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/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 */
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/kernel.h>
50#include <linux/export.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 <linux/igmp.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#include <linux/uio.h>
82
83struct raw_frag_vec {
84 struct msghdr *msg;
85 union {
86 struct icmphdr icmph;
87 char c[1];
88 } hdr;
89 int hlen;
90};
91
92static struct raw_hashinfo raw_v4_hashinfo = {
93 .lock = __RW_LOCK_UNLOCKED(raw_v4_hashinfo.lock),
94};
95
96int raw_hash_sk(struct sock *sk)
97{
98 struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
99 struct hlist_head *head;
100
101 head = &h->ht[inet_sk(sk)->inet_num & (RAW_HTABLE_SIZE - 1)];
102
103 write_lock_bh(&h->lock);
104 sk_add_node(sk, head);
105 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
106 write_unlock_bh(&h->lock);
107
108 return 0;
109}
110EXPORT_SYMBOL_GPL(raw_hash_sk);
111
112void raw_unhash_sk(struct sock *sk)
113{
114 struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
115
116 write_lock_bh(&h->lock);
117 if (sk_del_node_init(sk))
118 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
119 write_unlock_bh(&h->lock);
120}
121EXPORT_SYMBOL_GPL(raw_unhash_sk);
122
123static struct sock *__raw_v4_lookup(struct net *net, struct sock *sk,
124 unsigned short num, __be32 raddr, __be32 laddr, int dif)
125{
126 sk_for_each_from(sk) {
127 struct inet_sock *inet = inet_sk(sk);
128
129 if (net_eq(sock_net(sk), net) && inet->inet_num == num &&
130 !(inet->inet_daddr && inet->inet_daddr != raddr) &&
131 !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
132 !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
133 goto found; /* gotcha */
134 }
135 sk = NULL;
136found:
137 return sk;
138}
139
140/*
141 * 0 - deliver
142 * 1 - block
143 */
144static int icmp_filter(const struct sock *sk, const struct sk_buff *skb)
145{
146 struct icmphdr _hdr;
147 const struct icmphdr *hdr;
148
149 hdr = skb_header_pointer(skb, skb_transport_offset(skb),
150 sizeof(_hdr), &_hdr);
151 if (!hdr)
152 return 1;
153
154 if (hdr->type < 32) {
155 __u32 data = raw_sk(sk)->filter.data;
156
157 return ((1U << hdr->type) & data) != 0;
158 }
159
160 /* Do not block unknown ICMP types */
161 return 0;
162}
163
164/* IP input processing comes here for RAW socket delivery.
165 * Caller owns SKB, so we must make clones.
166 *
167 * RFC 1122: SHOULD pass TOS value up to the transport layer.
168 * -> It does. And not only TOS, but all IP header.
169 */
170static int raw_v4_input(struct sk_buff *skb, const struct iphdr *iph, int hash)
171{
172 struct sock *sk;
173 struct hlist_head *head;
174 int delivered = 0;
175 struct net *net;
176
177 read_lock(&raw_v4_hashinfo.lock);
178 head = &raw_v4_hashinfo.ht[hash];
179 if (hlist_empty(head))
180 goto out;
181
182 net = dev_net(skb->dev);
183 sk = __raw_v4_lookup(net, __sk_head(head), iph->protocol,
184 iph->saddr, iph->daddr,
185 skb->dev->ifindex);
186
187 while (sk) {
188 delivered = 1;
189 if ((iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) &&
190 ip_mc_sf_allow(sk, iph->daddr, iph->saddr,
191 skb->dev->ifindex)) {
192 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
193
194 /* Not releasing hash table! */
195 if (clone)
196 raw_rcv(sk, clone);
197 }
198 sk = __raw_v4_lookup(net, sk_next(sk), iph->protocol,
199 iph->saddr, iph->daddr,
200 skb->dev->ifindex);
201 }
202out:
203 read_unlock(&raw_v4_hashinfo.lock);
204 return delivered;
205}
206
207int raw_local_deliver(struct sk_buff *skb, int protocol)
208{
209 int hash;
210 struct sock *raw_sk;
211
212 hash = protocol & (RAW_HTABLE_SIZE - 1);
213 raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
214
215 /* If there maybe a raw socket we must check - if not we
216 * don't care less
217 */
218 if (raw_sk && !raw_v4_input(skb, ip_hdr(skb), hash))
219 raw_sk = NULL;
220
221 return raw_sk != NULL;
222
223}
224
225static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
226{
227 struct inet_sock *inet = inet_sk(sk);
228 const int type = icmp_hdr(skb)->type;
229 const int code = icmp_hdr(skb)->code;
230 int err = 0;
231 int harderr = 0;
232
233 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
234 ipv4_sk_update_pmtu(skb, sk, info);
235 else if (type == ICMP_REDIRECT) {
236 ipv4_sk_redirect(skb, sk);
237 return;
238 }
239
240 /* Report error on raw socket, if:
241 1. User requested ip_recverr.
