Loading...
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * RAW - implementation of IP "raw" sockets.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 *
11 * Fixes:
12 * Alan Cox : verify_area() fixed up
13 * Alan Cox : ICMP error handling
14 * Alan Cox : EMSGSIZE if you send too big a packet
15 * Alan Cox : Now uses generic datagrams and shared
16 * skbuff library. No more peek crashes,
17 * no more backlogs
18 * Alan Cox : Checks sk->broadcast.
19 * Alan Cox : Uses skb_free_datagram/skb_copy_datagram
20 * Alan Cox : Raw passes ip options too
21 * Alan Cox : Setsocketopt added
22 * Alan Cox : Fixed error return for broadcasts
23 * Alan Cox : Removed wake_up calls
24 * Alan Cox : Use ttl/tos
25 * Alan Cox : Cleaned up old debugging
26 * Alan Cox : Use new kernel side addresses
27 * Arnt Gulbrandsen : Fixed MSG_DONTROUTE in raw sockets.
28 * Alan Cox : BSD style RAW socket demultiplexing.
29 * Alan Cox : Beginnings of mrouted support.
30 * Alan Cox : Added IP_HDRINCL option.
31 * Alan Cox : Skip broadcast check if BSDism set.
32 * David S. Miller : New socket lookup architecture.
33 *
34 * This program is free software; you can redistribute it and/or
35 * modify it under the terms of the GNU General Public License
36 * as published by the Free Software Foundation; either version
37 * 2 of the License, or (at your option) any later version.
38 */
39
40#include <linux/types.h>
41#include <linux/atomic.h>
42#include <asm/byteorder.h>
43#include <asm/current.h>
44#include <asm/uaccess.h>
45#include <asm/ioctls.h>
46#include <linux/stddef.h>
47#include <linux/slab.h>
48#include <linux/errno.h>
49#include <linux/aio.h>
50#include <linux/kernel.h>
51#include <linux/export.h>
52#include <linux/spinlock.h>
53#include <linux/sockios.h>
54#include <linux/socket.h>
55#include <linux/in.h>
56#include <linux/mroute.h>
57#include <linux/netdevice.h>
58#include <linux/in_route.h>
59#include <linux/route.h>
60#include <linux/skbuff.h>
61#include <net/net_namespace.h>
62#include <net/dst.h>
63#include <net/sock.h>
64#include <linux/ip.h>
65#include <linux/net.h>
66#include <net/ip.h>
67#include <net/icmp.h>
68#include <net/udp.h>
69#include <net/raw.h>
70#include <net/snmp.h>
71#include <net/tcp_states.h>
72#include <net/inet_common.h>
73#include <net/checksum.h>
74#include <net/xfrm.h>
75#include <linux/rtnetlink.h>
76#include <linux/proc_fs.h>
77#include <linux/seq_file.h>
78#include <linux/netfilter.h>
79#include <linux/netfilter_ipv4.h>
80#include <linux/compat.h>
81
82static struct raw_hashinfo raw_v4_hashinfo = {
83 .lock = __RW_LOCK_UNLOCKED(raw_v4_hashinfo.lock),
84};
85
86void raw_hash_sk(struct sock *sk)
87{
88 struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
89 struct hlist_head *head;
90
91 head = &h->ht[inet_sk(sk)->inet_num & (RAW_HTABLE_SIZE - 1)];
92
93 write_lock_bh(&h->lock);
94 sk_add_node(sk, head);
95 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
96 write_unlock_bh(&h->lock);
97}
98EXPORT_SYMBOL_GPL(raw_hash_sk);
99
100void raw_unhash_sk(struct sock *sk)
101{
102 struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
103
104 write_lock_bh(&h->lock);
105 if (sk_del_node_init(sk))
106 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
107 write_unlock_bh(&h->lock);
108}
109EXPORT_SYMBOL_GPL(raw_unhash_sk);
110
111static struct sock *__raw_v4_lookup(struct net *net, struct sock *sk,
112 unsigned short num, __be32 raddr, __be32 laddr, int dif)
113{
114 struct hlist_node *node;
115
116 sk_for_each_from(sk, node) {
117 struct inet_sock *inet = inet_sk(sk);
118
119 if (net_eq(sock_net(sk), net) && inet->inet_num == num &&
120 !(inet->inet_daddr && inet->inet_daddr != raddr) &&
121 !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
122 !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
123 goto found; /* gotcha */
124 }
125 sk = NULL;
126found:
127 return sk;
128}
129
130/*
131 * 0 - deliver
132 * 1 - block
133 */
134static __inline__ int icmp_filter(struct sock *sk, struct sk_buff *skb)
135{
136 int type;
137
138 if (!pskb_may_pull(skb, sizeof(struct icmphdr)))
139 return 1;
140
141 type = icmp_hdr(skb)->type;
142 if (type < 32) {
143 __u32 data = raw_sk(sk)->filter.data;
144
145 return ((1 << type) & data) != 0;
146 }
147
148 /* Do not block unknown ICMP types */
149 return 0;
150}
151
152/* IP input processing comes here for RAW socket delivery.
153 * Caller owns SKB, so we must make clones.
154 *
155 * RFC 1122: SHOULD pass TOS value up to the transport layer.
156 * -> It does. And not only TOS, but all IP header.
157 */
158static int raw_v4_input(struct sk_buff *skb, const struct iphdr *iph, int hash)
159{
160 struct sock *sk;
161 struct hlist_head *head;
162 int delivered = 0;
163 struct net *net;
164
165 read_lock(&raw_v4_hashinfo.lock);
166 head = &raw_v4_hashinfo.ht[hash];
167 if (hlist_empty(head))
168 goto out;
169
170 net = dev_net(skb->dev);
171 sk = __raw_v4_lookup(net, __sk_head(head), iph->protocol,
172 iph->saddr, iph->daddr,
173 skb->dev->ifindex);
174
175 while (sk) {
176 delivered = 1;
177 if (iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) {
178 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
179
180 /* Not releasing hash table! */
181 if (clone)
182 raw_rcv(sk, clone);
183 }
184 sk = __raw_v4_lookup(net, sk_next(sk), iph->protocol,
185 iph->saddr, iph->daddr,
186 skb->dev->ifindex);
187 }
188out:
189 read_unlock(&raw_v4_hashinfo.lock);
190 return delivered;
191}
192
193int raw_local_deliver(struct sk_buff *skb, int protocol)
194{
195 int hash;
196 struct sock *raw_sk;
197
198 hash = protocol & (RAW_HTABLE_SIZE - 1);
199 raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
200
201 /* If there maybe a raw socket we must check - if not we
202 * don't care less
203 */
204 if (raw_sk && !raw_v4_input(skb, ip_hdr(skb), hash))
205 raw_sk = NULL;
206
207 return raw_sk != NULL;
208
209}
210
211static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
212{
213 struct inet_sock *inet = inet_sk(sk);
214 const int type = icmp_hdr(skb)->type;
215 const int code = icmp_hdr(skb)->code;
216 int err = 0;
217 int harderr = 0;
218
219 /* Report error on raw socket, if:
220 1. User requested ip_recverr.
221 2. Socket is connected (otherwise the error indication
222 is useless without ip_recverr and error is hard.
