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.
6 * RAW - implementation of IP "raw" sockets.
8 * Version: $Id: raw.c,v 1.64 2002/02/01 22:01:04 davem Exp $
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Alan Cox : verify_area() fixed up
15 * Alan Cox : ICMP error handling
16 * Alan Cox : EMSGSIZE if you send too big a packet
17 * Alan Cox : Now uses generic datagrams and shared
18 * skbuff library. No more peek crashes,
20 * Alan Cox : Checks sk->broadcast.
21 * Alan Cox : Uses skb_free_datagram/skb_copy_datagram
22 * Alan Cox : Raw passes ip options too
23 * Alan Cox : Setsocketopt added
24 * Alan Cox : Fixed error return for broadcasts
25 * Alan Cox : Removed wake_up calls
26 * Alan Cox : Use ttl/tos
27 * Alan Cox : Cleaned up old debugging
28 * Alan Cox : Use new kernel side addresses
29 * Arnt Gulbrandsen : Fixed MSG_DONTROUTE in raw sockets.
30 * Alan Cox : BSD style RAW socket demultiplexing.
31 * Alan Cox : Beginnings of mrouted support.
32 * Alan Cox : Added IP_HDRINCL option.
33 * Alan Cox : Skip broadcast check if BSDism set.
34 * David S. Miller : New socket lookup architecture.
36 * This program is free software; you can redistribute it and/or
37 * modify it under the terms of the GNU General Public License
38 * as published by the Free Software Foundation; either version
39 * 2 of the License, or (at your option) any later version.
42 #include <linux/types.h>
43 #include <asm/atomic.h>
44 #include <asm/byteorder.h>
45 #include <asm/current.h>
46 #include <asm/uaccess.h>
47 #include <asm/ioctls.h>
48 #include <linux/stddef.h>
49 #include <linux/slab.h>
50 #include <linux/errno.h>
51 #include <linux/aio.h>
52 #include <linux/kernel.h>
53 #include <linux/spinlock.h>
54 #include <linux/sockios.h>
55 #include <linux/socket.h>
57 #include <linux/mroute.h>
58 #include <linux/netdevice.h>
59 #include <linux/in_route.h>
60 #include <linux/route.h>
61 #include <linux/skbuff.h>
62 #include <net/net_namespace.h>
65 #include <linux/gfp.h>
67 #include <linux/net.h>
73 #include <net/tcp_states.h>
74 #include <net/inet_common.h>
75 #include <net/checksum.h>
77 #include <linux/rtnetlink.h>
78 #include <linux/proc_fs.h>
79 #include <linux/seq_file.h>
80 #include <linux/netfilter.h>
81 #include <linux/netfilter_ipv4.h>
83 static struct raw_hashinfo raw_v4_hashinfo = {
84 .lock = __RW_LOCK_UNLOCKED(),
87 static void raw_v4_hash(struct sock *sk)
89 struct hlist_head *head = &raw_v4_hashinfo.ht[inet_sk(sk)->num &
90 (RAW_HTABLE_SIZE - 1)];
92 write_lock_bh(&raw_v4_hashinfo.lock);
93 sk_add_node(sk, head);
94 sock_prot_inc_use(sk->sk_prot);
95 write_unlock_bh(&raw_v4_hashinfo.lock);
98 static void raw_v4_unhash(struct sock *sk)
100 write_lock_bh(&raw_v4_hashinfo.lock);
101 if (sk_del_node_init(sk))
102 sock_prot_dec_use(sk->sk_prot);
103 write_unlock_bh(&raw_v4_hashinfo.lock);
106 static struct sock *__raw_v4_lookup(struct sock *sk, unsigned short num,
107 __be32 raddr, __be32 laddr,
110 struct hlist_node *node;
112 sk_for_each_from(sk, node) {
113 struct inet_sock *inet = inet_sk(sk);
115 if (inet->num == num &&
116 !(inet->daddr && inet->daddr != raddr) &&
117 !(inet->rcv_saddr && inet->rcv_saddr != laddr) &&
118 !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
119 goto found; /* gotcha */
130 static __inline__ int icmp_filter(struct sock *sk, struct sk_buff *skb)
134 if (!pskb_may_pull(skb, sizeof(struct icmphdr)))
137 type = icmp_hdr(skb)->type;
139 __u32 data = raw_sk(sk)->filter.data;
141 return ((1 << type) & data) != 0;
144 /* Do not block unknown ICMP types */
148 /* IP input processing comes here for RAW socket delivery.
