net/ipv4/tcp_ao.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP Authentication Option (TCP-AO).
   4  *              See RFC5925.
   5  *
   6  * Authors:     Dmitry Safonov <[email protected]>
   7  *              Francesco Ruggeri <[email protected]>
   8  *              Salam Noureddine <[email protected]>
   9  */
  10 #define pr_fmt(fmt) "TCP: " fmt
  11
  12 #include <crypto/hash.h>
  13 #include <linux/inetdevice.h>
  14 #include <linux/tcp.h>
  15
  16 #include <net/tcp.h>
  17 #include <net/ipv6.h>
  18 #include <net/icmp.h>
  19
  20 DEFINE_STATIC_KEY_DEFERRED_FALSE(tcp_ao_needed, HZ);
  21
  22 int tcp_ao_calc_traffic_key(struct tcp_ao_key *mkt, u8 *key, void *ctx,
  23                             unsigned int len, struct tcp_sigpool *hp)
  24 {
  25         struct scatterlist sg;
  26         int ret;
  27
  28         if (crypto_ahash_setkey(crypto_ahash_reqtfm(hp->req),
  29                                 mkt->key, mkt->keylen))
  30                 goto clear_hash;
  31
  32         ret = crypto_ahash_init(hp->req);
  33         if (ret)
  34                 goto clear_hash;
  35
  36         sg_init_one(&sg, ctx, len);
  37         ahash_request_set_crypt(hp->req, &sg, key, len);
  38         crypto_ahash_update(hp->req);
  39
  40         ret = crypto_ahash_final(hp->req);
  41         if (ret)
  42                 goto clear_hash;
  43
  44         return 0;
  45 clear_hash:
  46         memset(key, 0, tcp_ao_digest_size(mkt));
  47         return 1;
  48 }
  49
  50 bool tcp_ao_ignore_icmp(const struct sock *sk, int family, int type, int code)
  51 {
  52         bool ignore_icmp = false;
  53         struct tcp_ao_info *ao;
  54
  55         if (!static_branch_unlikely(&tcp_ao_needed.key))
  56                 return false;
  57
  58         /* RFC5925, 7.8:
  59          * >> A TCP-AO implementation MUST default to ignore incoming ICMPv4
  60          * messages of Type 3 (destination unreachable), Codes 2-4 (protocol
  61          * unreachable, port unreachable, and fragmentation needed -- ’hard
  62          * errors’), and ICMPv6 Type 1 (destination unreachable), Code 1
  63          * (administratively prohibited) and Code 4 (port unreachable) intended
  64          * for connections in synchronized states (ESTABLISHED, FIN-WAIT-1, FIN-
  65          * WAIT-2, CLOSE-WAIT, CLOSING, LAST-ACK, TIME-WAIT) that match MKTs.
  66          */
  67         if (family == AF_INET) {
  68                 if (type != ICMP_DEST_UNREACH)
  69                         return false;
  70                 if (code < ICMP_PROT_UNREACH || code > ICMP_FRAG_NEEDED)
  71                         return false;
  72         } else {
  73                 if (type != ICMPV6_DEST_UNREACH)
  74                         return false;
  75                 if (code != ICMPV6_ADM_PROHIBITED && code != ICMPV6_PORT_UNREACH)
  76                         return false;
  77         }
  78
  79         rcu_read_lock();
  80         switch (sk->sk_state) {
  81         case TCP_TIME_WAIT:
  82                 ao = rcu_dereference(tcp_twsk(sk)->ao_info);
  83                 break;
  84         case TCP_SYN_SENT:
  85         case TCP_SYN_RECV:
  86         case TCP_LISTEN:
  87         case TCP_NEW_SYN_RECV:
  88                 /* RFC5925 specifies to ignore ICMPs *only* on connections
  89                  * in synchronized states.
  90                  */
  91                 rcu_read_unlock();
  92                 return false;
  93         default:
  94                 ao = rcu_dereference(tcp_sk(sk)->ao_info);
  95         }
  96
  97         if (ao && !ao->accept_icmps) {
  98                 ignore_icmp = true;
  99                 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAODROPPEDICMPS);
 100                 atomic64_inc(&ao->counters.dropped_icmp);
 101         }
 102         rcu_read_unlock();
 103
 104         return ignore_icmp;
 105 }
 106
 107 /* Optimized version of tcp_ao_do_lookup(): only for sockets for which
 108  * it's known that the keys in ao_info are matching peer's
 109  * family/address/VRF/etc.
 110  */
 111 struct tcp_ao_key *tcp_ao_established_key(struct tcp_ao_info *ao,
 112                                           int sndid, int rcvid)
 113 {
 114         struct tcp_ao_key *key;
 115
 116         hlist_for_each_entry_rcu(key, &ao->head, node) {
 117                 if ((sndid >= 0 && key->sndid != sndid) ||
 118                     (rcvid >= 0 && key->rcvid != rcvid))
 119                         continue;
 120                 return key;
 121         }
 122
 123         return NULL;
 124 }
 125
 126 static int ipv4_prefix_cmp(const struct in_addr *addr1,
 127                            const struct in_addr *addr2,
 128                            unsigned int prefixlen)
 129 {
 130         __be32 mask = inet_make_mask(prefixlen);
 131         __be32 a1 = addr1->s_addr & mask;
 132         __be32 a2 = addr2->s_addr & mask;
 133
 134         if (a1 == a2)
 135                 return 0;
 136         return memcmp(&a1, &a2, sizeof(a1));
 137 }
 138
 139 static int __tcp_ao_key_cmp(const struct tcp_ao_key *key, int l3index,
 140                             const union tcp_ao_addr *addr, u8 prefixlen,
 141                             int family, int sndid, int rcvid)
 142 {
 143         if (sndid >= 0 && key->sndid != sndid)
 144                 return (key->sndid > sndid) ? 1 : -1;
 145         if (rcvid >= 0 && key->rcvid != rcvid)
 146                 return (key->rcvid > rcvid) ? 1 : -1;
 147         if (l3index >= 0 && (key->keyflags & TCP_AO_KEYF_IFINDEX)) {
 148                 if (key->l3index != l3index)
 149                         return (key->l3index > l3index) ? 1 : -1;
 150         }
 151
 152         if (family == AF_UNSPEC)
 153                 return 0;
 154         if (key->family != family)
 155                 return (key->family > family) ? 1 : -1;
 156
 157         if (family == AF_INET) {
 158                 if (ntohl(key->addr.a4.s_addr) == INADDR_ANY)
 159                         return 0;
 160                 if (ntohl(addr->a4.s_addr) == INADDR_ANY)
 161                         return 0;
 162                 return ipv4_prefix_cmp(&key->addr.a4, &addr->a4, prefixlen);
 163 #if IS_ENABLED(CONFIG_IPV6)
 164         } else {
 165                 if (ipv6_addr_any(&key->addr.a6) || ipv6_addr_any(&addr->a6))
 166                         return 0;
 167                 if (ipv6_prefix_equal(&key->addr.a6, &addr->a6, prefixlen))
 168                         return 0;
 169                 return memcmp(&key->addr.a6, &addr->a6, sizeof(addr->a6));
 170 #endif
 171         }
 172         return -1;
 173 }
 174
 175 static int tcp_ao_key_cmp(const struct tcp_ao_key *key, int l3index,
 176                           const union tcp_ao_addr *addr, u8 prefixlen,
 177                           int family, int sndid, int rcvid)
 178 {
 179 #if IS_ENABLED(CONFIG_IPV6)
 180         if (family == AF_INET6 && ipv6_addr_v4mapped(&addr->a6)) {
 181                 __be32 addr4 = addr->a6.s6_addr32[3];
 182
 183                 return __tcp_ao_key_cmp(key, l3index,
 184                                         (union tcp_ao_addr *)&addr4,
 185                                         prefixlen, AF_INET, sndid, rcvid);
 186         }
 187 #endif
 188         return __tcp_ao_key_cmp(key, l3index, addr,
 189                                 prefixlen, family, sndid, rcvid);
 190 }
 191
 192 static struct tcp_ao_key *__tcp_ao_do_lookup(const struct sock *sk, int l3index,
 193                 const union tcp_ao_addr *addr, int family, u8 prefix,
 194                 int sndid, int rcvid)
 195 {
 196         struct tcp_ao_key *key;
 197         struct tcp_ao_info *ao;
 198
 199         if (!static_branch_unlikely(&tcp_ao_needed.key))
 200                 return NULL;
 201
 202         ao = rcu_dereference_check(tcp_sk(sk)->ao_info,
 203                                    lockdep_sock_is_held(sk));
 204         if (!ao)
 205                 return NULL;
 206
 207         hlist_for_each_entry_rcu(key, &ao->head, node) {
 208                 u8 prefixlen = min(prefix, key->prefixlen);
 209
 210                 if (!tcp_ao_key_cmp(key, l3index, addr, prefixlen,
 211                                     family, sndid, rcvid))
 212                         return key;
 213         }
 214         return NULL;
 215 }
 216
 217 struct tcp_ao_key *tcp_ao_do_lookup(const struct sock *sk, int l3index,
 218                                     const union tcp_ao_addr *addr,
 219                                     int family, int sndid, int rcvid)
 220 {
 221         return __tcp_ao_do_lookup(sk, l3index, addr, family, U8_MAX, sndid, rcvid);
 222 }
 223
 224 static struct tcp_ao_info *tcp_ao_alloc_info(gfp_t flags)
 225 {
 226         struct tcp_ao_info *ao;
 227
 228         ao = kzalloc(sizeof(*ao), flags);
 229         if (!ao)
 230                 return NULL;
 231         INIT_HLIST_HEAD(&ao->head);
 232         refcount_set(&ao->refcnt, 1);
 233
 234         return ao;
 235 }
 236
 237 static void tcp_ao_link_mkt(struct tcp_ao_info *ao, struct tcp_ao_key *mkt)
 238 {
 239         hlist_add_head_rcu(&mkt->node, &ao->head);
 240 }
 241
 242 static struct tcp_ao_key *tcp_ao_copy_key(struct sock *sk,
 243                                           struct tcp_ao_key *key)
 244 {
 245         struct tcp_ao_key *new_key;
 246
 247         new_key = sock_kmalloc(sk, tcp_ao_sizeof_key(key),
 248                                GFP_ATOMIC);
 249         if (!new_key)
 250                 return NULL;
 251
 252         *new_key = *key;
 253         INIT_HLIST_NODE(&new_key->node);
 254         tcp_sigpool_get(new_key->tcp_sigpool_id);
 255         atomic64_set(&new_key->pkt_good, 0);
 256         atomic64_set(&new_key->pkt_bad, 0);
 257
 258         return new_key;
 259 }
 260
 261 static void tcp_ao_key_free_rcu(struct rcu_head *head)
 262 {
 263         struct tcp_ao_key *key = container_of(head, struct tcp_ao_key, rcu);
 264
 265         tcp_sigpool_release(key->tcp_sigpool_id);
 266         kfree_sensitive(key);
 267 }
 268
 269 void tcp_ao_destroy_sock(struct sock *sk, bool twsk)
 270 {
 271         struct tcp_ao_info *ao;
 272         struct tcp_ao_key *key;
 273         struct hlist_node *n;
 274
 275         if (twsk) {
 276                 ao = rcu_dereference_protected(tcp_twsk(sk)->ao_info, 1);
 277                 tcp_twsk(sk)->ao_info = NULL;
 278         } else {
 279                 ao = rcu_dereference_protected(tcp_sk(sk)->ao_info, 1);
 280                 tcp_sk(sk)->ao_info = NULL;
 281         }
 282
 283         if (!ao || !refcount_dec_and_test(&ao->refcnt))
 284                 return;
 285
 286         hlist_for_each_entry_safe(key, n, &ao->head, node) {
 287                 hlist_del_rcu(&key->node);
 288                 if (!twsk)
 289                         atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
 290                 call_rcu(&key->rcu, tcp_ao_key_free_rcu);
 291         }
 292
 293         kfree_rcu(ao, rcu);
 294         static_branch_slow_dec_deferred(&tcp_ao_needed);
 295 }
 296
 297 void tcp_ao_time_wait(struct tcp_timewait_sock *tcptw, struct tcp_sock *tp)
 298 {
 299         struct tcp_ao_info *ao_info = rcu_dereference_protected(tp->ao_info, 1);
 300
 301         if (ao_info) {
 302                 struct tcp_ao_key *key;
 303                 struct hlist_node *n;
 304                 int omem = 0;
 305
 306                 hlist_for_each_entry_safe(key, n, &ao_info->head, node) {