242 2. Socket is connected (otherwise the error indication
243 is useless without ip_recverr and error is hard.
244 */
245 if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
246 return;
247
248 switch (type) {
249 default:
250 case ICMP_TIME_EXCEEDED:
251 err = EHOSTUNREACH;
252 break;
253 case ICMP_SOURCE_QUENCH:
254 return;
255 case ICMP_PARAMETERPROB:
256 err = EPROTO;
257 harderr = 1;
258 break;
259 case ICMP_DEST_UNREACH:
260 err = EHOSTUNREACH;
261 if (code > NR_ICMP_UNREACH)
262 break;
263 err = icmp_err_convert[code].errno;
264 harderr = icmp_err_convert[code].fatal;
265 if (code == ICMP_FRAG_NEEDED) {
266 harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
267 err = EMSGSIZE;
268 }
269 }
270
271 if (inet->recverr) {
272 const struct iphdr *iph = (const struct iphdr *)skb->data;
273 u8 *payload = skb->data + (iph->ihl << 2);
274
275 if (inet->hdrincl)
276 payload = skb->data;
277 ip_icmp_error(sk, skb, err, 0, info, payload);
278 }
279
280 if (inet->recverr || harderr) {
281 sk->sk_err = err;
282 sk->sk_error_report(sk);
283 }
284}
285
286void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
287{
288 int hash;
289 struct sock *raw_sk;
290 const struct iphdr *iph;
291 struct net *net;
292
293 hash = protocol & (RAW_HTABLE_SIZE - 1);
294
295 read_lock(&raw_v4_hashinfo.lock);
296 raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
297 if (raw_sk) {
298 iph = (const struct iphdr *)skb->data;
299 net = dev_net(skb->dev);
300
301 while ((raw_sk = __raw_v4_lookup(net, raw_sk, protocol,
302 iph->daddr, iph->saddr,
303 skb->dev->ifindex)) != NULL) {
304 raw_err(raw_sk, skb, info);
305 raw_sk = sk_next(raw_sk);
306 iph = (const struct iphdr *)skb->data;
307 }
308 }
309 read_unlock(&raw_v4_hashinfo.lock);
310}
311
312static int raw_rcv_skb(struct sock *sk, struct sk_buff *skb)
313{
314 /* Charge it to the socket. */
315
316 ipv4_pktinfo_prepare(sk, skb);
317 if (sock_queue_rcv_skb(sk, skb) < 0) {
318 kfree_skb(skb);
319 return NET_RX_DROP;
320 }
321
322 return NET_RX_SUCCESS;
323}
324
325int raw_rcv(struct sock *sk, struct sk_buff *skb)
326{
327 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
328 atomic_inc(&sk->sk_drops);
329 kfree_skb(skb);
330 return NET_RX_DROP;
331 }
332 nf_reset(skb);
333
334 skb_push(skb, skb->data - skb_network_header(skb));
335
336 raw_rcv_skb(sk, skb);
337 return 0;
338}
339
340static int raw_send_hdrinc(struct sock *sk, struct flowi4 *fl4,
341 struct msghdr *msg, size_t length,
342 struct rtable **rtp,
343 unsigned int flags)
344{
345 struct inet_sock *inet = inet_sk(sk);
346 struct net *net = sock_net(sk);
347 struct iphdr *iph;
348 struct sk_buff *skb;
349 unsigned int iphlen;
350 int err;
351 struct rtable *rt = *rtp;
352 int hlen, tlen;
353
354 if (length > rt->dst.dev->mtu) {
355 ip_local_error(sk, EMSGSIZE, fl4->daddr, inet->inet_dport,
356 rt->dst.dev->mtu);
357 return -EMSGSIZE;
358 }
359 if (flags&MSG_PROBE)
360 goto out;
361
362 hlen = LL_RESERVED_SPACE(rt->dst.dev);
363 tlen = rt->dst.dev->needed_tailroom;
364 skb = sock_alloc_send_skb(sk,
365 length + hlen + tlen + 15,
366 flags & MSG_DONTWAIT, &err);
367 if (!skb)
368 goto error;
369 skb_reserve(skb, hlen);
370
371 skb->priority = sk->sk_priority;
372 skb->mark = sk->sk_mark;