149 * Caller owns SKB, so we must make clones.
151 * RFC 1122: SHOULD pass TOS value up to the transport layer.
152 * -> It does. And not only TOS, but all IP header.
154 static int raw_v4_input(struct sk_buff *skb, struct iphdr *iph, int hash)
157 struct hlist_head *head;
160 read_lock(&raw_v4_hashinfo.lock);
161 head = &raw_v4_hashinfo.ht[hash];
162 if (hlist_empty(head))
164 sk = __raw_v4_lookup(__sk_head(head), iph->protocol,
165 iph->saddr, iph->daddr,
170 if (iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) {
171 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
173 /* Not releasing hash table! */
177 sk = __raw_v4_lookup(sk_next(sk), iph->protocol,
178 iph->saddr, iph->daddr,
182 read_unlock(&raw_v4_hashinfo.lock);
186 int raw_local_deliver(struct sk_buff *skb, int protocol)
191 hash = protocol & (RAW_HTABLE_SIZE - 1);
192 raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
194 /* If there maybe a raw socket we must check - if not we
197 if (raw_sk && !raw_v4_input(skb, ip_hdr(skb), hash))
200 return raw_sk != NULL;
204 static void raw_err(struct sock *sk, struct sk_buff *skb, u32 info)
206 struct inet_sock *inet = inet_sk(sk);
207 const int type = icmp_hdr(skb)->type;
208 const int code = icmp_hdr(skb)->code;
212 /* Report error on raw socket, if:
213 1. User requested ip_recverr.
214 2. Socket is connected (otherwise the error indication
215 is useless without ip_recverr and error is hard.
217 if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
222 case ICMP_TIME_EXCEEDED:
225 case ICMP_SOURCE_QUENCH:
227 case ICMP_PARAMETERPROB:
231 case ICMP_DEST_UNREACH:
233 if (code > NR_ICMP_UNREACH)
235 err = icmp_err_convert[code].errno;
236 harderr = icmp_err_convert[code].fatal;
237 if (code == ICMP_FRAG_NEEDED) {
238 harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
244 struct iphdr *iph = (struct iphdr*)skb->data;
245 u8 *payload = skb->data + (iph->ihl << 2);
249 ip_icmp_error(sk, skb, err, 0, info, payload);
252 if (inet->recverr || harderr) {
254 sk->sk_error_report(sk);
258 void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
264 hash = protocol & (RAW_HTABLE_SIZE - 1);
266 read_lock(&raw_v4_hashinfo.lock);
267 raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
268 if (raw_sk != NULL) {
269 iph = (struct iphdr *)skb->data;
270 while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr,
272 skb->dev->ifindex)) != NULL) {
273 raw_err(raw_sk, skb, info);
274 raw_sk = sk_next(raw_sk);
275 iph = (struct iphdr *)skb->data;
278 read_unlock(&raw_v4_hashinfo.lock);
281 static int raw_rcv_skb(struct sock * sk, struct sk_buff * skb)
283 /* Charge it to the socket. */
285 if (sock_queue_rcv_skb(sk, skb) < 0) {
286 atomic_inc(&sk->sk_drops);
291 return NET_RX_SUCCESS;
294 int raw_rcv(struct sock *sk, struct sk_buff *skb)
296 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
297 atomic_inc(&sk->sk_drops);
303 skb_push(skb, skb->data - skb_network_header(skb));
305 raw_rcv_skb(sk, skb);
309 static int raw_send_hdrinc(struct sock *sk, void *from, size_t length,
313 struct inet_sock *inet = inet_sk(sk);
320 if (length > rt->u.dst.dev->mtu) {
321 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport,
328 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
330 skb = sock_alloc_send_skb(sk, length+hh_len+15,
331 flags&MSG_DONTWAIT, &err);
334 skb_reserve(skb, hh_len);
336 skb->priority = sk->sk_priority;
337 skb->dst = dst_clone(&rt->u.dst);
