linux/net/ipv4/tcp_minisocks.c
<<
>>
Prefs
   1/*
   2 * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3 *              operating system.  INET is implemented using the  BSD Socket
   4 *              interface as the means of communication with the user level.
   5 *
   6 *              Implementation of the Transmission Control Protocol(TCP).
   7 *
   8 * Authors:     Ross Biro
   9 *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10 *              Mark Evans, <evansmp@uhura.aston.ac.uk>
  11 *              Corey Minyard <wf-rch!minyard@relay.EU.net>
  12 *              Florian La Roche, <flla@stud.uni-sb.de>
  13 *              Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
  14 *              Linus Torvalds, <torvalds@cs.helsinki.fi>
  15 *              Alan Cox, <gw4pts@gw4pts.ampr.org>
  16 *              Matthew Dillon, <dillon@apollo.west.oic.com>
  17 *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  18 *              Jorge Cwik, <jorge@laser.satlink.net>
  19 */
  20
  21#include <linux/mm.h>
  22#include <linux/module.h>
  23#include <linux/slab.h>
  24#include <linux/sysctl.h>
  25#include <linux/workqueue.h>
  26#include <net/tcp.h>
  27#include <net/inet_common.h>
  28#include <net/xfrm.h>
  29
  30int sysctl_tcp_syncookies __read_mostly = 1;
  31EXPORT_SYMBOL(sysctl_tcp_syncookies);
  32
  33int sysctl_tcp_abort_on_overflow __read_mostly;
  34
  35struct inet_timewait_death_row tcp_death_row = {
  36        .sysctl_max_tw_buckets = NR_FILE * 2,
  37        .period         = TCP_TIMEWAIT_LEN / INET_TWDR_TWKILL_SLOTS,
  38        .death_lock     = __SPIN_LOCK_UNLOCKED(tcp_death_row.death_lock),
  39        .hashinfo       = &tcp_hashinfo,
  40        .tw_timer       = TIMER_INITIALIZER(inet_twdr_hangman, 0,
  41                                            (unsigned long)&tcp_death_row),
  42        .twkill_work    = __WORK_INITIALIZER(tcp_death_row.twkill_work,
  43                                             inet_twdr_twkill_work),
  44/* Short-time timewait calendar */
  45
  46        .twcal_hand     = -1,
  47        .twcal_timer    = TIMER_INITIALIZER(inet_twdr_twcal_tick, 0,
  48                                            (unsigned long)&tcp_death_row),
  49};
  50EXPORT_SYMBOL_GPL(tcp_death_row);
  51
  52static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
  53{
  54        if (seq == s_win)
  55                return true;
  56        if (after(end_seq, s_win) && before(seq, e_win))
  57                return true;
  58        return seq == e_win && seq == end_seq;
  59}
  60
  61/*
  62 * * Main purpose of TIME-WAIT state is to close connection gracefully,
  63 *   when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
  64 *   (and, probably, tail of data) and one or more our ACKs are lost.
  65 * * What is TIME-WAIT timeout? It is associated with maximal packet
  66 *   lifetime in the internet, which results in wrong conclusion, that
  67 *   it is set to catch "old duplicate segments" wandering out of their path.
  68 *   It is not quite correct. This timeout is calculated so that it exceeds
  69 *   maximal retransmission timeout enough to allow to lose one (or more)
  70 *   segments sent by peer and our ACKs. This time may be calculated from RTO.
  71 * * When TIME-WAIT socket receives RST, it means that another end
  72 *   finally closed and we are allowed to kill TIME-WAIT too.
  73 * * Second purpose of TIME-WAIT is catching old duplicate segments.
  74 *   Well, certainly it is pure paranoia, but if we load TIME-WAIT
  75 *   with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
  76 * * If we invented some more clever way to catch duplicates
  77 *   (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
  78 *
  79 * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
  80 * When you compare it to RFCs, please, read section SEGMENT ARRIVES
  81 * from the very beginning.
