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 <linux/export.h>
  19#include <net/tcp.h>
  20#include <net/route.h>
  21
  22static u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS] __read_mostly;
  23
  24#define COOKIEBITS 24   /* Upper bits store count */
  25#define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
  26
  27/* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK
  28 * stores TCP options:
  29 *
  30 * MSB                               LSB
  31 * | 31 ...   6 |  5  |  4   | 3 2 1 0 |
  32 * |  Timestamp | ECN | SACK | WScale  |
  33 *
  34 * When we receive a valid cookie-ACK, we look at the echoed tsval (if
  35 * any) to figure out which TCP options we should use for the rebuilt
  36 * connection.
  37 *
  38 * A WScale setting of '0xf' (which is an invalid scaling value)
  39 * means that original syn did not include the TCP window scaling option.
  40 */
  41#define TS_OPT_WSCALE_MASK      0xf
  42#define TS_OPT_SACK             BIT(4)
  43#define TS_OPT_ECN              BIT(5)
  44/* There is no TS_OPT_TIMESTAMP:
  45 * if ACK contains timestamp option, we already know it was
  46 * requested/supported by the syn/synack exchange.
  47 */
  48#define TSBITS  6
  49#define TSMASK  (((__u32)1 << TSBITS) - 1)
  50
  51static DEFINE_PER_CPU(__u32 [16 + 5 + SHA_WORKSPACE_WORDS], ipv4_cookie_scratch);
  52
  53static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
  54                       u32 count, int c)
  55{
  56        __u32 *tmp;
  57
  58        net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
  59
  60        tmp  = this_cpu_ptr(ipv4_cookie_scratch);
  61        memcpy(tmp + 4, syncookie_secret[c], sizeof(syncookie_secret[c]));
  62        tmp[0] = (__force u32)saddr;
  63        tmp[1] = (__force u32)daddr;
  64        tmp[2] = ((__force u32)sport << 16) + (__force u32)dport;
  65        tmp[3] = count;
  66        sha_transform(tmp + 16, (__u8 *)tmp, tmp + 16 + 5);
  67
  68        return tmp[17];
  69}
  70
  71
  72/*
  73 * when syncookies are in effect and tcp timestamps are enabled we encode
  74 * tcp options in the lower bits of the timestamp value that will be
  75 * sent in the syn-ack.
  76 * Since subsequent timestamps use the normal tcp_time_stamp value, we
  77 * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
  78 */
  79__u32 cookie_init_timestamp(struct request_sock *req)
  80{
  81        struct inet_request_sock *ireq;
  82        u32 ts, ts_now = tcp_time_stamp;
  83        u32 options = 0;
  84
  85        ireq = inet_rsk(req);
  86
  87        options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
  88        if (ireq->sack_ok)
  89                options |= TS_OPT_SACK;
  90        if (ireq->ecn_ok)
  91                options |= TS_OPT_ECN;
  92
  93        ts = ts_now & ~TSMASK;
  94        ts |= options;
  95        if (ts > ts_now) {
  96                ts >>= TSBITS;
  97                ts--;
  98                ts <<= TSBITS;
  99                ts |= options;
 100        }
 101        return ts;
 102}
 103
 104
 105static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
 106                                   __be16 dport, __u32 sseq, __u32 data)
 107{
 108        /*
 109         * Compute the secure sequence number.
 110         * The output should be:
 111         *   HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
 112         *      + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
 113         * Where sseq is their sequence number and count increases every
 114         * minute by 1.
 115         * As an extra hack, we add a small "data" value that encodes the
 116         * MSS into the second hash value.
 117         */
 118        u32 count = tcp_cookie_time();
 119        return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
 120                sseq + (count << COOKIEBITS) +
 121                ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
 122                 & COOKIEMASK));
 123}
 124
 125/*
 126 * This retrieves the small "data" value from the syncookie.
 127 * If the syncookie is bad, the data returned will be out of
 128 * range.  This must be checked by the caller.
 129 *
 130 * The count value used to generate the cookie must be less than
 131 * MAX_SYNCOOKIE_AGE minutes in the past.
 132 * The return value (__u32)-1 if this test fails.
 133 */
 134static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
 135                                  __be16 sport, __be16 dport, __u32 sseq)
 136{
 137        u32 diff, count = tcp_cookie_time();
 138
 139        /* Strip away the layers from the cookie */
 140        cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
 141
 142        /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
 143        diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
 144        if (diff >= MAX_SYNCOOKIE_AGE)
 145                return (__u32)-1;
 146
 147        return (cookie -
 148                cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
 149                & COOKIEMASK;   /* Leaving the data behind */
 150}
 151
 152/*
 153 * MSS Values are chosen based on the 2011 paper
 154 * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
 155 * Values ..
