
202 return kmalloc(size, GFP_ATOMIC);
203 else
204 return (struct hlist_head *)
205 __get_free_pages(GFP_ATOMIC, get_order(size));
206 }
207
208 static inline void nl_pid_hash_free(struct hlist_head *table, size_t size)
209 {
210 if (size <= PAGE_SIZE)
211 kfree(table);
212 else
213 free_pages((unsigned long)table, get_order(size));
214 }
215
216 static int nl_pid_hash_rehash(struct nl_pid_hash *hash, int grow)
217 {
218 unsigned int omask, mask, shift;
219 size_t osize, size;
220 struct hlist_head *otable, *table;
221 int i;
222
223 omask = mask = hash->mask;
224 osize = size = (mask + 1) * sizeof(*table);
225 shift = hash->shift;
226
227 if (grow) {
228 if (++shift > hash->max_shift)
229 return 0;
230 mask = mask * 2 + 1;
231 size *= 2;
232 }
233
234 table = nl_pid_hash_alloc(size);
235 if (!table)
236 return 0;
237
238 memset(table, 0, size);
239 otable = hash->table;
240 hash->table = table;
241 hash->mask = mask;
242 hash->shift = shift;
243 get_random_bytes(&hash->rnd, sizeof(hash->rnd));
244
245 for (i = 0; i <= omask; i++) {
246 struct sock *sk;
247 struct hlist_node *node, *tmp;
248
249 sk_for_each_safe(sk, node, tmp, &otable[i])
250 __sk_add_node(sk, nl_pid_hashfn(hash, nlk_sk(sk)->pid));
251 }
252
253 nl_pid_hash_free(otable, osize);
254 hash->rehash_time = jiffies + 10 * 60 * HZ;
255 return 1;
256 }
257
258 static inline int nl_pid_hash_dilute(struct nl_pid_hash *hash, int len)
259 {
260 int avg = hash->entries >> hash->shift;
261
262 if (unlikely(avg > 1) && nl_pid_hash_rehash(hash, 1))
263 return 1;
264
265 if (unlikely(len > avg) && time_after(jiffies, hash->rehash_time)) {
266 nl_pid_hash_rehash(hash, 0);
267 return 1;
268 }
269
270 return 0;
271 }
272
273 static struct proto_ops netlink_ops;
274
275 static int netlink_insert(struct sock *sk, u32 pid)
276 {
277 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
278 struct hlist_head *head;
279 int err = -EADDRINUSE;
280 struct sock *osk;
281 struct hlist_node *node;
282 int len;
283
284 netlink_table_grab();
285 head = nl_pid_hashfn(hash, pid);
286 len = 0;
287 sk_for_each(osk, node, head) {
288 if (nlk_sk(osk)->pid == pid)
289 break;
290 len++;
291 }
292 if (node)
293 goto err;
294
295 err = -EBUSY;
296 if (nlk_sk(sk)->pid)
297 goto err;
298
299 err = -ENOMEM;
300 if (BITS_PER_LONG > 32 && unlikely(hash->entries >= UINT_MAX))
301 goto err;
302
303 if (len && nl_pid_hash_dilute(hash, len))
304 head = nl_pid_hashfn(hash, pid);
305 hash->entries++;
306 nlk_sk(sk)->pid = pid;
307 sk_add_node(sk, head);
308 err = 0;
309
310 err:
311 netlink_table_ungrab();
312 return err;
313 }
314
315 static void netlink_remove(struct sock *sk)
316 {
317 netlink_table_grab();
318 nl_table[sk->sk_protocol].hash.entries--;
319 sk_del_node_init(sk);
320 if (nlk_sk(sk)->groups)
321 __sk_del_bind_node(sk);
322 netlink_table_ungrab();
323 }
324
325 static struct proto netlink_proto = {
326 .name = "NETLINK",
327 .owner = THIS_MODULE,
328 .obj_size = sizeof(struct netlink_sock),
329 };
330
331 static int netlink_create(struct socket *sock, int protocol)
332 {
333 struct sock *sk;
334 struct netlink_sock *nlk;
335
336 sock->state = SS_UNCONNECTED;
337
338 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
339 return -ESOCKTNOSUPPORT;
340
341 if (protocol<0 || protocol >= MAX_LINKS)
342 return -EPROTONOSUPPORT;
343
344 sock->ops = &netlink_ops;
345
346 sk = sk_alloc(PF_NETLINK, GFP_KERNEL, &netlink_proto, 1);
347 if (!sk)
348 return -ENOMEM;
349
350 sock_init_data(sock, sk);
351
352 nlk = nlk_sk(sk);
353
354 spin_lock_init(&nlk->cb_lock);
355 init_waitqueue_head(&nlk->wait);
356 sk->sk_destruct = netlink_sock_destruct;
357
358 sk->sk_protocol = protocol;
359 return 0;
360 }
361
362 static int netlink_release(struct socket *sock)
363 {
364 struct sock *sk = sock->sk;
365 struct netlink_sock *nlk;
366
367 if (!sk)
368 return 0;
369
370 netlink_remove(sk);
371 nlk = nlk_sk(sk);
372
373 spin_lock(&nlk->cb_lock);
374 if (nlk->cb) {
