linux/net/ipv4/syncookies.c
<<
>>
Prefs
   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
  13#include <linux/tcp.h>
  14#include <linux/slab.h>
  15#include <linux/random.h>
  16#include <linux/cryptohash.h>
  17#include <linux/kernel.h>
  18#include <net/tcp.h>
  19#include <net/route.h>
  20
  21/* Timestamps: lowest 9 bits store TCP options */
  22#define TSBITS 9
  23#define TSMASK (((__u32)1 << TSBITS) - 1)
  24
  25extern int sysctl_tcp_syncookies;
  26
  27__u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
  28EXPORT_SYMBOL(syncookie_secret);
  29
  30static __init int init_syncookies(void)
  31{
  32        get_random_bytes(syncookie_secret, sizeof(syncookie_secret));
  33        return 0;
  34}
  35__initcall(init_syncookies);
  36
  37#define COOKIEBITS 24   /* Upper bits store count */
  38#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
  39
  40static DEFINE_PER_CPU(__u32, cookie_scratch)[16 + 5 + SHA_WORKSPACE_WORDS];
  41
  42static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
  43                       u32 count, int c)
  44{
  45        __u32 *tmp = __get_cpu_var(cookie_scratch);
  46
  47        memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c]));
  48        tmp[0] = (__force u32)saddr;
  49        tmp[1] = (__force u32)daddr;
  50        tmp[2] = ((__force u32)sport << 16) + (__force u32)dport;
  51        tmp[3] = count;
  52        sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
  53
  54        return tmp[17];
  55}
  56
  57
  58/*
  59 * when syncookies are in effect and tcp timestamps are enabled we encode
  60 * tcp options in the lowest 9 bits of the timestamp value that will be
  61 * sent in the syn-ack.
  62 * Since subsequent timestamps use the normal tcp_time_stamp value, we
  63 * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
  64 */
  65__u32 cookie_init_timestamp(struct request_sock *req)
  66{
  67        struct inet_request_sock *ireq;
  68        u32 ts, ts_now = tcp_time_stamp;
  69        u32 options = 0;
  70
  71        ireq = inet_rsk(req);
  72        if (ireq->wscale_ok) {
  73                options = ireq->snd_wscale;
  74                options |= ireq->rcv_wscale << 4;
  75        }
  76        options |= ireq->sack_ok << 8;
  77
  78        ts = ts_now & ~TSMASK;
  79        ts |= options;
  80        if (ts > ts_now) {
  81                ts >>= TSBITS;
  82                ts--;
  83                ts <<= TSBITS;
  84                ts |= options;
  85        }
  86        return ts;
  87}
  88
  89
  90static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
  91                                   __be16 dport, __u32 sseq, __u32 count,
  92                                   __u32 data)
  93{
  94        /*
  95         * Compute the secure sequence number.
  96         * The output should be:
  97         *   HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
  98         *      + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
  99         * Where sseq is their sequence number and count increases every
 100         * minute by 1.
 101         * As an extra hack, we add a small "data" value that encodes the
 102         * MSS into the second hash value.
 103         */
 104
 105        return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
 106                sseq + (count << COOKIEBITS) +
 107                ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
 108                 & COOKIEMASK));
 109}
 110
 111/*
 112 * This retrieves the small "data" value from the syncookie.
 113 * If the syncookie is bad, the data returned will be out of
 114 * range.  This must be checked by the caller.
 115 *
 116 * The count value used to generate the cookie must be within
 117 * "maxdiff" if the current (passed-in) "count".  The return value
 118 * is (__u32)-1 if this test fails.
 119 */
 120static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
 121                                  __be16 sport, __be16 dport, __u32 sseq,
 122                                  __u32 count, __u32 maxdiff)
 123{
 124        __u32 diff;
 125
 126        /* Strip away the layers from the cookie */
 127        cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
 128
 129        /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
 130        diff = (count - (cookie >> COOKIEBITS)) & ((__u32) - 1 >> COOKIEBITS);
 131        if (diff >= maxdiff)
 132                return (__u32)-1;
 133
 134        return (cookie -
 135                cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
 136                & COOKIEMASK;   /* Leaving the data behind */
 137}
 138
 139/*
 140 * This table has to be sorted and terminated with (__u16)-1.
 141 * XXX generate a better table.
 142 * Unresolved Issues: HIPPI with a 64k MSS is not well supported.
