[TCP]: MD5 Signature Option (RFC2385) support.
[safe/jmp/linux-2.6] / net / ipv4 / tcp_ipv4.c
index 0ad0904..8c8e811 100644 (file)
@@ -78,6 +78,9 @@
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
 
+#include <linux/crypto.h>
+#include <linux/scatterlist.h>
+
 int sysctl_tcp_tw_reuse __read_mostly;
 int sysctl_tcp_low_latency __read_mostly;
 
@@ -89,6 +92,13 @@ static struct socket *tcp_socket;
 
 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb);
 
+#ifdef CONFIG_TCP_MD5SIG
+static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr);
+static int tcp_v4_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
+                                  __be32 saddr, __be32 daddr, struct tcphdr *th,
+                                  int protocol, int tcplen);
+#endif
+
 struct inet_hashinfo __cacheline_aligned tcp_hashinfo = {
        .lhash_lock     = __RW_LOCK_UNLOCKED(tcp_hashinfo.lhash_lock),
        .lhash_users    = ATOMIC_INIT(0),
@@ -526,11 +536,19 @@ int tcp_v4_gso_send_check(struct sk_buff *skb)
  *     Exception: precedence violation. We do not implement it in any case.
  */
 
-static void tcp_v4_send_reset(struct sk_buff *skb)
+static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
 {
        struct tcphdr *th = skb->h.th;
-       struct tcphdr rth;
+       struct {
+               struct tcphdr th;
+#ifdef CONFIG_TCP_MD5SIG
+               u32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
+#endif
+       } rep;
        struct ip_reply_arg arg;
+#ifdef CONFIG_TCP_MD5SIG
+       struct tcp_md5sig_key *key;
+#endif
 
        /* Never send a reset in response to a reset. */
        if (th->rst)
@@ -540,29 +558,50 @@ static void tcp_v4_send_reset(struct sk_buff *skb)
                return;
 
        /* Swap the send and the receive. */
-       memset(&rth, 0, sizeof(struct tcphdr));
-       rth.dest   = th->source;
-       rth.source = th->dest;
-       rth.doff   = sizeof(struct tcphdr) / 4;
-       rth.rst    = 1;
+       memset(&rep, 0, sizeof(rep));
+       rep.th.dest   = th->source;
+       rep.th.source = th->dest;
+       rep.th.doff   = sizeof(struct tcphdr) / 4;
+       rep.th.rst    = 1;
 
        if (th->ack) {
-               rth.seq = th->ack_seq;
+               rep.th.seq = th->ack_seq;
        } else {
-               rth.ack = 1;
-               rth.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
-                                   skb->len - (th->doff << 2));
+               rep.th.ack = 1;
+               rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
+                                      skb->len - (th->doff << 2));
        }
 
        memset(&arg, 0, sizeof arg);
-       arg.iov[0].iov_base = (unsigned char *)&rth;
-       arg.iov[0].iov_len  = sizeof rth;
+       arg.iov[0].iov_base = (unsigned char *)&rep;
+       arg.iov[0].iov_len  = sizeof(rep.th);
+
+#ifdef CONFIG_TCP_MD5SIG
+       key = sk ? tcp_v4_md5_do_lookup(sk, skb->nh.iph->daddr) : NULL;
+       if (key) {
+               rep.opt[0] = htonl((TCPOPT_NOP << 24) |
+                                  (TCPOPT_NOP << 16) |
+                                  (TCPOPT_MD5SIG << 8) |
+                                  TCPOLEN_MD5SIG);
+               /* Update length and the length the header thinks exists */
+               arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
+               rep.th.doff = arg.iov[0].iov_len / 4;
+
+               tcp_v4_do_calc_md5_hash((__u8 *)&rep.opt[1],
+                                       key,
+                                       skb->nh.iph->daddr,
+                                       skb->nh.iph->saddr,
+                                       &rep.th, IPPROTO_TCP,
+                                       arg.iov[0].iov_len);
+       }
+#endif
+
        arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr,
                                      skb->nh.iph->saddr, /*XXX*/
                                      sizeof(struct tcphdr), IPPROTO_TCP, 0);
        arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 
-       ip_send_reply(tcp_socket->sk, skb, &arg, sizeof rth);
+       ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len);
 
        TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
        TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
@@ -572,15 +611,24 @@ static void tcp_v4_send_reset(struct sk_buff *skb)
    outside socket context is ugly, certainly. What can I do?
  */
 
-static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
+static void tcp_v4_send_ack(struct tcp_timewait_sock *twsk,
+                           struct sk_buff *skb, u32 seq, u32 ack,
                            u32 win, u32 ts)
 {
        struct tcphdr *th = skb->h.th;
        struct {
                struct tcphdr th;
-               u32 tsopt[TCPOLEN_TSTAMP_ALIGNED >> 2];
+               u32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
+#ifdef CONFIG_TCP_MD5SIG
+                       + (TCPOLEN_MD5SIG_ALIGNED >> 2)
+#endif
+                       ];
        } rep;
        struct ip_reply_arg arg;
+#ifdef CONFIG_TCP_MD5SIG
+       struct tcp_md5sig_key *key;
+       struct tcp_md5sig_key tw_key;
+#endif
 
        memset(&rep.th, 0, sizeof(struct tcphdr));
        memset(&arg, 0, sizeof arg);
@@ -588,12 +636,12 @@ static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
        arg.iov[0].iov_base = (unsigned char *)&rep;
        arg.iov[0].iov_len  = sizeof(rep.th);
        if (ts) {
-               rep.tsopt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
-                                    (TCPOPT_TIMESTAMP << 8) |
-                                    TCPOLEN_TIMESTAMP);
-               rep.tsopt[1] = htonl(tcp_time_stamp);
-               rep.tsopt[2] = htonl(ts);
-               arg.iov[0].iov_len = sizeof(rep);
+               rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
+                                  (TCPOPT_TIMESTAMP << 8) |
+                                  TCPOLEN_TIMESTAMP);
+               rep.opt[1] = htonl(tcp_time_stamp);
+               rep.opt[2] = htonl(ts);
+               arg.iov[0].iov_len = TCPOLEN_TSTAMP_ALIGNED;
        }
 
        /* Swap the send and the receive. */
@@ -605,6 +653,44 @@ static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
        rep.th.ack     = 1;
        rep.th.window  = htons(win);
 
+#ifdef CONFIG_TCP_MD5SIG
+       /*
+        * The SKB holds an imcoming packet, but may not have a valid ->sk
+        * pointer. This is especially the case when we're dealing with a
+        * TIME_WAIT ack, because the sk structure is long gone, and only
+        * the tcp_timewait_sock remains. So the md5 key is stashed in that
+        * structure, and we use it in preference.  I believe that (twsk ||
+        * skb->sk) holds true, but we program defensively.
+        */
+       if (!twsk && skb->sk) {
+               key = tcp_v4_md5_do_lookup(skb->sk, skb->nh.iph->daddr);
+       } else if (twsk && twsk->tw_md5_keylen) {
+               tw_key.key = twsk->tw_md5_key;
+               tw_key.keylen = twsk->tw_md5_keylen;
+               key = &tw_key;
+       } else {
+               key = NULL;
+       }
+
+       if (key) {
+               int offset = (ts) ? 3 : 0;
+
+               rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
+                                         (TCPOPT_NOP << 16) |
+                                         (TCPOPT_MD5SIG << 8) |
+                                         TCPOLEN_MD5SIG);
+               arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
+               rep.th.doff = arg.iov[0].iov_len/4;
+
+               tcp_v4_do_calc_md5_hash((__u8 *)&rep.opt[offset],
+                                       key,
+                                       skb->nh.iph->daddr,
+                                       skb->nh.iph->saddr,
+                                       &rep.th, IPPROTO_TCP,
+                                       arg.iov[0].iov_len);
+       }
+#endif
+
        arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr,
                                      skb->nh.iph->saddr, /*XXX*/
                                      arg.iov[0].iov_len, IPPROTO_TCP, 0);
@@ -618,9 +704,9 @@ static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
 {
        struct inet_timewait_sock *tw = inet_twsk(sk);
-       const struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
+       struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 
-       tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
+       tcp_v4_send_ack(tcptw, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
                        tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, tcptw->tw_ts_recent);
 