  82 *
  83 * NOTE. With recycling (and later with fin-wait-2) TW bucket
  84 * is _not_ stateless. It means, that strictly speaking we must
  85 * spinlock it. I do not want! Well, probability of misbehaviour
  86 * is ridiculously low and, seems, we could use some mb() tricks
  87 * to avoid misread sequence numbers, states etc.  --ANK
  88 *
  89 * We don't need to initialize tmp_out.sack_ok as we don't use the results
  90 */
  91enum tcp_tw_status
  92tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
  93                           const struct tcphdr *th)
  94{
  95        struct tcp_options_received tmp_opt;
  96        const u8 *hash_location;
  97        struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
  98        bool paws_reject = false;
  99
 100        tmp_opt.saw_tstamp = 0;
 101        if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
 102                tcp_parse_options(skb, &tmp_opt, &hash_location, 0, NULL);
 103
 104                if (tmp_opt.saw_tstamp) {
 105                        tmp_opt.ts_recent       = tcptw->tw_ts_recent;
 106                        tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
 107                        paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
 108                }
 109        }
 110
 111        if (tw->tw_substate == TCP_FIN_WAIT2) {
 112                /* Just repeat all the checks of tcp_rcv_state_process() */
 113
 114                /* Out of window, send ACK */
 115                if (paws_reject ||
 116                    !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
 117                                   tcptw->tw_rcv_nxt,
 118                                   tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
 119                        return TCP_TW_ACK;
 120
 121                if (th->rst)
 122                        goto kill;
 123
 124                if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
 125                        goto kill_with_rst;
 126
 127                /* Dup ACK? */
 128                if (!th->ack ||
 129                    !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
 130                    TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
 131                        inet_twsk_put(tw);
 132                        return TCP_TW_SUCCESS;
 133                }
 134
 135                /* New data or FIN. If new data arrive after half-duplex close,
 136                 * reset.
 137                 */
 138                if (!th->fin ||
 139                    TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) {
 140kill_with_rst:
 141                        inet_twsk_deschedule(tw, &tcp_death_row);
 142                        inet_twsk_put(tw);
 143                        return TCP_TW_RST;
 144                }
 145
 146                /* FIN arrived, enter true time-wait state. */
 147                tw->tw_substate   = TCP_TIME_WAIT;
 148                tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
 149                if (tmp_opt.saw_tstamp) {
 150                        tcptw->tw_ts_recent_stamp = get_seconds();
 151                        tcptw->tw_ts_recent       = tmp_opt.rcv_tsval;
 152                }
 153
 154                if (tcp_death_row.sysctl_tw_recycle &&
 155                    tcptw->tw_ts_recent_stamp &&
 156                    tcp_tw_remember_stamp(tw))
 157                        inet_twsk_schedule(tw, &tcp_death_row, tw->tw_timeout,
 158                                           TCP_TIMEWAIT_LEN);
 159                else
 160                        inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
 161                                           TCP_TIMEWAIT_LEN);
 162                return TCP_TW_ACK;
 163        }
 164
 165        /*
 166         *      Now real TIME-WAIT state.
 167         *
 168         *      RFC 1122:
 169         *      "When a connection is [...] on TIME-WAIT state [...]
 170         *      [a TCP] MAY accept a new SYN from the remote TCP to
 171         *      reopen the connection directly, if it:
 172         *
 173         *      (1)  assigns its initial sequence number for the new
 174         *      connection to be larger than the largest sequence
 175         *      number it used on the previous connection incarnation,
 176         *      and
 177         *
 178         *      (2)  returns to TIME-WAIT state if the SYN turns out
 179         *      to be an old duplicate".
 180         */
 181
 182        if (!paws_reject &&
 183            (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
 184             (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
 185                /* In window segment, it may be only reset or bare ack. */
 186
 187                if (th->rst) {
 188                        /* This is TIME_WAIT assassination, in two flavors.
 189                         * Oh well... nobody has a sufficient solution to this
 190                         * protocol bug yet.
 191                         */
 192                        if (sysctl_tcp_rfc1337 == 0) {
 193kill:
 194                                inet_twsk_deschedule(tw, &tcp_death_row);
 195                                inet_twsk_put(tw);
 196                                return TCP_TW_SUCCESS;
 197                        }
 198                }
 199                inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
 200                                   TCP_TIMEWAIT_LEN);
 201
 202                if (tmp_opt.saw_tstamp) {
 203                        tcptw->tw_ts_recent       = tmp_opt.rcv_tsval;
 204                        tcptw->tw_ts_recent_stamp = get_seconds();
 205                }
 206
 207                inet_twsk_put(tw);
 208                return TCP_TW_SUCCESS;
 209        }
 210
 211        /* Out of window segment.