373 skb_dst_set(skb, &rt->dst);
374 *rtp = NULL;
375
376 skb_reset_network_header(skb);
377 iph = ip_hdr(skb);
378 skb_put(skb, length);
379
380 skb->ip_summed = CHECKSUM_NONE;
381
382 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
383
384 skb->transport_header = skb->network_header;
385 err = -EFAULT;
386 if (memcpy_from_msg(iph, msg, length))
387 goto error_free;
388
389 iphlen = iph->ihl * 4;
390
391 /*
392 * We don't want to modify the ip header, but we do need to
393 * be sure that it won't cause problems later along the network
394 * stack. Specifically we want to make sure that iph->ihl is a
395 * sane value. If ihl points beyond the length of the buffer passed
396 * in, reject the frame as invalid
397 */
398 err = -EINVAL;
399 if (iphlen > length)
400 goto error_free;
401
402 if (iphlen >= sizeof(*iph)) {
403 if (!iph->saddr)
404 iph->saddr = fl4->saddr;
405 iph->check = 0;
406 iph->tot_len = htons(length);
407 if (!iph->id)
408 ip_select_ident(net, skb, NULL);
409
410 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
411 skb->transport_header += iphlen;
412 if (iph->protocol == IPPROTO_ICMP &&
413 length >= iphlen + sizeof(struct icmphdr))
414 icmp_out_count(net, ((struct icmphdr *)
415 skb_transport_header(skb))->type);
416 }
417
418 err = NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
419 net, sk, skb, NULL, rt->dst.dev,
420 dst_output);
421 if (err > 0)
422 err = net_xmit_errno(err);
423 if (err)
424 goto error;
425out:
426 return 0;
427
428error_free:
429 kfree_skb(skb);
430error:
431 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
432 if (err == -ENOBUFS && !inet->recverr)
433 err = 0;
434 return err;
435}
436
437static int raw_probe_proto_opt(struct raw_frag_vec *rfv, struct flowi4 *fl4)
438{
439 int err;
440
441 if (fl4->flowi4_proto != IPPROTO_ICMP)
442 return 0;
443
444 /* We only need the first two bytes. */
445 rfv->hlen = 2;
446
447 err = memcpy_from_msg(rfv->hdr.c, rfv->msg, rfv->hlen);
448 if (err)
449 return err;
450
451 fl4->fl4_icmp_type = rfv->hdr.icmph.type;
452 fl4->fl4_icmp_code = rfv->hdr.icmph.code;
453
454 return 0;
455}
456
457static int raw_getfrag(void *from, char *to, int offset, int len, int odd,
458 struct sk_buff *skb)
459{
460 struct raw_frag_vec *rfv = from;
461
462 if (offset < rfv->hlen) {
463 int copy = min(rfv->hlen - offset, len);
464
465 if (skb->ip_summed == CHECKSUM_PARTIAL)
466 memcpy(to, rfv->hdr.c + offset, copy);
467 else
468 skb->csum = csum_block_add(
469 skb->csum,
470 csum_partial_copy_nocheck(rfv->hdr.c + offset,
471 to, copy, 0),
472 odd);
473
474 odd = 0;
475 offset += copy;
476 to += copy;
477 len -= copy;
478
479 if (!len)
480 return 0;
481 }
482
483 offset -= rfv->hlen;
484
485 return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb);
486}
487
488static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
489{
490 struct inet_sock *inet = inet_sk(sk);
491 struct net *net = sock_net(sk);
492 struct ipcm_cookie ipc;
493 struct rtable *rt = NULL;
494 struct flowi4 fl4;
495 int free = 0;
496 __be32 daddr;
497 __be32 saddr;
498 u8 tos;
499 int err;
500 struct ip_options_data opt_copy;
501 struct raw_frag_vec rfv;
502
503 err = -EMSGSIZE;
504 if (len > 0xFFFF)
505 goto out;
506
507 /*
508 * Check the flags.