223 */
224 if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
225 return;
226
227 switch (type) {
228 default:
229 case ICMP_TIME_EXCEEDED:
230 err = EHOSTUNREACH;
231 break;
232 case ICMP_SOURCE_QUENCH:
233 return;
234 case ICMP_PARAMETERPROB:
235 err = EPROTO;
236 harderr = 1;
237 break;
238 case ICMP_DEST_UNREACH:
239 err = EHOSTUNREACH;
240 if (code > NR_ICMP_UNREACH)
241 break;
242 err = icmp_err_convert[code].errno;
243 harderr = icmp_err_convert[code].fatal;
244 if (code == ICMP_FRAG_NEEDED) {
245 harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
246 err = EMSGSIZE;
247 }
248 }
249
250 if (inet->recverr) {
251 const struct iphdr *iph = (const struct iphdr *)skb->data;
252 u8 *payload = skb->data + (iph->ihl << 2);
253
254 if (inet->hdrincl)
255 payload = skb->data;
256 ip_icmp_error(sk, skb, err, 0, info, payload);
257 }
258
259 if (inet->recverr || harderr) {
260 sk->sk_err = err;
261 sk->sk_error_report(sk);
262 }
263}
264
265void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
266{
267 int hash;
268 struct sock *raw_sk;
269 const struct iphdr *iph;
270 struct net *net;
271
272 hash = protocol & (RAW_HTABLE_SIZE - 1);
273
274 read_lock(&raw_v4_hashinfo.lock);
275 raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
276 if (raw_sk != NULL) {
277 iph = (const struct iphdr *)skb->data;
278 net = dev_net(skb->dev);
279
280 while ((raw_sk = __raw_v4_lookup(net, raw_sk, protocol,
281 iph->daddr, iph->saddr,
282 skb->dev->ifindex)) != NULL) {
283 raw_err(raw_sk, skb, info);
284 raw_sk = sk_next(raw_sk);
285 iph = (const struct iphdr *)skb->data;
286 }
287 }
288 read_unlock(&raw_v4_hashinfo.lock);
289}
290
291static int raw_rcv_skb(struct sock *sk, struct sk_buff *skb)
292{
293 /* Charge it to the socket. */
294
295 ipv4_pktinfo_prepare(skb);
296 if (sock_queue_rcv_skb(sk, skb) < 0) {
297 kfree_skb(skb);
298 return NET_RX_DROP;
299 }
300
301 return NET_RX_SUCCESS;
302}
303
304int raw_rcv(struct sock *sk, struct sk_buff *skb)
305{
306 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
307 atomic_inc(&sk->sk_drops);
308 kfree_skb(skb);
309 return NET_RX_DROP;
310 }
311 nf_reset(skb);
312
313 skb_push(skb, skb->data - skb_network_header(skb));
314
315 raw_rcv_skb(sk, skb);
316 return 0;
317}
318
319static int raw_send_hdrinc(struct sock *sk, struct flowi4 *fl4,
320 void *from, size_t length,
321 struct rtable **rtp,
322 unsigned int flags)
323{
324 struct inet_sock *inet = inet_sk(sk);
325 struct net *net = sock_net(sk);
326 struct iphdr *iph;
327 struct sk_buff *skb;
328 unsigned int iphlen;
329 int err;
330 struct rtable *rt = *rtp;
331 int hlen, tlen;
332
333 if (length > rt->dst.dev->mtu) {
334 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
335 rt->dst.dev->mtu);
336 return -EMSGSIZE;
337 }
338 if (flags&MSG_PROBE)
339 goto out;
340
341 hlen = LL_RESERVED_SPACE(rt->dst.dev);
342 tlen = rt->dst.dev->needed_tailroom;
343 skb = sock_alloc_send_skb(sk,
344 length + hlen + tlen + 15,
345 flags & MSG_DONTWAIT, &err);
346 if (skb == NULL)
347 goto error;
348 skb_reserve(skb, hlen);
349
350 skb->priority = sk->sk_priority;
351 skb->mark = sk->sk_mark;
352 skb_dst_set(skb, &rt->dst);
353 *rtp = NULL;
354
355 skb_reset_network_header(skb);
356 iph = ip_hdr(skb);
357 skb_put(skb, length);
358
359 skb->ip_summed = CHECKSUM_NONE;
360
361 skb->transport_header = skb->network_header;
362 err = -EFAULT;
363 if (memcpy_fromiovecend((void *)iph, from, 0, length))
364 goto error_free;
365
366 iphlen = iph->ihl * 4;
367
368 /*
369 * We don't want to modify the ip header, but we do need to
370 * be sure that it won't cause problems later along the network
371 * stack. Specifically we want to make sure that iph->ihl is a
372 * sane value. If ihl points beyond the length of the buffer passed
373 * in, reject the frame as invalid
374 */
375 err = -EINVAL;
376 if (iphlen > length)
377 goto error_free;
378
379 if (iphlen >= sizeof(*iph)) {
380 if (!iph->saddr)
381 iph->saddr = fl4->saddr;
382 iph->check = 0;
383 iph->tot_len = htons(length);
384 if (!iph->id)
385 ip_select_ident(iph, &rt->dst, NULL);
386
387 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
388 }
389 if (iph->protocol == IPPROTO_ICMP)
390 icmp_out_count(net, ((struct icmphdr *)
391 skb_transport_header(skb))->type);
392
393 err = NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
394 rt->dst.dev, dst_output);
395 if (err > 0)
396 err = net_xmit_errno(err);
397 if (err)
398 goto error;
399out:
400 return 0;
401
402error_free:
403 kfree_skb(skb);
404error:
405 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
406 if (err == -ENOBUFS && !inet->recverr)
407 err = 0;
408 return err;
409}
410
411static int raw_probe_proto_opt(struct flowi4 *fl4, struct msghdr *msg)
412{
413 struct iovec *iov;
414 u8 __user *type = NULL;
415 u8 __user *code = NULL;
416 int probed = 0;
417 unsigned int i;
418
419 if (!msg->msg_iov)
420 return 0;
421
422 for (i = 0; i < msg->msg_iovlen; i++) {
423 iov = &msg->msg_iov[i];
424 if (!iov)
425 continue;
426
427 switch (fl4->flowi4_proto) {
428 case IPPROTO_ICMP:
429 /* check if one-byte field is readable or not. */
430 if (iov->iov_base && iov->iov_len < 1)
431 break;
432
433 if (!type) {
434 type = iov->iov_base;
435 /* check if code field is readable or not. */
436 if (iov->iov_len > 1)
437 code = type + 1;
438 } else if (!code)
439 code = iov->iov_base;
440
441 if (type && code) {
442 if (get_user(fl4->fl4_icmp_type, type) ||
443 get_user(fl4->fl4_icmp_code, code))
444 return -EFAULT;
445 probed = 1;
446 }
447 break;
448 default:
449 probed = 1;
450 break;
451 }
452 if (probed)
453 break;
454 }
455 return 0;
456}
457
458static int raw_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
459 size_t len)
460{
461 struct inet_sock *inet = inet_sk(sk);
462 struct ipcm_cookie ipc;
463 struct rtable *rt = NULL;
464 struct flowi4 fl4;
465 int free = 0;
466 __be32 daddr;
467 __be32 saddr;
468 u8 tos;
469 int err;
470 struct ip_options_data opt_copy;
471
472 err = -EMSGSIZE;
473 if (len > 0xFFFF)
474 goto out;
475
476 /*
477 * Check the flags.
478 */
479
480 err = -EOPNOTSUPP;
481 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message */
482 goto out; /* compatibility */
483
484 /*
485 * Get and verify the address.
486 */
487
488 if (msg->msg_namelen) {
489 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
490 err = -EINVAL;
491 if (msg->msg_namelen < sizeof(*usin))
492 goto out;
493 if (usin->sin_family != AF_INET) {
494 pr_info_once("%s: %s forgot to set AF_INET. Fix it!\n",
495 __func__, current->comm);
496 err = -EAFNOSUPPORT;
497 if (usin->sin_family)
498 goto out;
499 }
500 daddr = usin->sin_addr.s_addr;
501 /* ANK: I did not forget to get protocol from port field.
502 * I just do not know, who uses this weirdness.
503 * IP_HDRINCL is much more convenient.
504 */
505 } else {
506 err = -EDESTADDRREQ;
507 if (sk->sk_state != TCP_ESTABLISHED)
508 goto out;
509 daddr = inet->inet_daddr;
510 }
511
512 ipc.addr = inet->inet_saddr;
513 ipc.opt = NULL;
514 ipc.tx_flags = 0;
515 ipc.oif = sk->sk_bound_dev_if;
516
517 if (msg->msg_controllen) {
518 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
519 if (err)
520 goto out;
521 if (ipc.opt)
522 free = 1;
523 }
524
525 saddr = ipc.addr;
526 ipc.addr = daddr;
527
528 if (!ipc.opt) {
529 struct ip_options_rcu *inet_opt;
530
531 rcu_read_lock();
532 inet_opt = rcu_dereference(inet->inet_opt);
533 if (inet_opt) {
534 memcpy(&opt_copy, inet_opt,
535 sizeof(*inet_opt) + inet_opt->opt.optlen);
536 ipc.opt = &opt_copy.opt;
537 }
538 rcu_read_unlock();
539 }
540
541 if (ipc.opt) {
542 err = -EINVAL;
543 /* Linux does not mangle headers on raw sockets,
544 * so that IP options + IP_HDRINCL is non-sense.
