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// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * RAW - implementation of IP "raw" sockets.
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 *
12 * Fixes:
13 * Alan Cox : verify_area() fixed up
14 * Alan Cox : ICMP error handling
15 * Alan Cox : EMSGSIZE if you send too big a packet
16 * Alan Cox : Now uses generic datagrams and shared
17 * skbuff library. No more peek crashes,
18 * no more backlogs
19 * Alan Cox : Checks sk->broadcast.
20 * Alan Cox : Uses skb_free_datagram/skb_copy_datagram
21 * Alan Cox : Raw passes ip options too
22 * Alan Cox : Setsocketopt added
23 * Alan Cox : Fixed error return for broadcasts
24 * Alan Cox : Removed wake_up calls
25 * Alan Cox : Use ttl/tos
26 * Alan Cox : Cleaned up old debugging
27 * Alan Cox : Use new kernel side addresses
28 * Arnt Gulbrandsen : Fixed MSG_DONTROUTE in raw sockets.
29 * Alan Cox : BSD style RAW socket demultiplexing.
30 * Alan Cox : Beginnings of mrouted support.
31 * Alan Cox : Added IP_HDRINCL option.
32 * Alan Cox : Skip broadcast check if BSDism set.
33 * David S. Miller : New socket lookup architecture.
34 */
35
36#include <linux/types.h>
37#include <linux/atomic.h>
38#include <asm/byteorder.h>
39#include <asm/current.h>
40#include <linux/uaccess.h>
41#include <asm/ioctls.h>
42#include <linux/stddef.h>
43#include <linux/slab.h>
44#include <linux/errno.h>
45#include <linux/kernel.h>
46#include <linux/export.h>
47#include <linux/spinlock.h>
48#include <linux/sockios.h>
49#include <linux/socket.h>
50#include <linux/in.h>
51#include <linux/mroute.h>
52#include <linux/netdevice.h>
53#include <linux/in_route.h>
54#include <linux/route.h>
55#include <linux/skbuff.h>
56#include <linux/igmp.h>
57#include <net/net_namespace.h>
58#include <net/dst.h>
59#include <net/sock.h>
60#include <linux/ip.h>
61#include <linux/net.h>
62#include <net/ip.h>
63#include <net/icmp.h>
64#include <net/udp.h>
65#include <net/raw.h>
66#include <net/snmp.h>
67#include <net/tcp_states.h>
68#include <net/inet_common.h>
69#include <net/checksum.h>
70#include <net/xfrm.h>
71#include <linux/rtnetlink.h>
72#include <linux/proc_fs.h>
73#include <linux/seq_file.h>
74#include <linux/netfilter.h>
75#include <linux/netfilter_ipv4.h>
76#include <linux/compat.h>
77#include <linux/uio.h>
78
79struct raw_frag_vec {
80 struct msghdr *msg;
81 union {
82 struct icmphdr icmph;
83 char c[1];
84 } hdr;
85 int hlen;
86};
87
88struct raw_hashinfo raw_v4_hashinfo;
89EXPORT_SYMBOL_GPL(raw_v4_hashinfo);
90
91int raw_hash_sk(struct sock *sk)
92{
93 struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
94 struct hlist_nulls_head *hlist;
95
96 hlist = &h->ht[inet_sk(sk)->inet_num & (RAW_HTABLE_SIZE - 1)];
97
98 spin_lock(&h->lock);
99 __sk_nulls_add_node_rcu(sk, hlist);
100 sock_set_flag(sk, SOCK_RCU_FREE);
101 spin_unlock(&h->lock);
102 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
103
104 return 0;
105}
106EXPORT_SYMBOL_GPL(raw_hash_sk);
107
108void raw_unhash_sk(struct sock *sk)
109{
110 struct raw_hashinfo *h = sk->sk_prot->h.raw_hash;
111
112 spin_lock(&h->lock);
113 if (__sk_nulls_del_node_init_rcu(sk))
114 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
115 spin_unlock(&h->lock);
116}
117EXPORT_SYMBOL_GPL(raw_unhash_sk);
118
119bool raw_v4_match(struct net *net, struct sock *sk, unsigned short num,
120 __be32 raddr, __be32 laddr, int dif, int sdif)
121{
122 struct inet_sock *inet = inet_sk(sk);
123
124 if (net_eq(sock_net(sk), net) && inet->inet_num == num &&
125 !(inet->inet_daddr && inet->inet_daddr != raddr) &&
126 !(inet->inet_rcv_saddr && inet->inet_rcv_saddr != laddr) &&
127 raw_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
128 return true;
129 return false;
130}
131EXPORT_SYMBOL_GPL(raw_v4_match);
132
133/*
134 * 0 - deliver
135 * 1 - block
136 */
137static int icmp_filter(const struct sock *sk, const struct sk_buff *skb)
138{
139 struct icmphdr _hdr;
140 const struct icmphdr *hdr;
141
142 hdr = skb_header_pointer(skb, skb_transport_offset(skb),
143 sizeof(_hdr), &_hdr);
144 if (!hdr)
145 return 1;
146
147 if (hdr->type < 32) {
148 __u32 data = raw_sk(sk)->filter.data;
149
150 return ((1U << hdr->type) & data) != 0;
151 }
152
153 /* Do not block unknown ICMP types */
154 return 0;
155}
156
157/* IP input processing comes here for RAW socket delivery.
