root/net/ipv4/syncookies.c

DEFINITIONS

1. cookie_hash
2. cookie_init_timestamp
3. secure_tcp_syn_cookie
4. check_tcp_syn_cookie
5. __cookie_v4_init_sequence
6. cookie_v4_init_sequence
7. __cookie_v4_check
8. tcp_get_cookie_sock
9. cookie_timestamp_decode
10. cookie_ecn_ok
11. cookie_v4_check

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  Syncookies implementation for the Linux kernel
   4  *
   5  *  Copyright (C) 1997 Andi Kleen
   6  *  Based on ideas by D.J.Bernstein and Eric Schenk.
   7  */
   8 
   9 #include <linux/tcp.h>
  10 #include <linux/slab.h>
  11 #include <linux/random.h>
  12 #include <linux/siphash.h>
  13 #include <linux/kernel.h>
  14 #include <linux/export.h>
  15 #include <net/secure_seq.h>
  16 #include <net/tcp.h>
  17 #include <net/route.h>
  18 
  19 static siphash_key_t syncookie_secret[2] __read_mostly;
  20 
  21 #define COOKIEBITS 24   /* Upper bits store count */
  22 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
  23 
  24 /* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK
  25  * stores TCP options:
  26  *
  27  * MSB                               LSB
  28  * | 31 ...   6 |  5  |  4   | 3 2 1 0 |
  29  * |  Timestamp | ECN | SACK | WScale  |
  30  *
  31  * When we receive a valid cookie-ACK, we look at the echoed tsval (if
  32  * any) to figure out which TCP options we should use for the rebuilt
  33  * connection.
  34  *
  35  * A WScale setting of '0xf' (which is an invalid scaling value)
  36  * means that original syn did not include the TCP window scaling option.
  37  */
  38 #define TS_OPT_WSCALE_MASK      0xf
  39 #define TS_OPT_SACK             BIT(4)
  40 #define TS_OPT_ECN              BIT(5)
  41 /* There is no TS_OPT_TIMESTAMP:
  42  * if ACK contains timestamp option, we already know it was
  43  * requested/supported by the syn/synack exchange.
  44  */
  45 #define TSBITS  6
  46 #define TSMASK  (((__u32)1 << TSBITS) - 1)
  47 
  48 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
  49                        u32 count, int c)
  50 {
  51         net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
  52         return siphash_4u32((__force u32)saddr, (__force u32)daddr,
  53                             (__force u32)sport << 16 | (__force u32)dport,
  54                             count, &syncookie_secret[c]);
  55 }
  56 
  57 
  58 /*
  59  * when syncookies are in effect and tcp timestamps are enabled we encode
  60  * tcp options in the lower 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 u64 cookie_init_timestamp(struct request_sock *req)
  66 {
  67         struct inet_request_sock *ireq;
  68         u32 ts, ts_now = tcp_time_stamp_raw();
  69         u32 options = 0;
  70 
  71         ireq = inet_rsk(req);
  72 
  73         options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
  74         if (ireq->sack_ok)
  75                 options |= TS_OPT_SACK;
  76         if (ireq->ecn_ok)
  77                 options |= TS_OPT_ECN;
  78 
  79         ts = ts_now & ~TSMASK;
  80         ts |= options;
  81         if (ts > ts_now) {
  82                 ts >>= TSBITS;
  83                 ts--;
  84                 ts <<= TSBITS;
  85                 ts |= options;
  86         }
  87         return (u64)ts * (NSEC_PER_SEC / TCP_TS_HZ);
  88 }
  89 
  90 
  91 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
  92                                    __be16 dport, __u32 sseq, __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         u32 count = tcp_cookie_time();
 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 less than
 117  * MAX_SYNCOOKIE_AGE minutes in the past.
 118  * The return value (__u32)-1 if this test fails.
 119  */
 120 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
 121                                   __be16 sport, __be16 dport, __u32 sseq)
 122 {
 123         u32 diff, count = tcp_cookie_time();
 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 >= MAX_SYNCOOKIE_AGE)
 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  * MSS Values are chosen based on the 2011 paper
 140  * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
 141  * Values ..
 142  *  .. lower than 536 are rare (< 0.2%)
 143  *  .. between 537 and 1299 account for less than < 1.5% of observed values
 144  *  .. in the 1300-1349 range account for about 15 to 20% of observed mss values
 145  *  .. exceeding 1460 are very rare (< 0.04%)
 146  *
 147  *  1460 is the single most frequently announced mss value (30 to 46% depending
 148  *  on monitor location).  Table must be sorted.