375 nlk->cb->done(nlk->cb);
376 netlink_destroy_callback(nlk->cb);
377 nlk->cb = NULL;
378 }
379 spin_unlock(&nlk->cb_lock);
380
381 /* OK. Socket is unlinked, and, therefore,
382 no new packets will arrive */
383
384 sock_orphan(sk);
385 sock->sk = NULL;
386 wake_up_interruptible_all(&nlk->wait);
387
388 skb_queue_purge(&sk->sk_write_queue);
389
390 if (nlk->pid && !nlk->groups) {
391 struct netlink_notify n = {
392 .protocol = sk->sk_protocol,
393 .pid = nlk->pid,
394 };
395 notifier_call_chain(&netlink_chain, NETLINK_URELEASE, &n);
396 }
397
398 sock_put(sk);
399 return 0;
400 }
401
402 static int netlink_autobind(struct socket *sock)
403 {
404 struct sock *sk = sock->sk;
405 struct nl_pid_hash *hash = &nl_table[sk->sk_protocol].hash;
406 struct hlist_head *head;
407 struct sock *osk;
408 struct hlist_node *node;
409 s32 pid = current->pid;
410 int err;
411 static s32 rover = -4097;
412
413 retry:
414 cond_resched();
415 netlink_table_grab();
416 head = nl_pid_hashfn(hash, pid);
417 sk_for_each(osk, node, head) {
418 if (nlk_sk(osk)->pid == pid) {
419 /* Bind collision, search negative pid values. */
420 pid = rover--;
421 if (rover > -4097)
422 rover = -4097;
423 netlink_table_ungrab();
424 goto retry;
425 }
426 }
427 netlink_table_ungrab();
428
429 err = netlink_insert(sk, pid);
430 if (err == -EADDRINUSE)
431 goto retry;
432 return 0;
433 }
434
435 static inline int netlink_capable(struct socket *sock, unsigned int flag)
436 {
437 return (nl_table[sock->sk->sk_protocol].nl_nonroot & flag) ||
438 capable(CAP_NET_ADMIN);
439 }
440
441 static int netlink_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
442 {
443 struct sock *sk = sock->sk;
444 struct netlink_sock *nlk = nlk_sk(sk);
445 struct sockaddr_nl *nladdr = (struct sockaddr_nl *)addr;
446 int err;
447
448 if (nladdr->nl_family != AF_NETLINK)
449 return -EINVAL;
450
451 /* Only superuser is allowed to listen multicasts */
452 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_RECV))
453 return -EPERM;
454
455 if (nlk->pid) {
456 if (nladdr->nl_pid != nlk->pid)
457 return -EINVAL;
458 } else {
459 err = nladdr->nl_pid ?
460 netlink_insert(sk, nladdr->nl_pid) :
461 netlink_autobind(sock);
462 if (err)
463 return err;
464 }
465
466 if (!nladdr->nl_groups && !nlk->groups)
467 return 0;
468
469 netlink_table_grab();
470 if (nlk->groups && !nladdr->nl_groups)
471 __sk_del_bind_node(sk);
472 else if (!nlk->groups && nladdr->nl_groups)
473 sk_add_bind_node(sk, &nl_table[sk->sk_protocol].mc_list);
474 nlk->groups = nladdr->nl_groups;
475 netlink_table_ungrab();
476
477 return 0;
478 }
479
480 static int netlink_connect(struct socket *sock, struct sockaddr *addr,
481 int alen, int flags)
482 {
483 int err = 0;
484 struct sock *sk = sock->sk;
485 struct netlink_sock *nlk = nlk_sk(sk);
486 struct sockaddr_nl *nladdr=(struct sockaddr_nl*)addr;
487
488 if (addr->sa_family == AF_UNSPEC) {
489 sk->sk_state = NETLINK_UNCONNECTED;
490 nlk->dst_pid = 0;
491 nlk->dst_groups = 0;
492 return 0;
493 }
494 if (addr->sa_family != AF_NETLINK)
495 return -EINVAL;
496
497 /* Only superuser is allowed to send multicasts */
498 if (nladdr->nl_groups && !netlink_capable(sock, NL_NONROOT_SEND))
499 return -EPERM;
500
501 if (!nlk->pid)
502 err = netlink_autobind(sock);
503
504 if (err == 0) {
505 sk->sk_state = NETLINK_CONNECTED;
506 nlk->dst_pid = nladdr->nl_pid;
507 nlk->dst_groups = nladdr->nl_groups;
508 }
509
510 return err;
511 }
512
513 static int netlink_getname(struct socket *sock, struct sockaddr *addr, int *addr_len, int peer)
514 {
515 struct sock *sk = sock->sk;
516 struct netlink_sock *nlk = nlk_sk(sk);
517 struct sockaddr_nl *nladdr=(struct sockaddr_nl *)addr;
518
519 nladdr->nl_family = AF_NETLINK;
520 nladdr->nl_pad = 0;
521 *addr_len = sizeof(*nladdr);
522
523 if (peer) {
524 nladdr->nl_pid = nlk->dst_pid;
525 nladdr->nl_groups = nlk->dst_groups;
526 } else {
527 nladdr->nl_pid = nlk->pid;
