net/ipv4/syncookies.c

   1 /*
   2  *  Syncookies implementation for the Linux kernel
   3  *
   4  *  Copyright (C) 1997 Andi Kleen
   5  *  Based on ideas by D.J.Bernstein and Eric Schenk.
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  *
  12  *  $Id: syncookies.c,v 1.18 2002/02/01 22:01:04 davem Exp $
  13  *
  14  *  Missing: IPv6 support.
  15  */
  16
  17 #include <linux/tcp.h>
  18 #include <linux/slab.h>
  19 #include <linux/random.h>
  20 #include <linux/kernel.h>
  21 #include <net/tcp.h>
  22
  23 extern int sysctl_tcp_syncookies;
  24
  25 /*
  26  * This table has to be sorted and terminated with (__u16)-1.
  27  * XXX generate a better table.
  28  * Unresolved Issues: HIPPI with a 64k MSS is not well supported.
  29  */
  30 static __u16 const msstab[] = {
  31         64 - 1,
  32         256 - 1,
  33         512 - 1,
  34         536 - 1,
  35         1024 - 1,
  36         1440 - 1,
  37         1460 - 1,
  38         4312 - 1,
  39         (__u16)-1
  40 };
  41 /* The number doesn't include the -1 terminator */
  42 #define NUM_MSS (ARRAY_SIZE(msstab) - 1)
  43
  44 /*
  45  * Generate a syncookie.  mssp points to the mss, which is returned
  46  * rounded down to the value encoded in the cookie.
  47  */
  48 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
  49 {
  50         struct tcp_opt *tp = tcp_sk(sk);
  51         int mssind;
  52         const __u16 mss = *mssp;
  53
  54
  55         tp->last_synq_overflow = jiffies;
  56
  57         /* XXX sort msstab[] by probability?  Binary search? */
  58         for (mssind = 0; mss > msstab[mssind + 1]; mssind++)
  59                 ;
  60         *mssp = msstab[mssind] + 1;
  61
  62         NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESSENT);
  63
  64         return secure_tcp_syn_cookie(skb->nh.iph->saddr, skb->nh.iph->daddr,
  65                                      skb->h.th->source, skb->h.th->dest,
  66                                      ntohl(skb->h.th->seq),
  67                                      jiffies / (HZ * 60), mssind);
  68 }
  69
  70 /*
  71  * This (misnamed) value is the age of syncookie which is permitted.
  72  * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
  73  * sysctl_tcp_retries1. It's a rather complicated formula (exponential
  74  * backoff) to compute at runtime so it's currently hardcoded here.
  75  */
  76 #define COUNTER_TRIES 4
  77 /*
  78  * Check if a ack sequence number is a valid syncookie.
  79  * Return the decoded mss if it is, or 0 if not.