545 */
546 if (inet->hdrincl)
547 goto done;
548 if (ipc.opt->opt.srr) {
549 if (!daddr)
550 goto done;
551 daddr = ipc.opt->opt.faddr;
552 }
553 }
554 tos = RT_CONN_FLAGS(sk);
555 if (msg->msg_flags & MSG_DONTROUTE)
556 tos |= RTO_ONLINK;
557
558 if (ipv4_is_multicast(daddr)) {
559 if (!ipc.oif)
560 ipc.oif = inet->mc_index;
561 if (!saddr)
562 saddr = inet->mc_addr;
563 } else if (!ipc.oif)
564 ipc.oif = inet->uc_index;
565
566 flowi4_init_output(&fl4, ipc.oif, sk->sk_mark, tos,
567 RT_SCOPE_UNIVERSE,
568 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
569 inet_sk_flowi_flags(sk) | FLOWI_FLAG_CAN_SLEEP,
570 daddr, saddr, 0, 0);
571
572 if (!inet->hdrincl) {
573 err = raw_probe_proto_opt(&fl4, msg);
574 if (err)
575 goto done;
576 }
577
578 security_sk_classify_flow(sk, flowi4_to_flowi(&fl4));
579 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
580 if (IS_ERR(rt)) {
581 err = PTR_ERR(rt);
582 rt = NULL;
583 goto done;
584 }
585
586 err = -EACCES;
587 if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
588 goto done;
589
590 if (msg->msg_flags & MSG_CONFIRM)
591 goto do_confirm;
592back_from_confirm:
593
594 if (inet->hdrincl)
595 err = raw_send_hdrinc(sk, &fl4, msg->msg_iov, len,
596 &rt, msg->msg_flags);
597
598 else {
599 if (!ipc.addr)
600 ipc.addr = fl4.daddr;
601 lock_sock(sk);
602 err = ip_append_data(sk, &fl4, ip_generic_getfrag,
603 msg->msg_iov, len, 0,
604 &ipc, &rt, msg->msg_flags);
605 if (err)
606 ip_flush_pending_frames(sk);
607 else if (!(msg->msg_flags & MSG_MORE)) {
608 err = ip_push_pending_frames(sk, &fl4);
609 if (err == -ENOBUFS && !inet->recverr)
610 err = 0;
611 }
612 release_sock(sk);
613 }
614done:
615 if (free)
616 kfree(ipc.opt);
617 ip_rt_put(rt);
618
619out:
620 if (err < 0)
621 return err;
622 return len;
623
624do_confirm:
625 dst_confirm(&rt->dst);
626 if (!(msg->msg_flags & MSG_PROBE) || len)
627 goto back_from_confirm;
628 err = 0;
629 goto done;
630}
631
632static void raw_close(struct sock *sk, long timeout)
633{
634 /*
635 * Raw sockets may have direct kernel references. Kill them.
636 */
637 ip_ra_control(sk, 0, NULL);
638
639 sk_common_release(sk);
640}
641
642static void raw_destroy(struct sock *sk)
643{
644 lock_sock(sk);
645 ip_flush_pending_frames(sk);
646 release_sock(sk);
647}
648
649/* This gets rid of all the nasties in af_inet. -DaveM */
650static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
651{
652 struct inet_sock *inet = inet_sk(sk);
653 struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
654 int ret = -EINVAL;
655 int chk_addr_ret;
656
657 if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
658 goto out;
659 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
660 ret = -EADDRNOTAVAIL;
661 if (addr->sin_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
662 chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
663 goto out;
664 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
665 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
666 inet->inet_saddr = 0; /* Use device */
667 sk_dst_reset(sk);
668 ret = 0;
669out: return ret;
670}
671
672/*
673 * This should be easy, if there is something there
674 * we return it, otherwise we block.
675 */
676
677static int raw_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
678 size_t len, int noblock, int flags, int *addr_len)
679{
680 struct inet_sock *inet = inet_sk(sk);
681 size_t copied = 0;
682 int err = -EOPNOTSUPP;
683 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
684 struct sk_buff *skb;
685
686 if (flags & MSG_OOB)
687 goto out;
688
689 if (addr_len)
690 *addr_len = sizeof(*sin);
691
692 if (flags & MSG_ERRQUEUE) {
693 err = ip_recv_error(sk, msg, len);
694 goto out;
695 }
696
697 skb = skb_recv_datagram(sk, flags, noblock, &err);
698 if (!skb)
699 goto out;
700
701 copied = skb->len;
702 if (len < copied) {
703 msg->msg_flags |= MSG_TRUNC;
704 copied = len;
705 }
706
707 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
708 if (err)
709 goto done;
710
711 sock_recv_ts_and_drops(msg, sk, skb);
712
713 /* Copy the address. */
714 if (sin) {
715 sin->sin_family = AF_INET;
716 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
717 sin->sin_port = 0;
718 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
719 }
720 if (inet->cmsg_flags)
721 ip_cmsg_recv(msg, skb);
722 if (flags & MSG_TRUNC)
723 copied = skb->len;
724done:
725 skb_free_datagram(sk, skb);
726out:
727 if (err)
728 return err;
729 return copied;
730}
731
732static int raw_init(struct sock *sk)
733{
734 struct raw_sock *rp = raw_sk(sk);
735
736 if (inet_sk(sk)->inet_num == IPPROTO_ICMP)
737 memset(&rp->filter, 0, sizeof(rp->filter));
738 return 0;
739}
740
741static int raw_seticmpfilter(struct sock *sk, char __user *optval, int optlen)
742{
743 if (optlen > sizeof(struct icmp_filter))
744 optlen = sizeof(struct icmp_filter);
745 if (copy_from_user(&raw_sk(sk)->filter, optval, optlen))
746 return -EFAULT;
747 return 0;
748}
749
750static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
751{
752 int len, ret = -EFAULT;
753
754 if (get_user(len, optlen))
755 goto out;
756 ret = -EINVAL;
757 if (len < 0)
758 goto out;
759 if (len > sizeof(struct icmp_filter))
760 len = sizeof(struct icmp_filter);
761 ret = -EFAULT;
762 if (put_user(len, optlen) ||
763 copy_to_user(optval, &raw_sk(sk)->filter, len))
764 goto out;
765 ret = 0;
766out: return ret;
767}
768
769static int do_raw_setsockopt(struct sock *sk, int level, int optname,
770 char __user *optval, unsigned int optlen)
771{
772 if (optname == ICMP_FILTER) {
773 if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
774 return -EOPNOTSUPP;
775 else
776 return raw_seticmpfilter(sk, optval, optlen);
777 }
778 return -ENOPROTOOPT;
779}
780
781static int raw_setsockopt(struct sock *sk, int level, int optname,
782 char __user *optval, unsigned int optlen)
783{
784 if (level != SOL_RAW)
785 return ip_setsockopt(sk, level, optname, optval, optlen);
786 return do_raw_setsockopt(sk, level, optname, optval, optlen);
787}
788
789#ifdef CONFIG_COMPAT
790static int compat_raw_setsockopt(struct sock *sk, int level, int optname,
791 char __user *optval, unsigned int optlen)
792{
793 if (level != SOL_RAW)
794 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
795 return do_raw_setsockopt(sk, level, optname, optval, optlen);
796}
797#endif
798
799static int do_raw_getsockopt(struct sock *sk, int level, int optname,
800 char __user *optval, int __user *optlen)
801{
802 if (optname == ICMP_FILTER) {
803 if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
804 return -EOPNOTSUPP;
805 else
806 return raw_geticmpfilter(sk, optval, optlen);
807 }
808 return -ENOPROTOOPT;
809}
810
811static int raw_getsockopt(struct sock *sk, int level, int optname,
812 char __user *optval, int __user *optlen)
813{
814 if (level != SOL_RAW)
815 return ip_getsockopt(sk, level, optname, optval, optlen);
