net/ipv6/syncookies.c

0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  *  IPv6 Syncookies implementation for the Linux kernel
0004  *
0005  *  Authors:
0006  *  Glenn Griffin   <ggriffin.kernel@gmail.com>
0007  *
0008  *  Based on IPv4 implementation by Andi Kleen
0009  *  linux/net/ipv4/syncookies.c
0010  */
0011
0012 #include <linux/tcp.h>
0013 #include <linux/random.h>
0014 #include <linux/siphash.h>
0015 #include <linux/kernel.h>
0016 #include <net/secure_seq.h>
0017 #include <net/ipv6.h>
0018 #include <net/tcp.h>
0019
0020 #define COOKIEBITS 24   /* Upper bits store count */
0021 #define COOKIEMASK (((__u32)1 << COOKIEBITS) - 1)
0022
0023 static siphash_aligned_key_t syncookie6_secret[2];
0024
0025 /* RFC 2460, Section 8.3:
0026  * [ipv6 tcp] MSS must be computed as the maximum packet size minus 60 [..]
0027  *
0028  * Due to IPV6_MIN_MTU=1280 the lowest possible MSS is 1220, which allows
0029  * using higher values than ipv4 tcp syncookies.
0030  * The other values are chosen based on ethernet (1500 and 9k MTU), plus
0031  * one that accounts for common encap (PPPoe) overhead. Table must be sorted.
0032  */
0033 static __u16 const msstab[] = {
0034     1280 - 60, /* IPV6_MIN_MTU - 60 */
0035     1480 - 60,
0036     1500 - 60,
0037     9000 - 60,
0038 };
0039
0040 static u32 cookie_hash(const struct in6_addr *saddr,
0041                const struct in6_addr *daddr,
0042                __be16 sport, __be16 dport, u32 count, int c)
0043 {
0044     const struct {
0045         struct in6_addr saddr;
0046         struct in6_addr daddr;
0047         u32 count;
0048         __be16 sport;
0049         __be16 dport;
0050     } __aligned(SIPHASH_ALIGNMENT) combined = {
0051         .saddr = *saddr,
0052         .daddr = *daddr,
0053         .count = count,
0054         .sport = sport,
0055         .dport = dport
0056     };
0057
0058     net_get_random_once(syncookie6_secret, sizeof(syncookie6_secret));
0059     return siphash(&combined, offsetofend(typeof(combined), dport),
0060                &syncookie6_secret[c]);
0061 }
0062
0063 static __u32 secure_tcp_syn_cookie(const struct in6_addr *saddr,
0064                    const struct in6_addr *daddr,
0065                    __be16 sport, __be16 dport, __u32 sseq,
0066                    __u32 data)
0067 {
0068     u32 count = tcp_cookie_time();
0069     return (cookie_hash(saddr, daddr, sport, dport, 0, 0) +
0070         sseq + (count << COOKIEBITS) +
0071         ((cookie_hash(saddr, daddr, sport, dport, count, 1) + data)
0072         & COOKIEMASK));
0073 }
0074
0075 static __u32 check_tcp_syn_cookie(__u32 cookie, const struct in6_addr *saddr,
0076                   const struct in6_addr *daddr, __be16 sport,
0077                   __be16 dport, __u32 sseq)
0078 {
0079     __u32 diff, count = tcp_cookie_time();
0080
0081     cookie -= cookie_hash(saddr, daddr, sport, dport, 0, 0) + sseq;
0082
0083     diff = (count - (cookie >> COOKIEBITS)) & ((__u32) -1 >> COOKIEBITS);
0084     if (diff >= MAX_SYNCOOKIE_AGE)
0085         return (__u32)-1;
0086
0087     return (cookie -
0088         cookie_hash(saddr, daddr, sport, dport, count - diff, 1))
0089         & COOKIEMASK;
0090 }
0091
0092 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
0093                   const struct tcphdr *th, __u16 *mssp)
0094 {
0095     int mssind;
0096     const __u16 mss = *mssp;
0097
0098     for (mssind = ARRAY_SIZE(msstab) - 1; mssind ; mssind--)
0099         if (mss >= msstab[mssind])
0100             break;
0101
0102     *mssp = msstab[mssind];
0103
0104     return secure_tcp_syn_cookie(&iph->saddr, &iph->daddr, th->source,
0105                      th->dest, ntohl(th->seq), mssind);
0106 }
0107 EXPORT_SYMBOL_GPL(__cookie_v6_init_sequence);
0108
0109 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mssp)
0110 {
0111     const struct ipv6hdr *iph = ipv6_hdr(skb);
0112     const struct tcphdr *th = tcp_hdr(skb);
0113
0114     return __cookie_v6_init_sequence(iph, th, mssp);
0115 }
0116
0117 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
0118               __u32 cookie)
0119 {
0120     __u32 seq = ntohl(th->seq) - 1;
0121     __u32 mssind = check_tcp_syn_cookie(cookie, &iph->saddr, &iph->daddr,
0122                         th->source, th->dest, seq);
0123
0124     return mssind < ARRAY_SIZE(msstab) ? msstab[mssind] : 0;
0125 }
0126 EXPORT_SYMBOL_GPL(__cookie_v6_check);
0127
0128 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb)
0129 {
0130     struct tcp_options_received tcp_opt;
0131     struct inet_request_sock *ireq;
0132     struct tcp_request_sock *treq;
0133     struct ipv6_pinfo *np = inet6_sk(sk);
0134     struct tcp_sock *tp = tcp_sk(sk);
0135     const struct tcphdr *th = tcp_hdr(skb);
0136     __u32 cookie = ntohl(th->ack_seq) - 1;
0137     struct sock *ret = sk;
0138     struct request_sock *req;
0139     int full_space, mss;