528 nladdr->nl_groups = nlk->groups;
529 }
530 return 0;
531 }
532
533 static void netlink_overrun(struct sock *sk)
534 {
535 if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
536 sk->sk_err = ENOBUFS;
537 sk->sk_error_report(sk);
538 }
539 }
540
541 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
542 {
543 int protocol = ssk->sk_protocol;
544 struct sock *sock;
545 struct netlink_sock *nlk;
546
547 sock = netlink_lookup(protocol, pid);
548 if (!sock)
549 return ERR_PTR(-ECONNREFUSED);
550
551 /* Don't bother queuing skb if kernel socket has no input function */
552 nlk = nlk_sk(sock);
553 if ((nlk->pid == 0 && !nlk->data_ready) ||
554 (sock->sk_state == NETLINK_CONNECTED &&
555 nlk->dst_pid != nlk_sk(ssk)->pid)) {
556 sock_put(sock);
557 return ERR_PTR(-ECONNREFUSED);
558 }
559 return sock;
560 }
561
562 struct sock *netlink_getsockbyfilp(struct file *filp)
563 {
564 struct inode *inode = filp->f_dentry->d_inode;
565 struct sock *sock;
566
567 if (!S_ISSOCK(inode->i_mode))
568 return ERR_PTR(-ENOTSOCK);
569
570 sock = SOCKET_I(inode)->sk;
571 if (sock->sk_family != AF_NETLINK)
572 return ERR_PTR(-EINVAL);
573
574 sock_hold(sock);
575 return sock;
576 }
577
578 /*
579 * Attach a skb to a netlink socket.
580 * The caller must hold a reference to the destination socket. On error, the
581 * reference is dropped. The skb is not send to the destination, just all
582 * all error checks are performed and memory in the queue is reserved.
583 * Return values:
584 * < 0: error. skb freed, reference to sock dropped.
585 * 0: continue
586 * 1: repeat lookup - reference dropped while waiting for socket memory.
587 */
588 int netlink_attachskb(struct sock *sk, struct sk_buff *skb, int nonblock, long timeo)
589 {
590 struct netlink_sock *nlk;
591
592 nlk = nlk_sk(sk);
593
594 if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
595 test_bit(0, &nlk->state)) {
596 DECLARE_WAITQUEUE(wait, current);
597 if (!timeo) {
598 if (!nlk->pid)
599 netlink_overrun(sk);
600 sock_put(sk);
601 kfree_skb(skb);
602 return -EAGAIN;
603 }
604
605 __set_current_state(TASK_INTERRUPTIBLE);
606 add_wait_queue(&nlk->wait, &wait);
607
608 if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
609 test_bit(0, &nlk->state)) &&
610 !sock_flag(sk, SOCK_DEAD))
611 timeo = schedule_timeout(timeo);
612
613 __set_current_state(TASK_RUNNING);
614 remove_wait_queue(&nlk->wait, &wait);
615 sock_put(sk);
616
617 if (signal_pending(current)) {
618 kfree_skb(skb);
619 return sock_intr_errno(timeo);
620 }
621 return 1;
622 }
623 skb_set_owner_r(skb, sk);
624 return 0;
625 }
626
627 int netlink_sendskb(struct sock *sk, struct sk_buff *skb, int protocol)
628 {
629 struct netlink_sock *nlk;
630 int len = skb->len;
631
632 nlk = nlk_sk(sk);
633
634 skb_queue_tail(&sk->sk_receive_queue, skb);
635 sk->sk_data_ready(sk, len);
636 sock_put(sk);
637 return len;
638 }
639
640 void netlink_detachskb(struct sock *sk, struct sk_buff *skb)
641 {
642 kfree_skb(skb);
643 sock_put(sk);
644 }
645
646 static inline struct sk_buff *netlink_trim(struct sk_buff *skb, int allocation)
647 {
648 int delta;
649
650 skb_orphan(skb);
651
652 delta = skb->end - skb->tail;
653 if (delta * 2 < skb->truesize)
654 return skb;
655
656 if (skb_shared(skb)) {
657 struct sk_buff *nskb = skb_clone(skb, allocation);
658 if (!nskb)
659 return skb;
660 kfree_skb(skb);
661 skb = nskb;
662 }
663
664 if (!pskb_expand_head(skb, 0, -delta, allocation))
665 skb->truesize -= delta;
666
667 return skb;
668 }
669
670 int netlink_unicast(struct sock *ssk, struct sk_buff *skb, u32 pid, int nonblock)
671 {
672 struct sock *sk;
673 int err;
674 long timeo;
675
676 skb = netlink_trim(skb, gfp_any());
677
678 timeo = sock_sndtimeo(ssk, nonblock);
679 retry:
680 sk = netlink_getsockbypid(ssk, pid);
681 if (IS_ERR(sk)) {
682 kfree_skb(skb);
683 return PTR_ERR(sk);
684 }
685 err = netlink_attachskb(sk, skb, nonblock, timeo);
686 if (err == 1)
687 goto retry;
688 if (err)
