net/ipv4/tcp_metrics.c

   1 #include <linux/rcupdate.h>
   2 #include <linux/spinlock.h>
   3 #include <linux/jiffies.h>
   4 #include <linux/module.h>
   5 #include <linux/cache.h>
   6 #include <linux/slab.h>
   7 #include <linux/init.h>
   8 #include <linux/tcp.h>
   9 #include <linux/hash.h>
  10 #include <linux/tcp_metrics.h>
  11 #include <linux/vmalloc.h>
  12
  13 #include <net/inet_connection_sock.h>
  14 #include <net/net_namespace.h>
  15 #include <net/request_sock.h>
  16 #include <net/inetpeer.h>
  17 #include <net/sock.h>
  18 #include <net/ipv6.h>
  19 #include <net/dst.h>
  20 #include <net/tcp.h>
  21 #include <net/genetlink.h>
  22
  23 int sysctl_tcp_nometrics_save __read_mostly;
  24
  25 static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr,
  26                                                    const struct inetpeer_addr *daddr,
  27                                                    struct net *net, unsigned int hash);
  28
  29 struct tcp_fastopen_metrics {
  30         u16     mss;
  31         u16     syn_loss:10;            /* Recurring Fast Open SYN losses */
  32         unsigned long   last_syn_loss;  /* Last Fast Open SYN loss */
  33         struct  tcp_fastopen_cookie     cookie;
  34 };
  35
  36 /* TCP_METRIC_MAX includes 2 extra fields for userspace compatibility
  37  * Kernel only stores RTT and RTTVAR in usec resolution
  38  */
  39 #define TCP_METRIC_MAX_KERNEL (TCP_METRIC_MAX - 2)
  40
  41 struct tcp_metrics_block {
  42         struct tcp_metrics_block __rcu  *tcpm_next;
  43         struct inetpeer_addr            tcpm_saddr;
  44         struct inetpeer_addr            tcpm_daddr;
  45         unsigned long                   tcpm_stamp;
  46         u32                             tcpm_ts;
  47         u32                             tcpm_ts_stamp;
  48         u32                             tcpm_lock;
  49         u32                             tcpm_vals[TCP_METRIC_MAX_KERNEL + 1];
  50         struct tcp_fastopen_metrics     tcpm_fastopen;
  51
  52         struct rcu_head                 rcu_head;
  53 };
  54
  55 static bool tcp_metric_locked(struct tcp_metrics_block *tm,
  56                               enum tcp_metric_index idx)
  57 {
  58         return tm->tcpm_lock & (1 << idx);
  59 }
  60
  61 static u32 tcp_metric_get(struct tcp_metrics_block *tm,
  62                           enum tcp_metric_index idx)
  63 {
  64         return tm->tcpm_vals[idx];
  65 }
  66
  67 static void tcp_metric_set(struct tcp_metrics_block *tm,
  68                            enum tcp_metric_index idx,
  69                            u32 val)
  70 {
  71         tm->tcpm_vals[idx] = val;
  72 }
  73
  74 static bool addr_same(const struct inetpeer_addr *a,
  75                       const struct inetpeer_addr *b)
  76 {
  77         const struct in6_addr *a6, *b6;
  78
  79         if (a->family != b->family)
  80                 return false;
  81         if (a->family == AF_INET)
  82                 return a->addr.a4 == b->addr.a4;
  83
  84         a6 = (const struct in6_addr *) &a->addr.a6[0];
  85         b6 = (const struct in6_addr *) &b->addr.a6[0];
  86
  87         return ipv6_addr_equal(a6, b6);
  88 }
  89
  90 struct tcpm_hash_bucket {
  91         struct tcp_metrics_block __rcu  *chain;
  92 };
  93
  94 static DEFINE_SPINLOCK(tcp_metrics_lock);
  95
  96 static void tcpm_suck_dst(struct tcp_metrics_block *tm,
  97                           const struct dst_entry *dst,
  98                           bool fastopen_clear)
  99 {
 100         u32 msval;
 101         u32 val;
 102
 103         tm->tcpm_stamp = jiffies;
 104
 105         val = 0;
 106         if (dst_metric_locked(dst, RTAX_RTT))
 107                 val |= 1 << TCP_METRIC_RTT;
 108         if (dst_metric_locked(dst, RTAX_RTTVAR))
 109                 val |= 1 << TCP_METRIC_RTTVAR;
 110         if (dst_metric_locked(dst, RTAX_SSTHRESH))
 111                 val |= 1 << TCP_METRIC_SSTHRESH;
 112         if (dst_metric_locked(dst, RTAX_CWND))
 113                 val |= 1 << TCP_METRIC_CWND;
 114         if (dst_metric_locked(dst, RTAX_REORDERING))
 115                 val |= 1 << TCP_METRIC_REORDERING;
 116         tm->tcpm_lock = val;
 117
 118         msval = dst_metric_raw(dst, RTAX_RTT);
 119         tm->tcpm_vals[TCP_METRIC_RTT] = msval * USEC_PER_MSEC;
 120
 121         msval = dst_metric_raw(dst, RTAX_RTTVAR);
 122         tm->tcpm_vals[TCP_METRIC_RTTVAR] = msval * USEC_PER_MSEC;
 123         tm->tcpm_vals[TCP_METRIC_SSTHRESH] = dst_metric_raw(dst, RTAX_SSTHRESH);
 124         tm->tcpm_vals[TCP_METRIC_CWND] = dst_metric_raw(dst, RTAX_CWND);
 125         tm->tcpm_vals[TCP_METRIC_REORDERING] = dst_metric_raw(dst, RTAX_REORDERING);
 126         tm->tcpm_ts = 0;
 127         tm->tcpm_ts_stamp = 0;
 128         if (fastopen_clear) {
 129                 tm->tcpm_fastopen.mss = 0;
 130                 tm->tcpm_fastopen.syn_loss = 0;
 131                 tm->tcpm_fastopen.cookie.len = 0;
 132         }
 133 }
 134
 135 #define TCP_METRICS_TIMEOUT             (60 * 60 * HZ)
 136
 137 static void tcpm_check_stamp(struct tcp_metrics_block *tm, struct dst_entry *dst)
 138 {
 139         if (tm && unlikely(time_after(jiffies, tm->tcpm_stamp + TCP_METRICS_TIMEOUT)))
 140                 tcpm_suck_dst(tm, dst, false);
 141 }
 142
 143 #define TCP_METRICS_RECLAIM_DEPTH       5
 144 #define TCP_METRICS_RECLAIM_PTR         (struct tcp_metrics_block *) 0x1UL
 145
 146 static struct tcp_metrics_block *tcpm_new(struct dst_entry *dst,
 147                                           struct inetpeer_addr *saddr,
 148                                           struct inetpeer_addr *daddr,
 149                                           unsigned int hash)
 150 {
 151         struct tcp_metrics_block *tm;
 152         struct net *net;
 153         bool reclaim = false;
 154
 155         spin_lock_bh(&tcp_metrics_lock);
 156         net = dev_net(dst->dev);
 157
 158         /* While waiting for the spin-lock the cache might have been populated
 159          * with this entry and so we have to check again.