509 */
510
511 err = -EOPNOTSUPP;
512 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message */
513 goto out; /* compatibility */
514
515 /*
516 * Get and verify the address.
517 */
518
519 if (msg->msg_namelen) {
520 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
521 err = -EINVAL;
522 if (msg->msg_namelen < sizeof(*usin))
523 goto out;
524 if (usin->sin_family != AF_INET) {
525 pr_info_once("%s: %s forgot to set AF_INET. Fix it!\n",
526 __func__, current->comm);
527 err = -EAFNOSUPPORT;
528 if (usin->sin_family)
529 goto out;
530 }
531 daddr = usin->sin_addr.s_addr;
532 /* ANK: I did not forget to get protocol from port field.
533 * I just do not know, who uses this weirdness.
534 * IP_HDRINCL is much more convenient.
535 */
536 } else {
537 err = -EDESTADDRREQ;
538 if (sk->sk_state != TCP_ESTABLISHED)
539 goto out;
540 daddr = inet->inet_daddr;
541 }
542
543 ipc.addr = inet->inet_saddr;
544 ipc.opt = NULL;
545 ipc.tx_flags = 0;
546 ipc.ttl = 0;
547 ipc.tos = -1;
548 ipc.oif = sk->sk_bound_dev_if;
549
550 if (msg->msg_controllen) {
551 err = ip_cmsg_send(net, msg, &ipc, false);
552 if (unlikely(err)) {
553 kfree(ipc.opt);
554 goto out;
555 }
556 if (ipc.opt)
557 free = 1;
558 }
559
560 saddr = ipc.addr;
561 ipc.addr = daddr;
562
563 if (!ipc.opt) {
564 struct ip_options_rcu *inet_opt;
565
566 rcu_read_lock();
567 inet_opt = rcu_dereference(inet->inet_opt);
568 if (inet_opt) {
569 memcpy(&opt_copy, inet_opt,
570 sizeof(*inet_opt) + inet_opt->opt.optlen);
571 ipc.opt = &opt_copy.opt;
572 }
573 rcu_read_unlock();
574 }
575
576 if (ipc.opt) {
577 err = -EINVAL;
578 /* Linux does not mangle headers on raw sockets,
579 * so that IP options + IP_HDRINCL is non-sense.
