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);
104 struct sock *__raw_v4_lookup(struct sock *sk, unsigned short num,
105 __be32 raddr, __be32 laddr,
108 struct hlist_node *node;
110 sk_for_each_from(sk, node) {
111 struct inet_sock *inet = inet_sk(sk);
113 if (inet->num == num &&
114 !(inet->daddr && inet->daddr != raddr) &&
115 !(inet->rcv_saddr && inet->rcv_saddr != laddr) &&
116 !(sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
117 goto found; /* gotcha */
128 static __inline__ int icmp_filter(struct sock *sk, struct sk_buff *skb)
132 if (!pskb_may_pull(skb, sizeof(struct icmphdr)))
135 type = skb->h.icmph->type;
137 __u32 data = raw_sk(sk)->filter.data;
139 return ((1 << type) & data) != 0;
142 /* Do not block unknown ICMP types */
146 /* IP input processing comes here for RAW socket delivery.
147 * Caller owns SKB, so we must make clones.
149 * RFC 1122: SHOULD pass TOS value up to the transport layer.
150 * -> It does. And not only TOS, but all IP header.
152 int raw_v4_input(struct sk_buff *skb, struct iphdr *iph, int hash)
155 struct hlist_head *head;
158 read_lock(&raw_v4_lock);
159 head = &raw_v4_htable[hash];
160 if (hlist_empty(head))
162 sk = __raw_v4_lookup(__sk_head(head), iph->protocol,
163 iph->saddr, iph->daddr,
168 if (iph->protocol != IPPROTO_ICMP || !icmp_filter(sk, skb)) {
169 struct sk_buff *clone = skb_clone(skb, GFP_ATOMIC);
171 /* Not releasing hash table! */
175 sk = __raw_v4_lookup(sk_next(sk), iph->protocol,
176 iph->saddr, iph->daddr,
180 read_unlock(&raw_v4_lock);
184 void raw_err (struct sock *sk, struct sk_buff *skb, u32 info)
186 struct inet_sock *inet = inet_sk(sk);
187 int type = skb->h.icmph->type;
188 int code = skb->h.icmph->code;
192 /* Report error on raw socket, if:
193 1. User requested ip_recverr.
194 2. Socket is connected (otherwise the error indication
195 is useless without ip_recverr and error is hard.
197 if (!inet->recverr && sk->sk_state != TCP_ESTABLISHED)
202 case ICMP_TIME_EXCEEDED:
205 case ICMP_SOURCE_QUENCH:
207 case ICMP_PARAMETERPROB:
211 case ICMP_DEST_UNREACH:
213 if (code > NR_ICMP_UNREACH)
215 err = icmp_err_convert[code].errno;
216 harderr = icmp_err_convert[code].fatal;
217 if (code == ICMP_FRAG_NEEDED) {
218 harderr = inet->pmtudisc != IP_PMTUDISC_DONT;
224 struct iphdr *iph = (struct iphdr*)skb->data;
225 u8 *payload = skb->data + (iph->ihl << 2);
229 ip_icmp_error(sk, skb, err, 0, info, payload);
232 if (inet->recverr || harderr) {
234 sk->sk_error_report(sk);
238 static int raw_rcv_skb(struct sock * sk, struct sk_buff * skb)
240 /* Charge it to the socket. */
242 if (sock_queue_rcv_skb(sk, skb) < 0) {
243 /* FIXME: increment a raw drops counter here */
248 return NET_RX_SUCCESS;
251 int raw_rcv(struct sock *sk, struct sk_buff *skb)
253 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
259 skb_push(skb, skb->data - skb->nh.raw);
261 raw_rcv_skb(sk, skb);
265 static int raw_send_hdrinc(struct sock *sk, void *from, size_t length,
269 struct inet_sock *inet = inet_sk(sk);
275 if (length > rt->u.dst.dev->mtu) {
276 ip_local_error(sk, EMSGSIZE, rt->rt_dst, inet->dport,
283 hh_len = LL_RESERVED_SPACE(rt->u.dst.dev);
285 skb = sock_alloc_send_skb(sk, length+hh_len+15,
286 flags&MSG_DONTWAIT, &err);
289 skb_reserve(skb, hh_len);
291 skb->priority = sk->sk_priority;