        inet_twsk_put(tw);
@@ -628,7 +714,8 @@ static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
 
 static void tcp_v4_reqsk_send_ack(struct sk_buff *skb, struct request_sock *req)
 {
-       tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
+       tcp_v4_send_ack(NULL, skb, tcp_rsk(req)->snt_isn + 1,
+                       tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
                        req->ts_recent);
 }
 
@@ -714,6 +801,461 @@ static struct ip_options *tcp_v4_save_options(struct sock *sk,
        return dopt;
 }
 
+#ifdef CONFIG_TCP_MD5SIG
+/*
+ * RFC2385 MD5 checksumming requires a mapping of
+ * IP address->MD5 Key.
+ * We need to maintain these in the sk structure.
+ */
+
+/* Find the Key structure for an address.  */
+static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
+{
+       struct tcp_sock *tp = tcp_sk(sk);
+       int i;
+
+       if (!tp->md5sig_info || !tp->md5sig_info->entries4)
+               return NULL;
+       for (i = 0; i < tp->md5sig_info->entries4; i++) {
+               if (tp->md5sig_info->keys4[i].addr == addr)
+                       return (struct tcp_md5sig_key *)&tp->md5sig_info->keys4[i];
+       }
+       return NULL;
+}
+
+struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
+                                        struct sock *addr_sk)
+{
+       return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->daddr);
+}
+
+EXPORT_SYMBOL(tcp_v4_md5_lookup);
+
+struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
+                                              struct request_sock *req)
+{
+       return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
+}
+
+/* This can be called on a newly created socket, from other files */
+int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
+                     u8 *newkey, u8 newkeylen)
+{
+       /* Add Key to the list */
+       struct tcp4_md5sig_key *key;
+       struct tcp_sock *tp = tcp_sk(sk);
+       struct tcp4_md5sig_key *keys;
+
+       key = (struct tcp4_md5sig_key *) tcp_v4_md5_do_lookup(sk, addr);
+       if (key) {
+               /* Pre-existing entry - just update that one. */
+               kfree (key->key);
+               key->key = newkey;
+               key->keylen = newkeylen;
+       } else {
+               if (!tp->md5sig_info) {
+                       tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
+                       if (!tp->md5sig_info) {
+                               kfree(newkey);
+                               return -ENOMEM;
+                       }
+               }
+               if (tcp_alloc_md5sig_pool() == NULL) {
+                       kfree(newkey);
+                       return -ENOMEM;
+               }
+               if (tp->md5sig_info->alloced4 == tp->md5sig_info->entries4) {
+                       keys = kmalloc((sizeof(struct tcp4_md5sig_key) *
+                                      (tp->md5sig_info->entries4 + 1)), GFP_ATOMIC);
+                       if (!keys) {
+                               kfree(newkey);
+                               tcp_free_md5sig_pool();
+                               return -ENOMEM;
+                       }
+
+                       if (tp->md5sig_info->entries4)
+                               memcpy(keys, tp->md5sig_info->keys4,
+                                      (sizeof (struct tcp4_md5sig_key) *
+                                       tp->md5sig_info->entries4));
+
+                       /* Free old key list, and reference new one */
+                       if (tp->md5sig_info->keys4)
+                               kfree(tp->md5sig_info->keys4);
+                       tp->md5sig_info->keys4 = keys;
+                       tp->md5sig_info->alloced4++;
+               }
+               tp->md5sig_info->entries4++;
+               tp->md5sig_info->keys4[tp->md5sig_info->entries4 - 1].addr = addr;
+               tp->md5sig_info->keys4[tp->md5sig_info->entries4 - 1].key = newkey;
+               tp->md5sig_info->keys4[tp->md5sig_info->entries4 - 1].keylen = newkeylen;
+       }
+       return 0;
+}
+
+EXPORT_SYMBOL(tcp_v4_md5_do_add);
+
+static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
+                              u8 *newkey, u8 newkeylen)
+{