580 */
581 if (inet->hdrincl)
582 goto done;
583 if (ipc.opt->opt.srr) {
584 if (!daddr)
585 goto done;
586 daddr = ipc.opt->opt.faddr;
587 }
588 }
589 tos = get_rtconn_flags(&ipc, sk);
590 if (msg->msg_flags & MSG_DONTROUTE)
591 tos |= RTO_ONLINK;
592
593 if (ipv4_is_multicast(daddr)) {
594 if (!ipc.oif)
595 ipc.oif = inet->mc_index;
596 if (!saddr)
597 saddr = inet->mc_addr;
598 } else if (!ipc.oif)
599 ipc.oif = inet->uc_index;
600
601 flowi4_init_output(&fl4, ipc.oif, sk->sk_mark, tos,
602 RT_SCOPE_UNIVERSE,
603 inet->hdrincl ? IPPROTO_RAW : sk->sk_protocol,
604 inet_sk_flowi_flags(sk) |
605 (inet->hdrincl ? FLOWI_FLAG_KNOWN_NH : 0),
606 daddr, saddr, 0, 0);
607
608 if (!saddr && ipc.oif) {
609 err = l3mdev_get_saddr(net, ipc.oif, &fl4);
610 if (err < 0)
611 goto done;
612 }
613
614 if (!inet->hdrincl) {
615 rfv.msg = msg;
616 rfv.hlen = 0;
617
618 err = raw_probe_proto_opt(&rfv, &fl4);
619 if (err)
620 goto done;
621 }
622
623 security_sk_classify_flow(sk, flowi4_to_flowi(&fl4));
624 rt = ip_route_output_flow(net, &fl4, sk);
625 if (IS_ERR(rt)) {
626 err = PTR_ERR(rt);
627 rt = NULL;
628 goto done;
629 }
630
631 err = -EACCES;
632 if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
633 goto done;
634
635 if (msg->msg_flags & MSG_CONFIRM)
636 goto do_confirm;
637back_from_confirm:
638
639 if (inet->hdrincl)
640 err = raw_send_hdrinc(sk, &fl4, msg, len,
641 &rt, msg->msg_flags);
642
643 else {
644 sock_tx_timestamp(sk, &ipc.tx_flags);
645
646 if (!ipc.addr)
647 ipc.addr = fl4.daddr;
648 lock_sock(sk);
649 err = ip_append_data(sk, &fl4, raw_getfrag,
650 &rfv, len, 0,
651 &ipc, &rt, msg->msg_flags);
652 if (err)
653 ip_flush_pending_frames(sk);
654 else if (!(msg->msg_flags & MSG_MORE)) {
655 err = ip_push_pending_frames(sk, &fl4);
656 if (err == -ENOBUFS && !inet->recverr)
657 err = 0;
658 }
659 release_sock(sk);
660 }
661done:
662 if (free)
663 kfree(ipc.opt);
664 ip_rt_put(rt);
665
666out:
667 if (err < 0)
668 return err;
669 return len;
670
671do_confirm:
672 dst_confirm(&rt->dst);
673 if (!(msg->msg_flags & MSG_PROBE) || len)
674 goto back_from_confirm;
675 err = 0;
676 goto done;
677}
678
679static void raw_close(struct sock *sk, long timeout)
680{
681 /*
682 * Raw sockets may have direct kernel references. Kill them.
683 */
684 ip_ra_control(sk, 0, NULL);
685
686 sk_common_release(sk);
687}
688
689static void raw_destroy(struct sock *sk)
690{
691 lock_sock(sk);
692 ip_flush_pending_frames(sk);
693 release_sock(sk);
694}
695
696/* This gets rid of all the nasties in af_inet. -DaveM */
697static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
698{
699 struct inet_sock *inet = inet_sk(sk);
700 struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
701 int ret = -EINVAL;
702 int chk_addr_ret;
703
704 if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
705 goto out;
706 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
707 ret = -EADDRNOTAVAIL;
708 if (addr->sin_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
709 chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
710 goto out;
711 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
712 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
713 inet->inet_saddr = 0; /* Use device */
714 sk_dst_reset(sk);
715 ret = 0;
716out: return ret;
717}
718
719/*
720 * This should be easy, if there is something there
721 * we return it, otherwise we block.
