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>
64 #include <linux/gfp.h>
66 #include <linux/net.h>
72 #include <net/tcp_states.h>
73 #include <net/inet_common.h>
74 #include <net/checksum.h>
76 #include <linux/rtnetlink.h>
77 #include <linux/proc_fs.h>
78 #include <linux/seq_file.h>
79 #include <linux/netfilter.h>
80 #include <linux/netfilter_ipv4.h>
82 struct hlist_head raw_v4_htable[RAWV4_HTABLE_SIZE];
83 DEFINE_RWLOCK(raw_v4_lock);
85 static void raw_v4_hash(struct sock *sk)
87 struct hlist_head *head = &raw_v4_htable[inet_sk(sk)->num &
88 (RAWV4_HTABLE_SIZE - 1)];
90 write_lock_bh(&raw_v4_lock);
91 sk_add_node(sk, head);
92 sock_prot_inc_use(sk->sk_prot);
93 write_unlock_bh(&raw_v4_lock);
96 static void raw_v4_unhash(struct sock *sk)
98 write_lock_bh(&raw_v4_lock);
99 if (sk_del_node_init(sk))
100 sock_prot_dec_use(sk->sk_prot);
101 write_unlock_bh(&raw_v4_lock);
106 * Check if a given address matches for a socket
108 * nxi: the socket's nx_info if any
109 * addr: to be verified address
110 * saddr/baddr: socket addresses
112 static inline int raw_addr_match (
119 vxdprintk(VXD_CBIT(net, 8),
120 "raw_addr_match: %p[#%u] #%d"
121 " %u.%u.%u.%u/%u.%u.%u.%u/%u.%u.%u.%u",
122 nxi, nxi?nxi->nx_id:0, tag,
123 NIPQUAD(addr), NIPQUAD(saddr), NIPQUAD(baddr));
125 if (nxi && !((tag == 1) || (nxi->nx_id == tag)))
127 if (addr && (saddr == addr || baddr == addr))
130 return addr_in_nx_info(nxi, addr);
134 struct sock *__raw_v4_lookup(struct sock *sk, unsigned short num,
135 __be32 raddr, __be32 laddr,
138 struct hlist_node *node;
140 sk_for_each_from(sk, node) {
141 struct inet_sock *inet = inet_sk(sk);
143 if (inet->num == num &&
144 !(inet->daddr && inet->daddr != raddr) &&
145 raw_addr_match(sk->sk_nx_info, laddr,
146 inet->rcv_saddr, inet->rcv_saddr2, tag) &&
147 !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
148 goto found; /* gotcha */
159 static __inline__ int icmp_filter(struct sock *sk, struct sk_buff *skb)
163 if (!pskb_may_pull(skb, sizeof(struct icmphdr)))
166 type = skb->h.icmph->type;
168 __u32 data = raw_sk(sk)->filter.data;
170 return ((1 << type) & data) != 0;
173 /* Do not block unknown ICMP types */
177 /* IP input processing comes here for RAW socket delivery.
178 * Caller owns SKB, so we must make clones.
180 * RFC 1122: SHOULD pass TOS value up to the transport layer.
181 * -> It does. And not only TOS, but all IP header.
183 int raw_v4_input(struct sk_buff *skb, struct iphdr *iph, int hash)
186 struct hlist_head *head;
189 read_lock(&raw_v4_lock);
190 head = &raw_v4_htable[hash];
191 if (hlist_empty(head))
193 sk = __raw_v4_lookup(__sk_head(head), iph->protocol,
194 iph->saddr, iph->daddr,
195 skb->dev->ifindex, skb->skb_tag);
199 if (iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) {
200 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
202 /* Not releasing hash table! */
206 sk = __raw_v4_lookup(sk_next(sk), iph->protocol,
207 iph->saddr, iph->daddr,
208 skb->dev->ifindex, skb->skb_tag);
211 read_unlock(&raw_v4_lock);
215 void raw_err (struct sock *sk, struct sk_buff *skb, u32 info)
217 struct inet_sock *inet = inet_sk(sk);
218 int type = skb->h.icmph->type;
219 int code = skb->h.icmph->code;
223 /* Report error on raw socket, if:
224 1. User requested ip_recverr.
225 2. Socket is connected (otherwise the error indication
226 is useless without ip_recverr and error is hard.