 307                         omem += tcp_ao_sizeof_key(key);
 308                 }
 309
 310                 refcount_inc(&ao_info->refcnt);
 311                 atomic_sub(omem, &(((struct sock *)tp)->sk_omem_alloc));
 312                 rcu_assign_pointer(tcptw->ao_info, ao_info);
 313         } else {
 314                 tcptw->ao_info = NULL;
 315         }
 316 }
 317
 318 /* 4 tuple and ISNs are expected in NBO */
 319 static int tcp_v4_ao_calc_key(struct tcp_ao_key *mkt, u8 *key,
 320                               __be32 saddr, __be32 daddr,
 321                               __be16 sport, __be16 dport,
 322                               __be32 sisn,  __be32 disn)
 323 {
 324         /* See RFC5926 3.1.1 */
 325         struct kdf_input_block {
 326                 u8                      counter;
 327                 u8                      label[6];
 328                 struct tcp4_ao_context  ctx;
 329                 __be16                  outlen;
 330         } __packed * tmp;
 331         struct tcp_sigpool hp;
 332         int err;
 333
 334         err = tcp_sigpool_start(mkt->tcp_sigpool_id, &hp);
 335         if (err)
 336                 return err;
 337
 338         tmp = hp.scratch;
 339         tmp->counter    = 1;
 340         memcpy(tmp->label, "TCP-AO", 6);
 341         tmp->ctx.saddr  = saddr;
 342         tmp->ctx.daddr  = daddr;
 343         tmp->ctx.sport  = sport;
 344         tmp->ctx.dport  = dport;
 345         tmp->ctx.sisn   = sisn;
 346         tmp->ctx.disn   = disn;
 347         tmp->outlen     = htons(tcp_ao_digest_size(mkt) * 8); /* in bits */
 348
 349         err = tcp_ao_calc_traffic_key(mkt, key, tmp, sizeof(*tmp), &hp);
 350         tcp_sigpool_end(&hp);
 351
 352         return err;
 353 }
 354
 355 int tcp_v4_ao_calc_key_sk(struct tcp_ao_key *mkt, u8 *key,
 356                           const struct sock *sk,
 357                           __be32 sisn, __be32 disn, bool send)
 358 {
 359         if (send)
 360                 return tcp_v4_ao_calc_key(mkt, key, sk->sk_rcv_saddr,
 361                                           sk->sk_daddr, htons(sk->sk_num),
 362                                           sk->sk_dport, sisn, disn);
 363         else
 364                 return tcp_v4_ao_calc_key(mkt, key, sk->sk_daddr,
 365                                           sk->sk_rcv_saddr, sk->sk_dport,
 366                                           htons(sk->sk_num), disn, sisn);
 367 }
 368
 369 static int tcp_ao_calc_key_sk(struct tcp_ao_key *mkt, u8 *key,
 370                               const struct sock *sk,
 371                               __be32 sisn, __be32 disn, bool send)
 372 {
 373         if (mkt->family == AF_INET)
 374                 return tcp_v4_ao_calc_key_sk(mkt, key, sk, sisn, disn, send);
 375 #if IS_ENABLED(CONFIG_IPV6)
 376         else if (mkt->family == AF_INET6)
 377                 return tcp_v6_ao_calc_key_sk(mkt, key, sk, sisn, disn, send);
 378 #endif
 379         else
 380                 return -EOPNOTSUPP;
 381 }
 382
 383 int tcp_v4_ao_calc_key_rsk(struct tcp_ao_key *mkt, u8 *key,
 384                            struct request_sock *req)
 385 {
 386         struct inet_request_sock *ireq = inet_rsk(req);
 387
 388         return tcp_v4_ao_calc_key(mkt, key,
 389                                   ireq->ir_loc_addr, ireq->ir_rmt_addr,
 390                                   htons(ireq->ir_num), ireq->ir_rmt_port,
 391                                   htonl(tcp_rsk(req)->snt_isn),
 392                                   htonl(tcp_rsk(req)->rcv_isn));
 393 }
 394
 395 static int tcp_v4_ao_calc_key_skb(struct tcp_ao_key *mkt, u8 *key,
 396                                   const struct sk_buff *skb,
 397                                   __be32 sisn, __be32 disn)
 398 {
 399         const struct iphdr *iph = ip_hdr(skb);
 400         const struct tcphdr *th = tcp_hdr(skb);
 401
 402         return tcp_v4_ao_calc_key(mkt, key, iph->saddr, iph->daddr,
 403                                   th->source, th->dest, sisn, disn);
 404 }
 405
 406 static int tcp_ao_calc_key_skb(struct tcp_ao_key *mkt, u8 *key,
 407                                const struct sk_buff *skb,
 408                                __be32 sisn, __be32 disn, int family)
 409 {
 410         if (family == AF_INET)
 411                 return tcp_v4_ao_calc_key_skb(mkt, key, skb, sisn, disn);
 412 #if IS_ENABLED(CONFIG_IPV6)
 413         else if (family == AF_INET6)
 414                 return tcp_v6_ao_calc_key_skb(mkt, key, skb, sisn, disn);
 415 #endif
 416         return -EAFNOSUPPORT;
 417 }
 418
 419 static int tcp_v4_ao_hash_pseudoheader(struct tcp_sigpool *hp,
 420                                        __be32 daddr, __be32 saddr,
 421                                        int nbytes)
 422 {
 423         struct tcp4_pseudohdr *bp;
 424         struct scatterlist sg;
 425
 426         bp = hp->scratch;
 427         bp->saddr = saddr;
 428         bp->daddr = daddr;
 429         bp->pad = 0;
 430         bp->protocol = IPPROTO_TCP;
 431         bp->len = cpu_to_be16(nbytes);
 432
 433         sg_init_one(&sg, bp, sizeof(*bp));
 434         ahash_request_set_crypt(hp->req, &sg, NULL, sizeof(*bp));
 435         return crypto_ahash_update(hp->req);
 436 }
 437
 438 static int tcp_ao_hash_pseudoheader(unsigned short int family,
 439                                     const struct sock *sk,
 440                                     const struct sk_buff *skb,
 441                                     struct tcp_sigpool *hp, int nbytes)
 442 {
 443         const struct tcphdr *th = tcp_hdr(skb);
 444
 445         /* TODO: Can we rely on checksum being zero to mean outbound pkt? */
 446         if (!th->check) {
 447                 if (family == AF_INET)
 448                         return tcp_v4_ao_hash_pseudoheader(hp, sk->sk_daddr,
 449                                         sk->sk_rcv_saddr, skb->len);
 450 #if IS_ENABLED(CONFIG_IPV6)
 451                 else if (family == AF_INET6)
 452                         return tcp_v6_ao_hash_pseudoheader(hp, &sk->sk_v6_daddr,
 453                                         &sk->sk_v6_rcv_saddr, skb->len);
 454 #endif
 455                 else
 456                         return -EAFNOSUPPORT;
 457         }
 458
 459         if (family == AF_INET) {
 460                 const struct iphdr *iph = ip_hdr(skb);
 461
 462                 return tcp_v4_ao_hash_pseudoheader(hp, iph->daddr,
 463                                 iph->saddr, skb->len);
 464 #if IS_ENABLED(CONFIG_IPV6)
 465         } else if (family == AF_INET6) {
 466                 const struct ipv6hdr *iph = ipv6_hdr(skb);
 467
 468                 return tcp_v6_ao_hash_pseudoheader(hp, &iph->daddr,
 469                                 &iph->saddr, skb->len);
 470 #endif
 471         }
 472         return -EAFNOSUPPORT;
 473 }
 474
 475 u32 tcp_ao_compute_sne(u32 next_sne, u32 next_seq, u32 seq)
 476 {
 477         u32 sne = next_sne;
 478
 479         if (before(seq, next_seq)) {
 480                 if (seq > next_seq)
 481                         sne--;
 482         } else {
 483                 if (seq < next_seq)
 484                         sne++;
 485         }
 486
 487         return sne;
 488 }
 489
 490 /* tcp_ao_hash_sne(struct tcp_sigpool *hp)
 491  * @hp  - used for hashing
 492  * @sne - sne value
 493  */
 494 static int tcp_ao_hash_sne(struct tcp_sigpool *hp, u32 sne)
 495 {
 496         struct scatterlist sg;
 497         __be32 *bp;
 498
 499         bp = (__be32 *)hp->scratch;
 500         *bp = htonl(sne);
 501
 502         sg_init_one(&sg, bp, sizeof(*bp));
 503         ahash_request_set_crypt(hp->req, &sg, NULL, sizeof(*bp));
 504         return crypto_ahash_update(hp->req);
 505 }
 506
 507 static int tcp_ao_hash_header(struct tcp_sigpool *hp,
 508                               const struct tcphdr *th,
 509                               bool exclude_options, u8 *hash,
 510                               int hash_offset, int hash_len)
 511 {
 512         struct scatterlist sg;
 513         u8 *hdr = hp->scratch;
 514         int err, len;
 515
 516         /* We are not allowed to change tcphdr, make a local copy */
 517         if (exclude_options) {
 518                 len = sizeof(*th) + sizeof(struct tcp_ao_hdr) + hash_len;
 519                 memcpy(hdr, th, sizeof(*th));
 520                 memcpy(hdr + sizeof(*th),
 521                        (u8 *)th + hash_offset - sizeof(struct tcp_ao_hdr),
 522                        sizeof(struct tcp_ao_hdr));
 523                 memset(hdr + sizeof(*th) + sizeof(struct tcp_ao_hdr),
 524                        0, hash_len);
 525                 ((struct tcphdr *)hdr)->check = 0;
 526         } else {
 527                 len = th->doff << 2;
 528                 memcpy(hdr, th, len);
 529                 /* zero out tcp-ao hash */
 530                 ((struct tcphdr *)hdr)->check = 0;
 531                 memset(hdr + hash_offset, 0, hash_len);
 532         }
 533
 534         sg_init_one(&sg, hdr, len);
 535         ahash_request_set_crypt(hp->req, &sg, NULL, len);
 536         err = crypto_ahash_update(hp->req);
 537         WARN_ON_ONCE(err != 0);
 538         return err;
 539 }
 540
 541 int tcp_ao_hash_hdr(unsigned short int family, char *ao_hash,
 542                     struct tcp_ao_key *key, const u8 *tkey,
 543                     const union tcp_ao_addr *daddr,
 544                     const union tcp_ao_addr *saddr,
 545                     const struct tcphdr *th, u32 sne)
 546 {
 547         int tkey_len = tcp_ao_digest_size(key);
 548         int hash_offset = ao_hash - (char *)th;
 549         struct tcp_sigpool hp;
 550         void *hash_buf = NULL;
 551
 552         hash_buf = kmalloc(tkey_len, GFP_ATOMIC);
 553         if (!hash_buf)
 554                 goto clear_hash_noput;
 555
 556         if (tcp_sigpool_start(key->tcp_sigpool_id, &hp))
 557                 goto clear_hash_noput;
 558
 559         if (crypto_ahash_setkey(crypto_ahash_reqtfm(hp.req), tkey, tkey_len))
 560                 goto clear_hash;
 561
 562         if (crypto_ahash_init(hp.req))
 563                 goto clear_hash;
 564
 565         if (tcp_ao_hash_sne(&hp, sne))
 566                 goto clear_hash;
 567         if (family == AF_INET) {
 568                 if (tcp_v4_ao_hash_pseudoheader(&hp, daddr->a4.s_addr,
 569                                                 saddr->a4.s_addr, th->doff * 4))
 570                         goto clear_hash;
 571 #if IS_ENABLED(CONFIG_IPV6)
 572         } else if (family == AF_INET6) {
 573                 if (tcp_v6_ao_hash_pseudoheader(&hp, &daddr->a6,
 574                                                 &saddr->a6, th->doff * 4))
 575                         goto clear_hash;
 576 #endif
 577         } else {
 578                 WARN_ON_ONCE(1);
 579                 goto clear_hash;
 580         }
 581         if (tcp_ao_hash_header(&hp, th,
 582                                !!(key->keyflags & TCP_AO_KEYF_EXCLUDE_OPT),
 583                                ao_hash, hash_offset, tcp_ao_maclen(key)))
 584                 goto clear_hash;
 585         ahash_request_set_crypt(hp.req, NULL, hash_buf, 0);
 586         if (crypto_ahash_final(hp.req))
 587                 goto clear_hash;
 588
 589         memcpy(ao_hash, hash_buf, tcp_ao_maclen(key));
 590         tcp_sigpool_end(&hp);
 591         kfree(hash_buf);
 592         return 0;
 593