 156 *  .. lower than 536 are rare (< 0.2%)
 157 *  .. between 537 and 1299 account for less than < 1.5% of observed values
 158 *  .. in the 1300-1349 range account for about 15 to 20% of observed mss values
 159 *  .. exceeding 1460 are very rare (< 0.04%)
 160 *
 161 *  1460 is the single most frequently announced mss value (30 to 46% depending
 162 *  on monitor location).  Table must be sorted.
 163 */
 164static __u16 const msstab[] = {
 165        536,
 166        1300,
 167        1440,   /* 1440, 1452: PPPoE */
 168        1460,
 169};
 170
 171/*
 172 * Generate a syncookie.  mssp points to the mss, which is returned
 173 * rounded down to the value encoded in the cookie.
 174 */
 175u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
 176                              u16 *mssp)
 177{
 178        int mssind;
 179        const __u16 mss = *mssp;
 180
 181        for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
 182                if (mss >= msstab[mssind])
 183                        break;
 184        *mssp = msstab[mssind];
 185
 186        return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
 187                                     th->source, th->dest, ntohl(th->seq),
 188                                     mssind);
 189}
 190EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
 191
 192__u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
 193{
 194        const struct iphdr *iph = ip_hdr(skb);
 195        const struct tcphdr *th = tcp_hdr(skb);
 196
 197        return __cookie_v4_init_sequence(iph, th, mssp);
 198}
 199
 200/*
 201 * Check if a ack sequence number is a valid syncookie.
 202 * Return the decoded mss if it is, or 0 if not.
 203 */
 204int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
 205                      u32 cookie)
 206{
 207        __u32 seq = ntohl(th->seq) - 1;
 208        __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
 209                                            th->source, th->dest, seq);
 210
 211        return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
 212}
 213EXPORT_SYMBOL_GPL(__cookie_v4_check);
 214
 215struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
 216                                 struct request_sock *req,
 217                                 struct dst_entry *dst)
 218{
 219        struct inet_connection_sock *icsk = inet_csk(sk);
 220        struct sock *child;
 221        bool own_req;
 222
 223        child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 224                                                 NULL, &own_req);
 225        if (child) {
 226                atomic_set(&req->rsk_refcnt, 1);
 227                sock_rps_save_rxhash(child, skb);
 228                inet_csk_reqsk_queue_add(sk, req, child);
 229        } else {
 230                reqsk_free(req);
 231        }
 232        return child;
 233}
 234EXPORT_SYMBOL(tcp_get_cookie_sock);
 235
 236/*
 237 * when syncookies are in effect and tcp timestamps are enabled we stored
 238 * additional tcp options in the timestamp.
 239 * This extracts these options from the timestamp echo.
 240 *
 241 * return false if we decode a tcp option that is disabled
 242 * on the host.
 243 */
 244bool cookie_timestamp_decode(struct tcp_options_received *tcp_opt)
 245{
 246        /* echoed timestamp, lowest bits contain options */
 247        u32 options = tcp_opt->rcv_tsecr;
 248
 249        if (!tcp_opt->saw_tstamp)  {
 250                tcp_clear_options(tcp_opt);
 251                return true;
 252        }
 253
 254        if (!sysctl_tcp_timestamps)
 255                return false;
 256
 257        tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
 258
 259        if (tcp_opt->sack_ok && !sysctl_tcp_sack)
 260                return false;
 261
 262        if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
 263                return true; /* no window scaling */
 264
 265        tcp_opt->wscale_ok = 1;
 266        tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
 267
 268        return sysctl_tcp_window_scaling != 0;
 269}
 270EXPORT_SYMBOL(cookie_timestamp_decode);
 271
 272bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
 273                   const struct net *net, const struct dst_entry *dst)
 274{
 275        bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN;
 276
 277        if (!ecn_ok)
 278                return false;
 279
 280        if (net->ipv4.sysctl_tcp_ecn)
 281                return true;
 282
 283        return dst_feature(dst, RTAX_FEATURE_ECN);
 284}
 285EXPORT_SYMBOL(cookie_ecn_ok);
 286
 287/* On input, sk is a listener.
 288 * Output is listener if incoming packet would not create a child
 289 *           NULL if memory could not be allocated.