816 return do_raw_getsockopt(sk, level, optname, optval, optlen);
817}
818
819#ifdef CONFIG_COMPAT
820static int compat_raw_getsockopt(struct sock *sk, int level, int optname,
821 char __user *optval, int __user *optlen)
822{
823 if (level != SOL_RAW)
824 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
825 return do_raw_getsockopt(sk, level, optname, optval, optlen);
826}
827#endif
828
829static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
830{
831 switch (cmd) {
832 case SIOCOUTQ: {
833 int amount = sk_wmem_alloc_get(sk);
834
835 return put_user(amount, (int __user *)arg);
836 }
837 case SIOCINQ: {
838 struct sk_buff *skb;
839 int amount = 0;
840
841 spin_lock_bh(&sk->sk_receive_queue.lock);
842 skb = skb_peek(&sk->sk_receive_queue);
843 if (skb != NULL)
844 amount = skb->len;
845 spin_unlock_bh(&sk->sk_receive_queue.lock);
846 return put_user(amount, (int __user *)arg);
847 }
848
849 default:
850#ifdef CONFIG_IP_MROUTE
851 return ipmr_ioctl(sk, cmd, (void __user *)arg);
852#else
853 return -ENOIOCTLCMD;
854#endif
855 }
856}
857
858#ifdef CONFIG_COMPAT
859static int compat_raw_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
860{
861 switch (cmd) {
862 case SIOCOUTQ:
863 case SIOCINQ:
864 return -ENOIOCTLCMD;
865 default:
866#ifdef CONFIG_IP_MROUTE
867 return ipmr_compat_ioctl(sk, cmd, compat_ptr(arg));
868#else
869 return -ENOIOCTLCMD;
870#endif
871 }
872}
873#endif
874
875struct proto raw_prot = {
876 .name = "RAW",
877 .owner = THIS_MODULE,
878 .close = raw_close,
879 .destroy = raw_destroy,
880 .connect = ip4_datagram_connect,
881 .disconnect = udp_disconnect,
882 .ioctl = raw_ioctl,
883 .init = raw_init,
884 .setsockopt = raw_setsockopt,
885 .getsockopt = raw_getsockopt,
886 .sendmsg = raw_sendmsg,
887 .recvmsg = raw_recvmsg,
888 .bind = raw_bind,
889 .backlog_rcv = raw_rcv_skb,
890 .hash = raw_hash_sk,
891 .unhash = raw_unhash_sk,
892 .obj_size = sizeof(struct raw_sock),
893 .h.raw_hash = &raw_v4_hashinfo,
894#ifdef CONFIG_COMPAT
895 .compat_setsockopt = compat_raw_setsockopt,
896 .compat_getsockopt = compat_raw_getsockopt,
897 .compat_ioctl = compat_raw_ioctl,
898#endif
899};
900
901#ifdef CONFIG_PROC_FS
902static struct sock *raw_get_first(struct seq_file *seq)
903{
904 struct sock *sk;
905 struct raw_iter_state *state = raw_seq_private(seq);
906
907 for (state->bucket = 0; state->bucket < RAW_HTABLE_SIZE;
908 ++state->bucket) {
909 struct hlist_node *node;
910
911 sk_for_each(sk, node, &state->h->ht[state->bucket])
912 if (sock_net(sk) == seq_file_net(seq))
913 goto found;
914 }
915 sk = NULL;
916found:
917 return sk;
918}
919
920static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
921{
922 struct raw_iter_state *state = raw_seq_private(seq);
923
924 do {
925 sk = sk_next(sk);
926try_again:
927 ;
928 } while (sk && sock_net(sk) != seq_file_net(seq));
929
930 if (!sk && ++state->bucket < RAW_HTABLE_SIZE) {
931 sk = sk_head(&state->h->ht[state->bucket]);
932 goto try_again;
933 }
934 return sk;
935}
936
937static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
938{
939 struct sock *sk = raw_get_first(seq);
940
941 if (sk)
942 while (pos && (sk = raw_get_next(seq, sk)) != NULL)
943 --pos;
944 return pos ? NULL : sk;
945}
946
947void *raw_seq_start(struct seq_file *seq, loff_t *pos)
948{
949 struct raw_iter_state *state = raw_seq_private(seq);
950
951 read_lock(&state->h->lock);
952 return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
953}
954EXPORT_SYMBOL_GPL(raw_seq_start);
955
956void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
957{
958 struct sock *sk;
959
960 if (v == SEQ_START_TOKEN)
961 sk = raw_get_first(seq);
962 else
963 sk = raw_get_next(seq, v);
964 ++*pos;
965 return sk;
966}
967EXPORT_SYMBOL_GPL(raw_seq_next);
968
969void raw_seq_stop(struct seq_file *seq, void *v)
970{
971 struct raw_iter_state *state = raw_seq_private(seq);
972
973 read_unlock(&state->h->lock);
974}
975EXPORT_SYMBOL_GPL(raw_seq_stop);
976
977static void raw_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
978{
979 struct inet_sock *inet = inet_sk(sp);
980 __be32 dest = inet->inet_daddr,
981 src = inet->inet_rcv_saddr;
982 __u16 destp = 0,
983 srcp = inet->inet_num;
984
985 seq_printf(seq, "%4d: %08X:%04X %08X:%04X"
986 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
987 i, src, srcp, dest, destp, sp->sk_state,
988 sk_wmem_alloc_get(sp),
989 sk_rmem_alloc_get(sp),
990 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
991 atomic_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
992}
993
994static int raw_seq_show(struct seq_file *seq, void *v)
995{
996 if (v == SEQ_START_TOKEN)
997 seq_printf(seq, " sl local_address rem_address st tx_queue "
998 "rx_queue tr tm->when retrnsmt uid timeout "
999 "inode ref pointer drops\n");
1000 else
1001 raw_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
1002 return 0;
1003}
1004
1005static const struct seq_operations raw_seq_ops = {
1006 .start = raw_seq_start,
1007 .next = raw_seq_next,
1008 .stop = raw_seq_stop,
1009 .show = raw_seq_show,
1010};
1011
1012int raw_seq_open(struct inode *ino, struct file *file,
1013 struct raw_hashinfo *h, const struct seq_operations *ops)
1014{
1015 int err;
1016 struct raw_iter_state *i;
1017
1018 err = seq_open_net(ino, file, ops, sizeof(struct raw_iter_state));
1019 if (err < 0)
1020 return err;
1021
1022 i = raw_seq_private((struct seq_file *)file->private_data);
1023 i->h = h;
1024 return 0;
1025}
1026EXPORT_SYMBOL_GPL(raw_seq_open);
1027
1028static int raw_v4_seq_open(struct inode *inode, struct file *file)
1029{
1030 return raw_seq_open(inode, file, &raw_v4_hashinfo, &raw_seq_ops);
1031}
1032
1033static const struct file_operations raw_seq_fops = {
1034 .owner = THIS_MODULE,
1035 .open = raw_v4_seq_open,
1036 .read = seq_read,
1037 .llseek = seq_lseek,
1038 .release = seq_release_net,
1039};
1040
1041static __net_init int raw_init_net(struct net *net)
1042{
1043 if (!proc_net_fops_create(net, "raw", S_IRUGO, &raw_seq_fops))
1044 return -ENOMEM;
1045
1046 return 0;
1047}
1048
1049static __net_exit void raw_exit_net(struct net *net)
1050{
1051 proc_net_remove(net, "raw");
1052}
1053
1054static __net_initdata struct pernet_operations raw_net_ops = {
1055 .init = raw_init_net,
1056 .exit = raw_exit_net,
1057};
1058
1059int __init raw_proc_init(void)
1060{
1061 return register_pernet_subsys(&raw_net_ops);
1062}
1063
1064void __init raw_proc_exit(void)
1065{
1066 unregister_pernet_subsys(&raw_net_ops);
1067}
1068#endif /* CONFIG_PROC_FS */
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * RAW - implementation of IP "raw" sockets.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 *
11 * Fixes:
12 * Alan Cox : verify_area() fixed up
13 * Alan Cox : ICMP error handling
14 * Alan Cox : EMSGSIZE if you send too big a packet
15 * Alan Cox : Now uses generic datagrams and shared
16 * skbuff library. No more peek crashes,
17 * no more backlogs
18 * Alan Cox : Checks sk->broadcast.