158 * Caller owns SKB, so we must make clones.
159 *
160 * RFC 1122: SHOULD pass TOS value up to the transport layer.
161 * -> It does. And not only TOS, but all IP header.
162 */
163static int raw_v4_input(struct sk_buff *skb, const struct iphdr *iph, int hash)
164{
165 struct net *net = dev_net(skb->dev);
166 struct hlist_nulls_head *hlist;
167 struct hlist_nulls_node *hnode;
168 int sdif = inet_sdif(skb);
169 int dif = inet_iif(skb);
170 int delivered = 0;
171 struct sock *sk;
172
173 hlist = &raw_v4_hashinfo.ht[hash];
174 rcu_read_lock();
175 sk_nulls_for_each(sk, hnode, hlist) {
176 if (!raw_v4_match(net, sk, iph->protocol,
177 iph->saddr, iph->daddr, dif, sdif))
178 continue;
179 delivered = 1;
180 if ((iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) &&
181 ip_mc_sf_allow(sk, iph->daddr, iph->saddr,
182 skb->dev->ifindex, sdif)) {
183 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
184
185 /* Not releasing hash table! */
186 if (clone)
187 raw_rcv(sk, clone);
188 }
189 }
190 rcu_read_unlock();
191 return delivered;
192}
193
194int raw_local_deliver(struct sk_buff *skb, int protocol)
195{
196 int hash = protocol & (RAW_HTABLE_SIZE - 1);
197
198 return raw_v4_input(skb, ip_hdr(skb), hash);
199}
200
201static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
202{
203 struct inet_sock *inet = inet_sk(sk);
204 const int type = icmp_hdr(skb)->type;
205 const int code = icmp_hdr(skb)->code;
206 int err = 0;
207 int harderr = 0;
208
209 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
210 ipv4_sk_update_pmtu(skb, sk, info);
211 else if (type == ICMP_REDIRECT) {
212 ipv4_sk_redirect(skb, sk);
213 return;
214 }
215
216 /* Report error on raw socket, if:
217 1. User requested ip_recverr.
218 2. Socket is connected (otherwise the error indication
219 is useless without ip_recverr and error is hard.