 290 */
 291struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
 292{
 293        struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
 294        struct tcp_options_received tcp_opt;
 295        struct inet_request_sock *ireq;
 296        struct tcp_request_sock *treq;
 297        struct tcp_sock *tp = tcp_sk(sk);
 298        const struct tcphdr *th = tcp_hdr(skb);
 299        __u32 cookie = ntohl(th->ack_seq) - 1;
 300        struct sock *ret = sk;
 301        struct request_sock *req;
 302        int mss;
 303        struct rtable *rt;
 304        __u8 rcv_wscale;
 305        struct flowi4 fl4;
 306
 307        if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies || !th->ack || th->rst)
 308                goto out;
 309
 310        if (tcp_synq_no_recent_overflow(sk))
 311                goto out;
 312
 313        mss = __cookie_v4_check(ip_hdr(skb), th, cookie);
 314        if (mss == 0) {
 315                __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
 316                goto out;
 317        }
 318
 319        __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
 320
 321        /* check for timestamp cookie support */
 322        memset(&tcp_opt, 0, sizeof(tcp_opt));
 323        tcp_parse_options(skb, &tcp_opt, 0, NULL);
 324
 325        if (!cookie_timestamp_decode(&tcp_opt))
 326                goto out;
 327
 328        ret = NULL;
 329        req = inet_reqsk_alloc(&tcp_request_sock_ops, sk, false); /* for safety */
 330        if (!req)
 331                goto out;
 332
 333        ireq = inet_rsk(req);
 334        treq = tcp_rsk(req);
 335        treq->rcv_isn           = ntohl(th->seq) - 1;
 336        treq->snt_isn           = cookie;
 337        treq->ts_off            = 0;
 338        req->mss                = mss;
 339        ireq->ir_num            = ntohs(th->dest);
 340        ireq->ir_rmt_port       = th->source;
 341        sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 342        sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 343        ireq->ir_mark           = inet_request_mark(sk, skb);
 344        ireq->snd_wscale        = tcp_opt.snd_wscale;
 345        ireq->sack_ok           = tcp_opt.sack_ok;
 346        ireq->wscale_ok         = tcp_opt.wscale_ok;
 347        ireq->tstamp_ok         = tcp_opt.saw_tstamp;
 348        req->ts_recent          = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
 349        treq->snt_synack.v64    = 0;
 350        treq->tfo_listener      = false;
 351
 352        ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
 353
 354        /* We throwed the options of the initial SYN away, so we hope
 355         * the ACK carries the same options again (see RFC1122 4.2.3.8)
 356         */
 357        ireq->opt = tcp_v4_save_options(skb);
 358
 359        if (security_inet_conn_request(sk, skb, req)) {
 360                reqsk_free(req);
 361                goto out;
 362        }
 363
 364        req->num_retrans = 0;
 365
 366        /*
 367         * We need to lookup the route here to get at the correct
 368         * window size. We should better make sure that the window size
 369         * hasn't changed since we received the original syn, but I see
 370         * no easy way to do this.
 371         */
 372        flowi4_init_output(&fl4, ireq->ir_iif, ireq->ir_mark,
 373                           RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
 374                           inet_sk_flowi_flags(sk),
 375                           opt->srr ? opt->faddr : ireq->ir_rmt_addr,
 376                           ireq->ir_loc_addr, th->source, th->dest, sk->sk_uid);
 377        security_req_classify_flow(req, flowi4_to_flowi(&fl4));
 378        rt = ip_route_output_key(sock_net(sk), &fl4);
 379        if (IS_ERR(rt)) {
 380                reqsk_free(req);
 381                goto out;
 382        }
 383
 384        /* Try to redo what tcp_v4_send_synack did. */
 385        req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
 386
 387        tcp_select_initial_window(tcp_full_space(sk), req->mss,
 388                                  &req->rsk_rcv_wnd, &req->rsk_window_clamp,
 389                                  ireq->wscale_ok, &rcv_wscale,
 390                                  dst_metric(&rt->dst, RTAX_INITRWND));
 391
 392        ireq->rcv_wscale  = rcv_wscale;
 393        ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
 394
 395        ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst);
 396        /* ip_queue_xmit() depends on our flow being setup
 397         * Normal sockets get it right from inet_csk_route_child_sock()
 398         */
 399        if (ret)
 400                inet_sk(ret)->cork.fl.u.ip4 = fl4;
 401out:    return ret;
 402}
 403
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.