 594 clear_hash:
 595         tcp_sigpool_end(&hp);
 596 clear_hash_noput:
 597         memset(ao_hash, 0, tcp_ao_maclen(key));
 598         kfree(hash_buf);
 599         return 1;
 600 }
 601
 602 int tcp_ao_hash_skb(unsigned short int family,
 603                     char *ao_hash, struct tcp_ao_key *key,
 604                     const struct sock *sk, const struct sk_buff *skb,
 605                     const u8 *tkey, int hash_offset, u32 sne)
 606 {
 607         const struct tcphdr *th = tcp_hdr(skb);
 608         int tkey_len = tcp_ao_digest_size(key);
 609         struct tcp_sigpool hp;
 610         void *hash_buf = NULL;
 611
 612         hash_buf = kmalloc(tkey_len, GFP_ATOMIC);
 613         if (!hash_buf)
 614                 goto clear_hash_noput;
 615
 616         if (tcp_sigpool_start(key->tcp_sigpool_id, &hp))
 617                 goto clear_hash_noput;
 618
 619         if (crypto_ahash_setkey(crypto_ahash_reqtfm(hp.req), tkey, tkey_len))
 620                 goto clear_hash;
 621
 622         /* For now use sha1 by default. Depends on alg in tcp_ao_key */
 623         if (crypto_ahash_init(hp.req))
 624                 goto clear_hash;
 625
 626         if (tcp_ao_hash_sne(&hp, sne))
 627                 goto clear_hash;
 628         if (tcp_ao_hash_pseudoheader(family, sk, skb, &hp, skb->len))
 629                 goto clear_hash;
 630         if (tcp_ao_hash_header(&hp, th,
 631                                !!(key->keyflags & TCP_AO_KEYF_EXCLUDE_OPT),
 632                                ao_hash, hash_offset, tcp_ao_maclen(key)))
 633                 goto clear_hash;
 634         if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2))
 635                 goto clear_hash;
 636         ahash_request_set_crypt(hp.req, NULL, hash_buf, 0);
 637         if (crypto_ahash_final(hp.req))
 638                 goto clear_hash;
 639
 640         memcpy(ao_hash, hash_buf, tcp_ao_maclen(key));
 641         tcp_sigpool_end(&hp);
 642         kfree(hash_buf);
 643         return 0;
 644
 645 clear_hash:
 646         tcp_sigpool_end(&hp);
 647 clear_hash_noput:
 648         memset(ao_hash, 0, tcp_ao_maclen(key));
 649         kfree(hash_buf);
 650         return 1;
 651 }
 652
 653 int tcp_v4_ao_hash_skb(char *ao_hash, struct tcp_ao_key *key,
 654                        const struct sock *sk, const struct sk_buff *skb,
 655                        const u8 *tkey, int hash_offset, u32 sne)
 656 {
 657         return tcp_ao_hash_skb(AF_INET, ao_hash, key, sk, skb,
 658                                tkey, hash_offset, sne);
 659 }
 660
 661 int tcp_v4_ao_synack_hash(char *ao_hash, struct tcp_ao_key *ao_key,
 662                           struct request_sock *req, const struct sk_buff *skb,
 663                           int hash_offset, u32 sne)
 664 {
 665         void *hash_buf = NULL;
 666         int err;
 667
 668         hash_buf = kmalloc(tcp_ao_digest_size(ao_key), GFP_ATOMIC);
 669         if (!hash_buf)
 670                 return -ENOMEM;
 671
 672         err = tcp_v4_ao_calc_key_rsk(ao_key, hash_buf, req);
 673         if (err)
 674                 goto out;
 675
 676         err = tcp_ao_hash_skb(AF_INET, ao_hash, ao_key, req_to_sk(req), skb,
 677                               hash_buf, hash_offset, sne);
 678 out:
 679         kfree(hash_buf);
 680         return err;
 681 }
 682
 683 struct tcp_ao_key *tcp_v4_ao_lookup_rsk(const struct sock *sk,
 684                                         struct request_sock *req,
 685                                         int sndid, int rcvid)
 686 {
 687         struct inet_request_sock *ireq = inet_rsk(req);
 688         union tcp_ao_addr *addr = (union tcp_ao_addr *)&ireq->ir_rmt_addr;
 689         int l3index;
 690
 691         l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
 692         return tcp_ao_do_lookup(sk, l3index, addr, AF_INET, sndid, rcvid);
 693 }
 694
 695 struct tcp_ao_key *tcp_v4_ao_lookup(const struct sock *sk, struct sock *addr_sk,
 696                                     int sndid, int rcvid)
 697 {
 698         int l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
 699                                                      addr_sk->sk_bound_dev_if);
 700         union tcp_ao_addr *addr = (union tcp_ao_addr *)&addr_sk->sk_daddr;
 701
 702         return tcp_ao_do_lookup(sk, l3index, addr, AF_INET, sndid, rcvid);
 703 }
 704
 705 int tcp_ao_prepare_reset(const struct sock *sk, struct sk_buff *skb,
 706                          const struct tcp_ao_hdr *aoh, int l3index, u32 seq,
 707                          struct tcp_ao_key **key, char **traffic_key,
 708                          bool *allocated_traffic_key, u8 *keyid, u32 *sne)
 709 {
 710         const struct tcphdr *th = tcp_hdr(skb);
 711         struct tcp_ao_info *ao_info;
 712
 713         *allocated_traffic_key = false;
 714         /* If there's no socket - than initial sisn/disn are unknown.
 715          * Drop the segment. RFC5925 (7.7) advises to require graceful
 716          * restart [RFC4724]. Alternatively, the RFC5925 advises to
 717          * save/restore traffic keys before/after reboot.
 718          * Linux TCP-AO support provides TCP_AO_ADD_KEY and TCP_AO_REPAIR
 719          * options to restore a socket post-reboot.
 720          */
 721         if (!sk)
 722                 return -ENOTCONN;
 723
 724         if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV)) {
 725                 unsigned int family = READ_ONCE(sk->sk_family);
 726                 union tcp_ao_addr *addr;
 727                 __be32 disn, sisn;
 728
 729                 if (sk->sk_state == TCP_NEW_SYN_RECV) {
 730                         struct request_sock *req = inet_reqsk(sk);
 731
 732                         sisn = htonl(tcp_rsk(req)->rcv_isn);
 733                         disn = htonl(tcp_rsk(req)->snt_isn);
 734                         *sne = tcp_ao_compute_sne(0, tcp_rsk(req)->snt_isn, seq);
 735                 } else {
 736                         sisn = th->seq;
 737                         disn = 0;
 738                 }
 739                 if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
 740                         addr = (union tcp_md5_addr *)&ipv6_hdr(skb)->saddr;
 741                 else
 742                         addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 743 #if IS_ENABLED(CONFIG_IPV6)
 744                 if (family == AF_INET6 && ipv6_addr_v4mapped(&sk->sk_v6_daddr))
 745                         family = AF_INET;
 746 #endif
 747
 748                 sk = sk_const_to_full_sk(sk);
 749                 ao_info = rcu_dereference(tcp_sk(sk)->ao_info);
 750                 if (!ao_info)
 751                         return -ENOENT;
 752                 *key = tcp_ao_do_lookup(sk, l3index, addr, family,
 753                                         -1, aoh->rnext_keyid);
 754                 if (!*key)
 755                         return -ENOENT;
 756                 *traffic_key = kmalloc(tcp_ao_digest_size(*key), GFP_ATOMIC);
 757                 if (!*traffic_key)
 758                         return -ENOMEM;
 759                 *allocated_traffic_key = true;
 760                 if (tcp_ao_calc_key_skb(*key, *traffic_key, skb,
 761                                         sisn, disn, family))
 762                         return -1;
 763                 *keyid = (*key)->rcvid;
 764         } else {
 765                 struct tcp_ao_key *rnext_key;
 766                 u32 snd_basis;
 767
 768                 if (sk->sk_state == TCP_TIME_WAIT) {
 769                         ao_info = rcu_dereference(tcp_twsk(sk)->ao_info);
 770                         snd_basis = tcp_twsk(sk)->tw_snd_nxt;
 771                 } else {
 772                         ao_info = rcu_dereference(tcp_sk(sk)->ao_info);
 773                         snd_basis = tcp_sk(sk)->snd_una;
 774                 }
 775                 if (!ao_info)
 776                         return -ENOENT;
 777
 778                 *key = tcp_ao_established_key(ao_info, aoh->rnext_keyid, -1);
 779                 if (!*key)
 780                         return -ENOENT;
 781                 *traffic_key = snd_other_key(*key);
 782                 rnext_key = READ_ONCE(ao_info->rnext_key);
 783                 *keyid = rnext_key->rcvid;
 784                 *sne = tcp_ao_compute_sne(READ_ONCE(ao_info->snd_sne),
 785                                           snd_basis, seq);
 786         }
 787         return 0;
 788 }
 789
 790 int tcp_ao_transmit_skb(struct sock *sk, struct sk_buff *skb,
 791                         struct tcp_ao_key *key, struct tcphdr *th,
 792                         __u8 *hash_location)
 793 {
 794         struct tcp_skb_cb *tcb = TCP_SKB_CB(skb);
 795         struct tcp_sock *tp = tcp_sk(sk);
 796         struct tcp_ao_info *ao;
 797         void *tkey_buf = NULL;
 798         u8 *traffic_key;
 799         u32 sne;
 800
 801         ao = rcu_dereference_protected(tcp_sk(sk)->ao_info,
 802                                        lockdep_sock_is_held(sk));
 803         traffic_key = snd_other_key(key);
 804         if (unlikely(tcb->tcp_flags & TCPHDR_SYN)) {
 805                 __be32 disn;
 806
 807                 if (!(tcb->tcp_flags & TCPHDR_ACK)) {
 808                         disn = 0;
 809                         tkey_buf = kmalloc(tcp_ao_digest_size(key), GFP_ATOMIC);
 810                         if (!tkey_buf)
 811                                 return -ENOMEM;
 812                         traffic_key = tkey_buf;
 813                 } else {
 814                         disn = ao->risn;
 815                 }
 816                 tp->af_specific->ao_calc_key_sk(key, traffic_key,
 817                                                 sk, ao->lisn, disn, true);
 818         }
 819         sne = tcp_ao_compute_sne(READ_ONCE(ao->snd_sne), READ_ONCE(tp->snd_una),
 820                                  ntohl(th->seq));
 821         tp->af_specific->calc_ao_hash(hash_location, key, sk, skb, traffic_key,
 822                                       hash_location - (u8 *)th, sne);
 823         kfree(tkey_buf);
 824         return 0;
 825 }
 826
 827 static struct tcp_ao_key *tcp_ao_inbound_lookup(unsigned short int family,
 828                 const struct sock *sk, const struct sk_buff *skb,
 829                 int sndid, int rcvid, int l3index)
 830 {
 831         if (family == AF_INET) {
 832                 const struct iphdr *iph = ip_hdr(skb);
 833
 834                 return tcp_ao_do_lookup(sk, l3index,
 835                                         (union tcp_ao_addr *)&iph->saddr,
 836                                         AF_INET, sndid, rcvid);
 837         } else {
 838                 const struct ipv6hdr *iph = ipv6_hdr(skb);
 839
 840                 return tcp_ao_do_lookup(sk, l3index,
 841                                         (union tcp_ao_addr *)&iph->saddr,
 842                                         AF_INET6, sndid, rcvid);
 843         }
 844 }
 845
 846 void tcp_ao_syncookie(struct sock *sk, const struct sk_buff *skb,
 847                       struct request_sock *req, unsigned short int family)
 848 {
 849         struct tcp_request_sock *treq = tcp_rsk(req);
 850         const struct tcphdr *th = tcp_hdr(skb);
 851         const struct tcp_ao_hdr *aoh;
 852         struct tcp_ao_key *key;
 853         int l3index;
 854
 855         /* treq->af_specific is used to perform TCP_AO lookup
 856          * in tcp_create_openreq_child().