+       return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->daddr,
+                                newkey, newkeylen);
+}
+
+int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
+{
+       struct tcp_sock *tp = tcp_sk(sk);
+       int i;
+
+       for (i = 0; i < tp->md5sig_info->entries4; i++) {
+               if (tp->md5sig_info->keys4[i].addr == addr) {
+                       /* Free the key */
+                       kfree(tp->md5sig_info->keys4[i].key);
+                       tp->md5sig_info->entries4--;
+
+                       if (tp->md5sig_info->entries4 == 0) {
+                               kfree(tp->md5sig_info->keys4);
+                               tp->md5sig_info->keys4 = NULL;
+                       } else {
+                               /* Need to do some manipulation */
+                               if (tp->md5sig_info->entries4 != i)
+                                       memcpy(&tp->md5sig_info->keys4[i],
+                                              &tp->md5sig_info->keys4[i+1],
+                                              (tp->md5sig_info->entries4 - i)
+                                               * sizeof (struct tcp4_md5sig_key));
+                       }
+                       tcp_free_md5sig_pool();
+                       return 0;
+               }
+       }
+       return -ENOENT;
+}
+
+EXPORT_SYMBOL(tcp_v4_md5_do_del);
+
+static void tcp_v4_clear_md5_list (struct sock *sk)
+{
+       struct tcp_sock *tp = tcp_sk(sk);
+
+       /* Free each key, then the set of key keys,
+        * the crypto element, and then decrement our
+        * hold on the last resort crypto.
+        */
+       if (tp->md5sig_info->entries4) {
+               int i;
+               for (i = 0; i < tp->md5sig_info->entries4; i++)
+                       kfree(tp->md5sig_info->keys4[i].key);
+               tp->md5sig_info->entries4 = 0;
+               tcp_free_md5sig_pool();
+       }
+       if (tp->md5sig_info->keys4) {
+               kfree(tp->md5sig_info->keys4);
+               tp->md5sig_info->keys4 = NULL;
+               tp->md5sig_info->alloced4  = 0;
+       }
+}
+
+static int tcp_v4_parse_md5_keys (struct sock *sk, char __user *optval,
+                                 int optlen)
+{
+       struct tcp_md5sig cmd;
+       struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
+       u8 *newkey;
+
+       if (optlen < sizeof(cmd))
+               return -EINVAL;
+
+       if (copy_from_user (&cmd, optval, sizeof(cmd)))
+               return -EFAULT;
+
+       if (sin->sin_family != AF_INET)
+               return -EINVAL;
+
+       if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
+               if (!tcp_sk(sk)->md5sig_info)
+                       return -ENOENT;
+               return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
+       }
+
+       if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
+               return -EINVAL;
+
+       if (!tcp_sk(sk)->md5sig_info) {
+               struct tcp_sock *tp = tcp_sk(sk);
+               struct tcp_md5sig_info *p;
+
+               p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
+               if (!p)
+                       return -EINVAL;
+
+               tp->md5sig_info = p;
+
+       }
+
+       newkey = kmalloc(cmd.tcpm_keylen, GFP_KERNEL);
+       if (!newkey)
+               return -ENOMEM;
+       memcpy(newkey, cmd.tcpm_key, cmd.tcpm_keylen);
+       return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
+                                newkey, cmd.tcpm_keylen);
+}
+
+static int tcp_v4_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
+                                  __be32 saddr, __be32 daddr,
+                                  struct tcphdr *th, int protocol,
+                                  int tcplen)
+{
+       struct scatterlist sg[4];
+       __u16 data_len;
+       int block = 0;
+#ifdef CONFIG_TCP_MD5SIG_DEBUG
+       int i;
+#endif
+       __u16 old_checksum;
+       struct tcp_md5sig_pool *hp;
+       struct tcp4_pseudohdr *bp;
+       struct hash_desc *desc;
+       int err;
+       unsigned int nbytes = 0;
+
+       /*
+        * Okay, so RFC2385 is turned on for this connection,
+        * so we need to generate the MD5 hash for the packet now.