339 skb_reset_network_header(skb);
341 skb_put(skb, length);
343 skb->ip_summed = CHECKSUM_NONE;
345 skb->transport_header = skb->network_header;
346 err = memcpy_fromiovecend((void *)iph, from, 0, length);
350 /* We don't modify invalid header */
351 iphlen = iph->ihl * 4;
352 if (iphlen >= sizeof(*iph) && iphlen <= length) {
354 iph->saddr = rt->rt_src;
356 iph->tot_len = htons(length);
358 ip_select_ident(iph, &rt->u.dst, NULL);
360 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
362 if (iph->protocol == IPPROTO_ICMP)
363 icmp_out_count(((struct icmphdr *)
364 skb_transport_header(skb))->type);
366 err = NF_HOOK(PF_INET, NF_INET_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
369 err = inet->recverr ? net_xmit_errno(err) : 0;
379 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
383 static int raw_probe_proto_opt(struct flowi *fl, struct msghdr *msg)
386 u8 __user *type = NULL;
387 u8 __user *code = NULL;
394 for (i = 0; i < msg->msg_iovlen; i++) {
395 iov = &msg->msg_iov[i];
401 /* check if one-byte field is readable or not. */
402 if (iov->iov_base && iov->iov_len < 1)
406 type = iov->iov_base;
407 /* check if code field is readable or not. */
408 if (iov->iov_len > 1)
411 code = iov->iov_base;
414 if (get_user(fl->fl_icmp_type, type) ||
415 get_user(fl->fl_icmp_code, code))
430 static int raw_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
433 struct inet_sock *inet = inet_sk(sk);
434 struct ipcm_cookie ipc;
435 struct rtable *rt = NULL;
451 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message */
452 goto out; /* compatibility */
455 * Get and verify the address.
458 if (msg->msg_namelen) {
459 struct sockaddr_in *usin = (struct sockaddr_in*)msg->msg_name;
461 if (msg->msg_namelen < sizeof(*usin))
463 if (usin->sin_family != AF_INET) {
464 static int complained;
466 printk(KERN_INFO "%s forgot to set AF_INET in "
467 "raw sendmsg. Fix it!\n",
470 if (usin->sin_family)
473 daddr = usin->sin_addr.s_addr;
474 /* ANK: I did not forget to get protocol from port field.
475 * I just do not know, who uses this weirdness.
476 * IP_HDRINCL is much more convenient.
480 if (sk->sk_state != TCP_ESTABLISHED)
485 ipc.addr = inet->saddr;
487 ipc.oif = sk->sk_bound_dev_if;
489 if (msg->msg_controllen) {
490 err = ip_cmsg_send(msg, &ipc);
505 /* Linux does not mangle headers on raw sockets,
506 * so that IP options + IP_HDRINCL is non-sense.
513 daddr = ipc.opt->faddr;
516 tos = RT_CONN_FLAGS(sk);
517 if (msg->msg_flags & MSG_DONTROUTE)
520 if (MULTICAST(daddr)) {
522 ipc.oif = inet->mc_index;
524 saddr = inet->mc_addr;
528 struct flowi fl = { .oif = ipc.oif,
533 .proto = inet->hdrincl ? IPPROTO_RAW :
536 if (!inet->hdrincl) {
537 err = raw_probe_proto_opt(&fl, msg);
542 security_sk_classify_flow(sk, &fl);
543 err = ip_route_output_flow(&rt, &fl, sk, 1);
549 if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
552 if (msg->msg_flags & MSG_CONFIRM)
557 err = raw_send_hdrinc(sk, msg->msg_iov, len,
562 ipc.addr = rt->rt_dst;
564 err = ip_append_data(sk, ip_generic_getfrag, msg->msg_iov, len, 0,
565 &ipc, rt, msg->msg_flags);
567 ip_flush_pending_frames(sk);
568 else if (!(msg->msg_flags & MSG_MORE))
569 err = ip_push_pending_frames(sk);
583 dst_confirm(&rt->u.dst);
584 if (!(msg->msg_flags & MSG_PROBE) || len)
585 goto back_from_confirm;
590 static void raw_close(struct sock *sk, long timeout)
593 * Raw sockets may have direct kernel refereneces. Kill them.