220 */
221 if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
222 return;
223
224 switch (type) {
225 default:
226 case ICMP_TIME_EXCEEDED:
227 err = EHOSTUNREACH;
228 break;
229 case ICMP_SOURCE_QUENCH:
230 return;
231 case ICMP_PARAMETERPROB:
232 err = EPROTO;
233 harderr = 1;
234 break;
235 case ICMP_DEST_UNREACH:
236 err = EHOSTUNREACH;
237 if (code > NR_ICMP_UNREACH)
238 break;
239 if (code == ICMP_FRAG_NEEDED) {
240 harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
241 err = EMSGSIZE;
242 } else {
243 err = icmp_err_convert[code].errno;
244 harderr = icmp_err_convert[code].fatal;
245 }
246 }
247
248 if (inet->recverr) {
249 const struct iphdr *iph = (const struct iphdr *)skb->data;
250 u8 *payload = skb->data + (iph->ihl << 2);
251
252 if (inet->hdrincl)
253 payload = skb->data;
254 ip_icmp_error(sk, skb, err, 0, info, payload);
255 }
256
257 if (inet->recverr || harderr) {
258 sk->sk_err = err;
259 sk_error_report(sk);
260 }
261}
262
263void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
264{
265 struct net *net = dev_net(skb->dev);
266 struct hlist_nulls_head *hlist;
267 struct hlist_nulls_node *hnode;
268 int dif = skb->dev->ifindex;
269 int sdif = inet_sdif(skb);
270 const struct iphdr *iph;
271 struct sock *sk;
272 int hash;
273
274 hash = protocol & (RAW_HTABLE_SIZE - 1);
275 hlist = &raw_v4_hashinfo.ht[hash];
276
277 rcu_read_lock();
278 sk_nulls_for_each(sk, hnode, hlist) {
279 iph = (const struct iphdr *)skb->data;
280 if (!raw_v4_match(net, sk, iph->protocol,
281 iph->daddr, iph->saddr, dif, sdif))
282 continue;
283 raw_err(sk, skb, info);
284 }
285 rcu_read_unlock();
286}
287
288static int raw_rcv_skb(struct sock *sk, struct sk_buff *skb)
289{
290 /* Charge it to the socket. */
291
292 ipv4_pktinfo_prepare(sk, skb);
293 if (sock_queue_rcv_skb(sk, skb) < 0) {
294 kfree_skb(skb);
295 return NET_RX_DROP;
296 }
297
298 return NET_RX_SUCCESS;
299}
300
301int raw_rcv(struct sock *sk, struct sk_buff *skb)
302{
303 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
304 atomic_inc(&sk->sk_drops);
305 kfree_skb(skb);
306 return NET_RX_DROP;
307 }
308 nf_reset_ct(skb);
309
310 skb_push(skb, skb->data - skb_network_header(skb));
311
312 raw_rcv_skb(sk, skb);
313 return 0;
314}
315
316static int raw_send_hdrinc(struct sock *sk, struct flowi4 *fl4,
317 struct msghdr *msg, size_t length,
318 struct rtable **rtp, unsigned int flags,
319 const struct sockcm_cookie *sockc)
320{
321 struct inet_sock *inet = inet_sk(sk);
322 struct net *net = sock_net(sk);
323 struct iphdr *iph;
324 struct sk_buff *skb;
325 unsigned int iphlen;
326 int err;
327 struct rtable *rt = *rtp;
328 int hlen, tlen;
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 (length < sizeof(struct iphdr))
336 return -EINVAL;
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)
347 goto error;
348 skb_reserve(skb, hlen);
349
350 skb->priority = sk->sk_priority;
351 skb->mark = sockc->mark;
352 skb->tstamp = sockc->transmit_time;
353 skb_dst_set(skb, &rt->dst);
354 *rtp = NULL;
355
356 skb_reset_network_header(skb);
357 iph = ip_hdr(skb);
358 skb_put(skb, length);
359
360 skb->ip_summed = CHECKSUM_NONE;
361
362 skb_setup_tx_timestamp(skb, sockc->tsflags);
363
364 if (flags & MSG_CONFIRM)
365 skb_set_dst_pending_confirm(skb, 1);
366
367 skb->transport_header = skb->network_header;
368 err = -EFAULT;
369 if (memcpy_from_msg(iph, msg, length))
370 goto error_free;
371
372 iphlen = iph->ihl * 4;
373
374 /*
375 * We don't want to modify the ip header, but we do need to
376 * be sure that it won't cause problems later along the network
377 * stack. Specifically we want to make sure that iph->ihl is a
378 * sane value. If ihl points beyond the length of the buffer passed
379 * in, reject the frame as invalid
380 */
381 err = -EINVAL;
382 if (iphlen > length)
383 goto error_free;
384
385 if (iphlen >= sizeof(*iph)) {
386 if (!iph->saddr)
387 iph->saddr = fl4->saddr;
388 iph->check = 0;
389 iph->tot_len = htons(length);
390 if (!iph->id)
391 ip_select_ident(net, skb, NULL);
392