0140     struct dst_entry *dst;
0141     __u8 rcv_wscale;
0142     u32 tsoff = 0;
0143
0144     if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) ||
0145         !th->ack || th->rst)
0146         goto out;
0147
0148     if (tcp_synq_no_recent_overflow(sk))
0149         goto out;
0150
0151     mss = __cookie_v6_check(ipv6_hdr(skb), th, cookie);
0152     if (mss == 0) {
0153         __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESFAILED);
0154         goto out;
0155     }
0156
0157     __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESRECV);
0158
0159     /* check for timestamp cookie support */
0160     memset(&tcp_opt, 0, sizeof(tcp_opt));
0161     tcp_parse_options(sock_net(sk), skb, &tcp_opt, 0, NULL);
0162
0163     if (tcp_opt.saw_tstamp && tcp_opt.rcv_tsecr) {
0164         tsoff = secure_tcpv6_ts_off(sock_net(sk),
0165                         ipv6_hdr(skb)->daddr.s6_addr32,
0166                         ipv6_hdr(skb)->saddr.s6_addr32);
0167         tcp_opt.rcv_tsecr -= tsoff;
0168     }
0169
0170     if (!cookie_timestamp_decode(sock_net(sk), &tcp_opt))
0171         goto out;
0172
0173     ret = NULL;
0174     req = cookie_tcp_reqsk_alloc(&tcp6_request_sock_ops,
0175                      &tcp_request_sock_ipv6_ops, sk, skb);
0176     if (!req)
0177         goto out;
0178
0179     ireq = inet_rsk(req);
0180     treq = tcp_rsk(req);
0181     treq->tfo_listener = false;
0182
0183     if (security_inet_conn_request(sk, skb, req))
0184         goto out_free;
0185
0186     req->mss = mss;
0187     ireq->ir_rmt_port = th->source;
0188     ireq->ir_num = ntohs(th->dest);
0189     ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
0190     ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
0191     if (ipv6_opt_accepted(sk, skb, &TCP_SKB_CB(skb)->header.h6) ||
0192         np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
0193         np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
0194         refcount_inc(&skb->users);
0195         ireq->pktopts = skb;
0196     }
0197
0198     ireq->ir_iif = inet_request_bound_dev_if(sk, skb);
0199     /* So that link locals have meaning */
0200     if (!sk->sk_bound_dev_if &&
0201         ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
0202         ireq->ir_iif = tcp_v6_iif(skb);
0203
0204     ireq->ir_mark = inet_request_mark(sk, skb);
0205
0206     req->num_retrans = 0;
0207     ireq->snd_wscale    = tcp_opt.snd_wscale;
0208     ireq->sack_ok       = tcp_opt.sack_ok;
0209     ireq->wscale_ok     = tcp_opt.wscale_ok;
0210     ireq->tstamp_ok     = tcp_opt.saw_tstamp;
0211     req->ts_recent      = tcp_opt.saw_tstamp ? tcp_opt.rcv_tsval : 0;
0212     treq->snt_synack    = 0;
0213     treq->rcv_isn = ntohl(th->seq) - 1;
0214     treq->snt_isn = cookie;
0215     treq->ts_off = 0;
0216     treq->txhash = net_tx_rndhash();
0217     if (IS_ENABLED(CONFIG_SMC))
0218         ireq->smc_ok = 0;
0219
0220     /*
0221      * We need to lookup the dst_entry to get the correct window size.
0222      * This is taken from tcp_v6_syn_recv_sock.  Somebody please enlighten
0223      * me if there is a preferred way.
0224      */
0225     {
0226         struct in6_addr *final_p, final;
0227         struct flowi6 fl6;
0228         memset(&fl6, 0, sizeof(fl6));
0229         fl6.flowi6_proto = IPPROTO_TCP;
0230         fl6.daddr = ireq->ir_v6_rmt_addr;
0231         final_p = fl6_update_dst(&fl6, rcu_dereference(np->opt), &final);
0232         fl6.saddr = ireq->ir_v6_loc_addr;
0233         fl6.flowi6_oif = ireq->ir_iif;
0234         fl6.flowi6_mark = ireq->ir_mark;
0235         fl6.fl6_dport = ireq->ir_rmt_port;
0236         fl6.fl6_sport = inet_sk(sk)->inet_sport;
0237         fl6.flowi6_uid = sk->sk_uid;
0238         security_req_classify_flow(req, flowi6_to_flowi_common(&fl6));
0239
0240         dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
0241         if (IS_ERR(dst))
0242             goto out_free;
0243     }
0244
0245     req->rsk_window_clamp = tp->window_clamp ? :dst_metric(dst, RTAX_WINDOW);
0246     /* limit the window selection if the user enforce a smaller rx buffer */
0247     full_space = tcp_full_space(sk);
0248     if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
0249         (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
0250         req->rsk_window_clamp = full_space;
0251
0252     tcp_select_initial_window(sk, full_space, req->mss,
0253                   &req->rsk_rcv_wnd, &req->rsk_window_clamp,
0254                   ireq->wscale_ok, &rcv_wscale,
0255                   dst_metric(dst, RTAX_INITRWND));
0256
0257     ireq->rcv_wscale = rcv_wscale;
0258     ireq->ecn_ok = cookie_ecn_ok(&tcp_opt, sock_net(sk), dst);
0259
0260     ret = tcp_get_cookie_sock(sk, skb, req, dst, tsoff);
0261 out:
0262     return ret;
0263 out_free:
0264     reqsk_free(req);
0265     return NULL;
0266 }