 143 */
 144static __u16 const msstab[] = {
 145        64 - 1,
 146        256 - 1,
 147        512 - 1,
 148        536 - 1,
 149        1024 - 1,
 150        1440 - 1,
 151        1460 - 1,
 152        4312 - 1,
 153        (__u16)-1
 154};
 155/* The number doesn't include the -1 terminator */
 156#define NUM_MSS (ARRAY_SIZE(msstab) - 1)
 157
 158/*
 159 * Generate a syncookie.  mssp points to the mss, which is returned
 160 * rounded down to the value encoded in the cookie.
 161 */
 162__u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, __u16 *mssp)
 163{
 164        struct tcp_sock *tp = tcp_sk(sk);
 165        const struct iphdr *iph = ip_hdr(skb);
 166        const struct tcphdr *th = tcp_hdr(skb);
 167        int mssind;
 168        const __u16 mss = *mssp;
 169
 170        tp->last_synq_overflow = jiffies;
 171
 172        /* XXX sort msstab[] by probability?  Binary search? */
 173        for (mssind = 0; mss > msstab[mssind + 1]; mssind++)
 174                ;
 175        *mssp = msstab[mssind] + 1;
 176
 177        NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
 178
 179        return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
 180                                     th->source, th->dest, ntohl(th->seq),
 181                                     jiffies / (HZ * 60), mssind);
 182}
 183
 184/*
 185 * This (misnamed) value is the age of syncookie which is permitted.
 186 * Its ideal value should be dependent on TCP_TIMEOUT_INIT and
 187 * sysctl_tcp_retries1. It's a rather complicated formula (exponential
 188 * backoff) to compute at runtime so it's currently hardcoded here.
 189 */
 190#define COUNTER_TRIES 4
 191/*
 192 * Check if a ack sequence number is a valid syncookie.
 193 * Return the decoded mss if it is, or 0 if not.
 194 */
 195static inline int cookie_check(struct sk_buff *skb, __u32 cookie)
 196{
 197        const struct iphdr *iph = ip_hdr(skb);
 198        const struct tcphdr *th = tcp_hdr(skb);
 199        __u32 seq = ntohl(th->seq) - 1;
 200        __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
 201                                            th->source, th->dest, seq,
 202                                            jiffies / (HZ * 60),
 203                                            COUNTER_TRIES);
 204
 205        return mssind < NUM_MSS ? msstab[mssind] + 1 : 0;
 206}
 207
 208static inline struct sock *get_cookie_sock(struct sock *sk, struct sk_buff *skb,
 209                                           struct request_sock *req,
 210                                           struct dst_entry *dst)
 211{
 212        struct inet_connection_sock *icsk = inet_csk(sk);
 213        struct sock *child;
 214
 215        child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst);
 216        if (child)
 217                inet_csk_reqsk_queue_add(sk, req, child);
 218        else
 219                reqsk_free(req);
 220
 221        return child;
 222}
 223
 224
 225/*
 226 * when syncookies are in effect and tcp timestamps are enabled we stored
 227 * additional tcp options in the timestamp.
 228 * This extracts these options from the timestamp echo.
 229 *
 230 * The lowest 4 bits are for snd_wscale
 231 * The next 4 lsb are for rcv_wscale
 232 * The next lsb is for sack_ok
 233 */
 234void cookie_check_timestamp(struct tcp_options_received *tcp_opt)
 235{
 236        /* echoed timestamp, 9 lowest bits contain options */
 237        u32 options = tcp_opt->rcv_tsecr & TSMASK;
 238
 239        tcp_opt->snd_wscale = options & 0xf;
 240        options >>= 4;
 241        tcp_opt->rcv_wscale = options & 0xf;
 242
 243        tcp_opt->sack_ok = (options >> 4) & 0x1;
 244
 245        if (tcp_opt->sack_ok)
 246                tcp_sack_reset(tcp_opt);
 247
 248        if (tcp_opt->snd_wscale || tcp_opt->rcv_wscale)
 249                tcp_opt->wscale_ok = 1;
 250}
 251EXPORT_SYMBOL(cookie_check_timestamp);
 252
 253struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
 254                             struct ip_options *opt)
 255{
 256        struct inet_request_sock *ireq;
 257        struct tcp_request_sock *treq;
 258        struct tcp_sock *tp = tcp_sk(sk);
 259        const struct tcphdr *th = tcp_hdr(skb);
 260        __u32 cookie = ntohl(th->ack_seq) - 1;
 261        struct sock *ret = sk;
 262        struct request_sock *req;
 263        int mss;
 264        struct rtable *rt;