393 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
394 skb->transport_header += iphlen;
395 if (iph->protocol == IPPROTO_ICMP &&
396 length >= iphlen + sizeof(struct icmphdr))
397 icmp_out_count(net, ((struct icmphdr *)
398 skb_transport_header(skb))->type);
399 }
400
401 err = NF_HOOK(NFPROTO_IPV4, NF_INET_LOCAL_OUT,
402 net, sk, skb, NULL, rt->dst.dev,
403 dst_output);
404 if (err > 0)
405 err = net_xmit_errno(err);
406 if (err)
407 goto error;
408out:
409 return 0;
410
411error_free:
412 kfree_skb(skb);
413error:
414 IP_INC_STATS(net, IPSTATS_MIB_OUTDISCARDS);
415 if (err == -ENOBUFS && !inet->recverr)
416 err = 0;
417 return err;
418}
419
420static int raw_probe_proto_opt(struct raw_frag_vec *rfv, struct flowi4 *fl4)
421{
422 int err;
423
424 if (fl4->flowi4_proto != IPPROTO_ICMP)
425 return 0;
426
427 /* We only need the first two bytes. */
428 rfv->hlen = 2;
429
430 err = memcpy_from_msg(rfv->hdr.c, rfv->msg, rfv->hlen);
431 if (err)
432 return err;
433
434 fl4->fl4_icmp_type = rfv->hdr.icmph.type;
435 fl4->fl4_icmp_code = rfv->hdr.icmph.code;
436
437 return 0;
438}
439
440static int raw_getfrag(void *from, char *to, int offset, int len, int odd,
441 struct sk_buff *skb)
442{
443 struct raw_frag_vec *rfv = from;
444
445 if (offset < rfv->hlen) {
446 int copy = min(rfv->hlen - offset, len);
447
448 if (skb->ip_summed == CHECKSUM_PARTIAL)
449 memcpy(to, rfv->hdr.c + offset, copy);
450 else
451 skb->csum = csum_block_add(
452 skb->csum,
453 csum_partial_copy_nocheck(rfv->hdr.c + offset,
454 to, copy),
455 odd);
456
457 odd = 0;
458 offset += copy;
459 to += copy;
460 len -= copy;
461
462 if (!len)
463 return 0;
464 }
465
466 offset -= rfv->hlen;
467
468 return ip_generic_getfrag(rfv->msg, to, offset, len, odd, skb);
469}
470
471static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
472{
473 struct inet_sock *inet = inet_sk(sk);
474 struct net *net = sock_net(sk);
475 struct ipcm_cookie ipc;
476 struct rtable *rt = NULL;
477 struct flowi4 fl4;
478 int free = 0;
479 __be32 daddr;
480 __be32 saddr;
481 u8 tos;
482 int err;
483 struct ip_options_data opt_copy;
484 struct raw_frag_vec rfv;
485 int hdrincl;
486
487 err = -EMSGSIZE;
488 if (len > 0xFFFF)
489 goto out;
490
491 /* hdrincl should be READ_ONCE(inet->hdrincl)
492 * but READ_ONCE() doesn't work with bit fields.
493 * Doing this indirectly yields the same result.
494 */
495 hdrincl = inet->hdrincl;
496 hdrincl = READ_ONCE(hdrincl);
497 /*
498 * Check the flags.
499 */
500
501 err = -EOPNOTSUPP;
502 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message */
503 goto out; /* compatibility */
504
505 /*
506 * Get and verify the address.
507 */
508
509 if (msg->msg_namelen) {
510 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
511 err = -EINVAL;
512 if (msg->msg_namelen < sizeof(*usin))
513 goto out;
514 if (usin->sin_family != AF_INET) {
515 pr_info_once("%s: %s forgot to set AF_INET. Fix it!\n",
516 __func__, current->comm);
517 err = -EAFNOSUPPORT;
518 if (usin->sin_family)
519 goto out;
520 }
521 daddr = usin->sin_addr.s_addr;
522 /* ANK: I did not forget to get protocol from port field.
523 * I just do not know, who uses this weirdness.
524 * IP_HDRINCL is much more convenient.
525 */
526 } else {
527 err = -EDESTADDRREQ;
528 if (sk->sk_state != TCP_ESTABLISHED)
529 goto out;
530 daddr = inet->inet_daddr;
531 }
532
533 ipcm_init_sk(&ipc, inet);
534
535 if (msg->msg_controllen) {
536 err = ip_cmsg_send(sk, msg, &ipc, false);
537 if (unlikely(err)) {
538 kfree(ipc.opt);
539 goto out;
540 }
541 if (ipc.opt)
542 free = 1;
543 }
544
545 saddr = ipc.addr;
546 ipc.addr = daddr;
547
548 if (!ipc.opt) {
549 struct ip_options_rcu *inet_opt;
550
551 rcu_read_lock();
552 inet_opt = rcu_dereference(inet->inet_opt);
553 if (inet_opt) {
554 memcpy(&opt_copy, inet_opt,
555 sizeof(*inet_opt) + inet_opt->opt.optlen);
556 ipc.opt = &opt_copy.opt;
557 }
558 rcu_read_unlock();
559 }
560
561 if (ipc.opt) {
562 err = -EINVAL;
563 /* Linux does not mangle headers on raw sockets,
564 * so that IP options + IP_HDRINCL is non-sense.