722 */
723
724static int raw_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
725 int noblock, int flags, int *addr_len)
726{
727 struct inet_sock *inet = inet_sk(sk);
728 size_t copied = 0;
729 int err = -EOPNOTSUPP;
730 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
731 struct sk_buff *skb;
732
733 if (flags & MSG_OOB)
734 goto out;
735
736 if (flags & MSG_ERRQUEUE) {
737 err = ip_recv_error(sk, msg, len, addr_len);
738 goto out;
739 }
740
741 skb = skb_recv_datagram(sk, flags, noblock, &err);
742 if (!skb)
743 goto out;
744
745 copied = skb->len;
746 if (len < copied) {
747 msg->msg_flags |= MSG_TRUNC;
748 copied = len;
749 }
750
751 err = skb_copy_datagram_msg(skb, 0, msg, copied);
752 if (err)
753 goto done;
754
755 sock_recv_ts_and_drops(msg, sk, skb);
756
757 /* Copy the address. */
758 if (sin) {
759 sin->sin_family = AF_INET;
760 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
761 sin->sin_port = 0;
762 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
763 *addr_len = sizeof(*sin);
764 }
765 if (inet->cmsg_flags)
766 ip_cmsg_recv(msg, skb);
767 if (flags & MSG_TRUNC)
768 copied = skb->len;
769done:
770 skb_free_datagram(sk, skb);
771out:
772 if (err)
773 return err;
774 return copied;
775}
776
777static int raw_init(struct sock *sk)
778{
779 struct raw_sock *rp = raw_sk(sk);
780
781 if (inet_sk(sk)->inet_num == IPPROTO_ICMP)
782 memset(&rp->filter, 0, sizeof(rp->filter));
783 return 0;
784}
785
786static int raw_seticmpfilter(struct sock *sk, char __user *optval, int optlen)
787{
788 if (optlen > sizeof(struct icmp_filter))
789 optlen = sizeof(struct icmp_filter);
790 if (copy_from_user(&raw_sk(sk)->filter, optval, optlen))
791 return -EFAULT;
792 return 0;
793}
794
795static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
796{
797 int len, ret = -EFAULT;
798
799 if (get_user(len, optlen))
800 goto out;
801 ret = -EINVAL;
802 if (len < 0)
803 goto out;
804 if (len > sizeof(struct icmp_filter))
805 len = sizeof(struct icmp_filter);
806 ret = -EFAULT;
807 if (put_user(len, optlen) ||
808 copy_to_user(optval, &raw_sk(sk)->filter, len))
809 goto out;
810 ret = 0;
811out: return ret;
812}
813
814static int do_raw_setsockopt(struct sock *sk, int level, int optname,
815 char __user *optval, unsigned int optlen)
816{
817 if (optname == ICMP_FILTER) {
818 if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
819 return -EOPNOTSUPP;
820 else
821 return raw_seticmpfilter(sk, optval, optlen);
822 }
823 return -ENOPROTOOPT;
824}
825
826static int raw_setsockopt(struct sock *sk, int level, int optname,
827 char __user *optval, unsigned int optlen)
828{
829 if (level != SOL_RAW)
830 return ip_setsockopt(sk, level, optname, optval, optlen);
831 return do_raw_setsockopt(sk, level, optname, optval, optlen);
832}
833
834#ifdef CONFIG_COMPAT
835static int compat_raw_setsockopt(struct sock *sk, int level, int optname,
836 char __user *optval, unsigned int optlen)
837{
838 if (level != SOL_RAW)
839 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
840 return do_raw_setsockopt(sk, level, optname, optval, optlen);
841}
842#endif
843
844static int do_raw_getsockopt(struct sock *sk, int level, int optname,
845 char __user *optval, int __user *optlen)
846{
847 if (optname == ICMP_FILTER) {
848 if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
849 return -EOPNOTSUPP;
850 else
851 return raw_geticmpfilter(sk, optval, optlen);
852 }