19 * Alan Cox : Uses skb_free_datagram/skb_copy_datagram
20 * Alan Cox : Raw passes ip options too
21 * Alan Cox : Setsocketopt added
22 * Alan Cox : Fixed error return for broadcasts
23 * Alan Cox : Removed wake_up calls
24 * Alan Cox : Use ttl/tos
25 * Alan Cox : Cleaned up old debugging
26 * Alan Cox : Use new kernel side addresses
27 * Arnt Gulbrandsen : Fixed MSG_DONTROUTE in raw sockets.
28 * Alan Cox : BSD style RAW socket demultiplexing.
29 * Alan Cox : Beginnings of mrouted support.
30 * Alan Cox : Added IP_HDRINCL option.
31 * Alan Cox : Skip broadcast check if BSDism set.
32 * David S. Miller : New socket lookup architecture.
33 *
34 * This program is free software; you can redistribute it and/or
35 * modify it under the terms of the GNU General Public License
36 * as published by the Free Software Foundation; either version
37 * 2 of the License, or (at your option) any later version.
38 */
39
40#include <linux/types.h>
41#include <linux/atomic.h>
42#include <asm/byteorder.h>
43#include <asm/current.h>
44#include <asm/uaccess.h>
45#include <asm/ioctls.h>
46#include <linux/stddef.h>
47#include <linux/slab.h>
48#include <linux/errno.h>
49#include <linux/aio.h>
50#include <linux/kernel.h>
51#include <linux/spinlock.h>
52#include <linux/sockios.h>
53#include <linux/socket.h>
54#include <linux/in.h>
55#include <linux/mroute.h>
56#include <linux/netdevice.h>
57#include <linux/in_route.h>
58#include <linux/route.h>
59#include <linux/skbuff.h>
60#include <net/net_namespace.h>
61#include <net/dst.h>
62#include <net/sock.h>
63#include <linux/ip.h>
64#include <linux/net.h>
65#include <net/ip.h>
66#include <net/icmp.h>
67#include <net/udp.h>
68#include <net/raw.h>
69#include <net/snmp.h>
70#include <net/tcp_states.h>
71#include <net/inet_common.h>
72#include <net/checksum.h>
73#include <net/xfrm.h>
74#include <linux/rtnetlink.h>
75#include <linux/proc_fs.h>
76#include <linux/seq_file.h>
77#include <linux/netfilter.h>
78#include <linux/netfilter_ipv4.h>
79#include <linux/compat.h>
80
81static struct raw_hashinfo raw_v4_hashinfo = {
82 .lock = __RW_LOCK_UNLOCKED(raw_v4_hashinfo.lock),
83};
84
85void raw_hash_sk(struct sock *sk)
86{
87 struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
88 struct hlist_head *head;
89
90 head = &h->ht[inet_sk(sk)->inet_num & (RAW_HTABLE_SIZE - 1)];
91
92 write_lock_bh(&h->lock);
93 sk_add_node(sk, head);
94 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
95 write_unlock_bh(&h->lock);
96}
97EXPORT_SYMBOL_GPL(raw_hash_sk);
98
99void raw_unhash_sk(struct sock *sk)
100{
101 struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
102
103 write_lock_bh(&h->lock);
104 if (sk_del_node_init(sk))
105 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
106 write_unlock_bh(&h->lock);
107}
108EXPORT_SYMBOL_GPL(raw_unhash_sk);
109
110static struct sock *__raw_v4_lookup(struct net *net, struct sock *sk,
111 unsigned short num, __be32 raddr, __be32 laddr, int dif)
112{
113 struct hlist_node *node;
114
115 sk_for_each_from(sk, node) {
116 struct inet_sock *inet = inet_sk(sk);
117
118 if (net_eq(sock_net(sk), net) && inet->inet_num == num &&
119 !(inet->inet_daddr && inet->inet_daddr != raddr) &&
120 !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
121 !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
122 goto found; /* gotcha */
123 }
124 sk = NULL;
125found:
126 return sk;
127}
128
129/*
130 * 0 - deliver
131 * 1 - block
132 */
133static __inline__ int icmp_filter(struct sock *sk, struct sk_buff *skb)
134{
135 int type;
136
137 if (!pskb_may_pull(skb, sizeof(struct icmphdr)))
138 return 1;
139
140 type = icmp_hdr(skb)->type;
141 if (type < 32) {
142 __u32 data = raw_sk(sk)->filter.data;
143
144 return ((1 << type) & data) != 0;
145 }
146
147 /* Do not block unknown ICMP types */
148 return 0;
149}
150
151/* IP input processing comes here for RAW socket delivery.
152 * Caller owns SKB, so we must make clones.
153 *
154 * RFC 1122: SHOULD pass TOS value up to the transport layer.
155 * -> It does. And not only TOS, but all IP header.
156 */
157static int raw_v4_input(struct sk_buff *skb, const struct iphdr *iph, int hash)
158{
159 struct sock *sk;
160 struct hlist_head *head;
161 int delivered = 0;
162 struct net *net;
163
164 read_lock(&raw_v4_hashinfo.lock);
165 head = &raw_v4_hashinfo.ht[hash];
166 if (hlist_empty(head))
167 goto out;
168
169 net = dev_net(skb->dev);
170 sk = __raw_v4_lookup(net, __sk_head(head), iph->protocol,
171 iph->saddr, iph->daddr,
172 skb->dev->ifindex);
173
174 while (sk) {
175 delivered = 1;
176 if (iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) {
177 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
178
179 /* Not releasing hash table! */
180 if (clone)
181 raw_rcv(sk, clone);
182 }
183 sk = __raw_v4_lookup(net, sk_next(sk), iph->protocol,
184 iph->saddr, iph->daddr,
185 skb->dev->ifindex);
186 }
187out:
188 read_unlock(&raw_v4_hashinfo.lock);
189 return delivered;
190}
191
192int raw_local_deliver(struct sk_buff *skb, int protocol)
193{
194 int hash;
195 struct sock *raw_sk;
196
197 hash = protocol & (RAW_HTABLE_SIZE - 1);
198 raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
199
200 /* If there maybe a raw socket we must check - if not we
201 * don't care less
202 */
203 if (raw_sk && !raw_v4_input(skb, ip_hdr(skb), hash))
204 raw_sk = NULL;
205
206 return raw_sk != NULL;
207
208}
209
210static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
211{
212 struct inet_sock *inet = inet_sk(sk);
213 const int type = icmp_hdr(skb)->type;
214 const int code = icmp_hdr(skb)->code;
215 int err = 0;
216 int harderr = 0;
217
218 /* Report error on raw socket, if:
219 1. User requested ip_recverr.
220 2. Socket is connected (otherwise the error indication
221 is useless without ip_recverr and error is hard.