228 if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
233 case ICMP_TIME_EXCEEDED:
236 case ICMP_SOURCE_QUENCH:
238 case ICMP_PARAMETERPROB:
242 case ICMP_DEST_UNREACH:
244 if (code > NR_ICMP_UNREACH)
246 err = icmp_err_convert[code].errno;
247 harderr = icmp_err_convert[code].fatal;
248 if (code == ICMP_FRAG_NEEDED) {
249 harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
255 struct iphdr *iph = (struct iphdr*)skb->data;
256 u8 *payload = skb->data + (iph->ihl << 2);
260 ip_icmp_error(sk, skb, err, 0, info, payload);
263 if (inet->recverr || harderr) {
265 sk->sk_error_report(sk);
269 static int raw_rcv_skb(struct sock * sk, struct sk_buff * skb)
271 /* Charge it to the socket. */
273 if (sock_queue_rcv_skb(sk, skb) < 0) {
274 /* FIXME: increment a raw drops counter here */
279 return NET_RX_SUCCESS;
282 int raw_rcv(struct sock *sk, struct sk_buff *skb)
284 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
290 skb_push(skb, skb->data - skb->nh.raw);
292 raw_rcv_skb(sk, skb);
296 static int raw_send_hdrinc(struct sock *sk, void *from, size_t length,
300 struct inet_sock *inet = inet_sk(sk);
306 if (length > rt->u.dst.dev->mtu) {
307 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport,
314 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
316 skb = sock_alloc_send_skb(sk, length+hh_len+15,
317 flags&MSG_DONTWAIT, &err);
320 skb_reserve(skb, hh_len);
322 skb->priority = sk->sk_priority;
323 skb->dst = dst_clone(&rt->u.dst);
325 skb->nh.iph = iph = (struct iphdr *)skb_put(skb, length);
327 skb->ip_summed = CHECKSUM_NONE;
329 skb->h.raw = skb->nh.raw;
330 err = memcpy_fromiovecend((void *)iph, from, 0, length);
334 /* We don't modify invalid header */
335 if (length >= sizeof(*iph) && iph->ihl * 4U <= length) {
337 iph->saddr = rt->rt_src;
339 iph->tot_len = htons(length);
341 ip_select_ident(iph, &rt->u.dst, NULL);
343 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
347 if (!nx_check(0, VS_ADMIN) && !nx_capable(CAP_NET_RAW, NXC_RAW_SOCKET)
348 && (!addr_in_nx_info(sk->sk_nx_info, iph->saddr)))
351 err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
354 err = inet->recverr ? net_xmit_errno(err) : 0;
365 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
369 static int raw_probe_proto_opt(struct flowi *fl, struct msghdr *msg)
372 u8 __user *type = NULL;
373 u8 __user *code = NULL;
380 for (i = 0; i < msg->msg_iovlen; i++) {
381 iov = &msg->msg_iov[i];
387 /* check if one-byte field is readable or not. */
388 if (iov->iov_base && iov->iov_len < 1)
392 type = iov->iov_base;
393 /* check if code field is readable or not. */
394 if (iov->iov_len > 1)
397 code = iov->iov_base;
400 if (get_user(fl->fl_icmp_type, type) ||
401 get_user(fl->fl_icmp_code, code))
416 static int raw_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
419 struct inet_sock *inet = inet_sk(sk);
420 struct ipcm_cookie ipc;
421 struct rtable *rt = NULL;
437 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message */
438 goto out; /* compatibility */
441 * Get and verify the address.
444 if (msg->msg_namelen) {
445 struct sockaddr_in *usin = (struct sockaddr_in*)msg->msg_name;
447 if (msg->msg_namelen < sizeof(*usin))
449 if (usin->sin_family != AF_INET) {
450 static int complained;
452 printk(KERN_INFO "%s forgot to set AF_INET in "
453 "raw sendmsg. Fix it!\n",
456 if (usin->sin_family)
459 daddr = usin->sin_addr.s_addr;
460 /* ANK: I did not forget to get protocol from port field.
461 * I just do not know, who uses this weirdness.
462 * IP_HDRINCL is much more convenient.
466 if (sk->sk_state != TCP_ESTABLISHED)
471 ipc.addr = inet->saddr;
473 ipc.oif = sk->sk_bound_dev_if;
475 if (msg->msg_controllen) {
476 err = ip_cmsg_send(msg, &ipc);
491 /* Linux does not mangle headers on raw sockets,
492 * so that IP options + IP_HDRINCL is non-sense.