 857          */
 858 #if IS_ENABLED(CONFIG_IPV6)
 859         if (family == AF_INET6)
 860                 treq->af_specific = &tcp_request_sock_ipv6_ops;
 861         else
 862 #endif
 863                 treq->af_specific = &tcp_request_sock_ipv4_ops;
 864
 865         treq->used_tcp_ao = false;
 866
 867         if (tcp_parse_auth_options(th, NULL, &aoh) || !aoh)
 868                 return;
 869
 870         l3index = l3mdev_master_ifindex_by_index(sock_net(sk), inet_rsk(req)->ir_iif);
 871         key = tcp_ao_inbound_lookup(family, sk, skb, -1, aoh->keyid, l3index);
 872         if (!key)
 873                 /* Key not found, continue without TCP-AO */
 874                 return;
 875
 876         treq->ao_rcv_next = aoh->keyid;
 877         treq->ao_keyid = aoh->rnext_keyid;
 878         treq->used_tcp_ao = true;
 879 }
 880
 881 static enum skb_drop_reason
 882 tcp_ao_verify_hash(const struct sock *sk, const struct sk_buff *skb,
 883                    unsigned short int family, struct tcp_ao_info *info,
 884                    const struct tcp_ao_hdr *aoh, struct tcp_ao_key *key,
 885                    u8 *traffic_key, u8 *phash, u32 sne, int l3index)
 886 {
 887         u8 maclen = aoh->length - sizeof(struct tcp_ao_hdr);
 888         const struct tcphdr *th = tcp_hdr(skb);
 889         void *hash_buf = NULL;
 890
 891         if (maclen != tcp_ao_maclen(key)) {
 892                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD);
 893                 atomic64_inc(&info->counters.pkt_bad);
 894                 atomic64_inc(&key->pkt_bad);
 895                 tcp_hash_fail("AO hash wrong length", family, skb,
 896                               "%u != %d L3index: %d", maclen,
 897                               tcp_ao_maclen(key), l3index);
 898                 return SKB_DROP_REASON_TCP_AOFAILURE;
 899         }
 900
 901         hash_buf = kmalloc(tcp_ao_digest_size(key), GFP_ATOMIC);
 902         if (!hash_buf)
 903                 return SKB_DROP_REASON_NOT_SPECIFIED;
 904
 905         /* XXX: make it per-AF callback? */
 906         tcp_ao_hash_skb(family, hash_buf, key, sk, skb, traffic_key,
 907                         (phash - (u8 *)th), sne);
 908         if (memcmp(phash, hash_buf, maclen)) {
 909                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOBAD);
 910                 atomic64_inc(&info->counters.pkt_bad);
 911                 atomic64_inc(&key->pkt_bad);
 912                 tcp_hash_fail("AO hash mismatch", family, skb,
 913                               "L3index: %d", l3index);
 914                 kfree(hash_buf);
 915                 return SKB_DROP_REASON_TCP_AOFAILURE;
 916         }
 917         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOGOOD);
 918         atomic64_inc(&info->counters.pkt_good);
 919         atomic64_inc(&key->pkt_good);
 920         kfree(hash_buf);
 921         return SKB_NOT_DROPPED_YET;
 922 }
 923
 924 enum skb_drop_reason
 925 tcp_inbound_ao_hash(struct sock *sk, const struct sk_buff *skb,
 926                     unsigned short int family, const struct request_sock *req,
 927                     int l3index, const struct tcp_ao_hdr *aoh)
 928 {
 929         const struct tcphdr *th = tcp_hdr(skb);
 930         u8 *phash = (u8 *)(aoh + 1); /* hash goes just after the header */
 931         struct tcp_ao_info *info;
 932         enum skb_drop_reason ret;
 933         struct tcp_ao_key *key;
 934         __be32 sisn, disn;
 935         u8 *traffic_key;
 936         u32 sne = 0;
 937
 938         info = rcu_dereference(tcp_sk(sk)->ao_info);
 939         if (!info) {
 940                 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOKEYNOTFOUND);
 941                 tcp_hash_fail("AO key not found", family, skb,
 942                               "keyid: %u L3index: %d", aoh->keyid, l3index);
 943                 return SKB_DROP_REASON_TCP_AOUNEXPECTED;
 944         }
 945
 946         if (unlikely(th->syn)) {
 947                 sisn = th->seq;
 948                 disn = 0;
 949         }
 950
 951         /* Fast-path */
 952         if (likely((1 << sk->sk_state) & TCP_AO_ESTABLISHED)) {
 953                 enum skb_drop_reason err;
 954                 struct tcp_ao_key *current_key;
 955
 956                 /* Check if this socket's rnext_key matches the keyid in the
 957                  * packet. If not we lookup the key based on the keyid
 958                  * matching the rcvid in the mkt.
 959                  */
 960                 key = READ_ONCE(info->rnext_key);
 961                 if (key->rcvid != aoh->keyid) {
 962                         key = tcp_ao_established_key(info, -1, aoh->keyid);
 963                         if (!key)
 964                                 goto key_not_found;
 965                 }
 966
 967                 /* Delayed retransmitted SYN */
 968                 if (unlikely(th->syn && !th->ack))
 969                         goto verify_hash;
 970
 971                 sne = tcp_ao_compute_sne(info->rcv_sne, tcp_sk(sk)->rcv_nxt,
 972                                          ntohl(th->seq));
 973                 /* Established socket, traffic key are cached */
 974                 traffic_key = rcv_other_key(key);
 975                 err = tcp_ao_verify_hash(sk, skb, family, info, aoh, key,
 976                                          traffic_key, phash, sne, l3index);
 977                 if (err)
 978                         return err;
 979                 current_key = READ_ONCE(info->current_key);
 980                 /* Key rotation: the peer asks us to use new key (RNext) */
 981                 if (unlikely(aoh->rnext_keyid != current_key->sndid)) {
 982                         /* If the key is not found we do nothing. */
 983                         key = tcp_ao_established_key(info, aoh->rnext_keyid, -1);
 984                         if (key)
 985                                 /* pairs with tcp_ao_del_cmd */
 986                                 WRITE_ONCE(info->current_key, key);
 987                 }
 988                 return SKB_NOT_DROPPED_YET;
 989         }
 990
 991         /* Lookup key based on peer address and keyid.
 992          * current_key and rnext_key must not be used on tcp listen
 993          * sockets as otherwise:
 994          * - request sockets would race on those key pointers
 995          * - tcp_ao_del_cmd() allows async key removal
 996          */
 997         key = tcp_ao_inbound_lookup(family, sk, skb, -1, aoh->keyid, l3index);
 998         if (!key)
 999                 goto key_not_found;
1000
1001         if (th->syn && !th->ack)
1002                 goto verify_hash;
1003
1004         if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV)) {
1005                 /* Make the initial syn the likely case here */
1006                 if (unlikely(req)) {
1007                         sne = tcp_ao_compute_sne(0, tcp_rsk(req)->rcv_isn,
1008                                                  ntohl(th->seq));
1009                         sisn = htonl(tcp_rsk(req)->rcv_isn);
1010                         disn = htonl(tcp_rsk(req)->snt_isn);
1011                 } else if (unlikely(th->ack && !th->syn)) {
1012                         /* Possible syncookie packet */
1013                         sisn = htonl(ntohl(th->seq) - 1);
1014                         disn = htonl(ntohl(th->ack_seq) - 1);
1015                         sne = tcp_ao_compute_sne(0, ntohl(sisn),
1016                                                  ntohl(th->seq));
1017                 } else if (unlikely(!th->syn)) {
1018                         /* no way to figure out initial sisn/disn - drop */
1019                         return SKB_DROP_REASON_TCP_FLAGS;
1020                 }
1021         } else if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1022                 disn = info->lisn;
1023                 if (th->syn || th->rst)
1024                         sisn = th->seq;
1025                 else
1026                         sisn = info->risn;
1027         } else {
1028                 WARN_ONCE(1, "TCP-AO: Unexpected sk_state %d", sk->sk_state);
1029                 return SKB_DROP_REASON_TCP_AOFAILURE;
1030         }
1031 verify_hash:
1032         traffic_key = kmalloc(tcp_ao_digest_size(key), GFP_ATOMIC);
1033         if (!traffic_key)
1034                 return SKB_DROP_REASON_NOT_SPECIFIED;
1035         tcp_ao_calc_key_skb(key, traffic_key, skb, sisn, disn, family);
1036         ret = tcp_ao_verify_hash(sk, skb, family, info, aoh, key,
1037                                  traffic_key, phash, sne, l3index);
1038         kfree(traffic_key);
1039         return ret;
1040
1041 key_not_found:
1042         NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOKEYNOTFOUND);
1043         atomic64_inc(&info->counters.key_not_found);
1044         tcp_hash_fail("Requested by the peer AO key id not found",
1045                       family, skb, "L3index: %d", l3index);
1046         return SKB_DROP_REASON_TCP_AOKEYNOTFOUND;
1047 }
1048
1049 static int tcp_ao_cache_traffic_keys(const struct sock *sk,
1050                                      struct tcp_ao_info *ao,
1051                                      struct tcp_ao_key *ao_key)
1052 {
1053         u8 *traffic_key = snd_other_key(ao_key);
1054         int ret;
1055
1056         ret = tcp_ao_calc_key_sk(ao_key, traffic_key, sk,
1057                                  ao->lisn, ao->risn, true);
1058         if (ret)
1059                 return ret;
1060
1061         traffic_key = rcv_other_key(ao_key);
1062         ret = tcp_ao_calc_key_sk(ao_key, traffic_key, sk,
1063                                  ao->lisn, ao->risn, false);
1064         return ret;
1065 }
1066
1067 void tcp_ao_connect_init(struct sock *sk)
1068 {
1069         struct tcp_sock *tp = tcp_sk(sk);
1070         struct tcp_ao_info *ao_info;
1071         union tcp_ao_addr *addr;
1072         struct tcp_ao_key *key;
1073         int family, l3index;
1074
1075         ao_info = rcu_dereference_protected(tp->ao_info,
1076                                             lockdep_sock_is_held(sk));
1077         if (!ao_info)
1078                 return;
1079
1080         /* Remove all keys that don't match the peer */
1081         family = sk->sk_family;
1082         if (family == AF_INET)
1083                 addr = (union tcp_ao_addr *)&sk->sk_daddr;
1084 #if IS_ENABLED(CONFIG_IPV6)
1085         else if (family == AF_INET6)
1086                 addr = (union tcp_ao_addr *)&sk->sk_v6_daddr;
1087 #endif
1088         else
1089                 return;
1090         l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1091                                                  sk->sk_bound_dev_if);
1092
1093         hlist_for_each_entry_rcu(key, &ao_info->head, node) {
1094                 if (!tcp_ao_key_cmp(key, l3index, addr, key->prefixlen, family, -1, -1))
1095                         continue;
1096
1097                 if (key == ao_info->current_key)
1098                         ao_info->current_key = NULL;
1099                 if (key == ao_info->rnext_key)
1100                         ao_info->rnext_key = NULL;
1101                 hlist_del_rcu(&key->node);
1102                 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1103                 call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1104         }
1105
1106         key = tp->af_specific->ao_lookup(sk, sk, -1, -1);
1107         if (key) {
1108                 /* if current_key or rnext_key were not provided,
1109                  * use the first key matching the peer
1110                  */
1111                 if (!ao_info->current_key)
1112                         ao_info->current_key = key;
1113                 if (!ao_info->rnext_key)
1114                         ao_info->rnext_key = key;
1115                 tp->tcp_header_len += tcp_ao_len_aligned(key);
1116
1117                 ao_info->lisn = htonl(tp->write_seq);
1118                 ao_info->snd_sne = 0;
1119         } else {
1120                 /* Can't happen: tcp_connect() verifies that there's
1121                  * at least one tcp-ao key that matches the remote peer.
1122                  */
1123                 WARN_ON_ONCE(1);
1124                 rcu_assign_pointer(tp->ao_info, NULL);
1125                 kfree(ao_info);
1126         }
1127 }
1128
1129 void tcp_ao_established(struct sock *sk)
1130 {
1131         struct tcp_ao_info *ao;
1132         struct tcp_ao_key *key;
1133
1134         ao = rcu_dereference_protected(tcp_sk(sk)->ao_info,
1135                                        lockdep_sock_is_held(sk));
1136         if (!ao)
1137                 return;
1138
1139         hlist_for_each_entry_rcu(key, &ao->head, node)
1140                 tcp_ao_cache_traffic_keys(sk, ao, key);
1141 }
1142
1143 void tcp_ao_finish_connect(struct sock *sk, struct sk_buff *skb)
1144 {
1145         struct tcp_ao_info *ao;
1146         struct tcp_ao_key *key;
1147
1148         ao = rcu_dereference_protected(tcp_sk(sk)->ao_info,
1149                                        lockdep_sock_is_held(sk));
1150         if (!ao)
1151                 return;
1152
1153         WRITE_ONCE(ao->risn, tcp_hdr(skb)->seq);
1154         ao->rcv_sne = 0;
1155
1156         hlist_for_each_entry_rcu(key, &ao->head, node)
1157                 tcp_ao_cache_traffic_keys(sk, ao, key);
1158 }
1159
1160 int tcp_ao_copy_all_matching(const struct sock *sk, struct sock *newsk,
1161                              struct request_sock *req, struct sk_buff *skb,
1162                              int family)
1163 {
1164         struct tcp_ao_key *key, *new_key, *first_key;
1165         struct tcp_ao_info *new_ao, *ao;
1166         struct hlist_node *key_head;
1167         int l3index, ret = -ENOMEM;
1168         union tcp_ao_addr *addr;
1169         bool match = false;
1170
1171         ao = rcu_dereference(tcp_sk(sk)->ao_info);
1172         if (!ao)
1173                 return 0;
1174
1175         /* New socket without TCP-AO on it */
1176         if (!tcp_rsk_used_ao(req))
1177                 return 0;
1178
1179         new_ao = tcp_ao_alloc_info(GFP_ATOMIC);
1180         if (!new_ao)
1181                 return -ENOMEM;
1182         new_ao->lisn = htonl(tcp_rsk(req)->snt_isn);
1183         new_ao->risn = htonl(tcp_rsk(req)->rcv_isn);
1184         new_ao->ao_required = ao->ao_required;
1185         new_ao->accept_icmps = ao->accept_icmps;
1186
1187         if (family == AF_INET) {
1188                 addr = (union tcp_ao_addr *)&newsk->sk_daddr;
1189 #if IS_ENABLED(CONFIG_IPV6)
1190         } else if (family == AF_INET6) {
1191                 addr = (union tcp_ao_addr *)&newsk->sk_v6_daddr;
1192 #endif
1193         } else {
1194                 ret = -EAFNOSUPPORT;
1195                 goto free_ao;
1196         }
1197         l3index = l3mdev_master_ifindex_by_index(sock_net(newsk),
1198                                                  newsk->sk_bound_dev_if);
1199
1200         hlist_for_each_entry_rcu(key, &ao->head, node) {
1201                 if (tcp_ao_key_cmp(key, l3index, addr, key->prefixlen, family, -1, -1))
1202                         continue;
1203
1204                 new_key = tcp_ao_copy_key(newsk, key);
1205                 if (!new_key)
1206                         goto free_and_exit;
1207
1208                 tcp_ao_cache_traffic_keys(newsk, new_ao, new_key);
1209                 tcp_ao_link_mkt(new_ao, new_key);
1210                 match = true;
1211         }
1212
1213         if (!match) {
1214                 /* RFC5925 (7.4.1) specifies that the TCP-AO status
1215                  * of a connection is determined on the initial SYN.