 212
 213           All the segments are ACKed immediately.
 214
 215           The only exception is new SYN. We accept it, if it is
 216           not old duplicate and we are not in danger to be killed
 217           by delayed old duplicates. RFC check is that it has
 218           newer sequence number works at rates <40Mbit/sec.
 219           However, if paws works, it is reliable AND even more,
 220           we even may relax silly seq space cutoff.
 221
 222           RED-PEN: we violate main RFC requirement, if this SYN will appear
 223           old duplicate (i.e. we receive RST in reply to SYN-ACK),
 224           we must return socket to time-wait state. It is not good,
 225           but not fatal yet.
 226         */
 227
 228        if (th->syn && !th->rst && !th->ack && !paws_reject &&
 229            (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
 230             (tmp_opt.saw_tstamp &&
 231              (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
 232                u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
 233                if (isn == 0)
 234                        isn++;
 235                TCP_SKB_CB(skb)->when = isn;
 236                return TCP_TW_SYN;
 237        }
 238
 239        if (paws_reject)
 240                NET_INC_STATS_BH(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
 241
 242        if (!th->rst) {
 243                /* In this case we must reset the TIMEWAIT timer.
 244                 *
 245                 * If it is ACKless SYN it may be both old duplicate
 246                 * and new good SYN with random sequence number <rcv_nxt.
 247                 * Do not reschedule in the last case.
 248                 */
 249                if (paws_reject || th->ack)
 250                        inet_twsk_schedule(tw, &tcp_death_row, TCP_TIMEWAIT_LEN,
 251                                           TCP_TIMEWAIT_LEN);
 252
 253                /* Send ACK. Note, we do not put the bucket,
 254                 * it will be released by caller.
 255                 */
 256                return TCP_TW_ACK;
 257        }
 258        inet_twsk_put(tw);
 259        return TCP_TW_SUCCESS;
 260}
 261EXPORT_SYMBOL(tcp_timewait_state_process);
 262
 263/*
 264 * Move a socket to time-wait or dead fin-wait-2 state.
 265 */
 266void tcp_time_wait(struct sock *sk, int state, int timeo)
 267{
 268        struct inet_timewait_sock *tw = NULL;
 269        const struct inet_connection_sock *icsk = inet_csk(sk);
 270        const struct tcp_sock *tp = tcp_sk(sk);
 271        bool recycle_ok = false;
 272
 273        if (tcp_death_row.sysctl_tw_recycle && tp->rx_opt.ts_recent_stamp)
 274                recycle_ok = tcp_remember_stamp(sk);
 275
 276        if (tcp_death_row.tw_count < tcp_death_row.sysctl_max_tw_buckets)
 277                tw = inet_twsk_alloc(sk, state);
 278
 279        if (tw != NULL) {
 280                struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
 281                const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
 282                struct inet_sock *inet = inet_sk(sk);
 283
 284                tw->tw_transparent      = inet->transparent;
 285                tw->tw_rcv_wscale       = tp->rx_opt.rcv_wscale;
 286                tcptw->tw_rcv_nxt       = tp->rcv_nxt;
 287                tcptw->tw_snd_nxt       = tp->snd_nxt;
 288                tcptw->tw_rcv_wnd       = tcp_receive_window(tp);
 289                tcptw->tw_ts_recent     = tp->rx_opt.ts_recent;
 290                tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
 291
 292#if IS_ENABLED(CONFIG_IPV6)
 293                if (tw->tw_family == PF_INET6) {
 294                        struct ipv6_pinfo *np = inet6_sk(sk);
 295                        struct inet6_timewait_sock *tw6;
 296
 297                        tw->tw_ipv6_offset = inet6_tw_offset(sk->sk_prot);
 298                        tw6 = inet6_twsk((struct sock *)tw);
 299                        tw6->tw_v6_daddr = np->daddr;
 300                        tw6->tw_v6_rcv_saddr = np->rcv_saddr;
 301                        tw->tw_tclass = np->tclass;
 302                        tw->tw_ipv6only = np->ipv6only;
 303                }
 304#endif
 305
 306#ifdef CONFIG_TCP_MD5SIG
 307                /*
 308                 * The timewait bucket does not have the key DB from the
 309                 * sock structure. We just make a quick copy of the
 310                 * md5 key being used (if indeed we are using one)
 311                 * so the timewait ack generating code has the key.
