net/ipv4/syncookies.c

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  *  Syncookies implementation for the Linux kernel
0004  *
0005  *  Copyright (C) 1997 Andi Kleen
0006  *  Based on ideas by D.J.Bernstein and Eric Schenk.
0007  */
0008
0009 #include <linux/tcp.h>
0010 #include <linux/siphash.h>
0011 #include <linux/kernel.h>
0012 #include <linux/export.h>
0013 #include <net/secure_seq.h>
0014 #include <net/tcp.h>
0015 #include <net/route.h>
0016
0017 static siphash_aligned_key_t syncookie_secret[2];
0018
0019 #define COOKIEBITS 24   /* Upper bits store count */
0020 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
0021
0022 /* TCP Timestamp: 6 lowest bits of timestamp sent in the cookie SYN-ACK
0023  * stores TCP options:
0024  *
0025  * MSB                               LSB
0026  * | 31 ...   6 |  5  |  4   | 3 2 1 0 |
0027  * |  Timestamp | ECN | SACK | WScale  |
0028  *
0029  * When we receive a valid cookie-ACK, we look at the echoed tsval (if
0030  * any) to figure out which TCP options we should use for the rebuilt
0031  * connection.
0032  *
0033  * A WScale setting of '0xf' (which is an invalid scaling value)
0034  * means that original syn did not include the TCP window scaling option.
0035  */
0036 #define TS_OPT_WSCALE_MASK  0xf
0037 #define TS_OPT_SACK     BIT(4)
0038 #define TS_OPT_ECN      BIT(5)
0039 /* There is no TS_OPT_TIMESTAMP:
0040  * if ACK contains timestamp option, we already know it was
0041  * requested/supported by the syn/synack exchange.
0042  */
0043 #define TSBITS  6
0044 #define TSMASK  (((__u32)1 << TSBITS) - 1)
0045
0046 static u32 cookie_hash(__be32 saddr, __be32 daddr, __be16 sport, __be16 dport,
0047                u32 count, int c)
0048 {
0049     net_get_random_once(syncookie_secret, sizeof(syncookie_secret));
0050     return siphash_4u32((__force u32)saddr, (__force u32)daddr,
0051                 (__force u32)sport << 16 | (__force u32)dport,
0052                 count, &syncookie_secret[c]);
0053 }
0054
0055
0056 /*
0057  * when syncookies are in effect and tcp timestamps are enabled we encode
0058  * tcp options in the lower bits of the timestamp value that will be
0059  * sent in the syn-ack.
0060  * Since subsequent timestamps use the normal tcp_time_stamp value, we
0061  * must make sure that the resulting initial timestamp is <= tcp_time_stamp.
0062  */
0063 u64 cookie_init_timestamp(struct request_sock *req, u64 now)
0064 {
0065     struct inet_request_sock *ireq;
0066     u32 ts, ts_now = tcp_ns_to_ts(now);
0067     u32 options = 0;
0068
0069     ireq = inet_rsk(req);
0070
0071     options = ireq->wscale_ok ? ireq->snd_wscale : TS_OPT_WSCALE_MASK;
0072     if (ireq->sack_ok)
0073         options |= TS_OPT_SACK;
0074     if (ireq->ecn_ok)
0075         options |= TS_OPT_ECN;
0076
0077     ts = ts_now & ~TSMASK;
0078     ts |= options;
0079     if (ts > ts_now) {
0080         ts >>= TSBITS;
0081         ts--;
0082         ts <<= TSBITS;
0083         ts |= options;
0084     }
0085     return (u64)ts * (NSEC_PER_SEC / TCP_TS_HZ);
0086 }
0087
0088
0089 static __u32 secure_tcp_syn_cookie(__be32 saddr, __be32 daddr, __be16 sport,
0090                    __be16 dport, __u32 sseq, __u32 data)
0091 {
0092     /*
0093      * Compute the secure sequence number.
0094      * The output should be:
0095      *   HASH(sec1,saddr,sport,daddr,dport,sec1) + sseq + (count * 2^24)
0096      *      + (HASH(sec2,saddr,sport,daddr,dport,count,sec2) % 2^24).
0097      * Where sseq is their sequence number and count increases every
0098      * minute by 1.
0099      * As an extra hack, we add a small "data" value that encodes the
0100      * MSS into the second hash value.
0101      */
0102     u32 count = tcp_cookie_time();
0103     return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
0104         sseq + (count << COOKIEBITS) +
0105         ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
0106          & COOKIEMASK));
0107 }
0108
0109 /*
0110  * This retrieves the small "data" value from the syncookie.