+        */
+
+       hp = tcp_get_md5sig_pool();
+       if (!hp)
+               goto clear_hash_noput;
+
+       bp = &hp->md5_blk.ip4;
+       desc = &hp->md5_desc;
+
+       /*
+        * 1. the TCP pseudo-header (in the order: source IP address,
+        * destination IP address, zero-padded protocol number, and
+        * segment length)
+        */
+       bp->saddr = saddr;
+       bp->daddr = daddr;
+       bp->pad = 0;
+       bp->protocol = protocol;
+       bp->len = htons(tcplen);
+       sg_set_buf(&sg[block++], bp, sizeof(*bp));
+       nbytes += sizeof(*bp);
+
+#ifdef CONFIG_TCP_MD5SIG_DEBUG
+       printk("Calcuating hash for: ");
+       for (i = 0; i < sizeof (*bp); i++)
+               printk ("%02x ", (unsigned int)((unsigned char *)bp)[i]);
+       printk(" ");
+#endif
+
+       /* 2. the TCP header, excluding options, and assuming a
+        * checksum of zero/
+        */
+       old_checksum = th->check;
+       th->check = 0;
+       sg_set_buf(&sg[block++], th, sizeof(struct tcphdr));
+       nbytes += sizeof(struct tcphdr);
+#ifdef CONFIG_TCP_MD5SIG_DEBUG
+       for (i = 0; i < sizeof (struct tcphdr); i++)
+               printk (" %02x", (unsigned int)((unsigned char *)th)[i]);
+#endif
+       /* 3. the TCP segment data (if any) */
+       data_len = tcplen - (th->doff << 2);
+       if (data_len > 0) {
+               unsigned char *data = (unsigned char *)th + (th->doff << 2);
+               sg_set_buf(&sg[block++], data, data_len);
+               nbytes += data_len;
+       }
+
+       /* 4. an independently-specified key or password, known to both
+        * TCPs and presumably connection-specific
+        */
+       sg_set_buf(&sg[block++], key->key, key->keylen);
+       nbytes += key->keylen;
+
+#ifdef CONFIG_TCP_MD5SIG_DEBUG
+       printk ("  and password: ");
+       for (i = 0; i < key->keylen; i++)
+               printk ("%02x ", (unsigned int)key->key[i]);
+#endif
+
+       /* Now store the Hash into the packet */
+       err = crypto_hash_init(desc);
+       if (err)
+               goto clear_hash;
+       err = crypto_hash_update(desc, sg, nbytes);
+       if (err)
+               goto clear_hash;
+       err = crypto_hash_final(desc, md5_hash);
+       if (err)
+               goto clear_hash;
+
+       /* Reset header, and free up the crypto */
+       tcp_put_md5sig_pool();
+       th->check = old_checksum;
+
+out:
+#ifdef CONFIG_TCP_MD5SIG_DEBUG
+       printk(" result:");
+       for (i = 0; i < 16; i++)
+               printk (" %02x", (unsigned int)(((u8*)md5_hash)[i]));
+       printk("\n");
+#endif
+       return 0;
+clear_hash:
+       tcp_put_md5sig_pool();
+clear_hash_noput:
+       memset(md5_hash, 0, 16);
+       goto out;
+}
+
+int tcp_v4_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
+                        struct sock *sk,
+                        struct dst_entry *dst,
+                        struct request_sock *req,
+                        struct tcphdr *th, int protocol,
+                        int tcplen)
+{
+       __be32 saddr, daddr;
+
+       if (sk) {
+               saddr = inet_sk(sk)->saddr;
+               daddr = inet_sk(sk)->daddr;
+       } else {
+               struct rtable *rt = (struct rtable *)dst;
+               BUG_ON(!rt);
+               saddr = rt->rt_src;
+               daddr = rt->rt_dst;
+       }
+       return tcp_v4_do_calc_md5_hash(md5_hash, key,
+                                      saddr, daddr,
+                                      th, protocol, tcplen);
+}
+
+EXPORT_SYMBOL(tcp_v4_calc_md5_hash);
+
+static int tcp_v4_inbound_md5_hash (struct sock *sk, struct sk_buff *skb)
+{
+       /*
+        * This gets called for each TCP segment that arrives
+        * so we want to be efficient.
+        * We have 3 drop cases:
+        * o No MD5 hash and one expected.
+        * o MD5 hash and we're not expecting one.
+        * o MD5 hash and its wrong.
+        */
+       __u8 *hash_location = NULL;
+       struct tcp_md5sig_key *hash_expected;
+       struct iphdr *iph = skb->nh.iph;
+       struct tcphdr *th = skb->h.th;
+       int length = (th->doff << 2) - sizeof (struct tcphdr);
+       int genhash;
+       unsigned char *ptr;
+       unsigned char newhash[16];