222 */
223 if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
224 return;
225
226 switch (type) {
227 default:
228 case ICMP_TIME_EXCEEDED:
229 err = EHOSTUNREACH;
230 break;
231 case ICMP_SOURCE_QUENCH:
232 return;
233 case ICMP_PARAMETERPROB:
234 err = EPROTO;
235 harderr = 1;
236 break;
237 case ICMP_DEST_UNREACH:
238 err = EHOSTUNREACH;
239 if (code > NR_ICMP_UNREACH)
240 break;
241 err = icmp_err_convert[code].errno;
242 harderr = icmp_err_convert[code].fatal;
243 if (code == ICMP_FRAG_NEEDED) {
244 harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
245 err = EMSGSIZE;
246 }
247 }
248
249 if (inet->recverr) {
250 const struct iphdr *iph = (const struct iphdr *)skb->data;
251 u8 *payload = skb->data + (iph->ihl << 2);
252
253 if (inet->hdrincl)
254 payload = skb->data;
255 ip_icmp_error(sk, skb, err, 0, info, payload);
256 }
257
258 if (inet->recverr || harderr) {
259 sk->sk_err = err;
260 sk->sk_error_report(sk);
261 }
262}
263
264void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
265{
266 int hash;
267 struct sock *raw_sk;
268 const struct iphdr *iph;
269 struct net *net;
270
271 hash = protocol & (RAW_HTABLE_SIZE - 1);
272
273 read_lock(&raw_v4_hashinfo.lock);
274 raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
275 if (raw_sk != NULL) {
276 iph = (const struct iphdr *)skb->data;
277 net = dev_net(skb->dev);
278
279 while ((raw_sk = __raw_v4_lookup(net, raw_sk, protocol,
280 iph->daddr, iph->saddr,
281 skb->dev->ifindex)) != NULL) {
282 raw_err(raw_sk, skb, info);
283 raw_sk = sk_next(raw_sk);
284 iph = (const struct iphdr *)skb->data;
285 }
286 }
287 read_unlock(&raw_v4_hashinfo.lock);
288}
289
290static int raw_rcv_skb(struct sock * sk, struct sk_buff * skb)
291{
292 /* Charge it to the socket. */
293
294 if (ip_queue_rcv_skb(sk, skb) < 0) {
295 kfree_skb(skb);
296 return NET_RX_DROP;
297 }
298
299 return NET_RX_SUCCESS;
300}
301
302int raw_rcv(struct sock *sk, struct sk_buff *skb)
303{
304 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
305 atomic_inc(&sk->sk_drops);
306 kfree_skb(skb);
307 return NET_RX_DROP;
308 }
309 nf_reset(skb);
310
311 skb_push(skb, skb->data - skb_network_header(skb));
312
313 raw_rcv_skb(sk, skb);
314 return 0;
315}
316
317static int raw_send_hdrinc(struct sock *sk, struct flowi4 *fl4,
318 void *from, size_t length,
319 struct rtable **rtp,
320 unsigned int flags)
321{
322 struct inet_sock *inet = inet_sk(sk);
323 struct net *net = sock_net(sk);
324 struct iphdr *iph;
325 struct sk_buff *skb;
326 unsigned int iphlen;
327 int err;
328 struct rtable *rt = *rtp;
329
330 if (length > rt->dst.dev->mtu) {
331 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
332 rt->dst.dev->mtu);
333 return -EMSGSIZE;
334 }
335 if (flags&MSG_PROBE)
336 goto out;
337
338 skb = sock_alloc_send_skb(sk,
339 length + LL_ALLOCATED_SPACE(rt->dst.dev) + 15,
340 flags & MSG_DONTWAIT, &err);
341 if (skb == NULL)
342 goto error;
343 skb_reserve(skb, LL_RESERVED_SPACE(rt->dst.dev));
344
345 skb->priority = sk->sk_priority;
346 skb->mark = sk->sk_mark;
347 skb_dst_set(skb, &rt->dst);
348 *rtp = NULL;
349
350 skb_reset_network_header(skb);
351 iph = ip_hdr(skb);
352 skb_put(skb, length);
353
354 skb->ip_summed = CHECKSUM_NONE;
355
356 skb->transport_header = skb->network_header;
357 err = -EFAULT;
358 if (memcpy_fromiovecend((void *)iph, from, 0, length))
359 goto error_free;
360
361 iphlen = iph->ihl * 4;
362
363 /*
364 * We don't want to modify the ip header, but we do need to
365 * be sure that it won't cause problems later along the network
366 * stack. Specifically we want to make sure that iph->ihl is a
367 * sane value. If ihl points beyond the length of the buffer passed
368 * in, reject the frame as invalid
369 */
370 err = -EINVAL;
371 if (iphlen > length)
372 goto error_free;
373
374 if (iphlen >= sizeof(*iph)) {
375 if (!iph->saddr)
376 iph->saddr = fl4->saddr;
377 iph->check = 0;
378 iph->tot_len = htons(length);
379 if (!iph->id)
380 ip_select_ident(iph, &rt->dst, NULL);
381
382 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
383 }
384 if (iph->protocol == IPPROTO_ICMP)
385 icmp_out_count(net, ((struct icmphdr *)
386 skb_transport_header(skb))->type);
387
388 err = NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_OUT, skb, NULL,
389 rt->dst.dev, dst_output);
390 if (err > 0)
391 err = net_xmit_errno(err);
392 if (err)
393 goto error;
394out:
395 return 0;
396
397error_free:
398 kfree_skb(skb);
399error:
400 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
401 if (err == -ENOBUFS && !inet->recverr)
402 err = 0;
403 return err;
404}
405
406static int raw_probe_proto_opt(struct flowi4 *fl4, struct msghdr *msg)
407{
408 struct iovec *iov;
409 u8 __user *type = NULL;
410 u8 __user *code = NULL;
411 int probed = 0;
412 unsigned int i;
413
414 if (!msg->msg_iov)
415 return 0;
416
417 for (i = 0; i < msg->msg_iovlen; i++) {
418 iov = &msg->msg_iov[i];
419 if (!iov)
420 continue;
421
422 switch (fl4->flowi4_proto) {
423 case IPPROTO_ICMP:
424 /* check if one-byte field is readable or not. */
425 if (iov->iov_base && iov->iov_len < 1)
426 break;
427
428 if (!type) {
429 type = iov->iov_base;
430 /* check if code field is readable or not. */
431 if (iov->iov_len > 1)
432 code = type + 1;
433 } else if (!code)
434 code = iov->iov_base;
435
436 if (type && code) {
437 if (get_user(fl4->fl4_icmp_type, type) ||
438 get_user(fl4->fl4_icmp_code, code))
439 return -EFAULT;
440 probed = 1;
441 }
442 break;
443 default:
444 probed = 1;
445 break;
446 }
447 if (probed)
448 break;
449 }
450 return 0;
451}
452
453static int raw_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
454 size_t len)
455{
456 struct inet_sock *inet = inet_sk(sk);
457 struct ipcm_cookie ipc;
458 struct rtable *rt = NULL;
459 struct flowi4 fl4;
460 int free = 0;
461 __be32 daddr;
462 __be32 saddr;
463 u8 tos;
464 int err;
465 struct ip_options_data opt_copy;
466
467 err = -EMSGSIZE;
468 if (len > 0xFFFF)
469 goto out;
470
471 /*
472 * Check the flags.
473 */
474
475 err = -EOPNOTSUPP;
476 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message */
477 goto out; /* compatibility */
478
479 /*
480 * Get and verify the address.
481 */
482
483 if (msg->msg_namelen) {
484 struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
485 err = -EINVAL;
486 if (msg->msg_namelen < sizeof(*usin))
487 goto out;
488 if (usin->sin_family != AF_INET) {
489 static int complained;
490 if (!complained++)
491 printk(KERN_INFO "%s forgot to set AF_INET in "
492 "raw sendmsg. Fix it!\n",
493 current->comm);
494 err = -EAFNOSUPPORT;
495 if (usin->sin_family)
496 goto out;
497 }
498 daddr = usin->sin_addr.s_addr;
499 /* ANK: I did not forget to get protocol from port field.
500 * I just do not know, who uses this weirdness.
501 * IP_HDRINCL is much more convenient.
502 */
503 } else {
504 err = -EDESTADDRREQ;
505 if (sk->sk_state != TCP_ESTABLISHED)
506 goto out;
507 daddr = inet->inet_daddr;
508 }
509
510 ipc.addr = inet->inet_saddr;
511 ipc.opt = NULL;
512 ipc.tx_flags = 0;
513 ipc.oif = sk->sk_bound_dev_if;
514
515 if (msg->msg_controllen) {
516 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
517 if (err)
518 goto out;
519 if (ipc.opt)
520 free = 1;
521 }
522
523 saddr = ipc.addr;
524 ipc.addr = daddr;
525
526 if (!ipc.opt) {
527 struct ip_options_rcu *inet_opt;
528
529 rcu_read_lock();
530 inet_opt = rcu_dereference(inet->inet_opt);
531 if (inet_opt) {
532 memcpy(&opt_copy, inet_opt,
533 sizeof(*inet_opt) + inet_opt->opt.optlen);
534 ipc.opt = &opt_copy.opt;
535 }
536 rcu_read_unlock();
537 }
538
539 if (ipc.opt) {
540 err = -EINVAL;
541 /* Linux does not mangle headers on raw sockets,
542 * so that IP options + IP_HDRINCL is non-sense.