0111  * If the syncookie is bad, the data returned will be out of
0112  * range.  This must be checked by the caller.
0113  *
0114  * The count value used to generate the cookie must be less than
0115  * MAX_SYNCOOKIE_AGE minutes in the past.
0116  * The return value (__u32)-1 if this test fails.
0117  */
0118 static __u32 check_tcp_syn_cookie(__u32 cookie, __be32 saddr, __be32 daddr,
0119                   __be16 sport, __be16 dport, __u32 sseq)
0120 {
0121     u32 diff, count = tcp_cookie_time();
0122
0123     /* Strip away the layers from the cookie */
0124     cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
0125
0126     /* Cookie is now reduced to (count * 2^24) ^ (hash % 2^24) */
0127     diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
0128     if (diff >= MAX_SYNCOOKIE_AGE)
0129         return (__u32)-1;
0130
0131     return (cookie -
0132         cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
0133         & COOKIEMASK;   /* Leaving the data behind */
0134 }
0135
0136 /*
0137  * MSS Values are chosen based on the 2011 paper
0138  * 'An Analysis of TCP Maximum Segement Sizes' by S. Alcock and R. Nelson.
0139  * Values ..
0140  *  .. lower than 536 are rare (< 0.2%)
0141  *  .. between 537 and 1299 account for less than < 1.5% of observed values
0142  *  .. in the 1300-1349 range account for about 15 to 20% of observed mss values
0143  *  .. exceeding 1460 are very rare (< 0.04%)
0144  *
0145  *  1460 is the single most frequently announced mss value (30 to 46% depending
0146  *  on monitor location).  Table must be sorted.
0147  */
0148 static __u16 const msstab[] = {
0149     536,
0150     1300,
0151     1440,   /* 1440, 1452: PPPoE */
0152     1460,
0153 };
0154
0155 /*
0156  * Generate a syncookie.  mssp points to the mss, which is returned
0157  * rounded down to the value encoded in the cookie.
0158  */
0159 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
0160                   u16 *mssp)
0161 {
0162     int mssind;
0163     const __u16 mss = *mssp;
0164
0165     for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
0166         if (mss >= msstab[mssind])
0167             break;
0168     *mssp = msstab[mssind];
0169
0170     return secure_tcp_syn_cookie(iph->saddr, iph->daddr,
0171                      th->source, th->dest, ntohl(th->seq),
0172                      mssind);
0173 }
0174 EXPORT_SYMBOL_GPL(__cookie_v4_init_sequence);
0175
0176 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mssp)
0177 {
0178     const struct iphdr *iph = ip_hdr(skb);
0179     const struct tcphdr *th = tcp_hdr(skb);
0180
0181     return __cookie_v4_init_sequence(iph, th, mssp);
0182 }
0183
0184 /*
0185  * Check if a ack sequence number is a valid syncookie.
0186  * Return the decoded mss if it is, or 0 if not.
0187  */
0188 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
0189               u32 cookie)
0190 {
0191     __u32 seq = ntohl(th->seq) - 1;
0192     __u32 mssind = check_tcp_syn_cookie(cookie, iph->saddr, iph->daddr,
0193                         th->source, th->dest, seq);
0194
0195     return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
0196 }
0197 EXPORT_SYMBOL_GPL(__cookie_v4_check);
0198
0199 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
0200                  struct request_sock *req,
0201                  struct dst_entry *dst, u32 tsoff)
0202 {
0203     struct inet_connection_sock *icsk = inet_csk(sk);
0204     struct sock *child;
0205     bool own_req;
0206
0207     child = icsk->icsk_af_ops->syn_recv_sock(sk, skb, req, dst,
0208                          NULL, &own_req);
0209     if (child) {
0210         refcount_set(&req->rsk_refcnt, 1);
0211         tcp_sk(child)->tsoffset = tsoff;
0212         sock_rps_save_rxhash(child, skb);
0213
0214         if (rsk_drop_req(req)) {
0215             reqsk_put(req);
0216             return child;
0217         }
0218
0219         if (inet_csk_reqsk_queue_add(sk, req, child))
0220             return child;
0221
0222         bh_unlock_sock(child);
0223         sock_put(child);
0224     }
0225     __reqsk_free(req);
0226
0227     return NULL;
0228 }
0229 EXPORT_SYMBOL(tcp_get_cookie_sock);
0230
0231 /*
0232  * when syncookies are in effect and tcp timestamps are enabled we stored
0233  * additional tcp options in the timestamp.