565 */
566 if (hdrincl)
567 goto done;
568 if (ipc.opt->opt.srr) {
569 if (!daddr)
570 goto done;
571 daddr = ipc.opt->opt.faddr;
572 }
573 }
574 tos = get_rtconn_flags(&ipc, sk);
575 if (msg->msg_flags & MSG_DONTROUTE)
576 tos |= RTO_ONLINK;
577
578 if (ipv4_is_multicast(daddr)) {
579 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
580 ipc.oif = inet->mc_index;
581 if (!saddr)
582 saddr = inet->mc_addr;
583 } else if (!ipc.oif) {
584 ipc.oif = inet->uc_index;
585 } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
586 /* oif is set, packet is to local broadcast
587 * and uc_index is set. oif is most likely set
588 * by sk_bound_dev_if. If uc_index != oif check if the
589 * oif is an L3 master and uc_index is an L3 slave.
590 * If so, we want to allow the send using the uc_index.
591 */
592 if (ipc.oif != inet->uc_index &&
593 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
594 inet->uc_index)) {
595 ipc.oif = inet->uc_index;
596 }
597 }
598
599 flowi4_init_output(&fl4, ipc.oif, ipc.sockc.mark, tos,
600 RT_SCOPE_UNIVERSE,
601 hdrincl ? IPPROTO_RAW : sk->sk_protocol,
602 inet_sk_flowi_flags(sk) |
603 (hdrincl ? FLOWI_FLAG_KNOWN_NH : 0),
604 daddr, saddr, 0, 0, sk->sk_uid);
605
606 if (!hdrincl) {
607 rfv.msg = msg;
608 rfv.hlen = 0;
609
610 err = raw_probe_proto_opt(&rfv, &fl4);
611 if (err)
612 goto done;
613 }
614
615 security_sk_classify_flow(sk, flowi4_to_flowi_common(&fl4));
616 rt = ip_route_output_flow(net, &fl4, sk);
617 if (IS_ERR(rt)) {
618 err = PTR_ERR(rt);
619 rt = NULL;
620 goto done;
621 }
622
623 err = -EACCES;
624 if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
625 goto done;
626
627 if (msg->msg_flags & MSG_CONFIRM)
628 goto do_confirm;
629back_from_confirm:
630
631 if (hdrincl)
632 err = raw_send_hdrinc(sk, &fl4, msg, len,
633 &rt, msg->msg_flags, &ipc.sockc);
634
635 else {
636 if (!ipc.addr)
637 ipc.addr = fl4.daddr;
638 lock_sock(sk);
639 err = ip_append_data(sk, &fl4, raw_getfrag,
640 &rfv, len, 0,
641 &ipc, &rt, msg->msg_flags);
642 if (err)
643 ip_flush_pending_frames(sk);
644 else if (!(msg->msg_flags & MSG_MORE)) {
645 err = ip_push_pending_frames(sk, &fl4);
646 if (err == -ENOBUFS && !inet->recverr)
647 err = 0;
648 }
649 release_sock(sk);
650 }
651done:
652 if (free)
653 kfree(ipc.opt);
654 ip_rt_put(rt);
655
656out:
657 if (err < 0)
658 return err;
659 return len;
660
661do_confirm:
662 if (msg->msg_flags & MSG_PROBE)
663 dst_confirm_neigh(&rt->dst, &fl4.daddr);
664 if (!(msg->msg_flags & MSG_PROBE) || len)
665 goto back_from_confirm;
666 err = 0;
667 goto done;
668}
669
670static void raw_close(struct sock *sk, long timeout)
671{
672 /*
673 * Raw sockets may have direct kernel references. Kill them.