853 return -ENOPROTOOPT;
854}
855
856static int raw_getsockopt(struct sock *sk, int level, int optname,
857 char __user *optval, int __user *optlen)
858{
859 if (level != SOL_RAW)
860 return ip_getsockopt(sk, level, optname, optval, optlen);
861 return do_raw_getsockopt(sk, level, optname, optval, optlen);
862}
863
864#ifdef CONFIG_COMPAT
865static int compat_raw_getsockopt(struct sock *sk, int level, int optname,
866 char __user *optval, int __user *optlen)
867{
868 if (level != SOL_RAW)
869 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
870 return do_raw_getsockopt(sk, level, optname, optval, optlen);
871}
872#endif
873
874static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
875{
876 switch (cmd) {
877 case SIOCOUTQ: {
878 int amount = sk_wmem_alloc_get(sk);
879
880 return put_user(amount, (int __user *)arg);
881 }
882 case SIOCINQ: {
883 struct sk_buff *skb;
884 int amount = 0;
885
886 spin_lock_bh(&sk->sk_receive_queue.lock);
887 skb = skb_peek(&sk->sk_receive_queue);
888 if (skb)
889 amount = skb->len;
890 spin_unlock_bh(&sk->sk_receive_queue.lock);
891 return put_user(amount, (int __user *)arg);
892 }
893
894 default:
895#ifdef CONFIG_IP_MROUTE
896 return ipmr_ioctl(sk, cmd, (void __user *)arg);
897#else
898 return -ENOIOCTLCMD;
899#endif
900 }
901}
902
903#ifdef CONFIG_COMPAT
904static int compat_raw_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
905{
906 switch (cmd) {
907 case SIOCOUTQ:
908 case SIOCINQ:
909 return -ENOIOCTLCMD;
910 default:
911#ifdef CONFIG_IP_MROUTE
912 return ipmr_compat_ioctl(sk, cmd, compat_ptr(arg));
913#else
914 return -ENOIOCTLCMD;
915#endif
916 }
917}
918#endif
919
920struct proto raw_prot = {
921 .name = "RAW",
922 .owner = THIS_MODULE,
923 .close = raw_close,
924 .destroy = raw_destroy,
925 .connect = ip4_datagram_connect,
926 .disconnect = udp_disconnect,
927 .ioctl = raw_ioctl,
928 .init = raw_init,
929 .setsockopt = raw_setsockopt,
930 .getsockopt = raw_getsockopt,
931 .sendmsg = raw_sendmsg,
932 .recvmsg = raw_recvmsg,
933 .bind = raw_bind,
934 .backlog_rcv = raw_rcv_skb,
935 .release_cb = ip4_datagram_release_cb,
936 .hash = raw_hash_sk,
937 .unhash = raw_unhash_sk,
938 .obj_size = sizeof(struct raw_sock),
939 .h.raw_hash = &raw_v4_hashinfo,
940#ifdef CONFIG_COMPAT
941 .compat_setsockopt = compat_raw_setsockopt,
942 .compat_getsockopt = compat_raw_getsockopt,
943 .compat_ioctl = compat_raw_ioctl,
944#endif
945};
946
947#ifdef CONFIG_PROC_FS
948static struct sock *raw_get_first(struct seq_file *seq)
949{
950 struct sock *sk;
951 struct raw_iter_state *state = raw_seq_private(seq);
952
953 for (state->bucket = 0; state->bucket < RAW_HTABLE_SIZE;
954 ++state->bucket) {
955 sk_for_each(sk, &state->h->ht[state->bucket])
956 if (sock_net(sk) == seq_file_net(seq))
957 goto found;
958 }
959 sk = NULL;
960found:
961 return sk;
962}
963
964static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
965{
966 struct raw_iter_state *state = raw_seq_private(seq);
967
968 do {
969 sk = sk_next(sk);
970try_again:
971 ;
972 } while (sk && sock_net(sk) != seq_file_net(seq));
973
974 if (!sk && ++state->bucket < RAW_HTABLE_SIZE) {
975 sk = sk_head(&state->h->ht[state->bucket]);
976 goto try_again;
977 }
978 return sk;
979}
980
981static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
982{
983 struct sock *sk = raw_get_first(seq);
984
985 if (sk)
986 while (pos && (sk = raw_get_next(seq, sk)) != NULL)
