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
  15 #include <linux/tcp.h>
  16 #include <linux/slab.h>
  17 #include <linux/random.h>
  18 #include <linux/cryptohash.h>
  19 #include <linux/kernel.h>
  20 #include <net/tcp.h>
  21
  22 extern int sysctl_tcp_syncookies;
  23
  24 __u32 syncookie_secret[2][16-3+SHA_DIGEST_WORDS];
  25 EXPORT_SYMBOL(syncookie_secret);
  26
  27 static __init int init_syncookies(void)
  28 {
  29         get_random_bytes(syncookie_secret, sizeof(syncookie_secret));
  30         return 0;
  31 }
  32 __initcall(init_syncookies);
  33
  34 #define COOKIEBITS 24   /* Upper bits store count */
  35 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
  36
  37 static DEFINE_PER_CPU(__u32, cookie_scratch)[16 + 5 + SHA_WORKSPACE_WORDS];
  38
  39 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
  40                        u32 count, int c)
  41 {
  42         __u32 *tmp = __get_cpu_var(cookie_scratch);
  43
  44         memcpy(tmp + 3, syncookie_secret[c], sizeof(syncookie_secret[c]));
  45         tmp[0] = (__force u32)saddr;
  46         tmp[1] = (__force u32)daddr;
  47         tmp[2] = ((__force u32)sport << 16) + (__force u32)dport;
  48         tmp[3] = count;
  49         sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
  50
  51         return tmp[17];
  52 }
  53
  54 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
  55                                    __be16 dport, __u32 sseq, __u32 count,
  56                                    __u32 data)
  57 {
  58         /*
  59          * Compute the secure sequence number.
  60          * The output should be:
  61          *   HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
  62          *      + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
  63          * Where sseq is their sequence number and count increases every
  64          * minute by 1.
  65          * As an extra hack, we add a small "data" value that encodes the
  66          * MSS into the second hash value.
  67          */
  68
  69         return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
  70                 sseq + (count << COOKIEBITS) +
  71                 ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
  72                  & COOKIEMASK));
  73 }
  74
  75 /*
  76  * This retrieves the small "data" value from the syncookie.
  77  * If the syncookie is bad, the data returned will be out of
  78  * range.  This must be checked by the caller.
  79  *
  80  * The count value used to generate the cookie must be within
  81  * "maxdiff" if the current (passed-in) "count".  The return value
  82  * is (__u32)-1 if this test fails.
  83  */
  84 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
  85                                   __be16 sport, __be16 dport, __u32 sseq,
  86                                   __u32 count, __u32 maxdiff)
  87 {
  88         __u32 diff;
  89
  90         /* Strip away the layers from the cookie */
  91         cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
  92
  93         /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
  94         diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS);
  95         if (diff >= maxdiff)
  96                 return (__u32)-1;
  97
  98         return (cookie -
  99                 cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
 100                 & COOKIEMASK;   /* Leaving the data behind */
 101 }
 102
 103 /*
 104  * This table has to be sorted and terminated with (__u16)-1.
 105  * XXX generate a better table.
 106  * Unresolved Issues: HIPPI with a 64k MSS is not well supported.
 107  */
 108 static __u16 const msstab[] = {
 109         64 - 1,
 110         256 - 1,
 111         512 - 1,
 112         536 - 1,
 113         1024 - 1,
 114         1440 - 1,
 115         1460 - 1,
 116         4312 - 1,
 117         (__u16)-1
 118 };
 119 /* The number doesn't include the -1 terminator */
 120 #define NUM_MSS (ARRAY_SIZE(msstab) - 1)
 121
 122 /*
 123  * Generate a syncookie.  mssp points to the mss, which is returned
 124  * rounded down to the value encoded in the cookie.
 125  */
 126 __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
 127 {
 128         struct tcp_sock *tp = tcp_sk(sk);
 129         const struct iphdr *iph = ip_hdr(skb);
 130         const struct tcphdr *th = tcp_hdr(skb);
 131         int mssind;
 132         const __u16 mss = *mssp;
 133
 134         tp->last_synq_overflow = jiffies;
 135
 136         /* XXX sort msstab[] by probability?  Binary search? */
 137         for (mssind = 0; mss > msstab[mssind + 1]; mssind++)
 138                 ;
 139         *mssp = msstab[mssind] + 1;
 140
 141         NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESSENT);
 142
 143         return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
 144                                      th->source, th->dest, ntohl(th->seq),
 145                                      jiffies / (HZ * 60), mssind);
 146 }
 147
 148 /*
 149  * This (misnamed) value is the age of syncookie which is permitted.
 150  * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
 151  * sysctl_tcp_retries1. It's a rather complicated formula (exponential
 152  * backoff) to compute at runtime so it's currently hardcoded here.
 153  */
 154 #define COUNTER_TRIES 4
 155 /*
 156  * Check if a ack sequence number is a valid syncookie.
 157  * Return the decoded mss if it is, or 0 if not.