674 */
675 ip_ra_control(sk, 0, NULL);
676
677 sk_common_release(sk);
678}
679
680static void raw_destroy(struct sock *sk)
681{
682 lock_sock(sk);
683 ip_flush_pending_frames(sk);
684 release_sock(sk);
685}
686
687/* This gets rid of all the nasties in af_inet. -DaveM */
688static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
689{
690 struct inet_sock *inet = inet_sk(sk);
691 struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
692 struct net *net = sock_net(sk);
693 u32 tb_id = RT_TABLE_LOCAL;
694 int ret = -EINVAL;
695 int chk_addr_ret;
696
697 lock_sock(sk);
698 if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
699 goto out;
700
701 if (sk->sk_bound_dev_if)
702 tb_id = l3mdev_fib_table_by_index(net,
703 sk->sk_bound_dev_if) ? : tb_id;
704
705 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
706
707 ret = -EADDRNOTAVAIL;
708 if (!inet_addr_valid_or_nonlocal(net, inet, addr->sin_addr.s_addr,
709 chk_addr_ret))
710 goto out;
711
712 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
713 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
714 inet->inet_saddr = 0; /* Use device */
715 sk_dst_reset(sk);
716 ret = 0;
717out:
718 release_sock(sk);
719 return ret;
720}
721
722/*
723 * This should be easy, if there is something there
724 * we return it, otherwise we block.
725 */
726
727static int raw_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
728 int flags, int *addr_len)
729{
730 struct inet_sock *inet = inet_sk(sk);
731 size_t copied = 0;
732 int err = -EOPNOTSUPP;
733 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
734 struct sk_buff *skb;
735
736 if (flags & MSG_OOB)
737 goto out;
738
739 if (flags & MSG_ERRQUEUE) {
740 err = ip_recv_error(sk, msg, len, addr_len);
741 goto out;
742 }
743
744 skb = skb_recv_datagram(sk, flags, &err);
745 if (!skb)
746 goto out;
747
748 copied = skb->len;
749 if (len < copied) {
750 msg->msg_flags |= MSG_TRUNC;
751 copied = len;
752 }
753
754 err = skb_copy_datagram_msg(skb, 0, msg, copied);
755 if (err)
756 goto done;
757
758 sock_recv_cmsgs(msg, sk, skb);
759
760 /* Copy the address. */
761 if (sin) {
762 sin->sin_family = AF_INET;
763 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
764 sin->sin_port = 0;
765 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
766 *addr_len = sizeof(*sin);
767 }
768 if (inet->cmsg_flags)
769 ip_cmsg_recv(msg, skb);
770 if (flags & MSG_TRUNC)
771 copied = skb->len;
772done:
773 skb_free_datagram(sk, skb);
774out:
775 if (err)
776 return err;
777 return copied;
778}
779
780static int raw_sk_init(struct sock *sk)
781{
782 struct raw_sock *rp = raw_sk(sk);
783
784 if (inet_sk(sk)->inet_num == IPPROTO_ICMP)
785 memset(&rp->filter, 0, sizeof(rp->filter));
786 return 0;
787}
788
789static int raw_seticmpfilter(struct sock *sk, sockptr_t optval, int optlen)
790{
791 if (optlen > sizeof(struct icmp_filter))
792 optlen = sizeof(struct icmp_filter);
793 if (copy_from_sockptr(&raw_sk(sk)->filter, optval, optlen))
794 return -EFAULT;
795 return 0;
796}
797
798static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
799{
800 int len, ret = -EFAULT;
801
802 if (get_user(len, optlen))
803 goto out;
804 ret = -EINVAL;
805 if (len < 0)
806 goto out;
807 if (len > sizeof(struct icmp_filter))
808 len = sizeof(struct icmp_filter);
809 ret = -EFAULT;
810 if (put_user(len, optlen) ||
811 copy_to_user(optval, &raw_sk(sk)->filter, len))
812 goto out;
813 ret = 0;
814out: return ret;
815}
816
817static int do_raw_setsockopt(struct sock *sk, int level, int optname,
818 sockptr_t optval, unsigned int optlen)
819{
820 if (optname == ICMP_FILTER) {
821 if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