1216                  * At this point the connection was TCP-AO enabled, so
1217                  * it can't switch to being unsigned if peer's key
1218                  * disappears on the listening socket.
1219                  */
1220                 ret = -EKEYREJECTED;
1221                 goto free_and_exit;
1222         }
1223
1224         if (!static_key_fast_inc_not_disabled(&tcp_ao_needed.key.key)) {
1225                 ret = -EUSERS;
1226                 goto free_and_exit;
1227         }
1228
1229         key_head = rcu_dereference(hlist_first_rcu(&new_ao->head));
1230         first_key = hlist_entry_safe(key_head, struct tcp_ao_key, node);
1231
1232         key = tcp_ao_established_key(new_ao, tcp_rsk(req)->ao_keyid, -1);
1233         if (key)
1234                 new_ao->current_key = key;
1235         else
1236                 new_ao->current_key = first_key;
1237
1238         /* set rnext_key */
1239         key = tcp_ao_established_key(new_ao, -1, tcp_rsk(req)->ao_rcv_next);
1240         if (key)
1241                 new_ao->rnext_key = key;
1242         else
1243                 new_ao->rnext_key = first_key;
1244
1245         sk_gso_disable(newsk);
1246         rcu_assign_pointer(tcp_sk(newsk)->ao_info, new_ao);
1247
1248         return 0;
1249
1250 free_and_exit:
1251         hlist_for_each_entry_safe(key, key_head, &new_ao->head, node) {
1252                 hlist_del(&key->node);
1253                 tcp_sigpool_release(key->tcp_sigpool_id);
1254                 atomic_sub(tcp_ao_sizeof_key(key), &newsk->sk_omem_alloc);
1255                 kfree_sensitive(key);
1256         }
1257 free_ao:
1258         kfree(new_ao);
1259         return ret;
1260 }
1261
1262 static bool tcp_ao_can_set_current_rnext(struct sock *sk)
1263 {
1264         /* There aren't current/rnext keys on TCP_LISTEN sockets */
1265         if (sk->sk_state == TCP_LISTEN)
1266                 return false;
1267         return true;
1268 }
1269
1270 static int tcp_ao_verify_ipv4(struct sock *sk, struct tcp_ao_add *cmd,
1271                               union tcp_ao_addr **addr)
1272 {
1273         struct sockaddr_in *sin = (struct sockaddr_in *)&cmd->addr;
1274         struct inet_sock *inet = inet_sk(sk);
1275
1276         if (sin->sin_family != AF_INET)
1277                 return -EINVAL;
1278
1279         /* Currently matching is not performed on port (or port ranges) */
1280         if (sin->sin_port != 0)
1281                 return -EINVAL;
1282
1283         /* Check prefix and trailing 0's in addr */
1284         if (cmd->prefix != 0) {
1285                 __be32 mask;
1286
1287                 if (ntohl(sin->sin_addr.s_addr) == INADDR_ANY)
1288                         return -EINVAL;
1289                 if (cmd->prefix > 32)
1290                         return -EINVAL;
1291
1292                 mask = inet_make_mask(cmd->prefix);
1293                 if (sin->sin_addr.s_addr & ~mask)
1294                         return -EINVAL;
1295
1296                 /* Check that MKT address is consistent with socket */
1297                 if (ntohl(inet->inet_daddr) != INADDR_ANY &&
1298                     (inet->inet_daddr & mask) != sin->sin_addr.s_addr)
1299                         return -EINVAL;
1300         } else {
1301                 if (ntohl(sin->sin_addr.s_addr) != INADDR_ANY)
1302                         return -EINVAL;
1303         }
1304
1305         *addr = (union tcp_ao_addr *)&sin->sin_addr;
1306         return 0;
1307 }
1308
1309 static int tcp_ao_parse_crypto(struct tcp_ao_add *cmd, struct tcp_ao_key *key)
1310 {
1311         unsigned int syn_tcp_option_space;
1312         bool is_kdf_aes_128_cmac = false;
1313         struct crypto_ahash *tfm;
1314         struct tcp_sigpool hp;
1315         void *tmp_key = NULL;
1316         int err;
1317
1318         /* RFC5926, 3.1.1.2. KDF_AES_128_CMAC */
1319         if (!strcmp("cmac(aes128)", cmd->alg_name)) {
1320                 strscpy(cmd->alg_name, "cmac(aes)", sizeof(cmd->alg_name));
1321                 is_kdf_aes_128_cmac = (cmd->keylen != 16);
1322                 tmp_key = kmalloc(cmd->keylen, GFP_KERNEL);
1323                 if (!tmp_key)
1324                         return -ENOMEM;
1325         }
1326
1327         key->maclen = cmd->maclen ?: 12; /* 12 is the default in RFC5925 */
1328
1329         /* Check: maclen + tcp-ao header <= (MAX_TCP_OPTION_SPACE - mss
1330          *                                      - tstamp (including sackperm)
1331          *                                      - wscale),
1332          * see tcp_syn_options(), tcp_synack_options(), commit 33ad798c924b.
1333          *
1334          * In order to allow D-SACK with TCP-AO, the header size should be:
1335          * (MAX_TCP_OPTION_SPACE - TCPOLEN_TSTAMP_ALIGNED
1336          *                      - TCPOLEN_SACK_BASE_ALIGNED
1337          *                      - 2 * TCPOLEN_SACK_PERBLOCK) = 8 (maclen = 4),
1338          * see tcp_established_options().
1339          *
1340          * RFC5925, 2.2:
1341          * Typical MACs are 96-128 bits (12-16 bytes), but any length
1342          * that fits in the header of the segment being authenticated
1343          * is allowed.
1344          *
1345          * RFC5925, 7.6:
1346          * TCP-AO continues to consume 16 bytes in non-SYN segments,
1347          * leaving a total of 24 bytes for other options, of which
1348          * the timestamp consumes 10.  This leaves 14 bytes, of which 10
1349          * are used for a single SACK block. When two SACK blocks are used,
1350          * such as to handle D-SACK, a smaller TCP-AO MAC would be required
1351          * to make room for the additional SACK block (i.e., to leave 18
1352          * bytes for the D-SACK variant of the SACK option) [RFC2883].
1353          * Note that D-SACK is not supportable in TCP MD5 in the presence
1354          * of timestamps, because TCP MD5’s MAC length is fixed and too
1355          * large to leave sufficient option space.
1356          */
1357         syn_tcp_option_space = MAX_TCP_OPTION_SPACE;
1358         syn_tcp_option_space -= TCPOLEN_MSS_ALIGNED;
1359         syn_tcp_option_space -= TCPOLEN_TSTAMP_ALIGNED;
1360         syn_tcp_option_space -= TCPOLEN_WSCALE_ALIGNED;
1361         if (tcp_ao_len_aligned(key) > syn_tcp_option_space) {
1362                 err = -EMSGSIZE;
1363                 goto err_kfree;
1364         }
1365
1366         key->keylen = cmd->keylen;
1367         memcpy(key->key, cmd->key, cmd->keylen);
1368
1369         err = tcp_sigpool_start(key->tcp_sigpool_id, &hp);
1370         if (err)
1371                 goto err_kfree;
1372
1373         tfm = crypto_ahash_reqtfm(hp.req);
1374         if (is_kdf_aes_128_cmac) {
1375                 void *scratch = hp.scratch;
1376                 struct scatterlist sg;
1377
1378                 memcpy(tmp_key, cmd->key, cmd->keylen);
1379                 sg_init_one(&sg, tmp_key, cmd->keylen);
1380
1381                 /* Using zero-key of 16 bytes as described in RFC5926 */
1382                 memset(scratch, 0, 16);
1383                 err = crypto_ahash_setkey(tfm, scratch, 16);
1384                 if (err)
1385                         goto err_pool_end;
1386
1387                 err = crypto_ahash_init(hp.req);
1388                 if (err)
1389                         goto err_pool_end;
1390
1391                 ahash_request_set_crypt(hp.req, &sg, key->key, cmd->keylen);
1392                 err = crypto_ahash_update(hp.req);
1393                 if (err)
1394                         goto err_pool_end;
1395
1396                 err |= crypto_ahash_final(hp.req);
1397                 if (err)
1398                         goto err_pool_end;
1399                 key->keylen = 16;
1400         }
1401
1402         err = crypto_ahash_setkey(tfm, key->key, key->keylen);
1403         if (err)
1404                 goto err_pool_end;
1405
1406         tcp_sigpool_end(&hp);
1407         kfree_sensitive(tmp_key);
1408
1409         if (tcp_ao_maclen(key) > key->digest_size)
1410                 return -EINVAL;
1411
1412         return 0;
1413
1414 err_pool_end:
1415         tcp_sigpool_end(&hp);
1416 err_kfree:
1417         kfree_sensitive(tmp_key);
1418         return err;
1419 }
1420
1421 #if IS_ENABLED(CONFIG_IPV6)
1422 static int tcp_ao_verify_ipv6(struct sock *sk, struct tcp_ao_add *cmd,
1423                               union tcp_ao_addr **paddr,
1424                               unsigned short int *family)
1425 {
1426         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd->addr;
1427         struct in6_addr *addr = &sin6->sin6_addr;
1428         u8 prefix = cmd->prefix;
1429
1430         if (sin6->sin6_family != AF_INET6)
1431                 return -EINVAL;
1432
1433         /* Currently matching is not performed on port (or port ranges) */
1434         if (sin6->sin6_port != 0)
1435                 return -EINVAL;
1436
1437         /* Check prefix and trailing 0's in addr */
1438         if (cmd->prefix != 0 && ipv6_addr_v4mapped(addr)) {
1439                 __be32 addr4 = addr->s6_addr32[3];
1440                 __be32 mask;
1441
1442                 if (prefix > 32 || ntohl(addr4) == INADDR_ANY)
1443                         return -EINVAL;
1444
1445                 mask = inet_make_mask(prefix);
1446                 if (addr4 & ~mask)
1447                         return -EINVAL;
1448
1449                 /* Check that MKT address is consistent with socket */
1450                 if (!ipv6_addr_any(&sk->sk_v6_daddr)) {
1451                         __be32 daddr4 = sk->sk_v6_daddr.s6_addr32[3];
1452
1453                         if (!ipv6_addr_v4mapped(&sk->sk_v6_daddr))
1454                                 return -EINVAL;
1455                         if ((daddr4 & mask) != addr4)
1456                                 return -EINVAL;
1457                 }
1458
1459                 *paddr = (union tcp_ao_addr *)&addr->s6_addr32[3];
1460                 *family = AF_INET;
1461                 return 0;
1462         } else if (cmd->prefix != 0) {
1463                 struct in6_addr pfx;
1464
1465                 if (ipv6_addr_any(addr) || prefix > 128)
1466                         return -EINVAL;
1467
1468                 ipv6_addr_prefix(&pfx, addr, prefix);
1469                 if (ipv6_addr_cmp(&pfx, addr))
1470                         return -EINVAL;
1471
1472                 /* Check that MKT address is consistent with socket */
1473                 if (!ipv6_addr_any(&sk->sk_v6_daddr) &&
1474                     !ipv6_prefix_equal(&sk->sk_v6_daddr, addr, prefix))
1475
1476                         return -EINVAL;
1477         } else {
1478                 if (!ipv6_addr_any(addr))
1479                         return -EINVAL;
1480         }
1481
1482         *paddr = (union tcp_ao_addr *)addr;
1483         return 0;
1484 }
1485 #else
1486 static int tcp_ao_verify_ipv6(struct sock *sk, struct tcp_ao_add *cmd,
1487                               union tcp_ao_addr **paddr,
1488                               unsigned short int *family)
1489 {
1490         return -EOPNOTSUPP;
1491 }
1492 #endif
1493
1494 static struct tcp_ao_info *setsockopt_ao_info(struct sock *sk)
1495 {
1496         if (sk_fullsock(sk)) {
1497                 return rcu_dereference_protected(tcp_sk(sk)->ao_info,
1498                                                  lockdep_sock_is_held(sk));
1499         } else if (sk->sk_state == TCP_TIME_WAIT) {
1500                 return rcu_dereference_protected(tcp_twsk(sk)->ao_info,
1501                                                  lockdep_sock_is_held(sk));
1502         }
1503         return ERR_PTR(-ESOCKTNOSUPPORT);
1504 }
1505
1506 static struct tcp_ao_info *getsockopt_ao_info(struct sock *sk)
1507 {
1508         if (sk_fullsock(sk))
1509                 return rcu_dereference(tcp_sk(sk)->ao_info);
1510         else if (sk->sk_state == TCP_TIME_WAIT)
1511                 return rcu_dereference(tcp_twsk(sk)->ao_info);
1512
1513         return ERR_PTR(-ESOCKTNOSUPPORT);
1514 }
1515
1516 #define TCP_AO_KEYF_ALL (TCP_AO_KEYF_IFINDEX | TCP_AO_KEYF_EXCLUDE_OPT)
1517 #define TCP_AO_GET_KEYF_VALID   (TCP_AO_KEYF_IFINDEX)
1518
1519 static struct tcp_ao_key *tcp_ao_key_alloc(struct sock *sk,
1520                                            struct tcp_ao_add *cmd)
1521 {
1522         const char *algo = cmd->alg_name;
1523         unsigned int digest_size;
1524         struct crypto_ahash *tfm;
1525         struct tcp_ao_key *key;
1526         struct tcp_sigpool hp;
1527         int err, pool_id;
1528         size_t size;
1529
1530         /* Force null-termination of alg_name */
1531         cmd->alg_name[ARRAY_SIZE(cmd->alg_name) - 1] = '\0';
1532
1533         /* RFC5926, 3.1.1.2. KDF_AES_128_CMAC */
1534         if (!strcmp("cmac(aes128)", algo))
1535                 algo = "cmac(aes)";
1536
1537         /* Full TCP header (th->doff << 2) should fit into scratch area,
1538          * see tcp_ao_hash_header().