543 */
544 if (inet->hdrincl)
545 goto done;
546 if (ipc.opt->opt.srr) {
547 if (!daddr)
548 goto done;
549 daddr = ipc.opt->opt.faddr;
550 }
551 }
552 tos = RT_CONN_FLAGS(sk);
553 if (msg->msg_flags & MSG_DONTROUTE)
554 tos |= RTO_ONLINK;
555
556 if (ipv4_is_multicast(daddr)) {
557 if (!ipc.oif)
558 ipc.oif = inet->mc_index;
559 if (!saddr)
560 saddr = inet->mc_addr;
561 }
562
563 flowi4_init_output(&fl4, ipc.oif, sk->sk_mark, tos,
564 RT_SCOPE_UNIVERSE,
565 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
566 inet_sk_flowi_flags(sk) | FLOWI_FLAG_CAN_SLEEP,
567 daddr, saddr, 0, 0);
568
569 if (!inet->hdrincl) {
570 err = raw_probe_proto_opt(&fl4, msg);
571 if (err)
572 goto done;
573 }
574
575 security_sk_classify_flow(sk, flowi4_to_flowi(&fl4));
576 rt = ip_route_output_flow(sock_net(sk), &fl4, sk);
577 if (IS_ERR(rt)) {
578 err = PTR_ERR(rt);
579 rt = NULL;
580 goto done;
581 }
582
583 err = -EACCES;
584 if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
585 goto done;
586
587 if (msg->msg_flags & MSG_CONFIRM)
588 goto do_confirm;
589back_from_confirm:
590
591 if (inet->hdrincl)
592 err = raw_send_hdrinc(sk, &fl4, msg->msg_iov, len,
593 &rt, msg->msg_flags);
594
595 else {
596 if (!ipc.addr)
597 ipc.addr = fl4.daddr;
598 lock_sock(sk);
599 err = ip_append_data(sk, &fl4, ip_generic_getfrag,
600 msg->msg_iov, len, 0,
601 &ipc, &rt, msg->msg_flags);
602 if (err)
603 ip_flush_pending_frames(sk);
604 else if (!(msg->msg_flags & MSG_MORE)) {
605 err = ip_push_pending_frames(sk, &fl4);
606 if (err == -ENOBUFS && !inet->recverr)
607 err = 0;
608 }
609 release_sock(sk);
610 }
611done:
612 if (free)
613 kfree(ipc.opt);
614 ip_rt_put(rt);
615
616out:
617 if (err < 0)
618 return err;
619 return len;
620
621do_confirm:
622 dst_confirm(&rt->dst);
623 if (!(msg->msg_flags & MSG_PROBE) || len)
624 goto back_from_confirm;
625 err = 0;
626 goto done;
627}
628
629static void raw_close(struct sock *sk, long timeout)
630{
631 /*
632 * Raw sockets may have direct kernel references. Kill them.
633 */
634 ip_ra_control(sk, 0, NULL);
635
636 sk_common_release(sk);
637}
638
639static void raw_destroy(struct sock *sk)
640{
641 lock_sock(sk);
642 ip_flush_pending_frames(sk);
643 release_sock(sk);
644}
645
646/* This gets rid of all the nasties in af_inet. -DaveM */
647static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
648{
649 struct inet_sock *inet = inet_sk(sk);
650 struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
651 int ret = -EINVAL;
652 int chk_addr_ret;
653
654 if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
655 goto out;
656 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
657 ret = -EADDRNOTAVAIL;
658 if (addr->sin_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
659 chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
660 goto out;
661 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
662 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
663 inet->inet_saddr = 0; /* Use device */
664 sk_dst_reset(sk);
665 ret = 0;
666out: return ret;
667}
668
669/*
670 * This should be easy, if there is something there
671 * we return it, otherwise we block.
672 */
673
674static int raw_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
675 size_t len, int noblock, int flags, int *addr_len)
676{
677 struct inet_sock *inet = inet_sk(sk);
678 size_t copied = 0;
679 int err = -EOPNOTSUPP;
680 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
681 struct sk_buff *skb;
682
683 if (flags & MSG_OOB)
684 goto out;
685
686 if (addr_len)
687 *addr_len = sizeof(*sin);
688
689 if (flags & MSG_ERRQUEUE) {
690 err = ip_recv_error(sk, msg, len);
691 goto out;
692 }
693
694 skb = skb_recv_datagram(sk, flags, noblock, &err);
695 if (!skb)
696 goto out;
697
698 copied = skb->len;
699 if (len < copied) {
700 msg->msg_flags |= MSG_TRUNC;
701 copied = len;
702 }
703
704 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
705 if (err)
706 goto done;
707
708 sock_recv_ts_and_drops(msg, sk, skb);
709
710 /* Copy the address. */
711 if (sin) {
712 sin->sin_family = AF_INET;
713 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
714 sin->sin_port = 0;
715 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
716 }
717 if (inet->cmsg_flags)
718 ip_cmsg_recv(msg, skb);
719 if (flags & MSG_TRUNC)
720 copied = skb->len;
721done:
722 skb_free_datagram(sk, skb);
723out:
724 if (err)
725 return err;
726 return copied;
727}
728
729static int raw_init(struct sock *sk)
730{
731 struct raw_sock *rp = raw_sk(sk);
732
733 if (inet_sk(sk)->inet_num == IPPROTO_ICMP)
734 memset(&rp->filter, 0, sizeof(rp->filter));
735 return 0;
736}
737
738static int raw_seticmpfilter(struct sock *sk, char __user *optval, int optlen)
739{
740 if (optlen > sizeof(struct icmp_filter))
741 optlen = sizeof(struct icmp_filter);
742 if (copy_from_user(&raw_sk(sk)->filter, optval, optlen))
743 return -EFAULT;
744 return 0;
745}
746
747static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
748{
749 int len, ret = -EFAULT;
750
751 if (get_user(len, optlen))
752 goto out;
753 ret = -EINVAL;
754 if (len < 0)
755 goto out;
756 if (len > sizeof(struct icmp_filter))
757 len = sizeof(struct icmp_filter);
758 ret = -EFAULT;
759 if (put_user(len, optlen) ||
760 copy_to_user(optval, &raw_sk(sk)->filter, len))
761 goto out;
762 ret = 0;
763out: return ret;
764}
765
766static int do_raw_setsockopt(struct sock *sk, int level, int optname,
767 char __user *optval, unsigned int optlen)
768{
769 if (optname == ICMP_FILTER) {
770 if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
771 return -EOPNOTSUPP;
772 else
773 return raw_seticmpfilter(sk, optval, optlen);
774 }
775 return -ENOPROTOOPT;
776}
777
778static int raw_setsockopt(struct sock *sk, int level, int optname,
779 char __user *optval, unsigned int optlen)
780{
781 if (level != SOL_RAW)
782 return ip_setsockopt(sk, level, optname, optval, optlen);
783 return do_raw_setsockopt(sk, level, optname, optval, optlen);
784}
785
786#ifdef CONFIG_COMPAT
787static int compat_raw_setsockopt(struct sock *sk, int level, int optname,
788 char __user *optval, unsigned int optlen)
789{
790 if (level != SOL_RAW)
791 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
792 return do_raw_setsockopt(sk, level, optname, optval, optlen);
793}
794#endif
795
796static int do_raw_getsockopt(struct sock *sk, int level, int optname,
797 char __user *optval, int __user *optlen)
798{
799 if (optname == ICMP_FILTER) {
800 if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
801 return -EOPNOTSUPP;
802 else
803 return raw_geticmpfilter(sk, optval, optlen);
804 }
805 return -ENOPROTOOPT;
806}
807
808static int raw_getsockopt(struct sock *sk, int level, int optname,