 312                 */
 313                do {
 314                        struct tcp_md5sig_key *key;
 315                        tcptw->tw_md5_key = NULL;
 316                        key = tp->af_specific->md5_lookup(sk, sk);
 317                        if (key != NULL) {
 318                                tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
 319                                if (tcptw->tw_md5_key && tcp_alloc_md5sig_pool(sk) == NULL)
 320                                        BUG();
 321                        }
 322                } while (0);
 323#endif
 324
 325                /* Linkage updates. */
 326                __inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
 327
 328                /* Get the TIME_WAIT timeout firing. */
 329                if (timeo < rto)
 330                        timeo = rto;
 331
 332                if (recycle_ok) {
 333                        tw->tw_timeout = rto;
 334                } else {
 335                        tw->tw_timeout = TCP_TIMEWAIT_LEN;
 336                        if (state == TCP_TIME_WAIT)
 337                                timeo = TCP_TIMEWAIT_LEN;
 338                }
 339
 340                inet_twsk_schedule(tw, &tcp_death_row, timeo,
 341                                   TCP_TIMEWAIT_LEN);
 342                inet_twsk_put(tw);
 343        } else {
 344                /* Sorry, if we're out of memory, just CLOSE this
 345                 * socket up.  We've got bigger problems than
 346                 * non-graceful socket closings.
 347                 */
 348                NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
 349        }
 350
 351        tcp_update_metrics(sk);
 352        tcp_done(sk);
 353}
 354
 355void tcp_twsk_destructor(struct sock *sk)
 356{
 357#ifdef CONFIG_TCP_MD5SIG
 358        struct tcp_timewait_sock *twsk = tcp_twsk(sk);
 359
 360        if (twsk->tw_md5_key) {
 361                tcp_free_md5sig_pool();
 362                kfree_rcu(twsk->tw_md5_key, rcu);
 363        }
 364#endif
 365}
 366EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
 367
 368static inline void TCP_ECN_openreq_child(struct tcp_sock *tp,
 369                                         struct request_sock *req)
 370{
 371        tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
 372}
 373
 374/* This is not only more efficient than what we used to do, it eliminates
 375 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
 376 *
 377 * Actually, we could lots of memory writes here. tp of listening
 378 * socket contains all necessary default parameters.
 379 */
 380struct sock *tcp_create_openreq_child(struct sock *sk, struct request_sock *req, struct sk_buff *skb)
 381{
 382        struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
 383
 384        if (newsk != NULL) {
 385                const struct inet_request_sock *ireq = inet_rsk(req);
 386                struct tcp_request_sock *treq = tcp_rsk(req);
 387                struct inet_connection_sock *newicsk = inet_csk(newsk);
 388                struct tcp_sock *newtp = tcp_sk(newsk);
 389                struct tcp_sock *oldtp = tcp_sk(sk);
 390                struct tcp_cookie_values *oldcvp = oldtp->cookie_values;
 391
 392                /* TCP Cookie Transactions require space for the cookie pair,
 393                 * as it differs for each connection.  There is no need to
 394                 * copy any s_data_payload stored at the original socket.
 395                 * Failure will prevent resuming the connection.