0234  * This extracts these options from the timestamp echo.
0235  *
0236  * return false if we decode a tcp option that is disabled
0237  * on the host.
0238  */
0239 bool cookie_timestamp_decode(const struct net *net,
0240                  struct tcp_options_received *tcp_opt)
0241 {
0242     /* echoed timestamp, lowest bits contain options */
0243     u32 options = tcp_opt->rcv_tsecr;
0244
0245     if (!tcp_opt->saw_tstamp)  {
0246         tcp_clear_options(tcp_opt);
0247         return true;
0248     }
0249
0250     if (!READ_ONCE(net->ipv4.sysctl_tcp_timestamps))
0251         return false;
0252
0253     tcp_opt->sack_ok = (options & TS_OPT_SACK) ? TCP_SACK_SEEN : 0;
0254
0255     if (tcp_opt->sack_ok && !READ_ONCE(net->ipv4.sysctl_tcp_sack))
0256         return false;
0257
0258     if ((options & TS_OPT_WSCALE_MASK) == TS_OPT_WSCALE_MASK)
0259         return true; /* no window scaling */
0260
0261     tcp_opt->wscale_ok = 1;
0262     tcp_opt->snd_wscale = options & TS_OPT_WSCALE_MASK;
0263
0264     return READ_ONCE(net->ipv4.sysctl_tcp_window_scaling) != 0;
0265 }
0266 EXPORT_SYMBOL(cookie_timestamp_decode);
0267
0268 bool cookie_ecn_ok(const struct tcp_options_received *tcp_opt,
0269            const struct net *net, const struct dst_entry *dst)
0270 {
0271     bool ecn_ok = tcp_opt->rcv_tsecr & TS_OPT_ECN;
0272
0273     if (!ecn_ok)
0274         return false;
0275
0276     if (READ_ONCE(net->ipv4.sysctl_tcp_ecn))
0277         return true;
0278
0279     return dst_feature(dst, RTAX_FEATURE_ECN);
0280 }
0281 EXPORT_SYMBOL(cookie_ecn_ok);
0282
0283 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
0284                         const struct tcp_request_sock_ops *af_ops,
0285                         struct sock *sk,
0286                         struct sk_buff *skb)
0287 {
0288     struct tcp_request_sock *treq;
0289     struct request_sock *req;
0290
0291 #ifdef CONFIG_MPTCP
0292     if (sk_is_mptcp(sk))
0293         ops = &mptcp_subflow_request_sock_ops;
0294 #endif
0295
0296     req = inet_reqsk_alloc(ops, sk, false);
0297     if (!req)
0298         return NULL;
0299
0300     treq = tcp_rsk(req);
0301
0302     /* treq->af_specific might be used to perform TCP_MD5 lookup */
0303     treq->af_specific = af_ops;
0304
0305     treq->syn_tos = TCP_SKB_CB(skb)->ip_dsfield;
0306 #if IS_ENABLED(CONFIG_MPTCP)
0307     treq->is_mptcp = sk_is_mptcp(sk);
0308     if (treq->is_mptcp) {
0309         int err = mptcp_subflow_init_cookie_req(req, sk, skb);
0310
0311         if (err) {
0312             reqsk_free(req);
0313             return NULL;
0314         }
0315     }
0316 #endif
0317
0318     return req;
0319 }
0320 EXPORT_SYMBOL_GPL(cookie_tcp_reqsk_alloc);
0321
0322 /* On input, sk is a listener.
0323  * Output is listener if incoming packet would not create a child
0324  *           NULL if memory could not be allocated.