822 return -EOPNOTSUPP;
823 else
824 return raw_seticmpfilter(sk, optval, optlen);
825 }
826 return -ENOPROTOOPT;
827}
828
829static int raw_setsockopt(struct sock *sk, int level, int optname,
830 sockptr_t optval, unsigned int optlen)
831{
832 if (level != SOL_RAW)
833 return ip_setsockopt(sk, level, optname, optval, optlen);
834 return do_raw_setsockopt(sk, level, optname, optval, optlen);
835}
836
837static int do_raw_getsockopt(struct sock *sk, int level, int optname,
838 char __user *optval, int __user *optlen)
839{
840 if (optname == ICMP_FILTER) {
841 if (inet_sk(sk)->inet_num != IPPROTO_ICMP)
842 return -EOPNOTSUPP;
843 else
844 return raw_geticmpfilter(sk, optval, optlen);
845 }
846 return -ENOPROTOOPT;
847}
848
849static int raw_getsockopt(struct sock *sk, int level, int optname,
850 char __user *optval, int __user *optlen)
851{
852 if (level != SOL_RAW)
853 return ip_getsockopt(sk, level, optname, optval, optlen);
854 return do_raw_getsockopt(sk, level, optname, optval, optlen);
855}
856
857static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
858{
859 switch (cmd) {
860 case SIOCOUTQ: {
861 int amount = sk_wmem_alloc_get(sk);
862
863 return put_user(amount, (int __user *)arg);
864 }
865 case SIOCINQ: {
866 struct sk_buff *skb;
867 int amount = 0;
868
869 spin_lock_bh(&sk->sk_receive_queue.lock);
870 skb = skb_peek(&sk->sk_receive_queue);
871 if (skb)
872 amount = skb->len;
873 spin_unlock_bh(&sk->sk_receive_queue.lock);
874 return put_user(amount, (int __user *)arg);
875 }
876
877 default:
878#ifdef CONFIG_IP_MROUTE
879 return ipmr_ioctl(sk, cmd, (void __user *)arg);
880#else
881 return -ENOIOCTLCMD;
882#endif
883 }
884}
885
886#ifdef CONFIG_COMPAT
887static int compat_raw_ioctl(struct sock *sk, unsigned int cmd, unsigned long arg)
888{
889 switch (cmd) {
890 case SIOCOUTQ:
891 case SIOCINQ:
892 return -ENOIOCTLCMD;
893 default:
894#ifdef CONFIG_IP_MROUTE
895 return ipmr_compat_ioctl(sk, cmd, compat_ptr(arg));
896#else
897 return -ENOIOCTLCMD;
898#endif
899 }
900}
901#endif
902
903int raw_abort(struct sock *sk, int err)
904{
905 lock_sock(sk);
906
907 sk->sk_err = err;
908 sk_error_report(sk);
909 __udp_disconnect(sk, 0);
910
911 release_sock(sk);
912
913 return 0;
914}
915EXPORT_SYMBOL_GPL(raw_abort);
916
917struct proto raw_prot = {
918 .name = "RAW",
919 .owner = THIS_MODULE,
920 .close = raw_close,
921 .destroy = raw_destroy,
922 .connect = ip4_datagram_connect,
923 .disconnect = __udp_disconnect,
924 .ioctl = raw_ioctl,
925 .init = raw_sk_init,
926 .setsockopt = raw_setsockopt,
927 .getsockopt = raw_getsockopt,
928 .sendmsg = raw_sendmsg,
929 .recvmsg = raw_recvmsg,
930 .bind = raw_bind,
931 .backlog_rcv = raw_rcv_skb,
932 .release_cb = ip4_datagram_release_cb,
933 .hash = raw_hash_sk,
934 .unhash = raw_unhash_sk,
935 .obj_size = sizeof(struct raw_sock),
936 .useroffset = offsetof(struct raw_sock, filter),
937 .usersize = sizeof_field(struct raw_sock, filter),
938 .h.raw_hash = &raw_v4_hashinfo,
939#ifdef CONFIG_COMPAT
940 .compat_ioctl = compat_raw_ioctl,
941#endif
942 .diag_destroy = raw_abort,
943};
944
945#ifdef CONFIG_PROC_FS
946static struct sock *raw_get_first(struct seq_file *seq, int bucket)
947{
948 struct raw_hashinfo *h = pde_data(file_inode(seq->file));
949 struct raw_iter_state *state = raw_seq_private(seq);
950 struct hlist_nulls_head *hlist;
951 struct hlist_nulls_node *hnode;
952 struct sock *sk;
953
954 for (state->bucket = bucket; state->bucket < RAW_HTABLE_SIZE;
955 ++state->bucket) {
956 hlist = &h->ht[state->bucket];
957 sk_nulls_for_each(sk, hnode, hlist) {
958 if (sock_net(sk) == seq_file_net(seq))
959 return sk;
960 }
961 }
962 return NULL;
963}