689 return err;
690
691 return netlink_sendskb(sk, skb, ssk->sk_protocol);
692 }
693
694 static __inline__ int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
695 {
696 struct netlink_sock *nlk = nlk_sk(sk);
697
698 if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
699 !test_bit(0, &nlk->state)) {
700 skb_set_owner_r(skb, sk);
701 skb_queue_tail(&sk->sk_receive_queue, skb);
702 sk->sk_data_ready(sk, skb->len);
703 return atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf;
704 }
705 return -1;
706 }
707
708 struct netlink_broadcast_data {
709 struct sock *exclude_sk;
710 u32 pid;
711 u32 group;
712 int failure;
713 int congested;
714 int delivered;
715 int allocation;
716 struct sk_buff *skb, *skb2;
717 };
718
719 static inline int do_one_broadcast(struct sock *sk,
720 struct netlink_broadcast_data *p)
721 {
722 struct netlink_sock *nlk = nlk_sk(sk);
723 int val;
724
725 if (p->exclude_sk == sk)
726 goto out;
727
728 if (nlk->pid == p->pid || !(nlk->groups & p->group))
729 goto out;
730
731 if (p->failure) {
732 netlink_overrun(sk);
733 goto out;
734 }
735
736 sock_hold(sk);
737 if (p->skb2 == NULL) {
738 if (skb_shared(p->skb)) {
739 p->skb2 = skb_clone(p->skb, p->allocation);
740 } else {
741 p->skb2 = skb_get(p->skb);
742 /*
743 * skb ownership may have been set when
744 * delivered to a previous socket.
745 */
746 skb_orphan(p->skb2);
747 }
748 }
749 if (p->skb2 == NULL) {
750 netlink_overrun(sk);
751 /* Clone failed. Notify ALL listeners. */
752 p->failure = 1;
753 } else if ((val = netlink_broadcast_deliver(sk, p->skb2)) < 0) {
754 netlink_overrun(sk);
755 } else {
756 p->congested |= val;
757 p->delivered = 1;
758 p->skb2 = NULL;
759 }
760 sock_put(sk);
761
762 out:
763 return 0;
764 }
765
766 int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, u32 pid,
767 u32 group, int allocation)
768 {
769 struct netlink_broadcast_data info;
770 struct hlist_node *node;
771 struct sock *sk;
772
773 skb = netlink_trim(skb, allocation);
774
775 info.exclude_sk = ssk;
776 info.pid = pid;
777 info.group = group;
778 info.failure = 0;
779 info.congested = 0;
780 info.delivered = 0;
781 info.allocation = allocation;
782 info.skb = skb;
783 info.skb2 = NULL;
784
785 /* While we sleep in clone, do not allow to change socket list */
786
787 netlink_lock_table();
788
789 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
790 do_one_broadcast(sk, &info);
791
792 kfree_skb(skb);
793
794 netlink_unlock_table();
795
796 if (info.skb2)
797 kfree_skb(info.skb2);
798
799 if (info.delivered) {
800 if (info.congested && (allocation & __GFP_WAIT))
801 yield();
802 return 0;
803 }
804 if (info.failure)
805 return -ENOBUFS;
806 return -ESRCH;
807 }
808
809 struct netlink_set_err_data {
810 struct sock *exclude_sk;
811 u32 pid;
812 u32 group;
813 int code;
814 };
815
816 static inline int do_one_set_err(struct sock *sk,
817 struct netlink_set_err_data *p)
818 {
819 struct netlink_sock *nlk = nlk_sk(sk);
820
821 if (sk == p->exclude_sk)
822 goto out;
823
824 if (nlk->pid == p->pid || !(nlk->groups & p->group))
825 goto out;
826
827 sk->sk_err = p->code;
828 sk->sk_error_report(sk);
829 out:
830 return 0;
831 }
832
833 void netlink_set_err(struct sock *ssk, u32 pid, u32 group, int code)
834 {
835 struct netlink_set_err_data info;
836 struct hlist_node *node;
837 struct sock *sk;
838
839 info.exclude_sk = ssk;
840 info.pid = pid;
841 info.group = group;
842 info.code = code;
843
844 read_lock(&nl_table_lock);
845
846 sk_for_each_bound(sk, node, &nl_table[ssk->sk_protocol].mc_list)
847 do_one_set_err(sk, &info);
848
849 read_unlock(&nl_table_lock);
850 }
851
852 static inline void netlink_rcv_wake(struct sock *sk)
853 {
854 struct netlink_sock *nlk = nlk_sk(sk);
855
856 if (!skb_queue_len(&sk->sk_receive_queue))
857 clear_bit(0, &nlk->state);
858 if (!test_bit(0, &nlk->state))
859 wake_up_interruptible(&nlk->wait);
860 }
861
862 static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