 160          */
 161         tm = __tcp_get_metrics(saddr, daddr, net, hash);
 162         if (tm == TCP_METRICS_RECLAIM_PTR) {
 163                 reclaim = true;
 164                 tm = NULL;
 165         }
 166         if (tm) {
 167                 tcpm_check_stamp(tm, dst);
 168                 goto out_unlock;
 169         }
 170
 171         if (unlikely(reclaim)) {
 172                 struct tcp_metrics_block *oldest;
 173
 174                 oldest = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain);
 175                 for (tm = rcu_dereference(oldest->tcpm_next); tm;
 176                      tm = rcu_dereference(tm->tcpm_next)) {
 177                         if (time_before(tm->tcpm_stamp, oldest->tcpm_stamp))
 178                                 oldest = tm;
 179                 }
 180                 tm = oldest;
 181         } else {
 182                 tm = kmalloc(sizeof(*tm), GFP_ATOMIC);
 183                 if (!tm)
 184                         goto out_unlock;
 185         }
 186         tm->tcpm_saddr = *saddr;
 187         tm->tcpm_daddr = *daddr;
 188
 189         tcpm_suck_dst(tm, dst, true);
 190
 191         if (likely(!reclaim)) {
 192                 tm->tcpm_next = net->ipv4.tcp_metrics_hash[hash].chain;
 193                 rcu_assign_pointer(net->ipv4.tcp_metrics_hash[hash].chain, tm);
 194         }
 195
 196 out_unlock:
 197         spin_unlock_bh(&tcp_metrics_lock);
 198         return tm;
 199 }
 200
 201 static struct tcp_metrics_block *tcp_get_encode(struct tcp_metrics_block *tm, int depth)
 202 {
 203         if (tm)
 204                 return tm;
 205         if (depth > TCP_METRICS_RECLAIM_DEPTH)
 206                 return TCP_METRICS_RECLAIM_PTR;
 207         return NULL;
 208 }
 209
 210 static struct tcp_metrics_block *__tcp_get_metrics(const struct inetpeer_addr *saddr,
 211                                                    const struct inetpeer_addr *daddr,
 212                                                    struct net *net, unsigned int hash)
 213 {
 214         struct tcp_metrics_block *tm;
 215         int depth = 0;
 216
 217         for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
 218              tm = rcu_dereference(tm->tcpm_next)) {
 219                 if (addr_same(&tm->tcpm_saddr, saddr) &&
 220                     addr_same(&tm->tcpm_daddr, daddr))
 221                         break;
 222                 depth++;
 223         }
 224         return tcp_get_encode(tm, depth);
 225 }
 226
 227 static struct tcp_metrics_block *__tcp_get_metrics_req(struct request_sock *req,
 228                                                        struct dst_entry *dst)
 229 {
 230         struct tcp_metrics_block *tm;
 231         struct inetpeer_addr saddr, daddr;
 232         unsigned int hash;
 233         struct net *net;
 234
 235         saddr.family = req->rsk_ops->family;
 236         daddr.family = req->rsk_ops->family;
 237         switch (daddr.family) {
 238         case AF_INET:
 239                 saddr.addr.a4 = inet_rsk(req)->ir_loc_addr;
 240                 daddr.addr.a4 = inet_rsk(req)->ir_rmt_addr;
 241                 hash = (__force unsigned int) daddr.addr.a4;
 242                 break;
 243 #if IS_ENABLED(CONFIG_IPV6)
 244         case AF_INET6:
 245                 *(struct in6_addr *)saddr.addr.a6 = inet_rsk(req)->ir_v6_loc_addr;
 246                 *(struct in6_addr *)daddr.addr.a6 = inet_rsk(req)->ir_v6_rmt_addr;
 247                 hash = ipv6_addr_hash(&inet_rsk(req)->ir_v6_rmt_addr);
 248                 break;
 249 #endif
 250         default:
 251                 return NULL;
 252         }
 253
 254         net = dev_net(dst->dev);
 255         hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
 256
 257         for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
 258              tm = rcu_dereference(tm->tcpm_next)) {
 259                 if (addr_same(&tm->tcpm_saddr, &saddr) &&
 260                     addr_same(&tm->tcpm_daddr, &daddr))
 261                         break;
 262         }
 263         tcpm_check_stamp(tm, dst);
 264         return tm;
 265 }
 266
 267 static struct tcp_metrics_block *__tcp_get_metrics_tw(struct inet_timewait_sock *tw)
 268 {
 269         struct tcp_metrics_block *tm;
 270         struct inetpeer_addr saddr, daddr;
 271         unsigned int hash;
 272         struct net *net;
 273
 274         if (tw->tw_family == AF_INET) {
 275                 saddr.family = AF_INET;
 276                 saddr.addr.a4 = tw->tw_rcv_saddr;
 277                 daddr.family = AF_INET;
 278                 daddr.addr.a4 = tw->tw_daddr;
 279                 hash = (__force unsigned int) daddr.addr.a4;
 280         }
 281 #if IS_ENABLED(CONFIG_IPV6)
 282         else if (tw->tw_family == AF_INET6) {
 283                 if (ipv6_addr_v4mapped(&tw->tw_v6_daddr)) {
 284                         saddr.family = AF_INET;
 285                         saddr.addr.a4 = tw->tw_rcv_saddr;
 286                         daddr.family = AF_INET;
 287                         daddr.addr.a4 = tw->tw_daddr;
 288                         hash = (__force unsigned int) daddr.addr.a4;
 289                 } else {
 290                         saddr.family = AF_INET6;
 291                         *(struct in6_addr *)saddr.addr.a6 = tw->tw_v6_rcv_saddr;
 292                         daddr.family = AF_INET6;
 293                         *(struct in6_addr *)daddr.addr.a6 = tw->tw_v6_daddr;
 294                         hash = ipv6_addr_hash(&tw->tw_v6_daddr);
 295                 }
 296         }
 297 #endif
 298         else
 299                 return NULL;
 300
 301         net = twsk_net(tw);
 302         hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
 303
 304         for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
 305              tm = rcu_dereference(tm->tcpm_next)) {
 306                 if (addr_same(&tm->tcpm_saddr, &saddr) &&
 307                     addr_same(&tm->tcpm_daddr, &daddr))
 308                         break;
 309         }
 310         return tm;
 311 }
 312
 313 static struct tcp_metrics_block *tcp_get_metrics(struct sock *sk,
 314                                                  struct dst_entry *dst,
 315                                                  bool create)