499 daddr = ipc.opt->faddr;
502 tos = RT_CONN_FLAGS(sk);
503 if (msg->msg_flags & MSG_DONTROUTE)
506 if (MULTICAST(daddr)) {
508 ipc.oif = inet->mc_index;
510 saddr = inet->mc_addr;
514 struct flowi fl = { .oif = ipc.oif,
519 .proto = inet->hdrincl ? IPPROTO_RAW :
522 if (!inet->hdrincl) {
523 err = raw_probe_proto_opt(&fl, msg);
528 security_sk_classify_flow(sk, &fl);
529 if (sk->sk_nx_info) {
530 err = ip_find_src(sk->sk_nx_info, &rt, &fl);
535 err = ip_route_output_flow(&rt, &fl, sk, !(msg->msg_flags&MSG_DONTWAIT));
541 if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
544 if (msg->msg_flags & MSG_CONFIRM)
549 err = raw_send_hdrinc(sk, msg->msg_iov, len,
554 ipc.addr = rt->rt_dst;
556 err = ip_append_data(sk, ip_generic_getfrag, msg->msg_iov, len, 0,
557 &ipc, rt, msg->msg_flags);
559 ip_flush_pending_frames(sk);
560 else if (!(msg->msg_flags & MSG_MORE))
561 err = ip_push_pending_frames(sk);
575 dst_confirm(&rt->u.dst);
576 if (!(msg->msg_flags & MSG_PROBE) || len)
577 goto back_from_confirm;
582 static void raw_close(struct sock *sk, long timeout)
585 * Raw sockets may have direct kernel refereneces. Kill them.
587 ip_ra_control(sk, 0, NULL);
589 sk_common_release(sk);
592 /* This gets rid of all the nasties in af_inet. -DaveM */
593 static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
595 struct inet_sock *inet = inet_sk(sk);
596 struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
600 if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
602 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
603 ret = -EADDRNOTAVAIL;
604 if (addr->sin_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
605 chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
607 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
608 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
609 inet->saddr = 0; /* Use device */
616 * This should be easy, if there is something there
617 * we return it, otherwise we block.
620 static int raw_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
621 size_t len, int noblock, int flags, int *addr_len)
623 struct inet_sock *inet = inet_sk(sk);
625 int err = -EOPNOTSUPP;
626 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
633 *addr_len = sizeof(*sin);
635 if (flags & MSG_ERRQUEUE) {
636 err = ip_recv_error(sk, msg, len);
640 skb = skb_recv_datagram(sk, flags, noblock, &err);
646 msg->msg_flags |= MSG_TRUNC;
650 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
654 sock_recv_timestamp(msg, sk, skb);
656 /* Copy the address. */
658 sin->sin_family = AF_INET;
659 sin->sin_addr.s_addr = skb->nh.iph->saddr;
661 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
663 if (inet->cmsg_flags)
664 ip_cmsg_recv(msg, skb);
665 if (flags & MSG_TRUNC)
668 skb_free_datagram(sk, skb);
675 static int raw_init(struct sock *sk)
677 struct raw_sock *rp = raw_sk(sk);
679 if (inet_sk(sk)->num == IPPROTO_ICMP)
680 memset(&rp->filter, 0, sizeof(rp->filter));
684 static int raw_seticmpfilter(struct sock *sk, char __user *optval, int optlen)
686 if (optlen > sizeof(struct icmp_filter))
687 optlen = sizeof(struct icmp_filter);
688 if (copy_from_user(&raw_sk(sk)->filter, optval, optlen))
693 static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
695 int len, ret = -EFAULT;
697 if (get_user(len, optlen))
702 if (len > sizeof(struct icmp_filter))
703 len = sizeof(struct icmp_filter);
705 if (put_user(len, optlen) ||
706 copy_to_user(optval, &raw_sk(sk)->filter, len))
712 static int do_raw_setsockopt(struct sock *sk, int level, int optname,
713 char __user *optval, int optlen)
715 if (optname == ICMP_FILTER) {
716 if (inet_sk(sk)->num != IPPROTO_ICMP)
719 return raw_seticmpfilter(sk, optval, optlen);
724 static int raw_setsockopt(struct sock *sk, int level, int optname,
725 char __user *optval, int optlen)
727 if (level != SOL_RAW)
728 return ip_setsockopt(sk, level, optname, optval, optlen);
729 return do_raw_setsockopt(sk, level, optname, optval, optlen);
733 static int compat_raw_setsockopt(struct sock *sk, int level, int optname,
734 char __user *optval, int optlen)
736 if (level != SOL_RAW)
737 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
738 return do_raw_setsockopt(sk, level, optname, optval, optlen);
742 static int do_raw_getsockopt(struct sock *sk, int level, int optname,
743 char __user *optval, int __user *optlen)
745 if (optname == ICMP_FILTER) {
746 if (inet_sk(sk)->num != IPPROTO_ICMP)
749 return raw_geticmpfilter(sk, optval, optlen);