1539          */
1540         pool_id = tcp_sigpool_alloc_ahash(algo, 60);
1541         if (pool_id < 0)
1542                 return ERR_PTR(pool_id);
1543
1544         err = tcp_sigpool_start(pool_id, &hp);
1545         if (err)
1546                 goto err_free_pool;
1547
1548         tfm = crypto_ahash_reqtfm(hp.req);
1549         digest_size = crypto_ahash_digestsize(tfm);
1550         tcp_sigpool_end(&hp);
1551
1552         size = sizeof(struct tcp_ao_key) + (digest_size << 1);
1553         key = sock_kmalloc(sk, size, GFP_KERNEL);
1554         if (!key) {
1555                 err = -ENOMEM;
1556                 goto err_free_pool;
1557         }
1558
1559         key->tcp_sigpool_id = pool_id;
1560         key->digest_size = digest_size;
1561         return key;
1562
1563 err_free_pool:
1564         tcp_sigpool_release(pool_id);
1565         return ERR_PTR(err);
1566 }
1567
1568 static int tcp_ao_add_cmd(struct sock *sk, unsigned short int family,
1569                           sockptr_t optval, int optlen)
1570 {
1571         struct tcp_ao_info *ao_info;
1572         union tcp_ao_addr *addr;
1573         struct tcp_ao_key *key;
1574         struct tcp_ao_add cmd;
1575         int ret, l3index = 0;
1576         bool first = false;
1577
1578         if (optlen < sizeof(cmd))
1579                 return -EINVAL;
1580
1581         ret = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1582         if (ret)
1583                 return ret;
1584
1585         if (cmd.keylen > TCP_AO_MAXKEYLEN)
1586                 return -EINVAL;
1587
1588         if (cmd.reserved != 0 || cmd.reserved2 != 0)
1589                 return -EINVAL;
1590
1591         if (family == AF_INET)
1592                 ret = tcp_ao_verify_ipv4(sk, &cmd, &addr);
1593         else
1594                 ret = tcp_ao_verify_ipv6(sk, &cmd, &addr, &family);
1595         if (ret)
1596                 return ret;
1597
1598         if (cmd.keyflags & ~TCP_AO_KEYF_ALL)
1599                 return -EINVAL;
1600
1601         if (cmd.set_current || cmd.set_rnext) {
1602                 if (!tcp_ao_can_set_current_rnext(sk))
1603                         return -EINVAL;
1604         }
1605
1606         if (cmd.ifindex && !(cmd.keyflags & TCP_AO_KEYF_IFINDEX))
1607                 return -EINVAL;
1608
1609         /* For cmd.tcp_ifindex = 0 the key will apply to the default VRF */
1610         if (cmd.keyflags & TCP_AO_KEYF_IFINDEX && cmd.ifindex) {
1611                 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
1612                 struct net_device *dev;
1613
1614                 rcu_read_lock();
1615                 dev = dev_get_by_index_rcu(sock_net(sk), cmd.ifindex);
1616                 if (dev && netif_is_l3_master(dev))
1617                         l3index = dev->ifindex;
1618                 rcu_read_unlock();
1619
1620                 if (!dev || !l3index)
1621                         return -EINVAL;
1622
1623                 if (!bound_dev_if || bound_dev_if != cmd.ifindex) {
1624                         /* tcp_ao_established_key() doesn't expect having
1625                          * non peer-matching key on an established TCP-AO
1626                          * connection.
1627                          */
1628                         if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)))
1629                                 return -EINVAL;
1630                 }
1631
1632                 /* It's still possible to bind after adding keys or even
1633                  * re-bind to a different dev (with CAP_NET_RAW).
1634                  * So, no reason to return error here, rather try to be
1635                  * nice and warn the user.
1636                  */
1637                 if (bound_dev_if && bound_dev_if != cmd.ifindex)
1638                         net_warn_ratelimited("AO key ifindex %d != sk bound ifindex %d\n",
1639                                              cmd.ifindex, bound_dev_if);
1640         }
1641
1642         /* Don't allow keys for peers that have a matching TCP-MD5 key */
1643         if (cmd.keyflags & TCP_AO_KEYF_IFINDEX) {
1644                 /* Non-_exact version of tcp_md5_do_lookup() will
1645                  * as well match keys that aren't bound to a specific VRF
1646                  * (that will make them match AO key with
1647                  * sysctl_tcp_l3dev_accept = 1
1648                  */
1649                 if (tcp_md5_do_lookup(sk, l3index, addr, family))
1650                         return -EKEYREJECTED;
1651         } else {
1652                 if (tcp_md5_do_lookup_any_l3index(sk, addr, family))
1653                         return -EKEYREJECTED;
1654         }
1655
1656         ao_info = setsockopt_ao_info(sk);
1657         if (IS_ERR(ao_info))
1658                 return PTR_ERR(ao_info);
1659
1660         if (!ao_info) {
1661                 ao_info = tcp_ao_alloc_info(GFP_KERNEL);
1662                 if (!ao_info)
1663                         return -ENOMEM;
1664                 first = true;
1665         } else {
1666                 /* Check that neither RecvID nor SendID match any
1667                  * existing key for the peer, RFC5925 3.1:
1668                  * > The IDs of MKTs MUST NOT overlap where their
1669                  * > TCP connection identifiers overlap.
1670                  */
1671                 if (__tcp_ao_do_lookup(sk, l3index, addr, family, cmd.prefix, -1, cmd.rcvid))
1672                         return -EEXIST;
1673                 if (__tcp_ao_do_lookup(sk, l3index, addr, family,
1674                                        cmd.prefix, cmd.sndid, -1))
1675                         return -EEXIST;
1676         }
1677
1678         key = tcp_ao_key_alloc(sk, &cmd);
1679         if (IS_ERR(key)) {
1680                 ret = PTR_ERR(key);
1681                 goto err_free_ao;
1682         }
1683
1684         INIT_HLIST_NODE(&key->node);
1685         memcpy(&key->addr, addr, (family == AF_INET) ? sizeof(struct in_addr) :
1686                                                        sizeof(struct in6_addr));
1687         key->prefixlen  = cmd.prefix;
1688         key->family     = family;
1689         key->keyflags   = cmd.keyflags;
1690         key->sndid      = cmd.sndid;
1691         key->rcvid      = cmd.rcvid;
1692         key->l3index    = l3index;
1693         atomic64_set(&key->pkt_good, 0);
1694         atomic64_set(&key->pkt_bad, 0);
1695
1696         ret = tcp_ao_parse_crypto(&cmd, key);
1697         if (ret < 0)
1698                 goto err_free_sock;
1699
1700         if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))) {
1701                 tcp_ao_cache_traffic_keys(sk, ao_info, key);
1702                 if (first) {
1703                         ao_info->current_key = key;
1704                         ao_info->rnext_key = key;
1705                 }
1706         }
1707
1708         tcp_ao_link_mkt(ao_info, key);
1709         if (first) {
1710                 if (!static_branch_inc(&tcp_ao_needed.key)) {
1711                         ret = -EUSERS;
1712                         goto err_free_sock;
1713                 }
1714                 sk_gso_disable(sk);
1715                 rcu_assign_pointer(tcp_sk(sk)->ao_info, ao_info);
1716         }
1717
1718         if (cmd.set_current)
1719                 WRITE_ONCE(ao_info->current_key, key);
1720         if (cmd.set_rnext)
1721                 WRITE_ONCE(ao_info->rnext_key, key);
1722         return 0;
1723
1724 err_free_sock:
1725         atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1726         tcp_sigpool_release(key->tcp_sigpool_id);
1727         kfree_sensitive(key);
1728 err_free_ao:
1729         if (first)
1730                 kfree(ao_info);
1731         return ret;
1732 }
1733
1734 static int tcp_ao_delete_key(struct sock *sk, struct tcp_ao_info *ao_info,
1735                              bool del_async, struct tcp_ao_key *key,
1736                              struct tcp_ao_key *new_current,
1737                              struct tcp_ao_key *new_rnext)
1738 {
1739         int err;
1740
1741         hlist_del_rcu(&key->node);
1742
1743         /* Support for async delete on listening sockets: as they don't
1744          * need current_key/rnext_key maintaining, we don't need to check
1745          * them and we can just free all resources in RCU fashion.
1746          */
1747         if (del_async) {
1748                 atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1749                 call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1750                 return 0;
1751         }
1752
1753         /* At this moment another CPU could have looked this key up
1754          * while it was unlinked from the list. Wait for RCU grace period,
1755          * after which the key is off-list and can't be looked up again;
1756          * the rx path [just before RCU came] might have used it and set it
1757          * as current_key (very unlikely).
1758          * Free the key with next RCU grace period (in case it was
1759          * current_key before tcp_ao_current_rnext() might have
1760          * changed it in forced-delete).
1761          */
1762         synchronize_rcu();
1763         if (new_current)
1764                 WRITE_ONCE(ao_info->current_key, new_current);
1765         if (new_rnext)
1766                 WRITE_ONCE(ao_info->rnext_key, new_rnext);
1767
1768         if (unlikely(READ_ONCE(ao_info->current_key) == key ||
1769                      READ_ONCE(ao_info->rnext_key) == key)) {
1770                 err = -EBUSY;
1771                 goto add_key;
1772         }
1773
1774         atomic_sub(tcp_ao_sizeof_key(key), &sk->sk_omem_alloc);
1775         call_rcu(&key->rcu, tcp_ao_key_free_rcu);
1776
1777         return 0;
1778 add_key:
1779         hlist_add_head_rcu(&key->node, &ao_info->head);
1780         return err;
1781 }
1782
1783 #define TCP_AO_DEL_KEYF_ALL (TCP_AO_KEYF_IFINDEX)
1784 static int tcp_ao_del_cmd(struct sock *sk, unsigned short int family,
1785                           sockptr_t optval, int optlen)
1786 {
1787         struct tcp_ao_key *key, *new_current = NULL, *new_rnext = NULL;
1788         int err, addr_len, l3index = 0;
1789         struct tcp_ao_info *ao_info;
1790         union tcp_ao_addr *addr;
1791         struct tcp_ao_del cmd;
1792         __u8 prefix;
1793         u16 port;
1794
1795         if (optlen < sizeof(cmd))
1796                 return -EINVAL;
1797
1798         err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1799         if (err)
1800                 return err;
1801
1802         if (cmd.reserved != 0 || cmd.reserved2 != 0)
1803                 return -EINVAL;
1804
1805         if (cmd.set_current || cmd.set_rnext) {
1806                 if (!tcp_ao_can_set_current_rnext(sk))
1807                         return -EINVAL;
1808         }
1809
1810         if (cmd.keyflags & ~TCP_AO_DEL_KEYF_ALL)
1811                 return -EINVAL;
1812
1813         /* No sanity check for TCP_AO_KEYF_IFINDEX as if a VRF
1814          * was destroyed, there still should be a way to delete keys,
1815          * that were bound to that l3intf. So, fail late at lookup stage
1816          * if there is no key for that ifindex.