 316 {
 317         struct tcp_metrics_block *tm;
 318         struct inetpeer_addr saddr, daddr;
 319         unsigned int hash;
 320         struct net *net;
 321
 322         if (sk->sk_family == AF_INET) {
 323                 saddr.family = AF_INET;
 324                 saddr.addr.a4 = inet_sk(sk)->inet_saddr;
 325                 daddr.family = AF_INET;
 326                 daddr.addr.a4 = inet_sk(sk)->inet_daddr;
 327                 hash = (__force unsigned int) daddr.addr.a4;
 328         }
 329 #if IS_ENABLED(CONFIG_IPV6)
 330         else if (sk->sk_family == AF_INET6) {
 331                 if (ipv6_addr_v4mapped(&sk->sk_v6_daddr)) {
 332                         saddr.family = AF_INET;
 333                         saddr.addr.a4 = inet_sk(sk)->inet_saddr;
 334                         daddr.family = AF_INET;
 335                         daddr.addr.a4 = inet_sk(sk)->inet_daddr;
 336                         hash = (__force unsigned int) daddr.addr.a4;
 337                 } else {
 338                         saddr.family = AF_INET6;
 339                         *(struct in6_addr *)saddr.addr.a6 = sk->sk_v6_rcv_saddr;
 340                         daddr.family = AF_INET6;
 341                         *(struct in6_addr *)daddr.addr.a6 = sk->sk_v6_daddr;
 342                         hash = ipv6_addr_hash(&sk->sk_v6_daddr);
 343                 }
 344         }
 345 #endif
 346         else
 347                 return NULL;
 348
 349         net = dev_net(dst->dev);
 350         hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
 351
 352         tm = __tcp_get_metrics(&saddr, &daddr, net, hash);
 353         if (tm == TCP_METRICS_RECLAIM_PTR)
 354                 tm = NULL;
 355         if (!tm && create)
 356                 tm = tcpm_new(dst, &saddr, &daddr, hash);
 357         else
 358                 tcpm_check_stamp(tm, dst);
 359
 360         return tm;
 361 }
 362
 363 /* Save metrics learned by this TCP session.  This function is called
 364  * only, when TCP finishes successfully i.e. when it enters TIME-WAIT
 365  * or goes from LAST-ACK to CLOSE.
 366  */
 367 void tcp_update_metrics(struct sock *sk)
 368 {
 369         const struct inet_connection_sock *icsk = inet_csk(sk);
 370         struct dst_entry *dst = __sk_dst_get(sk);
 371         struct tcp_sock *tp = tcp_sk(sk);
 372         struct tcp_metrics_block *tm;
 373         unsigned long rtt;
 374         u32 val;
 375         int m;
 376
 377         if (sysctl_tcp_nometrics_save || !dst)
 378                 return;
 379
 380         if (dst->flags & DST_HOST)
 381                 dst_confirm(dst);
 382
 383         rcu_read_lock();
 384         if (icsk->icsk_backoff || !tp->srtt_us) {
 385                 /* This session failed to estimate rtt. Why?
 386                  * Probably, no packets returned in time.  Reset our
 387                  * results.
 388                  */
 389                 tm = tcp_get_metrics(sk, dst, false);
 390                 if (tm && !tcp_metric_locked(tm, TCP_METRIC_RTT))
 391                         tcp_metric_set(tm, TCP_METRIC_RTT, 0);
 392                 goto out_unlock;
 393         } else
 394                 tm = tcp_get_metrics(sk, dst, true);
 395
 396         if (!tm)
 397                 goto out_unlock;
 398
 399         rtt = tcp_metric_get(tm, TCP_METRIC_RTT);
 400         m = rtt - tp->srtt_us;
 401
 402         /* If newly calculated rtt larger than stored one, store new
 403          * one. Otherwise, use EWMA. Remember, rtt overestimation is
 404          * always better than underestimation.
 405          */
 406         if (!tcp_metric_locked(tm, TCP_METRIC_RTT)) {
 407                 if (m <= 0)
 408                         rtt = tp->srtt_us;
 409                 else
 410                         rtt -= (m >> 3);
 411                 tcp_metric_set(tm, TCP_METRIC_RTT, rtt);
 412         }
 413
 414         if (!tcp_metric_locked(tm, TCP_METRIC_RTTVAR)) {
 415                 unsigned long var;
 416
 417                 if (m < 0)
 418                         m = -m;
 419
 420                 /* Scale deviation to rttvar fixed point */
 421                 m >>= 1;
 422                 if (m < tp->mdev_us)
 423                         m = tp->mdev_us;
 424
 425                 var = tcp_metric_get(tm, TCP_METRIC_RTTVAR);
 426                 if (m >= var)
 427                         var = m;
 428                 else
 429                         var -= (var - m) >> 2;
 430
 431                 tcp_metric_set(tm, TCP_METRIC_RTTVAR, var);
 432         }
 433
 434         if (tcp_in_initial_slowstart(tp)) {
 435                 /* Slow start still did not finish. */
 436                 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) {
 437                         val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
 438                         if (val && (tp->snd_cwnd >> 1) > val)
 439                                 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
 440                                                tp->snd_cwnd >> 1);
 441                 }
 442                 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
 443                         val = tcp_metric_get(tm, TCP_METRIC_CWND);
 444                         if (tp->snd_cwnd > val)
 445                                 tcp_metric_set(tm, TCP_METRIC_CWND,
 446                                                tp->snd_cwnd);
 447                 }
 448         } else if (tp->snd_cwnd > tp->snd_ssthresh &&
 449                    icsk->icsk_ca_state == TCP_CA_Open) {
 450                 /* Cong. avoidance phase, cwnd is reliable. */
 451                 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH))
 452                         tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
 453                                        max(tp->snd_cwnd >> 1, tp->snd_ssthresh));
 454                 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
 455                         val = tcp_metric_get(tm, TCP_METRIC_CWND);
 456                         tcp_metric_set(tm, TCP_METRIC_CWND, (val + tp->snd_cwnd) >> 1);
 457                 }
 458         } else {
 459                 /* Else slow start did not finish, cwnd is non-sense,
 460                  * ssthresh may be also invalid.