595 ip_ra_control(sk, 0, NULL);
597 sk_common_release(sk);
600 /* This gets rid of all the nasties in af_inet. -DaveM */
601 static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
603 struct inet_sock *inet = inet_sk(sk);
604 struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
608 if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
610 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
611 ret = -EADDRNOTAVAIL;
612 if (addr->sin_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
613 chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
615 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
616 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
617 inet->saddr = 0; /* Use device */
624 * This should be easy, if there is something there
625 * we return it, otherwise we block.
628 static int raw_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
629 size_t len, int noblock, int flags, int *addr_len)
631 struct inet_sock *inet = inet_sk(sk);
633 int err = -EOPNOTSUPP;
634 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
641 *addr_len = sizeof(*sin);
643 if (flags & MSG_ERRQUEUE) {
644 err = ip_recv_error(sk, msg, len);
648 skb = skb_recv_datagram(sk, flags, noblock, &err);
654 msg->msg_flags |= MSG_TRUNC;
658 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
662 sock_recv_timestamp(msg, sk, skb);
664 /* Copy the address. */
666 sin->sin_family = AF_INET;
667 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
669 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
671 if (inet->cmsg_flags)
672 ip_cmsg_recv(msg, skb);
673 if (flags & MSG_TRUNC)
676 skb_free_datagram(sk, skb);
683 static int raw_init(struct sock *sk)
685 struct raw_sock *rp = raw_sk(sk);
687 if (inet_sk(sk)->num == IPPROTO_ICMP)
688 memset(&rp->filter, 0, sizeof(rp->filter));
692 static int raw_seticmpfilter(struct sock *sk, char __user *optval, int optlen)
694 if (optlen > sizeof(struct icmp_filter))
695 optlen = sizeof(struct icmp_filter);
696 if (copy_from_user(&raw_sk(sk)->filter, optval, optlen))
701 static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
703 int len, ret = -EFAULT;
705 if (get_user(len, optlen))
710 if (len > sizeof(struct icmp_filter))
711 len = sizeof(struct icmp_filter);
713 if (put_user(len, optlen) ||
714 copy_to_user(optval, &raw_sk(sk)->filter, len))
720 static int do_raw_setsockopt(struct sock *sk, int level, int optname,
721 char __user *optval, int optlen)
723 if (optname == ICMP_FILTER) {
724 if (inet_sk(sk)->num != IPPROTO_ICMP)
727 return raw_seticmpfilter(sk, optval, optlen);
732 static int raw_setsockopt(struct sock *sk, int level, int optname,
733 char __user *optval, int optlen)
735 if (level != SOL_RAW)
736 return ip_setsockopt(sk, level, optname, optval, optlen);
737 return do_raw_setsockopt(sk, level, optname, optval, optlen);
741 static int compat_raw_setsockopt(struct sock *sk, int level, int optname,
742 char __user *optval, int optlen)
744 if (level != SOL_RAW)
745 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
746 return do_raw_setsockopt(sk, level, optname, optval, optlen);
750 static int do_raw_getsockopt(struct sock *sk, int level, int optname,
751 char __user *optval, int __user *optlen)
753 if (optname == ICMP_FILTER) {
754 if (inet_sk(sk)->num != IPPROTO_ICMP)
757 return raw_geticmpfilter(sk, optval, optlen);
762 static int raw_getsockopt(struct sock *sk, int level, int optname,
763 char __user *optval, int __user *optlen)
765 if (level != SOL_RAW)
766 return ip_getsockopt(sk, level, optname, optval, optlen);
767 return do_raw_getsockopt(sk, level, optname, optval, optlen);
771 static int compat_raw_getsockopt(struct sock *sk, int level, int optname,
772 char __user *optval, int __user *optlen)
774 if (level != SOL_RAW)
775 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
776 return do_raw_getsockopt(sk, level, optname, optval, optlen);