1817          */
1818         if (cmd.ifindex && !(cmd.keyflags & TCP_AO_KEYF_IFINDEX))
1819                 return -EINVAL;
1820
1821         ao_info = setsockopt_ao_info(sk);
1822         if (IS_ERR(ao_info))
1823                 return PTR_ERR(ao_info);
1824         if (!ao_info)
1825                 return -ENOENT;
1826
1827         /* For sockets in TCP_CLOSED it's possible set keys that aren't
1828          * matching the future peer (address/VRF/etc),
1829          * tcp_ao_connect_init() will choose a correct matching MKT
1830          * if there's any.
1831          */
1832         if (cmd.set_current) {
1833                 new_current = tcp_ao_established_key(ao_info, cmd.current_key, -1);
1834                 if (!new_current)
1835                         return -ENOENT;
1836         }
1837         if (cmd.set_rnext) {
1838                 new_rnext = tcp_ao_established_key(ao_info, -1, cmd.rnext);
1839                 if (!new_rnext)
1840                         return -ENOENT;
1841         }
1842         if (cmd.del_async && sk->sk_state != TCP_LISTEN)
1843                 return -EINVAL;
1844
1845         if (family == AF_INET) {
1846                 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.addr;
1847
1848                 addr = (union tcp_ao_addr *)&sin->sin_addr;
1849                 addr_len = sizeof(struct in_addr);
1850                 port = ntohs(sin->sin_port);
1851         } else {
1852                 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.addr;
1853                 struct in6_addr *addr6 = &sin6->sin6_addr;
1854
1855                 if (ipv6_addr_v4mapped(addr6)) {
1856                         addr = (union tcp_ao_addr *)&addr6->s6_addr32[3];
1857                         addr_len = sizeof(struct in_addr);
1858                         family = AF_INET;
1859                 } else {
1860                         addr = (union tcp_ao_addr *)addr6;
1861                         addr_len = sizeof(struct in6_addr);
1862                 }
1863                 port = ntohs(sin6->sin6_port);
1864         }
1865         prefix = cmd.prefix;
1866
1867         /* Currently matching is not performed on port (or port ranges) */
1868         if (port != 0)
1869                 return -EINVAL;
1870
1871         /* We could choose random present key here for current/rnext
1872          * but that's less predictable. Let's be strict and don't
1873          * allow removing a key that's in use. RFC5925 doesn't
1874          * specify how-to coordinate key removal, but says:
1875          * "It is presumed that an MKT affecting a particular
1876          * connection cannot be destroyed during an active connection"
1877          */
1878         hlist_for_each_entry_rcu(key, &ao_info->head, node) {
1879                 if (cmd.sndid != key->sndid ||
1880                     cmd.rcvid != key->rcvid)
1881                         continue;
1882
1883                 if (family != key->family ||
1884                     prefix != key->prefixlen ||
1885                     memcmp(addr, &key->addr, addr_len))
1886                         continue;
1887
1888                 if ((cmd.keyflags & TCP_AO_KEYF_IFINDEX) !=
1889                     (key->keyflags & TCP_AO_KEYF_IFINDEX))
1890                         continue;
1891
1892                 if (key->l3index != l3index)
1893                         continue;
1894
1895                 if (key == new_current || key == new_rnext)
1896                         continue;
1897
1898                 return tcp_ao_delete_key(sk, ao_info, cmd.del_async, key,
1899                                          new_current, new_rnext);
1900         }
1901         return -ENOENT;
1902 }
1903
1904 /* cmd.ao_required makes a socket TCP-AO only.
1905  * Don't allow any md5 keys for any l3intf on the socket together with it.
1906  * Restricting it early in setsockopt() removes a check for
1907  * ao_info->ao_required on inbound tcp segment fast-path.
1908  */
1909 static int tcp_ao_required_verify(struct sock *sk)
1910 {
1911 #ifdef CONFIG_TCP_MD5SIG
1912         const struct tcp_md5sig_info *md5sig;
1913
1914         if (!static_branch_unlikely(&tcp_md5_needed.key))
1915                 return 0;
1916
1917         md5sig = rcu_dereference_check(tcp_sk(sk)->md5sig_info,
1918                                        lockdep_sock_is_held(sk));
1919         if (!md5sig)
1920                 return 0;
1921
1922         if (rcu_dereference_check(hlist_first_rcu(&md5sig->head),
1923                                   lockdep_sock_is_held(sk)))
1924                 return 1;
1925 #endif
1926         return 0;
1927 }
1928
1929 static int tcp_ao_info_cmd(struct sock *sk, unsigned short int family,
1930                            sockptr_t optval, int optlen)
1931 {
1932         struct tcp_ao_key *new_current = NULL, *new_rnext = NULL;
1933         struct tcp_ao_info *ao_info;
1934         struct tcp_ao_info_opt cmd;
1935         bool first = false;
1936         int err;
1937
1938         if (optlen < sizeof(cmd))
1939                 return -EINVAL;
1940
1941         err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
1942         if (err)
1943                 return err;
1944
1945         if (cmd.set_current || cmd.set_rnext) {
1946                 if (!tcp_ao_can_set_current_rnext(sk))
1947                         return -EINVAL;
1948         }
1949
1950         if (cmd.reserved != 0 || cmd.reserved2 != 0)
1951                 return -EINVAL;
1952
1953         ao_info = setsockopt_ao_info(sk);
1954         if (IS_ERR(ao_info))
1955                 return PTR_ERR(ao_info);
1956         if (!ao_info) {
1957                 if (!((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)))
1958                         return -EINVAL;
1959                 ao_info = tcp_ao_alloc_info(GFP_KERNEL);
1960                 if (!ao_info)
1961                         return -ENOMEM;
1962                 first = true;
1963         }
1964
1965         if (cmd.ao_required && tcp_ao_required_verify(sk))
1966                 return -EKEYREJECTED;
1967
1968         /* For sockets in TCP_CLOSED it's possible set keys that aren't
1969          * matching the future peer (address/port/VRF/etc),
1970          * tcp_ao_connect_init() will choose a correct matching MKT
1971          * if there's any.
1972          */
1973         if (cmd.set_current) {
1974                 new_current = tcp_ao_established_key(ao_info, cmd.current_key, -1);
1975                 if (!new_current) {
1976                         err = -ENOENT;
1977                         goto out;
1978                 }
1979         }
1980         if (cmd.set_rnext) {
1981                 new_rnext = tcp_ao_established_key(ao_info, -1, cmd.rnext);
1982                 if (!new_rnext) {
1983                         err = -ENOENT;
1984                         goto out;
1985                 }
1986         }
1987         if (cmd.set_counters) {
1988                 atomic64_set(&ao_info->counters.pkt_good, cmd.pkt_good);
1989                 atomic64_set(&ao_info->counters.pkt_bad, cmd.pkt_bad);
1990                 atomic64_set(&ao_info->counters.key_not_found, cmd.pkt_key_not_found);
1991                 atomic64_set(&ao_info->counters.ao_required, cmd.pkt_ao_required);
1992                 atomic64_set(&ao_info->counters.dropped_icmp, cmd.pkt_dropped_icmp);
1993         }
1994
1995         ao_info->ao_required = cmd.ao_required;
1996         ao_info->accept_icmps = cmd.accept_icmps;
1997         if (new_current)
1998                 WRITE_ONCE(ao_info->current_key, new_current);
1999         if (new_rnext)
2000                 WRITE_ONCE(ao_info->rnext_key, new_rnext);
2001         if (first) {
2002                 if (!static_branch_inc(&tcp_ao_needed.key)) {
2003                         err = -EUSERS;
2004                         goto out;
2005                 }
2006                 sk_gso_disable(sk);
2007                 rcu_assign_pointer(tcp_sk(sk)->ao_info, ao_info);
2008         }
2009         return 0;
2010 out:
2011         if (first)
2012                 kfree(ao_info);
2013         return err;
2014 }
2015
2016 int tcp_parse_ao(struct sock *sk, int cmd, unsigned short int family,
2017                  sockptr_t optval, int optlen)
2018 {
2019         if (WARN_ON_ONCE(family != AF_INET && family != AF_INET6))
2020                 return -EAFNOSUPPORT;
2021
2022         switch (cmd) {
2023         case TCP_AO_ADD_KEY:
2024                 return tcp_ao_add_cmd(sk, family, optval, optlen);
2025         case TCP_AO_DEL_KEY:
2026                 return tcp_ao_del_cmd(sk, family, optval, optlen);
2027         case TCP_AO_INFO:
2028                 return tcp_ao_info_cmd(sk, family, optval, optlen);
2029         default:
2030                 WARN_ON_ONCE(1);
2031                 return -EINVAL;
2032         }
2033 }
2034
2035 int tcp_v4_parse_ao(struct sock *sk, int cmd, sockptr_t optval, int optlen)
2036 {
2037         return tcp_parse_ao(sk, cmd, AF_INET, optval, optlen);
2038 }
2039
2040 /* tcp_ao_copy_mkts_to_user(ao_info, optval, optlen)
2041  *
2042  * @ao_info:    struct tcp_ao_info on the socket that
2043  *              socket getsockopt(TCP_AO_GET_KEYS) is executed on
2044  * @optval:     pointer to array of tcp_ao_getsockopt structures in user space.
2045  *              Must be != NULL.
2046  * @optlen:     pointer to size of tcp_ao_getsockopt structure.
2047  *              Must be != NULL.
2048  *
2049  * Return value: 0 on success, a negative error number otherwise.
2050  *
2051  * optval points to an array of tcp_ao_getsockopt structures in user space.
2052  * optval[0] is used as both input and output to getsockopt. It determines
2053  * which keys are returned by the kernel.
2054  * optval[0].nkeys is the size of the array in user space. On return it contains
2055  * the number of keys matching the search criteria.
2056  * If tcp_ao_getsockopt::get_all is set, then all keys in the socket are
2057  * returned, otherwise only keys matching <addr, prefix, sndid, rcvid>
2058  * in optval[0] are returned.
2059  * optlen is also used as both input and output. The user provides the size
2060  * of struct tcp_ao_getsockopt in user space, and the kernel returns the size
2061  * of the structure in kernel space.
2062  * The size of struct tcp_ao_getsockopt may differ between user and kernel.
2063  * There are three cases to consider:
2064  *  * If usize == ksize, then keys are copied verbatim.
2065  *  * If usize < ksize, then the userspace has passed an old struct to a
2066  *    newer kernel. The rest of the trailing bytes in optval[0]
2067  *    (ksize - usize) are interpreted as 0 by the kernel.
2068  *  * If usize > ksize, then the userspace has passed a new struct to an
2069  *    older kernel. The trailing bytes unknown to the kernel (usize - ksize)
2070  *    are checked to ensure they are zeroed, otherwise -E2BIG is returned.
2071  * On return the kernel fills in min(usize, ksize) in each entry of the array.
2072  * The layout of the fields in the user and kernel structures is expected to
2073  * be the same (including in the 32bit vs 64bit case).