292 skb->dst = dst_clone(&rt->u.dst);
294 skb->nh.iph = iph = (struct iphdr *)skb_put(skb, length);
296 skb->ip_summed = CHECKSUM_NONE;
298 skb->h.raw = skb->nh.raw;
299 err = memcpy_fromiovecend((void *)iph, from, 0, length);
303 /* We don't modify invalid header */
304 if (length >= sizeof(*iph) && iph->ihl * 4U <= length) {
306 iph->saddr = rt->rt_src;
308 iph->tot_len = htons(length);
310 ip_select_ident(iph, &rt->u.dst, NULL);
312 iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);
315 err = NF_HOOK(PF_INET, NF_IP_LOCAL_OUT, skb, NULL, rt->u.dst.dev,
318 err = inet->recverr ? net_xmit_errno(err) : 0;
328 IP_INC_STATS(IPSTATS_MIB_OUTDISCARDS);
332 static int raw_probe_proto_opt(struct flowi *fl, struct msghdr *msg)
335 u8 __user *type = NULL;
336 u8 __user *code = NULL;
343 for (i = 0; i < msg->msg_iovlen; i++) {
344 iov = &msg->msg_iov[i];
350 /* check if one-byte field is readable or not. */
351 if (iov->iov_base && iov->iov_len < 1)
355 type = iov->iov_base;
356 /* check if code field is readable or not. */
357 if (iov->iov_len > 1)
360 code = iov->iov_base;
363 if (get_user(fl->fl_icmp_type, type) ||
364 get_user(fl->fl_icmp_code, code))
379 static int raw_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
382 struct inet_sock *inet = inet_sk(sk);
383 struct ipcm_cookie ipc;
384 struct rtable *rt = NULL;
400 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message */
401 goto out; /* compatibility */
404 * Get and verify the address.
407 if (msg->msg_namelen) {
408 struct sockaddr_in *usin = (struct sockaddr_in*)msg->msg_name;
410 if (msg->msg_namelen < sizeof(*usin))
412 if (usin->sin_family != AF_INET) {
413 static int complained;
415 printk(KERN_INFO "%s forgot to set AF_INET in "
416 "raw sendmsg. Fix it!\n",
419 if (usin->sin_family)
422 daddr = usin->sin_addr.s_addr;
423 /* ANK: I did not forget to get protocol from port field.
424 * I just do not know, who uses this weirdness.
425 * IP_HDRINCL is much more convenient.
429 if (sk->sk_state != TCP_ESTABLISHED)
434 ipc.addr = inet->saddr;
436 ipc.oif = sk->sk_bound_dev_if;
438 if (msg->msg_controllen) {
439 err = ip_cmsg_send(msg, &ipc);
454 /* Linux does not mangle headers on raw sockets,
455 * so that IP options + IP_HDRINCL is non-sense.
462 daddr = ipc.opt->faddr;
465 tos = RT_CONN_FLAGS(sk);
466 if (msg->msg_flags & MSG_DONTROUTE)
469 if (MULTICAST(daddr)) {
471 ipc.oif = inet->mc_index;
473 saddr = inet->mc_addr;
477 struct flowi fl = { .oif = ipc.oif,
482 .proto = inet->hdrincl ? IPPROTO_RAW :
485 if (!inet->hdrincl) {
486 err = raw_probe_proto_opt(&fl, msg);
491 security_sk_classify_flow(sk, &fl);
492 err = ip_route_output_flow(&rt, &fl, sk, 1);
498 if (rt->rt_flags & RTCF_BROADCAST && !sock_flag(sk, SOCK_BROADCAST))
501 if (msg->msg_flags & MSG_CONFIRM)
506 err = raw_send_hdrinc(sk, msg->msg_iov, len,
511 ipc.addr = rt->rt_dst;
513 err = ip_append_data(sk, ip_generic_getfrag, msg->msg_iov, len, 0,
514 &ipc, rt, msg->msg_flags);
516 ip_flush_pending_frames(sk);
517 else if (!(msg->msg_flags & MSG_MORE))
518 err = ip_push_pending_frames(sk);
532 dst_confirm(&rt->u.dst);
533 if (!(msg->msg_flags & MSG_PROBE) || len)
534 goto back_from_confirm;
539 static void raw_close(struct sock *sk, long timeout)
542 * Raw sockets may have direct kernel refereneces. Kill them.