 149  */
 150 static __u16 const msstab[] = {
 151         536,
 152         1300,
 153         1440,   /* 1440, 1452: PPPoE */
 154         1460,
 155 };
 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(const struct iphdr *iph, const struct tcphdr *th,
 162                               u16 *mssp)
 163 {
 164         int mssind;
 165         const __u16 mss = *mssp;
 166 
 167         for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
 168                 if (mss >= msstab[mssind])
 169                         break;
 170         *mssp = msstab[mssind];
 171 
 172         return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
 173                                      th->source, th->dest, ntohl(th->seq),
 174                                      mssind);
 175 }
 176 EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
 177 
 178 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
 179 {
 180         const struct iphdr *iph = ip_hdr(skb);
 181         const struct tcphdr *th = tcp_hdr(skb);
 182 
 183         return __cookie_v4_init_sequence(iph, th, mssp);
 184 }
 185 
 186 /*
 187  * Check if a ack sequence number is a valid syncookie.
 188  * Return the decoded mss if it is, or 0 if not.
 189  */
 190 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
 191                       u32 cookie)
 192 {
 193         __u32 seq = ntohl(th->seq) - 1;
 194         __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
 195                                             th->source, th->dest, seq);
 196 
 197         return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
 198 }
 199 EXPORT_SYMBOL_GPL(__cookie_v4_check);
 200 
 201 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
 202                                  struct request_sock *req,
 203                                  struct dst_entry *dst, u32 tsoff)
 204 {
 205         struct inet_connection_sock *icsk = inet_csk(sk);
 206         struct sock *child;
 207         bool own_req;
 208 
 209         child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
 210                                                  NULL, &own_req);
 211         if (child) {
 212                 refcount_set(&req->rsk_refcnt, 1);
 213                 tcp_sk(child)->tsoffset = tsoff;
 214                 sock_rps_save_rxhash(child, skb);
 215                 if (inet_csk_reqsk_queue_add(sk, req, child))
 216                         return child;
 217 
 218                 bh_unlock_sock(child);
 219                 sock_put(child);
 220         }
 221         __reqsk_free(req);
 222 
 223         return NULL;
 224 }
 225 EXPORT_SYMBOL(tcp_get_cookie_sock);
 226 
 227 /*
 228  * when syncookies are in effect and tcp timestamps are enabled we stored
 229  * additional tcp options in the timestamp.
 230  * This extracts these options from the timestamp echo.
 231  *
 232  * return false if we decode a tcp option that is disabled
 233  * on the host.
 234  */
 235 bool cookie_timestamp_decode(const struct net *net,
 236                              struct tcp_options_received *tcp_opt)
 237 {
 238         /* echoed timestamp, lowest bits contain options */
 239         u32 options = tcp_opt->rcv_tsecr;
 240 
 241         if (!tcp_opt->saw_tstamp)  {
 242                 tcp_clear_options(tcp_opt);
 243                 return true;
 244         }
 245 
 246         if (!net->ipv4.sysctl_tcp_timestamps)
 247                 return false;
 248 
 249         tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
 250 
 251         if (tcp_opt->sack_ok && !net->ipv4.sysctl_tcp_sack)
 252                 return false;
 253 
 254         if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
 255                 return true; /* no window scaling */
 256 
 257         tcp_opt->wscale_ok = 1;
 258         tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
 259 
 260         return net->ipv4.sysctl_tcp_window_scaling != 0;
 261 }
 262 EXPORT_SYMBOL(cookie_timestamp_decode);
 263 
 264 bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
 265                    const struct net *net, const struct dst_entry *dst)
 266 {
 267         bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN;
 268 
 269         if (!ecn_ok)
 270                 return false;
 271 
 272         if (net->ipv4.sysctl_tcp_ecn)
 273                 return true;
 274 
 275         return dst_feature(dst, RTAX_FEATURE_ECN);
 276 }
 277 EXPORT_SYMBOL(cookie_ecn_ok);
 278 
 279 /* On input, sk is a listener.
 280  * Output is listener if incoming packet would not create a child
 281  *           NULL if memory could not be allocated.