 461                  */
 462                 if (!tcp_metric_locked(tm, TCP_METRIC_CWND)) {
 463                         val = tcp_metric_get(tm, TCP_METRIC_CWND);
 464                         tcp_metric_set(tm, TCP_METRIC_CWND,
 465                                        (val + tp->snd_ssthresh) >> 1);
 466                 }
 467                 if (!tcp_metric_locked(tm, TCP_METRIC_SSTHRESH)) {
 468                         val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
 469                         if (val && tp->snd_ssthresh > val)
 470                                 tcp_metric_set(tm, TCP_METRIC_SSTHRESH,
 471                                                tp->snd_ssthresh);
 472                 }
 473                 if (!tcp_metric_locked(tm, TCP_METRIC_REORDERING)) {
 474                         val = tcp_metric_get(tm, TCP_METRIC_REORDERING);
 475                         if (val < tp->reordering &&
 476                             tp->reordering != sysctl_tcp_reordering)
 477                                 tcp_metric_set(tm, TCP_METRIC_REORDERING,
 478                                                tp->reordering);
 479                 }
 480         }
 481         tm->tcpm_stamp = jiffies;
 482 out_unlock:
 483         rcu_read_unlock();
 484 }
 485
 486 /* Initialize metrics on socket. */
 487
 488 void tcp_init_metrics(struct sock *sk)
 489 {
 490         struct dst_entry *dst = __sk_dst_get(sk);
 491         struct tcp_sock *tp = tcp_sk(sk);
 492         struct tcp_metrics_block *tm;
 493         u32 val, crtt = 0; /* cached RTT scaled by 8 */
 494
 495         if (dst == NULL)
 496                 goto reset;
 497
 498         dst_confirm(dst);
 499
 500         rcu_read_lock();
 501         tm = tcp_get_metrics(sk, dst, true);
 502         if (!tm) {
 503                 rcu_read_unlock();
 504                 goto reset;
 505         }
 506
 507         if (tcp_metric_locked(tm, TCP_METRIC_CWND))
 508                 tp->snd_cwnd_clamp = tcp_metric_get(tm, TCP_METRIC_CWND);
 509
 510         val = tcp_metric_get(tm, TCP_METRIC_SSTHRESH);
 511         if (val) {
 512                 tp->snd_ssthresh = val;
 513                 if (tp->snd_ssthresh > tp->snd_cwnd_clamp)
 514                         tp->snd_ssthresh = tp->snd_cwnd_clamp;
 515         } else {
 516                 /* ssthresh may have been reduced unnecessarily during.
 517                  * 3WHS. Restore it back to its initial default.
 518                  */
 519                 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
 520         }
 521         val = tcp_metric_get(tm, TCP_METRIC_REORDERING);
 522         if (val && tp->reordering != val) {
 523                 tcp_disable_fack(tp);
 524                 tcp_disable_early_retrans(tp);
 525                 tp->reordering = val;
 526         }
 527
 528         crtt = tcp_metric_get(tm, TCP_METRIC_RTT);
 529         rcu_read_unlock();
 530 reset:
 531         /* The initial RTT measurement from the SYN/SYN-ACK is not ideal
 532          * to seed the RTO for later data packets because SYN packets are
 533          * small. Use the per-dst cached values to seed the RTO but keep
 534          * the RTT estimator variables intact (e.g., srtt, mdev, rttvar).
 535          * Later the RTO will be updated immediately upon obtaining the first
 536          * data RTT sample (tcp_rtt_estimator()). Hence the cached RTT only
 537          * influences the first RTO but not later RTT estimation.
 538          *
 539          * But if RTT is not available from the SYN (due to retransmits or
 540          * syn cookies) or the cache, force a conservative 3secs timeout.
 541          *
 542          * A bit of theory. RTT is time passed after "normal" sized packet
 543          * is sent until it is ACKed. In normal circumstances sending small
 544          * packets force peer to delay ACKs and calculation is correct too.
 545          * The algorithm is adaptive and, provided we follow specs, it
 546          * NEVER underestimate RTT. BUT! If peer tries to make some clever
 547          * tricks sort of "quick acks" for time long enough to decrease RTT
 548          * to low value, and then abruptly stops to do it and starts to delay
 549          * ACKs, wait for troubles.
 550          */
 551         if (crtt > tp->srtt_us) {
 552                 /* Set RTO like tcp_rtt_estimator(), but from cached RTT. */
 553                 crtt /= 8 * USEC_PER_MSEC;
 554                 inet_csk(sk)->icsk_rto = crtt + max(2 * crtt, tcp_rto_min(sk));
 555         } else if (tp->srtt_us == 0) {
 556                 /* RFC6298: 5.7 We've failed to get a valid RTT sample from
 557                  * 3WHS. This is most likely due to retransmission,
 558                  * including spurious one. Reset the RTO back to 3secs
 559                  * from the more aggressive 1sec to avoid more spurious
 560                  * retransmission.
 561                  */
 562                 tp->rttvar_us = jiffies_to_usecs(TCP_TIMEOUT_FALLBACK);
 563                 tp->mdev_us = tp->mdev_max_us = tp->rttvar_us;
 564
 565                 inet_csk(sk)->icsk_rto = TCP_TIMEOUT_FALLBACK;
 566         }
 567         /* Cut cwnd down to 1 per RFC5681 if SYN or SYN-ACK has been
 568          * retransmitted. In light of RFC6298 more aggressive 1sec
 569          * initRTO, we only reset cwnd when more than 1 SYN/SYN-ACK
 570          * retransmission has occurred.