 396                 *
 397                 * Presumed copied, in order of appearance:
 398                 *      cookie_in_always, cookie_out_never
 399                 */
 400                if (oldcvp != NULL) {
 401                        struct tcp_cookie_values *newcvp =
 402                                kzalloc(sizeof(*newtp->cookie_values),
 403                                        GFP_ATOMIC);
 404
 405                        if (newcvp != NULL) {
 406                                kref_init(&newcvp->kref);
 407                                newcvp->cookie_desired =
 408                                                oldcvp->cookie_desired;
 409                                newtp->cookie_values = newcvp;
 410                        } else {
 411                                /* Not Yet Implemented */
 412                                newtp->cookie_values = NULL;
 413                        }
 414                }
 415
 416                /* Now setup tcp_sock */
 417                newtp->pred_flags = 0;
 418
 419                newtp->rcv_wup = newtp->copied_seq =
 420                newtp->rcv_nxt = treq->rcv_isn + 1;
 421
 422                newtp->snd_sml = newtp->snd_una =
 423                newtp->snd_nxt = newtp->snd_up =
 424                        treq->snt_isn + 1 + tcp_s_data_size(oldtp);
 425
 426                tcp_prequeue_init(newtp);
 427                INIT_LIST_HEAD(&newtp->tsq_node);
 428
 429                tcp_init_wl(newtp, treq->rcv_isn);
 430
 431                newtp->srtt = 0;
 432                newtp->mdev = TCP_TIMEOUT_INIT;
 433                newicsk->icsk_rto = TCP_TIMEOUT_INIT;
 434
 435                newtp->packets_out = 0;
 436                newtp->retrans_out = 0;
 437                newtp->sacked_out = 0;
 438                newtp->fackets_out = 0;
 439                newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
 440                tcp_enable_early_retrans(newtp);
 441
 442                /* So many TCP implementations out there (incorrectly) count the
 443                 * initial SYN frame in their delayed-ACK and congestion control
 444                 * algorithms that we must have the following bandaid to talk
 445                 * efficiently to them.  -DaveM
 446                 */
 447                newtp->snd_cwnd = TCP_INIT_CWND;
 448                newtp->snd_cwnd_cnt = 0;
 449                newtp->bytes_acked = 0;
 450
 451                newtp->frto_counter = 0;
 452                newtp->frto_highmark = 0;
 453
 454                if (newicsk->icsk_ca_ops != &tcp_init_congestion_ops &&
 455                    !try_module_get(newicsk->icsk_ca_ops->owner))
 456                        newicsk->icsk_ca_ops = &tcp_init_congestion_ops;
 457
 458                tcp_set_ca_state(newsk, TCP_CA_Open);
 459                tcp_init_xmit_timers(newsk);
 460                skb_queue_head_init(&newtp->out_of_order_queue);
 461                newtp->write_seq = newtp->pushed_seq =
 462                        treq->snt_isn + 1 + tcp_s_data_size(oldtp);
 463
 464                newtp->rx_opt.saw_tstamp = 0;
 465
 466                newtp->rx_opt.dsack = 0;
 467                newtp->rx_opt.num_sacks = 0;
 468
 469                newtp->urg_data = 0;
 470
 471                if (sock_flag(newsk, SOCK_KEEPOPEN))
 472                        inet_csk_reset_keepalive_timer(newsk,
 473                                                       keepalive_time_when(newtp));
 474
 475                newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
 476                if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
 477                        if (sysctl_tcp_fack)
 478                                tcp_enable_fack(newtp);
 479                }
 480                newtp->window_clamp = req->window_clamp;
 481                newtp->rcv_ssthresh = req->rcv_wnd;
 482                newtp->rcv_wnd = req->rcv_wnd;
 483                newtp->rx_opt.wscale_ok = ireq->wscale_ok;
 484                if (newtp->rx_opt.wscale_ok) {
 485                        newtp->rx_opt.snd_wscale = ireq->snd_wscale;
 486                        newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
 487                } else {
 488                        newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
 489                        newtp->window_clamp = min(newtp->window_clamp, 65535U);
 490                }
 491                newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
 492                                  newtp->rx_opt.snd_wscale);
 493                newtp->max_window = newtp->snd_wnd;
 494
 495                if (newtp->rx_opt.tstamp_ok) {
 496                        newtp->rx_opt.ts_recent = req->ts_recent;
 497                        newtp->rx_opt.ts_recent_stamp = get_seconds();
 498                        newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
 499                } else {
 500                        newtp->rx_opt.ts_recent_stamp = 0;
 501                        newtp->tcp_header_len = sizeof(struct tcphdr);
 502                }
 503#ifdef CONFIG_TCP_MD5SIG
 504                newtp->md5sig_info = NULL;      /*XXX*/
 505                if (newtp->af_specific->md5_lookup(sk, newsk))
 506                        newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
 507#endif
 508                if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
 509                        newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
 510                newtp->rx_opt.mss_clamp = req->mss;
 511                TCP_ECN_openreq_child(newtp, req);
 512                newtp->fastopen_rsk = NULL;
 513                newtp->syn_data_acked = 0;
 514
 515                TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_PASSIVEOPENS);
 516        }
 517        return newsk;
 518}
 519EXPORT_SYMBOL(tcp_create_openreq_child);
 520
 521/*
 522 * Process an incoming packet for SYN_RECV sockets represented as a
 523 * request_sock. Normally sk is the listener socket but for TFO it
 524 * points to the child socket.