 282  */
 283 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
 284 {
 285         struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
 286         struct tcp_options_received tcp_opt;
 287         struct inet_request_sock *ireq;
 288         struct tcp_request_sock *treq;
 289         struct tcp_sock *tp = tcp_sk(sk);
 290         const struct tcphdr *th = tcp_hdr(skb);
 291         __u32 cookie = ntohl(th->ack_seq) - 1;
 292         struct sock *ret = sk;
 293         struct request_sock *req;
 294         int mss;
 295         struct rtable *rt;
 296         __u8 rcv_wscale;
 297         struct flowi4 fl4;
 298         u32 tsoff = 0;
 299 
 300         if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies || !th->ack || th->rst)
 301                 goto out;
 302 
 303         if (tcp_synq_no_recent_overflow(sk))
 304                 goto out;
 305 
 306         mss = __cookie_v4_check(ip_hdr(skb), th, cookie);
 307         if (mss == 0) {
 308                 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
 309                 goto out;
 310         }
 311 
 312         __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
 313 
 314         /* check for timestamp cookie support */
 315         memset(&tcp_opt, 0, sizeof(tcp_opt));
 316         tcp_parse_options(sock_net(sk), skb, &tcp_opt, 0, NULL);
 317 
 318         if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
 319                 tsoff = secure_tcp_ts_off(sock_net(sk),
 320                                           ip_hdr(skb)->daddr,
 321                                           ip_hdr(skb)->saddr);
 322                 tcp_opt.rcv_tsecr -= tsoff;
 323         }
 324 
 325         if (!cookie_timestamp_decode(sock_net(sk), &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         treq->txhash            = net_tx_rndhash();
 339         req->mss                = mss;
 340         ireq->ir_num            = ntohs(th->dest);
 341         ireq->ir_rmt_port       = th->source;
 342         sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
 343         sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
 344         ireq->ir_mark           = inet_request_mark(sk, skb);
 345         ireq->snd_wscale        = tcp_opt.snd_wscale;
 346         ireq->sack_ok           = tcp_opt.sack_ok;
 347         ireq->wscale_ok         = tcp_opt.wscale_ok;
 348         ireq->tstamp_ok         = tcp_opt.saw_tstamp;
 349         req->ts_recent          = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
 350         treq->snt_synack        = 0;
 351         treq->tfo_listener      = false;
 352         if (IS_ENABLED(CONFIG_SMC))
 353                 ireq->smc_ok = 0;
 354 
 355         ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
 356 
 357         /* We throwed the options of the initial SYN away, so we hope
 358          * the ACK carries the same options again (see RFC1122 4.2.3.8)
 359          */
 360         RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(sock_net(sk), skb));
 361 
 362         if (security_inet_conn_request(sk, skb, req)) {
 363                 reqsk_free(req);
 364                 goto out;
 365         }
 366 
 367         req->num_retrans = 0;
 368 
 369         /*
 370          * We need to lookup the route here to get at the correct
 371          * window size. We should better make sure that the window size
 372          * hasn't changed since we received the original syn, but I see
 373          * no easy way to do this.
 374          */
 375         flowi4_init_output(&fl4, ireq->ir_iif, ireq->ir_mark,
 376                            RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
 377                            inet_sk_flowi_flags(sk),
 378                            opt->srr ? opt->faddr : ireq->ir_rmt_addr,
 379                            ireq->ir_loc_addr, th->source, th->dest, sk->sk_uid);
 380         security_req_classify_flow(req, flowi4_to_flowi(&fl4));
 381         rt = ip_route_output_key(sock_net(sk), &fl4);
 382         if (IS_ERR(rt)) {
 383                 reqsk_free(req);
 384                 goto out;
 385         }
 386 
 387         /* Try to redo what tcp_v4_send_synack did. */
 388         req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
 389 
 390         tcp_select_initial_window(sk, tcp_full_space(sk), req->mss,
 391                                   &req->rsk_rcv_wnd, &req->rsk_window_clamp,
 392                                   ireq->wscale_ok, &rcv_wscale,
 393                                   dst_metric(&rt->dst, RTAX_INITRWND));
 394 
 395         ireq->rcv_wscale  = rcv_wscale;
 396         ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
 397 
 398         ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst, tsoff);
 399         /* ip_queue_xmit() depends on our flow being setup
 400          * Normal sockets get it right from inet_csk_route_child_sock()
 401          */
 402         if (ret)
 403                 inet_sk(ret)->cork.fl.u.ip4 = fl4;
 404 out:    return ret;
 405 }