754 static int raw_getsockopt(struct sock *sk, int level, int optname,
755 char __user *optval, int __user *optlen)
757 if (level != SOL_RAW)
758 return ip_getsockopt(sk, level, optname, optval, optlen);
759 return do_raw_getsockopt(sk, level, optname, optval, optlen);
763 static int compat_raw_getsockopt(struct sock *sk, int level, int optname,
764 char __user *optval, int __user *optlen)
766 if (level != SOL_RAW)
767 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
768 return do_raw_getsockopt(sk, level, optname, optval, optlen);
772 static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
776 int amount = atomic_read(&sk->sk_wmem_alloc);
777 return put_user(amount, (int __user *)arg);
783 spin_lock_bh(&sk->sk_receive_queue.lock);
784 skb = skb_peek(&sk->sk_receive_queue);
787 spin_unlock_bh(&sk->sk_receive_queue.lock);
788 return put_user(amount, (int __user *)arg);
792 #ifdef CONFIG_IP_MROUTE
793 return ipmr_ioctl(sk, cmd, (void __user *)arg);
800 struct proto raw_prot = {
802 .owner = THIS_MODULE,
804 .connect = ip4_datagram_connect,
805 .disconnect = udp_disconnect,
808 .setsockopt = raw_setsockopt,
809 .getsockopt = raw_getsockopt,
810 .sendmsg = raw_sendmsg,
811 .recvmsg = raw_recvmsg,
813 .backlog_rcv = raw_rcv_skb,
815 .unhash = raw_v4_unhash,
816 .obj_size = sizeof(struct raw_sock),
818 .compat_setsockopt = compat_raw_setsockopt,
819 .compat_getsockopt = compat_raw_getsockopt,
823 #ifdef CONFIG_PROC_FS
824 struct raw_iter_state {
828 #define raw_seq_private(seq) ((struct raw_iter_state *)(seq)->private)
830 static struct sock *raw_get_first(struct seq_file *seq)
833 struct raw_iter_state* state = raw_seq_private(seq);
835 for (state->bucket = 0; state->bucket < RAWV4_HTABLE_SIZE; ++state->bucket) {
836 struct hlist_node *node;
838 sk_for_each(sk, node, &raw_v4_htable[state->bucket])
839 if (sk->sk_family == PF_INET &&
840 nx_check(sk->sk_nid, VS_WATCH_P|VS_IDENT))
848 static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
850 struct raw_iter_state* state = raw_seq_private(seq);
856 } while (sk && (sk->sk_family != PF_INET ||
857 !nx_check(sk->sk_nid, VS_WATCH_P|VS_IDENT)));
859 if (!sk && ++state->bucket < RAWV4_HTABLE_SIZE) {
860 sk = sk_head(&raw_v4_htable[state->bucket]);
866 static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
868 struct sock *sk = raw_get_first(seq);
871 while (pos && (sk = raw_get_next(seq, sk)) != NULL)
873 return pos ? NULL : sk;
876 static void *raw_seq_start(struct seq_file *seq, loff_t *pos)
878 read_lock(&raw_v4_lock);
879 return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
882 static void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
886 if (v == SEQ_START_TOKEN)
887 sk = raw_get_first(seq);
889 sk = raw_get_next(seq, v);
894 static void raw_seq_stop(struct seq_file *seq, void *v)
896 read_unlock(&raw_v4_lock);
899 static __inline__ char *get_raw_sock(struct sock *sp, char *tmpbuf, int i)
901 struct inet_sock *inet = inet_sk(sp);
902 __be32 dest = inet->daddr,
903 src = inet->rcv_saddr;
907 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
908 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
909 i, src, srcp, dest, destp, sp->sk_state,
910 atomic_read(&sp->sk_wmem_alloc),
911 atomic_read(&sp->sk_rmem_alloc),
912 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
913 atomic_read(&sp->sk_refcnt), sp);
917 static int raw_seq_show(struct seq_file *seq, void *v)
921 if (v == SEQ_START_TOKEN)
922 seq_printf(seq, "%-127s\n",
923 " sl local_address rem_address st tx_queue "
924 "rx_queue tr tm->when retrnsmt uid timeout "
927 struct raw_iter_state *state = raw_seq_private(seq);
929 seq_printf(seq, "%-127s\n",
930 get_raw_sock(v, tmpbuf, state->bucket));
935 static struct seq_operations raw_seq_ops = {
936 .start = raw_seq_start,
937 .next = raw_seq_next,
938 .stop = raw_seq_stop,
939 .show = raw_seq_show,
942 static int raw_seq_open(struct inode *inode, struct file *file)
944 struct seq_file *seq;
946 struct raw_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
950 rc = seq_open(file, &raw_seq_ops);
954 seq = file->private_data;
956 memset(s, 0, sizeof(*s));
964 static struct file_operations raw_seq_fops = {
965 .owner = THIS_MODULE,
966 .open = raw_seq_open,
969 .release = seq_release_private,
972 int __init raw_proc_init(void)
974 if (!proc_net_fops_create("raw", S_IRUGO, &raw_seq_fops))
979 void __init raw_proc_exit(void)
981 proc_net_remove("raw");
983 #endif /* CONFIG_PROC_FS */