863 struct msghdr *msg, size_t len)
864 {
865 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
866 struct sock *sk = sock->sk;
867 struct netlink_sock *nlk = nlk_sk(sk);
868 struct sockaddr_nl *addr=msg->msg_name;
869 u32 dst_pid;
870 u32 dst_groups;
871 struct sk_buff *skb;
872 int err;
873 struct scm_cookie scm;
874
875 if (msg->msg_flags&MSG_OOB)
876 return -EOPNOTSUPP;
877
878 if (NULL == siocb->scm)
879 siocb->scm = &scm;
880 err = scm_send(sock, msg, siocb->scm);
881 if (err < 0)
882 return err;
883
884 if (msg->msg_namelen) {
885 if (addr->nl_family != AF_NETLINK)
886 return -EINVAL;
887 dst_pid = addr->nl_pid;
888 dst_groups = addr->nl_groups;
889 if (dst_groups && !netlink_capable(sock, NL_NONROOT_SEND))
890 return -EPERM;
891 } else {
892 dst_pid = nlk->dst_pid;
893 dst_groups = nlk->dst_groups;
894 }
895
896 if (!nlk->pid) {
897 err = netlink_autobind(sock);
898 if (err)
899 goto out;
900 }
901
902 err = -EMSGSIZE;
903 if (len > sk->sk_sndbuf - 32)
904 goto out;
905 err = -ENOBUFS;
906 skb = alloc_skb(len, GFP_KERNEL);
907 if (skb==NULL)
908 goto out;
909
910 NETLINK_CB(skb).pid = nlk->pid;
911 NETLINK_CB(skb).groups = nlk->groups;
912 NETLINK_CB(skb).dst_pid = dst_pid;
913 NETLINK_CB(skb).dst_groups = dst_groups;
914 NETLINK_CB(skb).loginuid = audit_get_loginuid(current->audit_context);
915 memcpy(NETLINK_CREDS(skb), &siocb->scm->creds, sizeof(struct ucred));
916
917 /* What can I do? Netlink is asynchronous, so that
918 we will have to save current capabilities to
919 check them, when this message will be delivered
920 to corresponding kernel module. --ANK (980802)
921 */
922
923 err = -EFAULT;
924 if (memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len)) {
925 kfree_skb(skb);
926 goto out;
927 }
928
929 err = security_netlink_send(sk, skb);
930 if (err) {
931 kfree_skb(skb);
932 goto out;
933 }
934
935 if (dst_groups) {
936 atomic_inc(&skb->users);
937 netlink_broadcast(sk, skb, dst_pid, dst_groups, GFP_KERNEL);
938 }
939 err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags&MSG_DONTWAIT);
940
941 out:
942 return err;
943 }
944
945 static int netlink_recvmsg(struct kiocb *kiocb, struct socket *sock,
946 struct msghdr *msg, size_t len,
947 int flags)
948 {
949 struct sock_iocb *siocb = kiocb_to_siocb(kiocb);
950 struct scm_cookie scm;
951 struct sock *sk = sock->sk;
952 struct netlink_sock *nlk = nlk_sk(sk);
953 int noblock = flags&MSG_DONTWAIT;
954 size_t copied;
955 struct sk_buff *skb;
956 int err;
957
958 if (flags&MSG_OOB)
959 return -EOPNOTSUPP;
960
961 copied = 0;
962
963 skb = skb_recv_datagram(sk,flags,noblock,&err);
964 if (skb==NULL)
965 goto out;
966
967 msg->msg_namelen = 0;
968
969 copied = skb->len;
970 if (len < copied) {
971 msg->msg_flags |= MSG_TRUNC;
972 copied = len;
973 }
974
975 skb->h.raw = skb->data;
976 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
977
978 if (msg->msg_name) {
979 struct sockaddr_nl *addr = (struct sockaddr_nl*)msg->msg_name;
980 addr->nl_family = AF_NETLINK;
981 addr->nl_pad = 0;
982 addr->nl_pid = NETLINK_CB(skb).pid;
983 addr->nl_groups = NETLINK_CB(skb).dst_groups;
984 msg->msg_namelen = sizeof(*addr);
985 }
986
987 if (NULL == siocb->scm) {
988 memset(&scm, 0, sizeof(scm));
989 siocb->scm = &scm;
990 }
991 siocb->scm->creds = *NETLINK_CREDS(skb);
992 skb_free_datagram(sk, skb);
993
994 if (nlk->cb && atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf / 2)
995 netlink_dump(sk);
996
997 scm_recv(sock, msg, siocb->scm, flags);
998
999 out:
1000 netlink_rcv_wake(sk);
1001 return err ? : copied;
1002 }
1003
1004 static void netlink_data_ready(struct sock *sk, int len)
1005 {
1006 struct netlink_sock *nlk = nlk_sk(sk);
1007
1008 if (nlk->data_ready)
1009 nlk->data_ready(sk, len);
1010 netlink_rcv_wake(sk);
1011 }
1012
1013 /*
1014 * We export these functions to other modules. They provide a
1015 * complete set of kernel non-blocking support for message
1016 * queueing.