780 static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
784 int amount = atomic_read(&sk->sk_wmem_alloc);
785 return put_user(amount, (int __user *)arg);
791 spin_lock_bh(&sk->sk_receive_queue.lock);
792 skb = skb_peek(&sk->sk_receive_queue);
795 spin_unlock_bh(&sk->sk_receive_queue.lock);
796 return put_user(amount, (int __user *)arg);
800 #ifdef CONFIG_IP_MROUTE
801 return ipmr_ioctl(sk, cmd, (void __user *)arg);
808 DEFINE_PROTO_INUSE(raw)
810 struct proto raw_prot = {
812 .owner = THIS_MODULE,
814 .connect = ip4_datagram_connect,
815 .disconnect = udp_disconnect,
818 .setsockopt = raw_setsockopt,
819 .getsockopt = raw_getsockopt,
820 .sendmsg = raw_sendmsg,
821 .recvmsg = raw_recvmsg,
823 .backlog_rcv = raw_rcv_skb,
825 .unhash = raw_v4_unhash,
826 .obj_size = sizeof(struct raw_sock),
828 .compat_setsockopt = compat_raw_setsockopt,
829 .compat_getsockopt = compat_raw_getsockopt,
834 #ifdef CONFIG_PROC_FS
835 struct raw_iter_state {
839 #define raw_seq_private(seq) ((struct raw_iter_state *)(seq)->private)
841 static struct sock *raw_get_first(struct seq_file *seq)
844 struct raw_iter_state* state = raw_seq_private(seq);
846 for (state->bucket = 0; state->bucket < RAW_HTABLE_SIZE;
848 struct hlist_node *node;
850 sk_for_each(sk, node, &raw_v4_hashinfo.ht[state->bucket])
851 if (sk->sk_family == PF_INET)
859 static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
861 struct raw_iter_state* state = raw_seq_private(seq);
867 } while (sk && sk->sk_family != PF_INET);
869 if (!sk && ++state->bucket < RAW_HTABLE_SIZE) {
870 sk = sk_head(&raw_v4_hashinfo.ht[state->bucket]);
876 static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
878 struct sock *sk = raw_get_first(seq);
881 while (pos && (sk = raw_get_next(seq, sk)) != NULL)
883 return pos ? NULL : sk;
886 static void *raw_seq_start(struct seq_file *seq, loff_t *pos)
888 read_lock(&raw_v4_hashinfo.lock);
889 return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
892 static void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
896 if (v == SEQ_START_TOKEN)
897 sk = raw_get_first(seq);
899 sk = raw_get_next(seq, v);
904 static void raw_seq_stop(struct seq_file *seq, void *v)
906 read_unlock(&raw_v4_hashinfo.lock);
909 static __inline__ char *get_raw_sock(struct sock *sp, char *tmpbuf, int i)
911 struct inet_sock *inet = inet_sk(sp);
912 __be32 dest = inet->daddr,
913 src = inet->rcv_saddr;
917 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
918 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d",
919 i, src, srcp, dest, destp, sp->sk_state,
920 atomic_read(&sp->sk_wmem_alloc),
921 atomic_read(&sp->sk_rmem_alloc),
922 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
923 atomic_read(&sp->sk_refcnt), sp, atomic_read(&sp->sk_drops));
929 static int raw_seq_show(struct seq_file *seq, void *v)
931 char tmpbuf[TMPSZ+1];
933 if (v == SEQ_START_TOKEN)
934 seq_printf(seq, "%-*s\n", TMPSZ-1,
935 " sl local_address rem_address st tx_queue "
936 "rx_queue tr tm->when retrnsmt uid timeout "
939 struct raw_iter_state *state = raw_seq_private(seq);
941 seq_printf(seq, "%-*s\n", TMPSZ-1,
942 get_raw_sock(v, tmpbuf, state->bucket));
947 static const struct seq_operations raw_seq_ops = {
948 .start = raw_seq_start,
949 .next = raw_seq_next,
950 .stop = raw_seq_stop,
951 .show = raw_seq_show,
954 static int raw_seq_open(struct inode *inode, struct file *file)
956 return seq_open_private(file, &raw_seq_ops,
957 sizeof(struct raw_iter_state));
960 static const struct file_operations raw_seq_fops = {
961 .owner = THIS_MODULE,
962 .open = raw_seq_open,
965 .release = seq_release_private,
968 int __init raw_proc_init(void)
970 if (!proc_net_fops_create(&init_net, "raw", S_IRUGO, &raw_seq_fops))
975 void __init raw_proc_exit(void)
977 proc_net_remove(&init_net, "raw");
979 #endif /* CONFIG_PROC_FS */