+
+       hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
+
+       /*
+        * If the TCP option length is less than the TCP_MD5SIG
+        * option length, then we can shortcut
+        */
+       if (length < TCPOLEN_MD5SIG) {
+               if (hash_expected)
+                       return 1;
+               else
+                       return 0;
+       }
+
+       /* Okay, we can't shortcut - we have to grub through the options */
+       ptr = (unsigned char *)(th + 1);
+       while (length > 0) {
+               int opcode = *ptr++;
+               int opsize;
+
+               switch (opcode) {
+               case TCPOPT_EOL:
+                       goto done_opts;
+               case TCPOPT_NOP:
+                       length--;
+                       continue;
+               default:
+                       opsize = *ptr++;
+                       if (opsize < 2)
+                               goto done_opts;
+                       if (opsize > length)
+                               goto done_opts;
+
+                       if (opcode == TCPOPT_MD5SIG) {
+                               hash_location = ptr;
+                               goto done_opts;
+                       }
+               }
+               ptr += opsize-2;
+               length -= opsize;
+       }
+done_opts:
+       /* We've parsed the options - do we have a hash? */
+       if (!hash_expected && !hash_location)
+               return 0;
+
+       if (hash_expected && !hash_location) {
+               if (net_ratelimit()) {
+                       printk(KERN_INFO "MD5 Hash NOT expected but found "
+                              "(" NIPQUAD_FMT ", %d)->(" NIPQUAD_FMT ", %d)\n",
+                              NIPQUAD (iph->saddr), ntohs(th->source),
+                              NIPQUAD (iph->daddr), ntohs(th->dest));
+               }
+               return 1;
+       }
+
+       if (!hash_expected && hash_location) {
+               if (net_ratelimit()) {
+                       printk(KERN_INFO "MD5 Hash NOT expected but found "
+                              "(" NIPQUAD_FMT ", %d)->(" NIPQUAD_FMT ", %d)\n",
+                              NIPQUAD (iph->saddr), ntohs(th->source),
+                              NIPQUAD (iph->daddr), ntohs(th->dest));
+               }
+               return 1;
+       }
+
+       /* Okay, so this is hash_expected and hash_location -
+        * so we need to calculate the checksum.
+        */
+       genhash = tcp_v4_do_calc_md5_hash(newhash,
+                                         hash_expected,
+                                         iph->saddr, iph->daddr,
+                                         th, sk->sk_protocol,
+                                         skb->len);
+
+       if (genhash || memcmp(hash_location, newhash, 16) != 0) {
+               if (net_ratelimit()) {
+                       printk(KERN_INFO "MD5 Hash failed for "
+                              "(" NIPQUAD_FMT ", %d)->(" NIPQUAD_FMT ", %d)%s\n",
+                              NIPQUAD (iph->saddr), ntohs(th->source),
+                              NIPQUAD (iph->daddr), ntohs(th->dest),
+                              genhash ? " tcp_v4_calc_md5_hash failed" : "");
+#ifdef CONFIG_TCP_MD5SIG_DEBUG
+                       do {
+                               int i;
+                               printk("Received: ");
+                               for (i = 0; i < 16; i++)
+                                       printk("%02x ", 0xff & (int)hash_location[i]);
+                               printk("\n");
+                               printk("Calculated: ");
+                               for (i = 0; i < 16; i++)
+                                       printk("%02x ", 0xff & (int)newhash[i]);
+                               printk("\n");
+                       } while(0);
+#endif
+               }
+               return 1;
+       }
+       return 0;
+}
+
+#endif
+
 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
        .family         =       PF_INET,
        .obj_size       =       sizeof(struct tcp_request_sock),
@@ -723,9 +1265,16 @@ struct request_sock_ops tcp_request_sock_ops __read_mostly = {
        .send_reset     =       tcp_v4_send_reset,
 };
 
+struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
+#ifdef CONFIG_TCP_MD5SIG
+       .md5_lookup     =       tcp_v4_reqsk_md5_lookup,
+#endif
+};
+
 static struct timewait_sock_ops tcp_timewait_sock_ops = {
        .twsk_obj_size  = sizeof(struct tcp_timewait_sock),
        .twsk_unique    = tcp_twsk_unique,
+       .twsk_destructor= tcp_twsk_destructor,
 };
 
 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
@@ -773,6 +1322,10 @@ int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
        if (!req)
                goto drop;
 
+#ifdef CONFIG_TCP_MD5SIG
+       tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
+#endif
+
        tcp_clear_options(&tmp_opt);
        tmp_opt.mss_clamp = 536;
        tmp_opt.user_mss  = tcp_sk(sk)->rx_opt.user_mss;
@@ -891,6 +1444,9 @@ struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
        struct inet_sock *newinet;
        struct tcp_sock *newtp;
        struct sock *newsk;
+#ifdef CONFIG_TCP_MD5SIG
+       struct tcp_md5sig_key *key;
+#endif
 
        if (sk_acceptq_is_full(sk))
                goto exit_overflow;
@@ -925,6 +1481,24 @@ struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
        newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
        tcp_initialize_rcv_mss(newsk);
 
+#ifdef CONFIG_TCP_MD5SIG
+       /* Copy over the MD5 key from the original socket */
+       if ((key = tcp_v4_md5_do_lookup(sk, newinet->daddr)) != NULL) {
+               /*
+                * We're using one, so create a matching key
+                * on the newsk structure. If we fail to get
+                * memory, then we end up not copying the key
+                * across. Shucks.
+                */
+               char *newkey = kmalloc(key->keylen, GFP_ATOMIC);
+               if (newkey) {
+                       memcpy(newkey, key->key, key->keylen);
+                       tcp_v4_md5_do_add(newsk, inet_sk(sk)->daddr,
+                                         newkey, key->keylen);
+               }
+       }
+#endif
+
        __inet_hash(&tcp_hashinfo, newsk, 0);
        __inet_inherit_port(&tcp_hashinfo, sk, newsk);
 