 571          */
 572         if (tp->total_retrans > 1)
 573                 tp->snd_cwnd = 1;
 574         else
 575                 tp->snd_cwnd = tcp_init_cwnd(tp, dst);
 576         tp->snd_cwnd_stamp = tcp_time_stamp;
 577 }
 578
 579 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
 580                         bool paws_check, bool timestamps)
 581 {
 582         struct tcp_metrics_block *tm;
 583         bool ret;
 584
 585         if (!dst)
 586                 return false;
 587
 588         rcu_read_lock();
 589         tm = __tcp_get_metrics_req(req, dst);
 590         if (paws_check) {
 591                 if (tm &&
 592                     (u32)get_seconds() - tm->tcpm_ts_stamp < TCP_PAWS_MSL &&
 593                     ((s32)(tm->tcpm_ts - req->ts_recent) > TCP_PAWS_WINDOW ||
 594                      !timestamps))
 595                         ret = false;
 596                 else
 597                         ret = true;
 598         } else {
 599                 if (tm && tcp_metric_get(tm, TCP_METRIC_RTT) && tm->tcpm_ts_stamp)
 600                         ret = true;
 601                 else
 602                         ret = false;
 603         }
 604         rcu_read_unlock();
 605
 606         return ret;
 607 }
 608 EXPORT_SYMBOL_GPL(tcp_peer_is_proven);
 609
 610 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst)
 611 {
 612         struct tcp_metrics_block *tm;
 613
 614         rcu_read_lock();
 615         tm = tcp_get_metrics(sk, dst, true);
 616         if (tm) {
 617                 struct tcp_sock *tp = tcp_sk(sk);
 618
 619                 if ((u32)get_seconds() - tm->tcpm_ts_stamp <= TCP_PAWS_MSL) {
 620                         tp->rx_opt.ts_recent_stamp = tm->tcpm_ts_stamp;
 621                         tp->rx_opt.ts_recent = tm->tcpm_ts;
 622                 }
 623         }
 624         rcu_read_unlock();
 625 }
 626 EXPORT_SYMBOL_GPL(tcp_fetch_timewait_stamp);
 627
 628 /* VJ's idea. Save last timestamp seen from this destination and hold
 629  * it at least for normal timewait interval to use for duplicate
 630  * segment detection in subsequent connections, before they enter
 631  * synchronized state.
 632  */
 633 bool tcp_remember_stamp(struct sock *sk)
 634 {
 635         struct dst_entry *dst = __sk_dst_get(sk);
 636         bool ret = false;
 637
 638         if (dst) {
 639                 struct tcp_metrics_block *tm;
 640
 641                 rcu_read_lock();
 642                 tm = tcp_get_metrics(sk, dst, true);
 643                 if (tm) {
 644                         struct tcp_sock *tp = tcp_sk(sk);
 645
 646                         if ((s32)(tm->tcpm_ts - tp->rx_opt.ts_recent) <= 0 ||
 647                             ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL &&
 648                              tm->tcpm_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) {
 649                                 tm->tcpm_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp;
 650                                 tm->tcpm_ts = tp->rx_opt.ts_recent;
 651                         }
 652                         ret = true;
 653                 }
 654                 rcu_read_unlock();
 655         }
 656         return ret;
 657 }
 658
 659 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw)
 660 {
 661         struct tcp_metrics_block *tm;
 662         bool ret = false;
 663
 664         rcu_read_lock();
 665         tm = __tcp_get_metrics_tw(tw);
 666         if (tm) {
 667                 const struct tcp_timewait_sock *tcptw;
 668                 struct sock *sk = (struct sock *) tw;
 669
 670                 tcptw = tcp_twsk(sk);
 671                 if ((s32)(tm->tcpm_ts - tcptw->tw_ts_recent) <= 0 ||
 672                     ((u32)get_seconds() - tm->tcpm_ts_stamp > TCP_PAWS_MSL &&
 673                      tm->tcpm_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) {
 674                         tm->tcpm_ts_stamp = (u32)tcptw->tw_ts_recent_stamp;
 675                         tm->tcpm_ts        = tcptw->tw_ts_recent;
 676                 }
 677                 ret = true;
 678         }
 679         rcu_read_unlock();
 680
 681         return ret;
 682 }
 683
 684 static DEFINE_SEQLOCK(fastopen_seqlock);
 685
 686 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
 687                             struct tcp_fastopen_cookie *cookie,
 688                             int *syn_loss, unsigned long *last_syn_loss)
 689 {
 690         struct tcp_metrics_block *tm;
 691
 692         rcu_read_lock();
 693         tm = tcp_get_metrics(sk, __sk_dst_get(sk), false);
 694         if (tm) {
 695                 struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen;
 696                 unsigned int seq;
 697
 698                 do {
 699                         seq = read_seqbegin(&fastopen_seqlock);
 700                         if (tfom->mss)
 701                                 *mss = tfom->mss;
 702                         *cookie = tfom->cookie;
 703                         *syn_loss = tfom->syn_loss;
 704                         *last_syn_loss = *syn_loss ? tfom->last_syn_loss : 0;
 705                 } while (read_seqretry(&fastopen_seqlock, seq));
 706         }
 707         rcu_read_unlock();
 708 }
 709
 710 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
 711                             struct tcp_fastopen_cookie *cookie, bool syn_lost)
 712 {
 713         struct dst_entry *dst = __sk_dst_get(sk);
 714         struct tcp_metrics_block *tm;
 715
 716         if (!dst)
 717                 return;
 718         rcu_read_lock();
 719         tm = tcp_get_metrics(sk, dst, true);
 720         if (tm) {
 721                 struct tcp_fastopen_metrics *tfom = &tm->tcpm_fastopen;
 722
 723                 write_seqlock_bh(&fastopen_seqlock);
 724                 if (mss)
 725                         tfom->mss = mss;
 726                 if (cookie && cookie->len > 0)
 727                         tfom->cookie = *cookie;
 728                 if (syn_lost) {