1017 */
1018
1019 struct sock *
1020 netlink_kernel_create(int unit, void (*input)(struct sock *sk, int len))
1021 {
1022 struct socket *sock;
1023 struct sock *sk;
1024
1025 if (!nl_table)
1026 return NULL;
1027
1028 if (unit<0 || unit>=MAX_LINKS)
1029 return NULL;
1030
1031 if (sock_create_lite(PF_NETLINK, SOCK_DGRAM, unit, &sock))
1032 return NULL;
1033
1034 if (netlink_create(sock, unit) < 0) {
1035 sock_release(sock);
1036 return NULL;
1037 }
1038 sk = sock->sk;
1039 sk->sk_data_ready = netlink_data_ready;
1040 if (input)
1041 nlk_sk(sk)->data_ready = input;
1042
1043 if (netlink_insert(sk, 0)) {
1044 sock_release(sock);
1045 return NULL;
1046 }
1047 return sk;
1048 }
1049
1050 void netlink_set_nonroot(int protocol, unsigned int flags)
1051 {
1052 if ((unsigned int)protocol < MAX_LINKS)
1053 nl_table[protocol].nl_nonroot = flags;
1054 }
1055
1056 static void netlink_destroy_callback(struct netlink_callback *cb)
1057 {
1058 if (cb->skb)
1059 kfree_skb(cb->skb);
1060 kfree(cb);
1061 }
1062
1063 /*
1064 * It looks a bit ugly.
1065 * It would be better to create kernel thread.
1066 */
1067
1068 static int netlink_dump(struct sock *sk)
1069 {
1070 struct netlink_sock *nlk = nlk_sk(sk);
1071 struct netlink_callback *cb;
1072 struct sk_buff *skb;
1073 struct nlmsghdr *nlh;
1074 int len;
1075
1076 skb = sock_rmalloc(sk, NLMSG_GOODSIZE, 0, GFP_KERNEL);
1077 if (!skb)
1078 return -ENOBUFS;
1079
1080 spin_lock(&nlk->cb_lock);
1081
1082 cb = nlk->cb;
1083 if (cb == NULL) {
1084 spin_unlock(&nlk->cb_lock);
1085 kfree_skb(skb);
1086 return -EINVAL;
1087 }
1088
1089 len = cb->dump(skb, cb);
1090
1091 if (len > 0) {
1092 spin_unlock(&nlk->cb_lock);
1093 skb_queue_tail(&sk->sk_receive_queue, skb);
1094 sk->sk_data_ready(sk, len);
1095 return 0;
1096 }
1097
1098 nlh = NLMSG_NEW_ANSWER(skb, cb, NLMSG_DONE, sizeof(len), NLM_F_MULTI);
1099 memcpy(NLMSG_DATA(nlh), &len, sizeof(len));
1100 skb_queue_tail(&sk->sk_receive_queue, skb);
1101 sk->sk_data_ready(sk, skb->len);
1102
1103 cb->done(cb);
1104 nlk->cb = NULL;
1105 spin_unlock(&nlk->cb_lock);
1106
1107 netlink_destroy_callback(cb);
1108 return 0;
1109
1110 nlmsg_failure:
1111 return -ENOBUFS;
1112 }
1113
1114 int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
1115 struct nlmsghdr *nlh,
1116 int (*dump)(struct sk_buff *skb, struct netlink_callback*),
1117 int (*done)(struct netlink_callback*))
1118 {
1119 struct netlink_callback *cb;
1120 struct sock *sk;
1121 struct netlink_sock *nlk;
1122
1123 cb = kmalloc(sizeof(*cb), GFP_KERNEL);
1124 if (cb == NULL)
1125 return -ENOBUFS;
1126
1127 memset(cb, 0, sizeof(*cb));
1128 cb->dump = dump;
1129 cb->done = done;
1130 cb->nlh = nlh;
1131 atomic_inc(&skb->users);
1132 cb->skb = skb;
1133
1134 sk = netlink_lookup(ssk->sk_protocol, NETLINK_CB(skb).pid);
1135 if (sk == NULL) {
1136 netlink_destroy_callback(cb);
1137 return -ECONNREFUSED;
1138 }
1139 nlk = nlk_sk(sk);
1140 /* A dump is in progress... */
1141 spin_lock(&nlk->cb_lock);
1142 if (nlk->cb) {
1143 spin_unlock(&nlk->cb_lock);
1144 netlink_destroy_callback(cb);
1145 sock_put(sk);
1146 return -EBUSY;
1147 }
1148 nlk->cb = cb;
1149 spin_unlock(&nlk->cb_lock);
1150
1151 netlink_dump(sk);
1152 sock_put(sk);
1153 return 0;
1154 }
1155
1156 void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
1157 {
1158 struct sk_buff *skb;
1159 struct nlmsghdr *rep;
1160 struct nlmsgerr *errmsg;
1161 int size;
1162
1163 if (err == 0)
1164 size = NLMSG_SPACE(sizeof(struct nlmsgerr));
1165 else
1166 size = NLMSG_SPACE(4 + NLMSG_ALIGN(nlh->nlmsg_len));
1167
1168 skb = alloc_skb(size, GFP_KERNEL);
1169 if (!skb) {
1170 struct sock *sk;
1171