544 ip_ra_control(sk, 0, NULL);
546 sk_common_release(sk);
549 /* This gets rid of all the nasties in af_inet. -DaveM */
550 static int raw_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len)
552 struct inet_sock *inet = inet_sk(sk);
553 struct sockaddr_in *addr = (struct sockaddr_in *) uaddr;
557 if (sk->sk_state != TCP_CLOSE || addr_len < sizeof(struct sockaddr_in))
559 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
560 ret = -EADDRNOTAVAIL;
561 if (addr->sin_addr.s_addr && chk_addr_ret != RTN_LOCAL &&
562 chk_addr_ret != RTN_MULTICAST && chk_addr_ret != RTN_BROADCAST)
564 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
565 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
566 inet->saddr = 0; /* Use device */
573 * This should be easy, if there is something there
574 * we return it, otherwise we block.
577 static int raw_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
578 size_t len, int noblock, int flags, int *addr_len)
580 struct inet_sock *inet = inet_sk(sk);
582 int err = -EOPNOTSUPP;
583 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
590 *addr_len = sizeof(*sin);
592 if (flags & MSG_ERRQUEUE) {
593 err = ip_recv_error(sk, msg, len);
597 skb = skb_recv_datagram(sk, flags, noblock, &err);
603 msg->msg_flags |= MSG_TRUNC;
607 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
611 sock_recv_timestamp(msg, sk, skb);
613 /* Copy the address. */
615 sin->sin_family = AF_INET;
616 sin->sin_addr.s_addr = skb->nh.iph->saddr;
618 memset(&sin->sin_zero, 0, sizeof(sin->sin_zero));
620 if (inet->cmsg_flags)
621 ip_cmsg_recv(msg, skb);
622 if (flags & MSG_TRUNC)
625 skb_free_datagram(sk, skb);
632 static int raw_init(struct sock *sk)
634 struct raw_sock *rp = raw_sk(sk);
636 if (inet_sk(sk)->num == IPPROTO_ICMP)
637 memset(&rp->filter, 0, sizeof(rp->filter));
641 static int raw_seticmpfilter(struct sock *sk, char __user *optval, int optlen)
643 if (optlen > sizeof(struct icmp_filter))
644 optlen = sizeof(struct icmp_filter);
645 if (copy_from_user(&raw_sk(sk)->filter, optval, optlen))
650 static int raw_geticmpfilter(struct sock *sk, char __user *optval, int __user *optlen)
652 int len, ret = -EFAULT;
654 if (get_user(len, optlen))
659 if (len > sizeof(struct icmp_filter))
660 len = sizeof(struct icmp_filter);
662 if (put_user(len, optlen) ||
663 copy_to_user(optval, &raw_sk(sk)->filter, len))
669 static int do_raw_setsockopt(struct sock *sk, int level, int optname,
670 char __user *optval, int optlen)
672 if (optname == ICMP_FILTER) {
673 if (inet_sk(sk)->num != IPPROTO_ICMP)
676 return raw_seticmpfilter(sk, optval, optlen);
681 static int raw_setsockopt(struct sock *sk, int level, int optname,
682 char __user *optval, int optlen)
684 if (level != SOL_RAW)
685 return ip_setsockopt(sk, level, optname, optval, optlen);
686 return do_raw_setsockopt(sk, level, optname, optval, optlen);
690 static int compat_raw_setsockopt(struct sock *sk, int level, int optname,
691 char __user *optval, int optlen)
693 if (level != SOL_RAW)
694 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
695 return do_raw_setsockopt(sk, level, optname, optval, optlen);
699 static int do_raw_getsockopt(struct sock *sk, int level, int optname,
700 char __user *optval, int __user *optlen)
702 if (optname == ICMP_FILTER) {
703 if (inet_sk(sk)->num != IPPROTO_ICMP)
706 return raw_geticmpfilter(sk, optval, optlen);
711 static int raw_getsockopt(struct sock *sk, int level, int optname,
712 char __user *optval, int __user *optlen)
714 if (level != SOL_RAW)
715 return ip_getsockopt(sk, level, optname, optval, optlen);
716 return do_raw_getsockopt(sk, level, optname, optval, optlen);
720 static int compat_raw_getsockopt(struct sock *sk, int level, int optname,
721 char __user *optval, int __user *optlen)
723 if (level != SOL_RAW)
724 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