809 char __user *optval, int __user *optlen)
810{
811 if (level != SOL_RAW)
812 return ip_getsockopt(sk, level, optname, optval, optlen);
813 return do_raw_getsockopt(sk, level, optname, optval, optlen);
814}
815
816#ifdef CONFIG_COMPAT
817static int compat_raw_getsockopt(struct sock *sk, int level, int optname,
818 char __user *optval, int __user *optlen)
819{
820 if (level != SOL_RAW)
821 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
822 return do_raw_getsockopt(sk, level, optname, optval, optlen);
823}
824#endif
825
826static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
827{
828 switch (cmd) {
829 case SIOCOUTQ: {
830 int amount = sk_wmem_alloc_get(sk);
831
832 return put_user(amount, (int __user *)arg);
833 }
834 case SIOCINQ: {
835 struct sk_buff *skb;
836 int amount = 0;
837
838 spin_lock_bh(&sk->sk_receive_queue.lock);
839 skb = skb_peek(&sk->sk_receive_queue);
840 if (skb != NULL)
841 amount = skb->len;
842 spin_unlock_bh(&sk->sk_receive_queue.lock);
843 return put_user(amount, (int __user *)arg);
844 }
845
846 default:
847#ifdef CONFIG_IP_MROUTE
848 return ipmr_ioctl(sk, cmd, (void __user *)arg);
849#else
850 return -ENOIOCTLCMD;
851#endif
852 }
853}
854
855#ifdef CONFIG_COMPAT
856static int compat_raw_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
857{
858 switch (cmd) {
859 case SIOCOUTQ:
860 case SIOCINQ:
861 return -ENOIOCTLCMD;
862 default:
863#ifdef CONFIG_IP_MROUTE
864 return ipmr_compat_ioctl(sk, cmd, compat_ptr(arg));
865#else
866 return -ENOIOCTLCMD;
867#endif
868 }
869}
870#endif
871
872struct proto raw_prot = {
873 .name = "RAW",
874 .owner = THIS_MODULE,
875 .close = raw_close,
876 .destroy = raw_destroy,
877 .connect = ip4_datagram_connect,
878 .disconnect = udp_disconnect,
879 .ioctl = raw_ioctl,
880 .init = raw_init,
881 .setsockopt = raw_setsockopt,
882 .getsockopt = raw_getsockopt,
883 .sendmsg = raw_sendmsg,
884 .recvmsg = raw_recvmsg,
885 .bind = raw_bind,
886 .backlog_rcv = raw_rcv_skb,
887 .hash = raw_hash_sk,
888 .unhash = raw_unhash_sk,
889 .obj_size = sizeof(struct raw_sock),
890 .h.raw_hash = &raw_v4_hashinfo,
891#ifdef CONFIG_COMPAT
892 .compat_setsockopt = compat_raw_setsockopt,
893 .compat_getsockopt = compat_raw_getsockopt,
894 .compat_ioctl = compat_raw_ioctl,
895#endif
896};
897
898#ifdef CONFIG_PROC_FS
899static struct sock *raw_get_first(struct seq_file *seq)
900{
901 struct sock *sk;
902 struct raw_iter_state *state = raw_seq_private(seq);
903
904 for (state->bucket = 0; state->bucket < RAW_HTABLE_SIZE;
905 ++state->bucket) {
906 struct hlist_node *node;
907
908 sk_for_each(sk, node, &state->h->ht[state->bucket])
909 if (sock_net(sk) == seq_file_net(seq))
910 goto found;
911 }
912 sk = NULL;
913found:
914 return sk;
915}
916
917static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
918{
919 struct raw_iter_state *state = raw_seq_private(seq);
920
921 do {
922 sk = sk_next(sk);
923try_again:
924 ;
925 } while (sk && sock_net(sk) != seq_file_net(seq));
926
927 if (!sk && ++state->bucket < RAW_HTABLE_SIZE) {
928 sk = sk_head(&state->h->ht[state->bucket]);
929 goto try_again;
930 }
931 return sk;
932}
933
934static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
935{
936 struct sock *sk = raw_get_first(seq);
937
938 if (sk)
939 while (pos && (sk = raw_get_next(seq, sk)) != NULL)
940 --pos;
941 return pos ? NULL : sk;
942}
943
944void *raw_seq_start(struct seq_file *seq, loff_t *pos)
945{
946 struct raw_iter_state *state = raw_seq_private(seq);
947
948 read_lock(&state->h->lock);
949 return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
950}
951EXPORT_SYMBOL_GPL(raw_seq_start);
952
953void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
954{
955 struct sock *sk;
956
957 if (v == SEQ_START_TOKEN)
958 sk = raw_get_first(seq);
959 else
960 sk = raw_get_next(seq, v);
961 ++*pos;
962 return sk;
963}
964EXPORT_SYMBOL_GPL(raw_seq_next);
965
966void raw_seq_stop(struct seq_file *seq, void *v)
967{
968 struct raw_iter_state *state = raw_seq_private(seq);
969
970 read_unlock(&state->h->lock);
971}
972EXPORT_SYMBOL_GPL(raw_seq_stop);
973
974static void raw_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
975{
976 struct inet_sock *inet = inet_sk(sp);
977 __be32 dest = inet->inet_daddr,
978 src = inet->inet_rcv_saddr;
979 __u16 destp = 0,
980 srcp = inet->inet_num;
981
982 seq_printf(seq, "%4d: %08X:%04X %08X:%04X"
983 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d\n",
984 i, src, srcp, dest, destp, sp->sk_state,
985 sk_wmem_alloc_get(sp),
986 sk_rmem_alloc_get(sp),
987 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
988 atomic_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
989}
990
991static int raw_seq_show(struct seq_file *seq, void *v)
992{
993 if (v == SEQ_START_TOKEN)
994 seq_printf(seq, " sl local_address rem_address st tx_queue "
995 "rx_queue tr tm->when retrnsmt uid timeout "
996 "inode ref pointer drops\n");
997 else
998 raw_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
999 return 0;
1000}
1001
1002static const struct seq_operations raw_seq_ops = {
1003 .start = raw_seq_start,
1004 .next = raw_seq_next,
1005 .stop = raw_seq_stop,
1006 .show = raw_seq_show,
1007};
1008
1009int raw_seq_open(struct inode *ino, struct file *file,
1010 struct raw_hashinfo *h, const struct seq_operations *ops)
1011{
1012 int err;
1013 struct raw_iter_state *i;
1014
1015 err = seq_open_net(ino, file, ops, sizeof(struct raw_iter_state));
1016 if (err < 0)
1017 return err;
1018
1019 i = raw_seq_private((struct seq_file *)file->private_data);
1020 i->h = h;
1021 return 0;
1022}
1023EXPORT_SYMBOL_GPL(raw_seq_open);
1024
1025static int raw_v4_seq_open(struct inode *inode, struct file *file)
1026{
1027 return raw_seq_open(inode, file, &raw_v4_hashinfo, &raw_seq_ops);
1028}
1029
1030static const struct file_operations raw_seq_fops = {
1031 .owner = THIS_MODULE,
1032 .open = raw_v4_seq_open,
1033 .read = seq_read,
1034 .llseek = seq_lseek,
1035 .release = seq_release_net,
1036};
1037
1038static __net_init int raw_init_net(struct net *net)
1039{
1040 if (!proc_net_fops_create(net, "raw", S_IRUGO, &raw_seq_fops))
1041 return -ENOMEM;
1042
1043 return 0;
1044}
1045
1046static __net_exit void raw_exit_net(struct net *net)
1047{
1048 proc_net_remove(net, "raw");
1049}
1050
1051static __net_initdata struct pernet_operations raw_net_ops = {
1052 .init = raw_init_net,
1053 .exit = raw_exit_net,
1054};
1055
1056int __init raw_proc_init(void)
1057{
1058 return register_pernet_subsys(&raw_net_ops);
1059}
1060
1061void __init raw_proc_exit(void)
1062{
1063 unregister_pernet_subsys(&raw_net_ops);
1064}
1065#endif /* CONFIG_PROC_FS */