 729                         ++tfom->syn_loss;
 730                         tfom->last_syn_loss = jiffies;
 731                 } else
 732                         tfom->syn_loss = 0;
 733                 write_sequnlock_bh(&fastopen_seqlock);
 734         }
 735         rcu_read_unlock();
 736 }
 737
 738 static struct genl_family tcp_metrics_nl_family = {
 739         .id             = GENL_ID_GENERATE,
 740         .hdrsize        = 0,
 741         .name           = TCP_METRICS_GENL_NAME,
 742         .version        = TCP_METRICS_GENL_VERSION,
 743         .maxattr        = TCP_METRICS_ATTR_MAX,
 744         .netnsok        = true,
 745 };
 746
 747 static struct nla_policy tcp_metrics_nl_policy[TCP_METRICS_ATTR_MAX + 1] = {
 748         [TCP_METRICS_ATTR_ADDR_IPV4]    = { .type = NLA_U32, },
 749         [TCP_METRICS_ATTR_ADDR_IPV6]    = { .type = NLA_BINARY,
 750                                             .len = sizeof(struct in6_addr), },
 751         /* Following attributes are not received for GET/DEL,
 752          * we keep them for reference
 753          */
 754 #if 0
 755         [TCP_METRICS_ATTR_AGE]          = { .type = NLA_MSECS, },
 756         [TCP_METRICS_ATTR_TW_TSVAL]     = { .type = NLA_U32, },
 757         [TCP_METRICS_ATTR_TW_TS_STAMP]  = { .type = NLA_S32, },
 758         [TCP_METRICS_ATTR_VALS]         = { .type = NLA_NESTED, },
 759         [TCP_METRICS_ATTR_FOPEN_MSS]    = { .type = NLA_U16, },
 760         [TCP_METRICS_ATTR_FOPEN_SYN_DROPS]      = { .type = NLA_U16, },
 761         [TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS]    = { .type = NLA_MSECS, },
 762         [TCP_METRICS_ATTR_FOPEN_COOKIE] = { .type = NLA_BINARY,
 763                                             .len = TCP_FASTOPEN_COOKIE_MAX, },
 764 #endif
 765 };
 766
 767 /* Add attributes, caller cancels its header on failure */
 768 static int tcp_metrics_fill_info(struct sk_buff *msg,
 769                                  struct tcp_metrics_block *tm)
 770 {
 771         struct nlattr *nest;
 772         int i;
 773
 774         switch (tm->tcpm_daddr.family) {
 775         case AF_INET:
 776                 if (nla_put_be32(msg, TCP_METRICS_ATTR_ADDR_IPV4,
 777                                 tm->tcpm_daddr.addr.a4) < 0)
 778                         goto nla_put_failure;
 779                 if (nla_put_be32(msg, TCP_METRICS_ATTR_SADDR_IPV4,
 780                                 tm->tcpm_saddr.addr.a4) < 0)
 781                         goto nla_put_failure;
 782                 break;
 783         case AF_INET6:
 784                 if (nla_put(msg, TCP_METRICS_ATTR_ADDR_IPV6, 16,
 785                             tm->tcpm_daddr.addr.a6) < 0)
 786                         goto nla_put_failure;
 787                 if (nla_put(msg, TCP_METRICS_ATTR_SADDR_IPV6, 16,
 788                             tm->tcpm_saddr.addr.a6) < 0)
 789                         goto nla_put_failure;
 790                 break;
 791         default:
 792                 return -EAFNOSUPPORT;
 793         }
 794
 795         if (nla_put_msecs(msg, TCP_METRICS_ATTR_AGE,
 796                           jiffies - tm->tcpm_stamp) < 0)
 797                 goto nla_put_failure;
 798         if (tm->tcpm_ts_stamp) {
 799                 if (nla_put_s32(msg, TCP_METRICS_ATTR_TW_TS_STAMP,
 800                                 (s32) (get_seconds() - tm->tcpm_ts_stamp)) < 0)
 801                         goto nla_put_failure;
 802                 if (nla_put_u32(msg, TCP_METRICS_ATTR_TW_TSVAL,
 803                                 tm->tcpm_ts) < 0)
 804                         goto nla_put_failure;
 805         }
 806
 807         {
 808                 int n = 0;
 809
 810                 nest = nla_nest_start(msg, TCP_METRICS_ATTR_VALS);
 811                 if (!nest)
 812                         goto nla_put_failure;
 813                 for (i = 0; i < TCP_METRIC_MAX_KERNEL + 1; i++) {
 814                         u32 val = tm->tcpm_vals[i];
 815
 816                         if (!val)
 817                                 continue;
 818                         if (i == TCP_METRIC_RTT) {
 819                                 if (nla_put_u32(msg, TCP_METRIC_RTT_US + 1,
 820                                                 val) < 0)
 821                                         goto nla_put_failure;
 822                                 n++;
 823                                 val = max(val / 1000, 1U);
 824                         }
 825                         if (i == TCP_METRIC_RTTVAR) {
 826                                 if (nla_put_u32(msg, TCP_METRIC_RTTVAR_US + 1,
 827                                                 val) < 0)
 828                                         goto nla_put_failure;
 829                                 n++;
 830                                 val = max(val / 1000, 1U);
 831                         }
 832                         if (nla_put_u32(msg, i + 1, val) < 0)
 833                                 goto nla_put_failure;
 834                         n++;
 835                 }
 836                 if (n)
 837                         nla_nest_end(msg, nest);
 838                 else
 839                         nla_nest_cancel(msg, nest);
 840         }
 841
 842         {
 843                 struct tcp_fastopen_metrics tfom_copy[1], *tfom;
 844                 unsigned int seq;
 845
 846                 do {
 847                         seq = read_seqbegin(&fastopen_seqlock);
 848                         tfom_copy[0] = tm->tcpm_fastopen;
 849                 } while (read_seqretry(&fastopen_seqlock, seq));
 850
 851                 tfom = tfom_copy;
 852                 if (tfom->mss &&
 853                     nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_MSS,
 854                                 tfom->mss) < 0)
 855                         goto nla_put_failure;
 856                 if (tfom->syn_loss &&
 857                     (nla_put_u16(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROPS,