725 return do_raw_getsockopt(sk, level, optname, optval, optlen);
729 static int raw_ioctl(struct sock *sk, int cmd, unsigned long arg)
733 int amount = atomic_read(&sk->sk_wmem_alloc);
734 return put_user(amount, (int __user *)arg);
740 spin_lock_bh(&sk->sk_receive_queue.lock);
741 skb = skb_peek(&sk->sk_receive_queue);
744 spin_unlock_bh(&sk->sk_receive_queue.lock);
745 return put_user(amount, (int __user *)arg);
749 #ifdef CONFIG_IP_MROUTE
750 return ipmr_ioctl(sk, cmd, (void __user *)arg);
757 struct proto raw_prot = {
759 .owner = THIS_MODULE,
761 .connect = ip4_datagram_connect,
762 .disconnect = udp_disconnect,
765 .setsockopt = raw_setsockopt,
766 .getsockopt = raw_getsockopt,
767 .sendmsg = raw_sendmsg,
768 .recvmsg = raw_recvmsg,
770 .backlog_rcv = raw_rcv_skb,
772 .unhash = raw_v4_unhash,
773 .obj_size = sizeof(struct raw_sock),
775 .compat_setsockopt = compat_raw_setsockopt,
776 .compat_getsockopt = compat_raw_getsockopt,
780 #ifdef CONFIG_PROC_FS
781 struct raw_iter_state {
785 #define raw_seq_private(seq) ((struct raw_iter_state *)(seq)->private)
787 static struct sock *raw_get_first(struct seq_file *seq)
790 struct raw_iter_state* state = raw_seq_private(seq);
792 for (state->bucket = 0; state->bucket < RAWV4_HTABLE_SIZE; ++state->bucket) {
793 struct hlist_node *node;
795 sk_for_each(sk, node, &raw_v4_htable[state->bucket])
796 if (sk->sk_family == PF_INET)
804 static struct sock *raw_get_next(struct seq_file *seq, struct sock *sk)
806 struct raw_iter_state* state = raw_seq_private(seq);
812 } while (sk && sk->sk_family != PF_INET);
814 if (!sk && ++state->bucket < RAWV4_HTABLE_SIZE) {
815 sk = sk_head(&raw_v4_htable[state->bucket]);
821 static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
823 struct sock *sk = raw_get_first(seq);
826 while (pos && (sk = raw_get_next(seq, sk)) != NULL)
828 return pos ? NULL : sk;
831 static void *raw_seq_start(struct seq_file *seq, loff_t *pos)
833 read_lock(&raw_v4_lock);
834 return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
837 static void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
841 if (v == SEQ_START_TOKEN)
842 sk = raw_get_first(seq);
844 sk = raw_get_next(seq, v);
849 static void raw_seq_stop(struct seq_file *seq, void *v)
851 read_unlock(&raw_v4_lock);
854 static __inline__ char *get_raw_sock(struct sock *sp, char *tmpbuf, int i)
856 struct inet_sock *inet = inet_sk(sp);
857 __be32 dest = inet->daddr,
858 src = inet->rcv_saddr;
862 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
863 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
864 i, src, srcp, dest, destp, sp->sk_state,
865 atomic_read(&sp->sk_wmem_alloc),
866 atomic_read(&sp->sk_rmem_alloc),
867 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
868 atomic_read(&sp->sk_refcnt), sp);
872 static int raw_seq_show(struct seq_file *seq, void *v)
876 if (v == SEQ_START_TOKEN)
877 seq_printf(seq, "%-127s\n",
878 " sl local_address rem_address st tx_queue "
879 "rx_queue tr tm->when retrnsmt uid timeout "
882 struct raw_iter_state *state = raw_seq_private(seq);
884 seq_printf(seq, "%-127s\n",
885 get_raw_sock(v, tmpbuf, state->bucket));
890 static struct seq_operations raw_seq_ops = {
891 .start = raw_seq_start,
892 .next = raw_seq_next,
893 .stop = raw_seq_stop,
894 .show = raw_seq_show,
897 static int raw_seq_open(struct inode *inode, struct file *file)
899 struct seq_file *seq;
901 struct raw_iter_state *s = kmalloc(sizeof(*s), GFP_KERNEL);
905 rc = seq_open(file, &raw_seq_ops);
909 seq = file->private_data;
911 memset(s, 0, sizeof(*s));
919 static const struct file_operations raw_seq_fops = {
920 .owner = THIS_MODULE,
921 .open = raw_seq_open,
924 .release = seq_release_private,
927 int __init raw_proc_init(void)
929 if (!proc_net_fops_create("raw", S_IRUGO, &raw_seq_fops))
934 void __init raw_proc_exit(void)
936 proc_net_remove("raw");
938 #endif /* CONFIG_PROC_FS */