1172 sk = netlink_lookup(in_skb->sk->sk_protocol,
1173 NETLINK_CB(in_skb).pid);
1174 if (sk) {
1175 sk->sk_err = ENOBUFS;
1176 sk->sk_error_report(sk);
1177 sock_put(sk);
1178 }
1179 return;
1180 }
1181
1182 rep = __nlmsg_put(skb, NETLINK_CB(in_skb).pid, nlh->nlmsg_seq,
1183 NLMSG_ERROR, sizeof(struct nlmsgerr), 0);
1184 errmsg = NLMSG_DATA(rep);
1185 errmsg->error = err;
1186 memcpy(&errmsg->msg, nlh, err ? nlh->nlmsg_len : sizeof(struct nlmsghdr));
1187 netlink_unicast(in_skb->sk, skb, NETLINK_CB(in_skb).pid, MSG_DONTWAIT);
1188 }
1189
1190
1191 #ifdef CONFIG_PROC_FS
1192 struct nl_seq_iter {
1193 int link;
1194 int hash_idx;
1195 };
1196
1197 static struct sock *netlink_seq_socket_idx(struct seq_file *seq, loff_t pos)
1198 {
1199 struct nl_seq_iter *iter = seq->private;
1200 int i, j;
1201 struct sock *s;
1202 struct hlist_node *node;
1203 loff_t off = 0;
1204
1205 for (i=0; i<MAX_LINKS; i++) {
1206 struct nl_pid_hash *hash = &nl_table[i].hash;
1207
1208 for (j = 0; j <= hash->mask; j++) {
1209 sk_for_each(s, node, &hash->table[j]) {
1210 if (off == pos) {
1211 iter->link = i;
1212 iter->hash_idx = j;
1213 return s;
1214 }
1215 ++off;
1216 }
1217 }
1218 }
1219 return NULL;
1220 }
1221
1222 static void *netlink_seq_start(struct seq_file *seq, loff_t *pos)
1223 {
1224 read_lock(&nl_table_lock);
1225 return *pos ? netlink_seq_socket_idx(seq, *pos - 1) : SEQ_START_TOKEN;
1226 }
1227
1228 static void *netlink_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1229 {
1230 struct sock *s;
1231 struct nl_seq_iter *iter;
1232 int i, j;
1233
1234 ++*pos;
1235
1236 if (v == SEQ_START_TOKEN)
1237 return netlink_seq_socket_idx(seq, 0);
1238
1239 s = sk_next(v);
1240 if (s)
1241 return s;
1242
1243 iter = seq->private;
1244 i = iter->link;
1245 j = iter->hash_idx + 1;
1246
1247 do {
1248 struct nl_pid_hash *hash = &nl_table[i].hash;
1249
1250 for (; j <= hash->mask; j++) {
1251 s = sk_head(&hash->table[j]);
1252 if (s) {
1253 iter->link = i;
1254 iter->hash_idx = j;
1255 return s;
1256 }
1257 }
1258
1259 j = 0;
1260 } while (++i < MAX_LINKS);
1261
1262 return NULL;
1263 }
1264
1265 static void netlink_seq_stop(struct seq_file *seq, void *v)
1266 {
1267 read_unlock(&nl_table_lock);
1268 }
1269
1270
1271 static int netlink_seq_show(struct seq_file *seq, void *v)
1272 {
1273 if (v == SEQ_START_TOKEN)
1274 seq_puts(seq,
1275 "sk Eth Pid Groups "
1276 "Rmem Wmem Dump Locks\n");
1277 else {
1278 struct sock *s = v;
1279 struct netlink_sock *nlk = nlk_sk(s);
1280
1281 seq_printf(seq, "%p %-3d %-6d %08x %-8d %-8d %p %d\n",
1282 s,
1283 s->sk_protocol,
1284 nlk->pid,
1285 nlk->groups,
1286 atomic_read(&s->sk_rmem_alloc),
1287 atomic_read(&s->sk_wmem_alloc),
1288 nlk->cb,
1289 atomic_read(&s->sk_refcnt)
1290 );
1291
1292 }
1293 return 0;
1294 }
1295
1296 static struct seq_operations netlink_seq_ops = {
1297 .start = netlink_seq_start,
1298 .next = netlink_seq_next,
1299 .stop = netlink_seq_stop,
1300 .show = netlink_seq_show,
1301 };
1302
1303
1304 static int netlink_seq_open(struct inode *inode, struct file *file)
1305 {
1306 struct seq_file *seq;
1307 struct nl_seq_iter *iter;
1308 int err;
1309
1310 iter = kmalloc(sizeof(*iter), GFP_KERNEL);
1311 if (!iter)
1312 return -ENOMEM;
1313
1314 err = seq_open(file, &netlink_seq_ops);
1315 if (err) {
1316 kfree(iter);
1317 return err;
1318 }
1319
1320 memset(iter, 0, sizeof(*iter));
1321 seq = file->private_data;
1322 seq->private = iter;
1323 return 0;
1324 }
1325
1326 static struct file_operations netlink_seq_fops = {
1327 .owner = THIS_MODULE,
1328 .open = netlink_seq_open,
1329 .read = seq_read,
1330 .llseek = seq_lseek,
1331 .release = seq_release_private,
1332 };
1333