 858                                 tfom->syn_loss) < 0 ||
 859                      nla_put_msecs(msg, TCP_METRICS_ATTR_FOPEN_SYN_DROP_TS,
 860                                 jiffies - tfom->last_syn_loss) < 0))
 861                         goto nla_put_failure;
 862                 if (tfom->cookie.len > 0 &&
 863                     nla_put(msg, TCP_METRICS_ATTR_FOPEN_COOKIE,
 864                             tfom->cookie.len, tfom->cookie.val) < 0)
 865                         goto nla_put_failure;
 866         }
 867
 868         return 0;
 869
 870 nla_put_failure:
 871         return -EMSGSIZE;
 872 }
 873
 874 static int tcp_metrics_dump_info(struct sk_buff *skb,
 875                                  struct netlink_callback *cb,
 876                                  struct tcp_metrics_block *tm)
 877 {
 878         void *hdr;
 879
 880         hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
 881                           &tcp_metrics_nl_family, NLM_F_MULTI,
 882                           TCP_METRICS_CMD_GET);
 883         if (!hdr)
 884                 return -EMSGSIZE;
 885
 886         if (tcp_metrics_fill_info(skb, tm) < 0)
 887                 goto nla_put_failure;
 888
 889         return genlmsg_end(skb, hdr);
 890
 891 nla_put_failure:
 892         genlmsg_cancel(skb, hdr);
 893         return -EMSGSIZE;
 894 }
 895
 896 static int tcp_metrics_nl_dump(struct sk_buff *skb,
 897                                struct netlink_callback *cb)
 898 {
 899         struct net *net = sock_net(skb->sk);
 900         unsigned int max_rows = 1U << net->ipv4.tcp_metrics_hash_log;
 901         unsigned int row, s_row = cb->args[0];
 902         int s_col = cb->args[1], col = s_col;
 903
 904         for (row = s_row; row < max_rows; row++, s_col = 0) {
 905                 struct tcp_metrics_block *tm;
 906                 struct tcpm_hash_bucket *hb = net->ipv4.tcp_metrics_hash + row;
 907
 908                 rcu_read_lock();
 909                 for (col = 0, tm = rcu_dereference(hb->chain); tm;
 910                      tm = rcu_dereference(tm->tcpm_next), col++) {
 911                         if (col < s_col)
 912                                 continue;
 913                         if (tcp_metrics_dump_info(skb, cb, tm) < 0) {
 914                                 rcu_read_unlock();
 915                                 goto done;
 916                         }
 917                 }
 918                 rcu_read_unlock();
 919         }
 920
 921 done:
 922         cb->args[0] = row;
 923         cb->args[1] = col;
 924         return skb->len;
 925 }
 926
 927 static int __parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
 928                            unsigned int *hash, int optional, int v4, int v6)
 929 {
 930         struct nlattr *a;
 931
 932         a = info->attrs[v4];
 933         if (a) {
 934                 addr->family = AF_INET;
 935                 addr->addr.a4 = nla_get_be32(a);
 936                 if (hash)
 937                         *hash = (__force unsigned int) addr->addr.a4;
 938                 return 0;
 939         }
 940         a = info->attrs[v6];
 941         if (a) {
 942                 if (nla_len(a) != sizeof(struct in6_addr))
 943                         return -EINVAL;
 944                 addr->family = AF_INET6;
 945                 memcpy(addr->addr.a6, nla_data(a), sizeof(addr->addr.a6));
 946                 if (hash)
 947                         *hash = ipv6_addr_hash((struct in6_addr *) addr->addr.a6);
 948                 return 0;
 949         }
 950         return optional ? 1 : -EAFNOSUPPORT;
 951 }
 952
 953 static int parse_nl_addr(struct genl_info *info, struct inetpeer_addr *addr,
 954                          unsigned int *hash, int optional)
 955 {
 956         return __parse_nl_addr(info, addr, hash, optional,
 957                                TCP_METRICS_ATTR_ADDR_IPV4,
 958                                TCP_METRICS_ATTR_ADDR_IPV6);
 959 }
 960
 961 static int parse_nl_saddr(struct genl_info *info, struct inetpeer_addr *addr)
 962 {
 963         return __parse_nl_addr(info, addr, NULL, 0,
 964                                TCP_METRICS_ATTR_SADDR_IPV4,
 965                                TCP_METRICS_ATTR_SADDR_IPV6);
 966 }
 967
 968 static int tcp_metrics_nl_cmd_get(struct sk_buff *skb, struct genl_info *info)
 969 {
 970         struct tcp_metrics_block *tm;
 971         struct inetpeer_addr saddr, daddr;
 972         unsigned int hash;
 973         struct sk_buff *msg;
 974         struct net *net = genl_info_net(info);
 975         void *reply;
 976         int ret;
 977         bool src = true;
 978
 979         ret = parse_nl_addr(info, &daddr, &hash, 0);
 980         if (ret < 0)
 981                 return ret;
 982
 983         ret = parse_nl_saddr(info, &saddr);
 984         if (ret < 0)
 985                 src = false;
 986
 987         msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
 988         if (!msg)
 989                 return -ENOMEM;
 990
 991         reply = genlmsg_put_reply(msg, info, &tcp_metrics_nl_family, 0,
 992                                   info->genlhdr->cmd);
 993         if (!reply)
 994                 goto nla_put_failure;
 995
 996         hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
 997         ret = -ESRCH;
 998         rcu_read_lock();
 999         for (tm = rcu_dereference(net->ipv4.tcp_metrics_hash[hash].chain); tm;
1000              tm = rcu_dereference(tm->tcpm_next)) {
1001                 if (addr_same(&tm->tcpm_daddr, &daddr) &&
1002                     (!src || addr_same(&tm->tcpm_saddr, &saddr))) {
1003                         ret = tcp_metrics_fill_info(msg, tm);
1004                         break;
1005                 }
1006         }
1007         rcu_read_unlock();
1008         if (ret < 0)
1009                 goto out_free;
1010
1011         genlmsg_end(msg, reply);
1012         return genlmsg_reply(msg, info);
1013
1014 nla_put_failure:
1015         ret = -EMSGSIZE;
1016
1017 out_free:
1018         nlmsg_free(msg);
1019         return ret;
1020 }
1021
1022 #define deref_locked_genl(p)    \
1023         rcu_dereference_protected(p, lockdep_genl_is_held() && \
1024                                      lockdep_is_held(&tcp_metrics_lock))
1025
1026 #define deref_genl(p)   rcu_dereference_protected(p, lockdep_genl_is_held())
1027
1028 static int tcp_metrics_flush_all(struct net *net)
1029 {
1030         unsigned int max_rows = 1U << net->ipv4.tcp_metrics_hash_log;
1031         struct tcpm_hash_bucket *hb = net->ipv4.tcp_metrics_hash;
1032         struct tcp_metrics_block *tm;
1033         unsigned int row;
1034
1035         for (row = 0; row < max_rows; row++, hb++) {
1036                 spin_lock_bh(&tcp_metrics_lock);
1037                 tm = deref_locked_genl(hb->chain);
1038                 if (tm)
1039                         hb->chain = NULL;
1040                 spin_unlock_bh(&tcp_metrics_lock);
1041                 while (tm) {
1042                         struct tcp_metrics_block *next;
1043
1044                         next = deref_genl(tm->tcpm_next);
1045                         kfree_rcu(tm, rcu_head);
1046                         tm = next;
1047                 }
1048         }
1049         return 0;
1050 }
1051
1052 static int tcp_metrics_nl_cmd_del(struct sk_buff *skb, struct genl_info *info)
1053 {
1054         struct tcpm_hash_bucket *hb;
1055         struct tcp_metrics_block *tm;
1056         struct tcp_metrics_block __rcu **pp;
1057         struct inetpeer_addr saddr, daddr;
1058         unsigned int hash;
1059         struct net *net = genl_info_net(info);
1060         int ret;
1061         bool src = true, found = false;
1062
1063         ret = parse_nl_addr(info, &daddr, &hash, 1);
1064         if (ret < 0)
1065                 return ret;
1066         if (ret > 0)
1067                 return tcp_metrics_flush_all(net);
1068         ret = parse_nl_saddr(info, &saddr);
1069         if (ret < 0)
1070                 src = false;
1071
1072         hash = hash_32(hash, net->ipv4.tcp_metrics_hash_log);
1073         hb = net->ipv4.tcp_metrics_hash + hash;
1074         pp = &hb->chain;
1075         spin_lock_bh(&tcp_metrics_lock);
1076         for (tm = deref_locked_genl(*pp); tm; tm = deref_locked_genl(*pp)) {
1077                 if (addr_same(&tm->tcpm_daddr, &daddr) &&
1078                     (!src || addr_same(&tm->tcpm_saddr, &saddr))) {
1079                         *pp = tm->tcpm_next;
1080                         kfree_rcu(tm, rcu_head);
1081                         found = true;
1082                 } else {
1083                         pp = &tm->tcpm_next;
1084                 }
1085         }
1086         spin_unlock_bh(&tcp_metrics_lock);
1087         if (!found)
1088                 return -ESRCH;
1089         return 0;
1090 }
1091
1092 static const struct genl_ops tcp_metrics_nl_ops[] = {
1093         {
1094                 .cmd = TCP_METRICS_CMD_GET,
1095                 .doit = tcp_metrics_nl_cmd_get,
1096                 .dumpit = tcp_metrics_nl_dump,
1097                 .policy = tcp_metrics_nl_policy,
1098         },
1099         {
1100                 .cmd = TCP_METRICS_CMD_DEL,
1101                 .doit = tcp_metrics_nl_cmd_del,
1102                 .policy = tcp_metrics_nl_policy,
1103                 .flags = GENL_ADMIN_PERM,
1104         },
1105 };
1106
1107 static unsigned int tcpmhash_entries;
1108 static int __init set_tcpmhash_entries(char *str)
1109 {
1110         ssize_t ret;
1111
1112         if (!str)
1113                 return 0;
1114
1115         ret = kstrtouint(str, 0, &tcpmhash_entries);
1116         if (ret)
1117                 return 0;
1118
1119         return 1;
1120 }
1121 __setup("tcpmhash_entries=", set_tcpmhash_entries);
1122
1123 static int __net_init tcp_net_metrics_init(struct net *net)
1124 {
1125         size_t size;
1126         unsigned int slots;
1127
1128         slots = tcpmhash_entries;
1129         if (!slots) {
1130                 if (totalram_pages >= 128 * 1024)
1131                         slots = 16 * 1024;
1132                 else
1133                         slots = 8 * 1024;
1134         }
1135
1136         net->ipv4.tcp_metrics_hash_log = order_base_2(slots);
1137         size = sizeof(struct tcpm_hash_bucket) << net->ipv4.tcp_metrics_hash_log;
1138
1139         net->ipv4.tcp_metrics_hash = kzalloc(size, GFP_KERNEL | __GFP_NOWARN);
1140         if (!net->ipv4.tcp_metrics_hash)
1141                 net->ipv4.tcp_metrics_hash = vzalloc(size);
1142
1143         if (!net->ipv4.tcp_metrics_hash)
1144                 return -ENOMEM;
1145
1146         return 0;
1147 }
1148
1149 static void __net_exit tcp_net_metrics_exit(struct net *net)
1150 {
1151         unsigned int i;
1152
1153         for (i = 0; i < (1U << net->ipv4.tcp_metrics_hash_log) ; i++) {
1154                 struct tcp_metrics_block *tm, *next;
1155
1156                 tm = rcu_dereference_protected(net->ipv4.tcp_metrics_hash[i].chain, 1);
1157                 while (tm) {
1158                         next = rcu_dereference_protected(tm->tcpm_next, 1);
1159                         kfree(tm);
1160                         tm = next;
1161                 }
1162         }
1163         kvfree(net->ipv4.tcp_metrics_hash);
1164 }
1165
1166 static __net_initdata struct pernet_operations tcp_net_metrics_ops = {
1167         .init   =       tcp_net_metrics_init,
1168         .exit   =       tcp_net_metrics_exit,
1169 };
1170
1171 void __init tcp_metrics_init(void)
1172 {
1173         int ret;
1174
1175         ret = register_pernet_subsys(&tcp_net_metrics_ops);
1176         if (ret < 0)
1177                 goto cleanup;
1178         ret = genl_register_family_with_ops(&tcp_metrics_nl_family,
1179                                             tcp_metrics_nl_ops);
1180         if (ret < 0)
1181                 goto cleanup_subsys;
1182         return;
1183
1184 cleanup_subsys:
1185         unregister_pernet_subsys(&tcp_net_metrics_ops);
1186
1187 cleanup:
1188         return;
1189 }