0325  */
0326 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
0327 {
0328     struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
0329     struct tcp_options_received tcp_opt;
0330     struct inet_request_sock *ireq;
0331     struct tcp_request_sock *treq;
0332     struct tcp_sock *tp = tcp_sk(sk);
0333     const struct tcphdr *th = tcp_hdr(skb);
0334     __u32 cookie = ntohl(th->ack_seq) - 1;
0335     struct sock *ret = sk;
0336     struct request_sock *req;
0337     int full_space, mss;
0338     struct rtable *rt;
0339     __u8 rcv_wscale;
0340     struct flowi4 fl4;
0341     u32 tsoff = 0;
0342
0343     if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) ||
0344         !th->ack || th->rst)
0345         goto out;
0346
0347     if (tcp_synq_no_recent_overflow(sk))
0348         goto out;
0349
0350     mss = __cookie_v4_check(ip_hdr(skb), th, cookie);
0351     if (mss == 0) {
0352         __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
0353         goto out;
0354     }
0355
0356     __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
0357
0358     /* check for timestamp cookie support */
0359     memset(&tcp_opt, 0, sizeof(tcp_opt));
0360     tcp_parse_options(sock_net(sk), skb, &tcp_opt, 0, NULL);
0361
0362     if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
0363         tsoff = secure_tcp_ts_off(sock_net(sk),
0364                       ip_hdr(skb)->daddr,
0365                       ip_hdr(skb)->saddr);
0366         tcp_opt.rcv_tsecr -= tsoff;
0367     }
0368
0369     if (!cookie_timestamp_decode(sock_net(sk), &tcp_opt))
0370         goto out;
0371
0372     ret = NULL;
0373     req = cookie_tcp_reqsk_alloc(&tcp_request_sock_ops,
0374                      &tcp_request_sock_ipv4_ops, sk, skb);
0375     if (!req)
0376         goto out;
0377
0378     ireq = inet_rsk(req);
0379     treq = tcp_rsk(req);
0380     treq->rcv_isn       = ntohl(th->seq) - 1;
0381     treq->snt_isn       = cookie;
0382     treq->ts_off        = 0;
0383     treq->txhash        = net_tx_rndhash();
0384     req->mss        = mss;
0385     ireq->ir_num        = ntohs(th->dest);
0386     ireq->ir_rmt_port   = th->source;
0387     sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
0388     sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
0389     ireq->ir_mark       = inet_request_mark(sk, skb);
0390     ireq->snd_wscale    = tcp_opt.snd_wscale;
0391     ireq->sack_ok       = tcp_opt.sack_ok;
0392     ireq->wscale_ok     = tcp_opt.wscale_ok;
0393     ireq->tstamp_ok     = tcp_opt.saw_tstamp;
0394     req->ts_recent      = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
0395     treq->snt_synack    = 0;
0396     treq->tfo_listener  = false;
0397
0398     if (IS_ENABLED(CONFIG_SMC))
0399         ireq->smc_ok = 0;
0400
0401     ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
0402
0403     /* We throwed the options of the initial SYN away, so we hope
0404      * the ACK carries the same options again (see RFC1122 4.2.3.8)
0405      */
0406     RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(sock_net(sk), skb));
0407
0408     if (security_inet_conn_request(sk, skb, req)) {
0409         reqsk_free(req);
0410         goto out;
0411     }
0412
0413     req->num_retrans = 0;
0414
0415     /*
0416      * We need to lookup the route here to get at the correct
0417      * window size. We should better make sure that the window size
0418      * hasn't changed since we received the original syn, but I see
0419      * no easy way to do this.
0420      */
0421     flowi4_init_output(&fl4, ireq->ir_iif, ireq->ir_mark,
0422                RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, IPPROTO_TCP,
0423                inet_sk_flowi_flags(sk),
0424                opt->srr ? opt->faddr : ireq->ir_rmt_addr,
0425                ireq->ir_loc_addr, th->source, th->dest, sk->sk_uid);
0426     security_req_classify_flow(req, flowi4_to_flowi_common(&fl4));
0427     rt = ip_route_output_key(sock_net(sk), &fl4);
0428     if (IS_ERR(rt)) {
0429         reqsk_free(req);
0430         goto out;
0431     }
0432
0433     /* Try to redo what tcp_v4_send_synack did. */
0434     req->rsk_window_clamp = tp->window_clamp ? :dst_metric(&rt->dst, RTAX_WINDOW);
0435     /* limit the window selection if the user enforce a smaller rx buffer */
0436     full_space = tcp_full_space(sk);
0437     if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
0438         (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
0439         req->rsk_window_clamp = full_space;
0440
0441     tcp_select_initial_window(sk, full_space, req->mss,
0442                   &req->rsk_rcv_wnd, &req->rsk_window_clamp,
0443                   ireq->wscale_ok, &rcv_wscale,
0444                   dst_metric(&rt->dst, RTAX_INITRWND));
0445
0446     ireq->rcv_wscale  = rcv_wscale;
0447     ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), &rt->dst);
0448
0449     ret = tcp_get_cookie_sock(sk, skb, req, &rt->dst, tsoff);
0450     /* ip_queue_xmit() depends on our flow being setup
0451      * Normal sockets get it right from inet_csk_route_child_sock()
0452      */
0453     if (ret)
0454         inet_sk(ret)->cork.fl.u.ip4 = fl4;
0455 out:    return ret;
0456 }