1334 #endif
1335
1336 int netlink_register_notifier(struct notifier_block *nb)
1337 {
1338 return notifier_chain_register(&netlink_chain, nb);
1339 }
1340
1341 int netlink_unregister_notifier(struct notifier_block *nb)
1342 {
1343 return notifier_chain_unregister(&netlink_chain, nb);
1344 }
1345
1346 static struct proto_ops netlink_ops = {
1347 .family = PF_NETLINK,
1348 .owner = THIS_MODULE,
1349 .release = netlink_release,
1350 .bind = netlink_bind,
1351 .connect = netlink_connect,
1352 .socketpair = sock_no_socketpair,
1353 .accept = sock_no_accept,
1354 .getname = netlink_getname,
1355 .poll = datagram_poll,
1356 .ioctl = sock_no_ioctl,
1357 .listen = sock_no_listen,
1358 .shutdown = sock_no_shutdown,
1359 .setsockopt = sock_no_setsockopt,
1360 .getsockopt = sock_no_getsockopt,
1361 .sendmsg = netlink_sendmsg,
1362 .recvmsg = netlink_recvmsg,
1363 .mmap = sock_no_mmap,
1364 .sendpage = sock_no_sendpage,
1365 };
1366
1367 static struct net_proto_family netlink_family_ops = {
1368 .family = PF_NETLINK,
1369 .create = netlink_create,
1370 .owner = THIS_MODULE, /* for consistency 8) */
1371 };
1372
1373 extern void netlink_skb_parms_too_large(void);
1374
1375 static int __init netlink_proto_init(void)
1376 {
1377 struct sk_buff *dummy_skb;
1378 int i;
1379 unsigned long max;
1380 unsigned int order;
1381 int err = proto_register(&netlink_proto, 0);
1382
1383 if (err != 0)
1384 goto out;
1385
1386 if (sizeof(struct netlink_skb_parms) > sizeof(dummy_skb->cb))
1387 netlink_skb_parms_too_large();
1388
1389 nl_table = kmalloc(sizeof(*nl_table) * MAX_LINKS, GFP_KERNEL);
1390 if (!nl_table) {
1391 enomem:
1392 printk(KERN_CRIT "netlink_init: Cannot allocate nl_table\n");
1393 return -ENOMEM;
1394 }
1395
1396 memset(nl_table, 0, sizeof(*nl_table) * MAX_LINKS);
1397
1398 if (num_physpages >= (128 * 1024))
1399 max = num_physpages >> (21 - PAGE_SHIFT);
1400 else
1401 max = num_physpages >> (23 - PAGE_SHIFT);
1402
1403 order = get_bitmask_order(max) - 1 + PAGE_SHIFT;
1404 max = (1UL << order) / sizeof(struct hlist_head);
1405 order = get_bitmask_order(max > UINT_MAX ? UINT_MAX : max) - 1;
1406
1407 for (i = 0; i < MAX_LINKS; i++) {
1408 struct nl_pid_hash *hash = &nl_table[i].hash;
1409
1410 hash->table = nl_pid_hash_alloc(1 * sizeof(*hash->table));
1411 if (!hash->table) {
1412 while (i-- > 0)
1413 nl_pid_hash_free(nl_table[i].hash.table,
1414 1 * sizeof(*hash->table));
1415 kfree(nl_table);
1416 goto enomem;
1417 }
1418 memset(hash->table, 0, 1 * sizeof(*hash->table));
1419 hash->max_shift = order;
1420 hash->shift = 0;
1421 hash->mask = 0;
1422 hash->rehash_time = jiffies;
1423 }
1424
1425 sock_register(&netlink_family_ops);
1426 #ifdef CONFIG_PROC_FS
1427 proc_net_fops_create("netlink", 0, &netlink_seq_fops);
1428 #endif
1429 /* The netlink device handler may be needed early. */
1430 rtnetlink_init();
1431 out:
1432 return err;
1433 }
1434
1435 static void __exit netlink_proto_exit(void)
1436 {
1437 sock_unregister(PF_NETLINK);
1438 proc_net_remove("netlink");
1439 kfree(nl_table);
1440 nl_table = NULL;
1441 proto_unregister(&netlink_proto);
1442 }
1443
1444 core_initcall(netlink_proto_init);
1445 module_exit(netlink_proto_exit);
1446
1447 MODULE_LICENSE("GPL");
1448
1449 MODULE_ALIAS_NETPROTO(PF_NETLINK);
1450
1451 EXPORT_SYMBOL(netlink_ack);
1452 EXPORT_SYMBOL(netlink_broadcast);
1453 EXPORT_SYMBOL(netlink_dump_start);
1454 EXPORT_SYMBOL(netlink_kernel_create);
1455 EXPORT_SYMBOL(netlink_register_notifier);
1456 EXPORT_SYMBOL(netlink_set_err);
1457 EXPORT_SYMBOL(netlink_set_nonroot);
1458 EXPORT_SYMBOL(netlink_unicast);
1459 EXPORT_SYMBOL(netlink_unregister_notifier);
1460
Linus' kernel tree