987 --pos;
988 return pos ? NULL : sk;
989}
990
991void *raw_seq_start(struct seq_file *seq, loff_t *pos)
992{
993 struct raw_iter_state *state = raw_seq_private(seq);
994
995 read_lock(&state->h->lock);
996 return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
997}
998EXPORT_SYMBOL_GPL(raw_seq_start);
999
1000void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1001{
1002 struct sock *sk;
1003
1004 if (v == SEQ_START_TOKEN)
1005 sk = raw_get_first(seq);
1006 else
1007 sk = raw_get_next(seq, v);
1008 ++*pos;
1009 return sk;
1010}
1011EXPORT_SYMBOL_GPL(raw_seq_next);
1012
1013void raw_seq_stop(struct seq_file *seq, void *v)
1014{
1015 struct raw_iter_state *state = raw_seq_private(seq);
1016
1017 read_unlock(&state->h->lock);
1018}
1019EXPORT_SYMBOL_GPL(raw_seq_stop);
1020
1021static void raw_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
1022{
1023 struct inet_sock *inet = inet_sk(sp);
1024 __be32 dest = inet->inet_daddr,
1025 src = inet->inet_rcv_saddr;
1026 __u16 destp = 0,
1027 srcp = inet->inet_num;
1028
1029 seq_printf(seq, "%4d: %08X:%04X %08X:%04X"
1030 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d\n",
1031 i, src, srcp, dest, destp, sp->sk_state,
1032 sk_wmem_alloc_get(sp),
1033 sk_rmem_alloc_get(sp),
1034 0, 0L, 0,
1035 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1036 0, sock_i_ino(sp),
1037 atomic_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
1038}
1039
1040static int raw_seq_show(struct seq_file *seq, void *v)
1041{
1042 if (v == SEQ_START_TOKEN)
1043 seq_printf(seq, " sl local_address rem_address st tx_queue "
1044 "rx_queue tr tm->when retrnsmt uid timeout "
1045 "inode ref pointer drops\n");
1046 else
1047 raw_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
1048 return 0;
1049}
1050
1051static const struct seq_operations raw_seq_ops = {
1052 .start = raw_seq_start,
1053 .next = raw_seq_next,
1054 .stop = raw_seq_stop,
1055 .show = raw_seq_show,
1056};
1057
1058int raw_seq_open(struct inode *ino, struct file *file,
1059 struct raw_hashinfo *h, const struct seq_operations *ops)
1060{
1061 int err;
1062 struct raw_iter_state *i;
1063
1064 err = seq_open_net(ino, file, ops, sizeof(struct raw_iter_state));
1065 if (err < 0)
1066 return err;
1067
1068 i = raw_seq_private((struct seq_file *)file->private_data);
1069 i->h = h;
1070 return 0;
1071}
1072EXPORT_SYMBOL_GPL(raw_seq_open);
1073
1074static int raw_v4_seq_open(struct inode *inode, struct file *file)
1075{
1076 return raw_seq_open(inode, file, &raw_v4_hashinfo, &raw_seq_ops);
1077}
1078
1079static const struct file_operations raw_seq_fops = {
1080 .owner = THIS_MODULE,
1081 .open = raw_v4_seq_open,
1082 .read = seq_read,
1083 .llseek = seq_lseek,
1084 .release = seq_release_net,
1085};
1086
1087static __net_init int raw_init_net(struct net *net)
1088{
1089 if (!proc_create("raw", S_IRUGO, net->proc_net, &raw_seq_fops))
1090 return -ENOMEM;
1091
1092 return 0;
1093}
1094
1095static __net_exit void raw_exit_net(struct net *net)
1096{
1097 remove_proc_entry("raw", net->proc_net);
1098}
1099
1100static __net_initdata struct pernet_operations raw_net_ops = {
1101 .init = raw_init_net,
1102 .exit = raw_exit_net,
1103};
1104
1105int __init raw_proc_init(void)
1106{
1107 return register_pernet_subsys(&raw_net_ops);
1108}
1109
1110void __init raw_proc_exit(void)
1111{
1112 unregister_pernet_subsys(&raw_net_ops);
1113}
1114#endif /* CONFIG_PROC_FS */