 158  */
 159 static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
 160 {
 161         const struct iphdr *iph = ip_hdr(skb);
 162         const struct tcphdr *th = tcp_hdr(skb);
 163         __u32 seq = ntohl(th->seq) - 1;
 164         __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
 165                                             th->source, th->dest, seq,
 166                                             jiffies / (HZ * 60),
 167                                             COUNTER_TRIES);
 168
 169         return mssind < NUM_MSS ? msstab[mssind] + 1 : 0;
 170 }
 171
 172 static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
 173                                            struct request_sock *req,
 174                                            struct dst_entry *dst)
 175 {
 176         struct inet_connection_sock *icsk = inet_csk(sk);
 177         struct sock *child;
 178
 179         child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
 180         if (child)
 181                 inet_csk_reqsk_queue_add(sk, req, child);
 182         else
 183                 reqsk_free(req);
 184
 185         return child;
 186 }
 187
 188 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
 189                              struct ip_options *opt)
 190 {
 191         struct inet_request_sock *ireq;
 192         struct tcp_request_sock *treq;
 193         struct tcp_sock *tp = tcp_sk(sk);
 194         const struct tcphdr *th = tcp_hdr(skb);
 195         __u32 cookie = ntohl(th->ack_seq) - 1;
 196         struct sock *ret = sk;
 197         struct request_sock *req;
 198         int mss;
 199         struct rtable *rt;
 200         __u8 rcv_wscale;
 201
 202         if (!sysctl_tcp_syncookies || !th->ack)
 203                 goto out;
 204
 205         if (time_after(jiffies, tp->last_synq_overflow + TCP_TIMEOUT_INIT) ||
 206             (mss = cookie_check(skb, cookie)) == 0) {
 207                 NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESFAILED);
 208                 goto out;
 209         }
 210
 211         NET_INC_STATS_BH(LINUX_MIB_SYNCOOKIESRECV);
 212
 213         ret = NULL;
 214         req = reqsk_alloc(&tcp_request_sock_ops); /* for safety */
 215         if (!req)
 216                 goto out;
 217
 218         if (security_inet_conn_request(sk, skb, req)) {
 219                 reqsk_free(req);
 220                 goto out;
 221         }
 222         ireq = inet_rsk(req);
 223         treq = tcp_rsk(req);
 224         treq->rcv_isn           = ntohl(th->seq) - 1;
 225         treq->snt_isn           = cookie;
 226         req->mss                = mss;
 227         ireq->rmt_port          = th->source;
 228         ireq->loc_addr          = ip_hdr(skb)->daddr;
 229         ireq->rmt_addr          = ip_hdr(skb)->saddr;
 230         ireq->opt               = NULL;
 231
 232         /* We throwed the options of the initial SYN away, so we hope
 233          * the ACK carries the same options again (see RFC1122 4.2.3.8)
 234          */
 235         if (opt && opt->optlen) {
 236                 int opt_size = sizeof(struct ip_options) + opt->optlen;
 237
 238                 ireq->opt = kmalloc(opt_size, GFP_ATOMIC);
 239                 if (ireq->opt != NULL && ip_options_echo(ireq->opt, skb)) {
 240                         kfree(ireq->opt);
 241                         ireq->opt = NULL;
 242                 }
 243         }
 244
 245         ireq->snd_wscale = ireq->rcv_wscale = ireq->tstamp_ok = 0;
 246         ireq->wscale_ok  = ireq->sack_ok = 0;
 247         req->expires    = 0UL;
 248         req->retrans    = 0;
 249
 250         /*
 251          * We need to lookup the route here to get at the correct
 252          * window size. We should better make sure that the window size
 253          * hasn't changed since we received the original syn, but I see
 254          * no easy way to do this.
 255          */
 256         {
 257                 struct flowi fl = { .nl_u = { .ip4_u =
 258                                               { .daddr = ((opt && opt->srr) ?
 259                                                           opt->faddr :
 260                                                           ireq->rmt_addr),
 261                                                 .saddr = ireq->loc_addr,
 262                                                 .tos = RT_CONN_FLAGS(sk) } },
 263                                     .proto = IPPROTO_TCP,
 264                                     .uli_u = { .ports =
 265                                                { .sport = th->dest,
 266                                                  .dport = th->source } } };
 267                 security_req_classify_flow(req, &fl);
 268                 if (ip_route_output_key(&init_net, &rt, &fl)) {
 269                         reqsk_free(req);
 270                         goto out;
 271                 }
 272         }
 273
 274         /* Try to redo what tcp_v4_send_synack did. */
 275         req->window_clamp = dst_metric(&rt->u.dst, RTAX_WINDOW);
 276         tcp_select_initial_window(tcp_full_space(sk), req->mss,
 277                                   &req->rcv_wnd, &req->window_clamp,
 278                                   0, &rcv_wscale);
 279         /* BTW win scale with syncookies is 0 by definition */
 280         ireq->rcv_wscale  = rcv_wscale;
 281
 282         ret = get_cookie_sock(sk, skb, req, &rt->u.dst);
 283 out:    return ret;
 284 }