@@ -1000,10 +1574,24 @@ static int tcp_v4_checksum_init(struct sk_buff *skb)
  */
 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
 {
+       struct sock *rsk;
+#ifdef CONFIG_TCP_MD5SIG
+       /*
+        * We really want to reject the packet as early as possible
+        * if:
+        *  o We're expecting an MD5'd packet and this is no MD5 tcp option
+        *  o There is an MD5 option and we're not expecting one
+        */
+       if (tcp_v4_inbound_md5_hash (sk, skb))
+               goto discard;
+#endif
+
        if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
                TCP_CHECK_TIMER(sk);
-               if (tcp_rcv_established(sk, skb, skb->h.th, skb->len))
+               if (tcp_rcv_established(sk, skb, skb->h.th, skb->len)) {
+                       rsk = sk;
                        goto reset;
+               }
                TCP_CHECK_TIMER(sk);
                return 0;
        }
@@ -1017,20 +1605,24 @@ int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
                        goto discard;
 
                if (nsk != sk) {
-                       if (tcp_child_process(sk, nsk, skb))
+                       if (tcp_child_process(sk, nsk, skb)) {
+                               rsk = nsk;
                                goto reset;
+                       }
                        return 0;
                }
        }
 
        TCP_CHECK_TIMER(sk);
-       if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len))
+       if (tcp_rcv_state_process(sk, skb, skb->h.th, skb->len)) {
+               rsk = sk;
                goto reset;
+       }
        TCP_CHECK_TIMER(sk);
        return 0;
 
 reset:
-       tcp_v4_send_reset(skb);
+       tcp_v4_send_reset(rsk, skb);
 discard:
        kfree_skb(skb);
        /* Be careful here. If this function gets more complicated and
@@ -1139,7 +1731,7 @@ no_tcp_socket:
 bad_packet:
                TCP_INC_STATS_BH(TCP_MIB_INERRS);
        } else {
-               tcp_v4_send_reset(skb);
+               tcp_v4_send_reset(NULL, skb);
        }
 
 discard_it:
@@ -1262,6 +1854,15 @@ struct inet_connection_sock_af_ops ipv4_specific = {
 #endif
 };
 
+struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
+#ifdef CONFIG_TCP_MD5SIG
+       .md5_lookup             = tcp_v4_md5_lookup,
+       .calc_md5_hash          = tcp_v4_calc_md5_hash,
+       .md5_add                = tcp_v4_md5_add_func,
+       .md5_parse              = tcp_v4_parse_md5_keys,
+#endif
+};
+
 /* NOTE: A lot of things set to zero explicitly by call to
  *       sk_alloc() so need not be done here.
  */
@@ -1301,6 +1902,9 @@ static int tcp_v4_init_sock(struct sock *sk)
 
        icsk->icsk_af_ops = &ipv4_specific;
        icsk->icsk_sync_mss = tcp_sync_mss;
+#ifdef CONFIG_TCP_MD5SIG
+       tp->af_specific = &tcp_sock_ipv4_specific;
+#endif
 
        sk->sk_sndbuf = sysctl_tcp_wmem[1];
        sk->sk_rcvbuf = sysctl_tcp_rmem[1];
@@ -1324,6 +1928,15 @@ int tcp_v4_destroy_sock(struct sock *sk)
        /* Cleans up our, hopefully empty, out_of_order_queue. */
        __skb_queue_purge(&tp->out_of_order_queue);
 
+#ifdef CONFIG_TCP_MD5SIG
+       /* Clean up the MD5 key list, if any */
+       if (tp->md5sig_info) {
+               tcp_v4_clear_md5_list(sk);
+               kfree(tp->md5sig_info);
+               tp->md5sig_info = NULL;
+       }
+#endif
+
 #ifdef CONFIG_NET_DMA
        /* Cleans up our sk_async_wait_queue */
        __skb_queue_purge(&sk->sk_async_wait_queue);