 265        __u8 rcv_wscale;
 266        struct tcp_options_received tcp_opt;
 267
 268        if (!sysctl_tcp_syncookies || !th->ack)
 269                goto out;
 270
 271        if (time_after(jiffies, tp->last_synq_overflow + TCP_TIMEOUT_INIT) ||
 272            (mss = cookie_check(skb, cookie)) == 0) {
 273                NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
 274                goto out;
 275        }
 276
 277        NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
 278
 279        /* check for timestamp cookie support */
 280        memset(&tcp_opt, 0, sizeof(tcp_opt));
 281        tcp_parse_options(skb, &tcp_opt, 0);
 282
 283        if (tcp_opt.saw_tstamp)
 284                cookie_check_timestamp(&tcp_opt);
 285
 286        ret = NULL;
 287        req = inet_reqsk_alloc(&tcp_request_sock_ops); /* for safety */
 288        if (!req)
 289                goto out;
 290
 291        ireq = inet_rsk(req);
 292        treq = tcp_rsk(req);
 293        treq->rcv_isn           = ntohl(th->seq) - 1;
 294        treq->snt_isn           = cookie;
 295        req->mss                = mss;
 296        ireq->loc_port          = th->dest;
 297        ireq->rmt_port          = th->source;
 298        ireq->loc_addr          = ip_hdr(skb)->daddr;
 299        ireq->rmt_addr          = ip_hdr(skb)->saddr;
 300        ireq->ecn_ok            = 0;
 301        ireq->snd_wscale        = tcp_opt.snd_wscale;
 302        ireq->rcv_wscale        = tcp_opt.rcv_wscale;
 303        ireq->sack_ok           = tcp_opt.sack_ok;
 304        ireq->wscale_ok         = tcp_opt.wscale_ok;
 305        ireq->tstamp_ok         = tcp_opt.saw_tstamp;
 306        req->ts_recent          = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
 307
 308        /* We throwed the options of the initial SYN away, so we hope
 309         * the ACK carries the same options again (see RFC1122 4.2.3.8)
 310         */
 311        if (opt && opt->optlen) {
 312                int opt_size = sizeof(struct ip_options) + opt->optlen;
 313
 314                ireq->opt = kmalloc(opt_size, GFP_ATOMIC);
 315                if (ireq->opt != NULL && ip_options_echo(ireq->opt, skb)) {
 316                        kfree(ireq->opt);
 317                        ireq->opt = NULL;
 318                }
 319        }
 320
 321        if (security_inet_conn_request(sk, skb, req)) {
 322                reqsk_free(req);
 323                goto out;
 324        }
 325
 326        req->expires    = 0UL;
 327        req->retrans    = 0;
 328
 329        /*
 330         * We need to lookup the route here to get at the correct
 331         * window size. We should better make sure that the window size
 332         * hasn't changed since we received the original syn, but I see
 333         * no easy way to do this.
 334         */
 335        {
 336                struct flowi fl = { .nl_u = { .ip4_u =
 337                                              { .daddr = ((opt && opt->srr) ?
 338                                                          opt->faddr :
 339                                                          ireq->rmt_addr),
 340                                                .saddr = ireq->loc_addr,
 341                                                .tos = RT_CONN_FLAGS(sk) } },
 342                                    .proto = IPPROTO_TCP,
 343                                    .flags = inet_sk_flowi_flags(sk),
 344                                    .uli_u = { .ports =
 345                                               { .sport = th->dest,
 346                                                 .dport = th->source } } };
 347                security_req_classify_flow(req, &fl);
 348                if (ip_route_output_key(&init_net, &rt, &fl)) {
 349                        reqsk_free(req);
 350                        goto out;
 351                }
 352        }
 353
 354        /* Try to redo what tcp_v4_send_synack did. */
 355        req->window_clamp = tp->window_clamp ? :dst_metric(&rt->u.dst, RTAX_WINDOW);
 356
 357        tcp_select_initial_window(tcp_full_space(sk), req->mss,
 358                                  &req->rcv_wnd, &req->window_clamp,
 359                                  ireq->wscale_ok, &rcv_wscale);
 360
 361        ireq->rcv_wscale  = rcv_wscale;
 362
 363        ret = get_cookie_sock(sk, skb, req, &rt->u.dst);
 364out:    return ret;
 365}
 366