2074  */
2075 static int tcp_ao_copy_mkts_to_user(struct tcp_ao_info *ao_info,
2076                                     sockptr_t optval, sockptr_t optlen)
2077 {
2078         struct tcp_ao_getsockopt opt_in, opt_out;
2079         struct tcp_ao_key *key, *current_key;
2080         bool do_address_matching = true;
2081         union tcp_ao_addr *addr = NULL;
2082         int err, l3index, user_len;
2083         unsigned int max_keys;  /* maximum number of keys to copy to user */
2084         size_t out_offset = 0;
2085         size_t bytes_to_write;  /* number of bytes to write to user level */
2086         u32 matched_keys;       /* keys from ao_info matched so far */
2087         int optlen_out;
2088         __be16 port = 0;
2089
2090         if (copy_from_sockptr(&user_len, optlen, sizeof(int)))
2091                 return -EFAULT;
2092
2093         if (user_len <= 0)
2094                 return -EINVAL;
2095
2096         memset(&opt_in, 0, sizeof(struct tcp_ao_getsockopt));
2097         err = copy_struct_from_sockptr(&opt_in, sizeof(opt_in),
2098                                        optval, user_len);
2099         if (err < 0)
2100                 return err;
2101
2102         if (opt_in.pkt_good || opt_in.pkt_bad)
2103                 return -EINVAL;
2104         if (opt_in.keyflags & ~TCP_AO_GET_KEYF_VALID)
2105                 return -EINVAL;
2106         if (opt_in.ifindex && !(opt_in.keyflags & TCP_AO_KEYF_IFINDEX))
2107                 return -EINVAL;
2108
2109         if (opt_in.reserved != 0)
2110                 return -EINVAL;
2111
2112         max_keys = opt_in.nkeys;
2113         l3index = (opt_in.keyflags & TCP_AO_KEYF_IFINDEX) ? opt_in.ifindex : -1;
2114
2115         if (opt_in.get_all || opt_in.is_current || opt_in.is_rnext) {
2116                 if (opt_in.get_all && (opt_in.is_current || opt_in.is_rnext))
2117                         return -EINVAL;
2118                 do_address_matching = false;
2119         }
2120
2121         switch (opt_in.addr.ss_family) {
2122         case AF_INET: {
2123                 struct sockaddr_in *sin;
2124                 __be32 mask;
2125
2126                 sin = (struct sockaddr_in *)&opt_in.addr;
2127                 port = sin->sin_port;
2128                 addr = (union tcp_ao_addr *)&sin->sin_addr;
2129
2130                 if (opt_in.prefix > 32)
2131                         return -EINVAL;
2132
2133                 if (ntohl(sin->sin_addr.s_addr) == INADDR_ANY &&
2134                     opt_in.prefix != 0)
2135                         return -EINVAL;
2136
2137                 mask = inet_make_mask(opt_in.prefix);
2138                 if (sin->sin_addr.s_addr & ~mask)
2139                         return -EINVAL;
2140
2141                 break;
2142         }
2143         case AF_INET6: {
2144                 struct sockaddr_in6 *sin6;
2145                 struct in6_addr *addr6;
2146
2147                 sin6 = (struct sockaddr_in6 *)&opt_in.addr;
2148                 addr = (union tcp_ao_addr *)&sin6->sin6_addr;
2149                 addr6 = &sin6->sin6_addr;
2150                 port = sin6->sin6_port;
2151
2152                 /* We don't have to change family and @addr here if
2153                  * ipv6_addr_v4mapped() like in key adding:
2154                  * tcp_ao_key_cmp() does it. Do the sanity checks though.
2155                  */
2156                 if (opt_in.prefix != 0) {
2157                         if (ipv6_addr_v4mapped(addr6)) {
2158                                 __be32 mask, addr4 = addr6->s6_addr32[3];
2159
2160                                 if (opt_in.prefix > 32 ||
2161                                     ntohl(addr4) == INADDR_ANY)
2162                                         return -EINVAL;
2163                                 mask = inet_make_mask(opt_in.prefix);
2164                                 if (addr4 & ~mask)
2165                                         return -EINVAL;
2166                         } else {
2167                                 struct in6_addr pfx;
2168
2169                                 if (ipv6_addr_any(addr6) ||
2170                                     opt_in.prefix > 128)
2171                                         return -EINVAL;
2172
2173                                 ipv6_addr_prefix(&pfx, addr6, opt_in.prefix);
2174                                 if (ipv6_addr_cmp(&pfx, addr6))
2175                                         return -EINVAL;
2176                         }
2177                 } else if (!ipv6_addr_any(addr6)) {
2178                         return -EINVAL;
2179                 }
2180                 break;
2181         }
2182         case 0:
2183                 if (!do_address_matching)
2184                         break;
2185                 fallthrough;
2186         default:
2187                 return -EAFNOSUPPORT;
2188         }
2189
2190         if (!do_address_matching) {
2191                 /* We could just ignore those, but let's do stricter checks */
2192                 if (addr || port)
2193                         return -EINVAL;
2194                 if (opt_in.prefix || opt_in.sndid || opt_in.rcvid)
2195                         return -EINVAL;
2196         }
2197
2198         bytes_to_write = min_t(int, user_len, sizeof(struct tcp_ao_getsockopt));
2199         matched_keys = 0;
2200         /* May change in RX, while we're dumping, pre-fetch it */
2201         current_key = READ_ONCE(ao_info->current_key);
2202
2203         hlist_for_each_entry_rcu(key, &ao_info->head, node) {
2204                 if (opt_in.get_all)
2205                         goto match;
2206
2207                 if (opt_in.is_current || opt_in.is_rnext) {
2208                         if (opt_in.is_current && key == current_key)
2209                                 goto match;
2210                         if (opt_in.is_rnext && key == ao_info->rnext_key)
2211                                 goto match;
2212                         continue;
2213                 }
2214
2215                 if (tcp_ao_key_cmp(key, l3index, addr, opt_in.prefix,
2216                                    opt_in.addr.ss_family,
2217                                    opt_in.sndid, opt_in.rcvid) != 0)
2218                         continue;
2219 match:
2220                 matched_keys++;
2221                 if (matched_keys > max_keys)
2222                         continue;
2223
2224                 memset(&opt_out, 0, sizeof(struct tcp_ao_getsockopt));
2225
2226                 if (key->family == AF_INET) {
2227                         struct sockaddr_in *sin_out = (struct sockaddr_in *)&opt_out.addr;
2228
2229                         sin_out->sin_family = key->family;
2230                         sin_out->sin_port = 0;
2231                         memcpy(&sin_out->sin_addr, &key->addr, sizeof(struct in_addr));
2232                 } else {
2233                         struct sockaddr_in6 *sin6_out = (struct sockaddr_in6 *)&opt_out.addr;
2234
2235                         sin6_out->sin6_family = key->family;
2236                         sin6_out->sin6_port = 0;
2237                         memcpy(&sin6_out->sin6_addr, &key->addr, sizeof(struct in6_addr));
2238                 }
2239                 opt_out.sndid = key->sndid;
2240                 opt_out.rcvid = key->rcvid;
2241                 opt_out.prefix = key->prefixlen;
2242                 opt_out.keyflags = key->keyflags;
2243                 opt_out.is_current = (key == current_key);
2244                 opt_out.is_rnext = (key == ao_info->rnext_key);
2245                 opt_out.nkeys = 0;
2246                 opt_out.maclen = key->maclen;
2247                 opt_out.keylen = key->keylen;
2248                 opt_out.ifindex = key->l3index;
2249                 opt_out.pkt_good = atomic64_read(&key->pkt_good);
2250                 opt_out.pkt_bad = atomic64_read(&key->pkt_bad);
2251                 memcpy(&opt_out.key, key->key, key->keylen);
2252                 tcp_sigpool_algo(key->tcp_sigpool_id, opt_out.alg_name, 64);
2253
2254                 /* Copy key to user */
2255                 if (copy_to_sockptr_offset(optval, out_offset,
2256                                            &opt_out, bytes_to_write))
2257                         return -EFAULT;
2258                 out_offset += user_len;
2259         }
2260
2261         optlen_out = (int)sizeof(struct tcp_ao_getsockopt);
2262         if (copy_to_sockptr(optlen, &optlen_out, sizeof(int)))
2263                 return -EFAULT;
2264
2265         out_offset = offsetof(struct tcp_ao_getsockopt, nkeys);
2266         if (copy_to_sockptr_offset(optval, out_offset,
2267                                    &matched_keys, sizeof(u32)))
2268                 return -EFAULT;
2269
2270         return 0;
2271 }
2272
2273 int tcp_ao_get_mkts(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2274 {
2275         struct tcp_ao_info *ao_info;
2276
2277         ao_info = setsockopt_ao_info(sk);
2278         if (IS_ERR(ao_info))
2279                 return PTR_ERR(ao_info);
2280         if (!ao_info)
2281                 return -ENOENT;
2282
2283         return tcp_ao_copy_mkts_to_user(ao_info, optval, optlen);
2284 }
2285
2286 int tcp_ao_get_sock_info(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2287 {
2288         struct tcp_ao_info_opt out, in = {};
2289         struct tcp_ao_key *current_key;
2290         struct tcp_ao_info *ao;
2291         int err, len;
2292
2293         if (copy_from_sockptr(&len, optlen, sizeof(int)))
2294                 return -EFAULT;
2295
2296         if (len <= 0)
2297                 return -EINVAL;
2298
2299         /* Copying this "in" only to check ::reserved, ::reserved2,
2300          * that may be needed to extend (struct tcp_ao_info_opt) and
2301          * what getsockopt() provides in future.
2302          */
2303         err = copy_struct_from_sockptr(&in, sizeof(in), optval, len);
2304         if (err)
2305                 return err;
2306
2307         if (in.reserved != 0 || in.reserved2 != 0)
2308                 return -EINVAL;
2309
2310         ao = setsockopt_ao_info(sk);
2311         if (IS_ERR(ao))
2312                 return PTR_ERR(ao);
2313         if (!ao)
2314                 return -ENOENT;
2315
2316         memset(&out, 0, sizeof(out));
2317         out.ao_required         = ao->ao_required;
2318         out.accept_icmps        = ao->accept_icmps;
2319         out.pkt_good            = atomic64_read(&ao->counters.pkt_good);
2320         out.pkt_bad             = atomic64_read(&ao->counters.pkt_bad);
2321         out.pkt_key_not_found   = atomic64_read(&ao->counters.key_not_found);
2322         out.pkt_ao_required     = atomic64_read(&ao->counters.ao_required);
2323         out.pkt_dropped_icmp    = atomic64_read(&ao->counters.dropped_icmp);
2324
2325         current_key = READ_ONCE(ao->current_key);
2326         if (current_key) {
2327                 out.set_current = 1;
2328                 out.current_key = current_key->sndid;
2329         }
2330         if (ao->rnext_key) {
2331                 out.set_rnext = 1;
2332                 out.rnext = ao->rnext_key->rcvid;
2333         }
2334
2335         if (copy_to_sockptr(optval, &out, min_t(int, len, sizeof(out))))
2336                 return -EFAULT;
2337
2338         return 0;
2339 }
2340
2341 int tcp_ao_set_repair(struct sock *sk, sockptr_t optval, unsigned int optlen)
2342 {
2343         struct tcp_sock *tp = tcp_sk(sk);
2344         struct tcp_ao_repair cmd;
2345         struct tcp_ao_key *key;
2346         struct tcp_ao_info *ao;
2347         int err;
2348
2349         if (optlen < sizeof(cmd))
2350                 return -EINVAL;
2351
2352         err = copy_struct_from_sockptr(&cmd, sizeof(cmd), optval, optlen);
2353         if (err)
2354                 return err;
2355
2356         if (!tp->repair)
2357                 return -EPERM;
2358
2359         ao = setsockopt_ao_info(sk);
2360         if (IS_ERR(ao))
2361                 return PTR_ERR(ao);
2362         if (!ao)
2363                 return -ENOENT;
2364
2365         WRITE_ONCE(ao->lisn, cmd.snt_isn);
2366         WRITE_ONCE(ao->risn, cmd.rcv_isn);
2367         WRITE_ONCE(ao->snd_sne, cmd.snd_sne);
2368         WRITE_ONCE(ao->rcv_sne, cmd.rcv_sne);
2369
2370         hlist_for_each_entry_rcu(key, &ao->head, node)
2371                 tcp_ao_cache_traffic_keys(sk, ao, key);
2372
2373         return 0;
2374 }
2375
2376 int tcp_ao_get_repair(struct sock *sk, sockptr_t optval, sockptr_t optlen)
2377 {
2378         struct tcp_sock *tp = tcp_sk(sk);
2379         struct tcp_ao_repair opt;
2380         struct tcp_ao_info *ao;
2381         int len;
2382
2383         if (copy_from_sockptr(&len, optlen, sizeof(int)))
2384                 return -EFAULT;
2385
2386         if (len <= 0)
2387                 return -EINVAL;
2388
2389         if (!tp->repair)
2390                 return -EPERM;
2391
2392         rcu_read_lock();
2393         ao = getsockopt_ao_info(sk);
2394         if (IS_ERR_OR_NULL(ao)) {
2395                 rcu_read_unlock();
2396                 return ao ? PTR_ERR(ao) : -ENOENT;
2397         }
2398
2399         opt.snt_isn     = ao->lisn;
2400         opt.rcv_isn     = ao->risn;
2401         opt.snd_sne     = READ_ONCE(ao->snd_sne);
2402         opt.rcv_sne     = READ_ONCE(ao->rcv_sne);
2403         rcu_read_unlock();
2404
2405         if (copy_to_sockptr(optval, &opt, min_t(int, len, sizeof(opt))))
2406                 return -EFAULT;
2407         return 0;
2408 }