  80  */
  81 static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
  82 {
  83         __u32 seq;
  84         __u32 mssind;
  85
  86         seq = ntohl(skb->h.th->seq)-1;
  87         mssind = check_tcp_syn_cookie(cookie,
  88                                       skb->nh.iph->saddr, skb->nh.iph->daddr,
  89                                       skb->h.th->source, skb->h.th->dest,
  90                                       seq, jiffies / (HZ * 60), COUNTER_TRIES);
  91
  92         return mssind < NUM_MSS ? msstab[mssind] + 1 : 0;
  93 }
  94
  95 extern struct or_calltable or_ipv4;
  96
  97 static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
  98                                            struct open_request *req,
  99                                            struct dst_entry *dst)
 100 {
 101         struct tcp_opt *tp = tcp_sk(sk);
 102         struct sock *child;
 103
 104         child = tp->af_specific->syn_recv_sock(sk, skb, req, dst);
 105         if (child) {
 106                 sk_set_owner(child, sk->sk_owner);
 107                 tcp_acceptq_queue(sk, req, child);
 108         } else
 109                 tcp_openreq_free(req);
 110
 111         return child;
 112 }
 113
 114 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
 115                              struct ip_options *opt)
 116 {
 117         struct tcp_opt *tp = tcp_sk(sk);
 118         __u32 cookie = ntohl(skb->h.th->ack_seq) - 1;
 119         struct sock *ret = sk;
 120         struct open_request *req;
 121         int mss;
 122         struct rtable *rt;
 123         __u8 rcv_wscale;
 124
 125         if (!sysctl_tcp_syncookies || !skb->h.th->ack)
 126                 goto out;
 127
 128         if (time_after(jiffies, tp->last_synq_overflow + TCP_TIMEOUT_INIT) ||
 129             (mss = cookie_check(skb, cookie)) == 0) {
 130                 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESFAILED);
 131                 goto out;
 132         }
 133
 134         NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESRECV);
 135
 136         req = tcp_openreq_alloc();
 137         ret = NULL;
 138         if (!req)
 139                 goto out;
 140
 141         req->rcv_isn            = htonl(skb->h.th->seq) - 1;
 142         req->snt_isn            = cookie;
 143         req->mss                = mss;
 144         req->rmt_port           = skb->h.th->source;
 145         req->af.v4_req.loc_addr = skb->nh.iph->daddr;
 146         req->af.v4_req.rmt_addr = skb->nh.iph->saddr;
 147         req->class              = &or_ipv4; /* for savety */
 148         req->af.v4_req.opt      = NULL;
 149
 150         /* We throwed the options of the initial SYN away, so we hope
 151          * the ACK carries the same options again (see RFC1122 4.2.3.8)
 152          */
 153         if (opt && opt->optlen) {
 154                 int opt_size = sizeof(struct ip_options) + opt->optlen;
 155
 156                 req->af.v4_req.opt = kmalloc(opt_size, GFP_ATOMIC);
 157                 if (req->af.v4_req.opt) {
 158                         if (ip_options_echo(req->af.v4_req.opt, skb)) {
 159                                 kfree(req->af.v4_req.opt);
 160                                 req->af.v4_req.opt = NULL;
 161                         }
 162                 }
 163         }
 164
 165         req->snd_wscale = req->rcv_wscale = req->tstamp_ok = 0;
 166         req->wscale_ok  = req->sack_ok = 0;
 167         req->expires    = 0UL;
 168         req->retrans    = 0;
 169
 170         /*
 171          * We need to lookup the route here to get at the correct
 172          * window size. We should better make sure that the window size
 173          * hasn't changed since we received the original syn, but I see
 174          * no easy way to do this.
 175          */
 176         {
 177                 struct flowi fl = { .nl_u = { .ip4_u =
 178                                               { .daddr = ((opt && opt->srr) ?
 179                                                           opt->faddr :
 180                                                           req->af.v4_req.rmt_addr),
 181                                                 .saddr = req->af.v4_req.loc_addr,
 182                                                 .tos = RT_CONN_FLAGS(sk) } },
 183                                     .proto = IPPROTO_TCP,
 184                                     .uli_u = { .ports =
 185                                                { .sport = skb->h.th->dest,
 186                                                  .dport = skb->h.th->source } } };
 187                 if (ip_route_output_key(&rt, &fl)) {
 188                         tcp_openreq_free(req);
 189                         goto out;
 190                 }
 191         }
 192
 193         /* Try to redo what tcp_v4_send_synack did. */
 194         req->window_clamp = dst_metric(&rt->u.dst, RTAX_WINDOW);
 195         tcp_select_initial_window(tcp_full_space(sk), req->mss,
 196                                   &req->rcv_wnd, &req->window_clamp,
 197                                   0, &rcv_wscale);
 198         /* BTW win scale with syncookies is 0 by definition */
 199         req->rcv_wscale   = rcv_wscale;
 200
 201         ret = get_cookie_sock(sk, skb, req, &rt->u.dst);
 202 out:    return ret;
 203 }