964
965static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
966{
967 struct raw_iter_state *state = raw_seq_private(seq);
968
969 do {
970 sk = sk_nulls_next(sk);
971 } while (sk && sock_net(sk) != seq_file_net(seq));
972
973 if (!sk)
974 return raw_get_first(seq, state->bucket + 1);
975 return sk;
976}
977
978static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
979{
980 struct sock *sk = raw_get_first(seq, 0);
981
982 if (sk)
983 while (pos && (sk = raw_get_next(seq, sk)) != NULL)
984 --pos;
985 return pos ? NULL : sk;
986}
987
988void *raw_seq_start(struct seq_file *seq, loff_t *pos)
989 __acquires(RCU)
990{
991 rcu_read_lock();
992 return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
993}
994EXPORT_SYMBOL_GPL(raw_seq_start);
995
996void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
997{
998 struct sock *sk;
999
1000 if (v == SEQ_START_TOKEN)
1001 sk = raw_get_first(seq, 0);
1002 else
1003 sk = raw_get_next(seq, v);
1004 ++*pos;
1005 return sk;
1006}
1007EXPORT_SYMBOL_GPL(raw_seq_next);
1008
1009void raw_seq_stop(struct seq_file *seq, void *v)
1010 __releases(RCU)
1011{
1012 rcu_read_unlock();
1013}
1014EXPORT_SYMBOL_GPL(raw_seq_stop);
1015
1016static void raw_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)
1017{
1018 struct inet_sock *inet = inet_sk(sp);
1019 __be32 dest = inet->inet_daddr,
1020 src = inet->inet_rcv_saddr;
1021 __u16 destp = 0,
1022 srcp = inet->inet_num;
1023
1024 seq_printf(seq, "%4d: %08X:%04X %08X:%04X"
1025 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u\n",
1026 i, src, srcp, dest, destp, sp->sk_state,
1027 sk_wmem_alloc_get(sp),
1028 sk_rmem_alloc_get(sp),
1029 0, 0L, 0,
1030 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
1031 0, sock_i_ino(sp),
1032 refcount_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
1033}
1034
1035static int raw_seq_show(struct seq_file *seq, void *v)
1036{
1037 if (v == SEQ_START_TOKEN)
1038 seq_printf(seq, " sl local_address rem_address st tx_queue "
1039 "rx_queue tr tm->when retrnsmt uid timeout "
1040 "inode ref pointer drops\n");
1041 else
1042 raw_sock_seq_show(seq, v, raw_seq_private(seq)->bucket);
1043 return 0;
1044}
1045
1046static const struct seq_operations raw_seq_ops = {
1047 .start = raw_seq_start,
1048 .next = raw_seq_next,
1049 .stop = raw_seq_stop,
1050 .show = raw_seq_show,
1051};
1052
1053static __net_init int raw_init_net(struct net *net)
1054{
1055 if (!proc_create_net_data("raw", 0444, net->proc_net, &raw_seq_ops,
1056 sizeof(struct raw_iter_state), &raw_v4_hashinfo))
1057 return -ENOMEM;
1058
1059 return 0;
1060}
1061
1062static __net_exit void raw_exit_net(struct net *net)
1063{
1064 remove_proc_entry("raw", net->proc_net);
1065}
1066
1067static __net_initdata struct pernet_operations raw_net_ops = {
1068 .init = raw_init_net,
1069 .exit = raw_exit_net,
1070};
1071
1072int __init raw_proc_init(void)
1073{
1074
1075 return register_pernet_subsys(&raw_net_ops);
1076}
1077
1078void __init raw_proc_exit(void)
1079{
1080 unregister_pernet_subsys(&raw_net_ops);
1081}
1082#endif /* CONFIG_PROC_FS */
1083
1084static void raw_sysctl_init_net(struct net *net)
1085{
1086#ifdef CONFIG_NET_L3_MASTER_DEV
1087 net->ipv4.sysctl_raw_l3mdev_accept = 1;
1088#endif
1089}
1090
1091static int __net_init raw_sysctl_init(struct net *net)
1092{
1093 raw_sysctl_init_net(net);
1094 return 0;
1095}
1096
1097static struct pernet_operations __net_initdata raw_sysctl_ops = {
1098 .init = raw_sysctl_init,
1099};
1100
1101void __init raw_init(void)
1102{
1103 raw_sysctl_init_net(&init_net);
1104 if (register_pernet_subsys(&raw_sysctl_ops))
1105 panic("RAW: failed to init sysctl parameters.\n");
1106}