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