 525 *
 526 * XXX (TFO) - The current impl contains a special check for ack
 527 * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
 528 *
 529 * We don't need to initialize tmp_opt.sack_ok as we don't use the results
 530 */
 531
 532struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
 533                           struct request_sock *req,
 534                           struct request_sock **prev,
 535                           bool fastopen)
 536{
 537        struct tcp_options_received tmp_opt;
 538        const u8 *hash_location;
 539        struct sock *child;
 540        const struct tcphdr *th = tcp_hdr(skb);
 541        __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
 542        bool paws_reject = false;
 543
 544        BUG_ON(fastopen == (sk->sk_state == TCP_LISTEN));
 545
 546        tmp_opt.saw_tstamp = 0;
 547        if (th->doff > (sizeof(struct tcphdr)>>2)) {
 548                tcp_parse_options(skb, &tmp_opt, &hash_location, 0, NULL);
 549
 550                if (tmp_opt.saw_tstamp) {
 551                        tmp_opt.ts_recent = req->ts_recent;
 552                        /* We do not store true stamp, but it is not required,
 553                         * it can be estimated (approximately)
 554                         * from another data.
 555                         */
 556                        tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->retrans);
 557                        paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
 558                }
 559        }
 560
 561        /* Check for pure retransmitted SYN. */
 562        if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
 563            flg == TCP_FLAG_SYN &&
 564            !paws_reject) {
 565                /*
 566                 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
 567                 * this case on figure 6 and figure 8, but formal
 568                 * protocol description says NOTHING.
 569                 * To be more exact, it says that we should send ACK,
 570                 * because this segment (at least, if it has no data)
 571                 * is out of window.
 572                 *
 573                 *  CONCLUSION: RFC793 (even with RFC1122) DOES NOT
 574                 *  describe SYN-RECV state. All the description
 575                 *  is wrong, we cannot believe to it and should
 576                 *  rely only on common sense and implementation
 577                 *  experience.
 578                 *
 579                 * Enforce "SYN-ACK" according to figure 8, figure 6
 580                 * of RFC793, fixed by RFC1122.
 581                 *
 582                 * Note that even if there is new data in the SYN packet
 583                 * they will be thrown away too.
 584                 */
 585                req->rsk_ops->rtx_syn_ack(sk, req, NULL);
 586                return NULL;
 587        }
 588
 589        /* Further reproduces section "SEGMENT ARRIVES"
 590           for state SYN-RECEIVED of RFC793.
 591           It is broken, however, it does not work only
 592           when SYNs are crossed.
 593
 594           You would think that SYN crossing is impossible here, since
 595           we should have a SYN_SENT socket (from connect()) on our end,
 596           but this is not true if the crossed SYNs were sent to both
 597           ends by a malicious third party.  We must defend against this,
 598           and to do that we first verify the ACK (as per RFC793, page
 599           36) and reset if it is invalid.  Is this a true full defense?
 600           To convince ourselves, let us consider a way in which the ACK
 601           test can still pass in this 'malicious crossed SYNs' case.
 602           Malicious sender sends identical SYNs (and thus identical sequence
 603           numbers) to both A and B:
 604
 605                A: gets SYN, seq=7
 606                B: gets SYN, seq=7
 607
 608           By our good fortune, both A and B select the same initial
 609           send sequence number of seven :-)
 610
 611                A: sends SYN|ACK, seq=7, ack_seq=8
 612                B: sends SYN|ACK, seq=7, ack_seq=8
 613
 614           So we are now A eating this SYN|ACK, ACK test passes.  So
 615           does sequence test, SYN is truncated, and thus we consider
 616           it a bare ACK.
 617
 618           If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
 619           bare ACK.  Otherwise, we create an established connection.  Both
 620           ends (listening sockets) accept the new incoming connection and try
 621           to talk to each other. 8-)
 622
 623           Note: This case is both harmless, and rare.  Possibility is about the
 624           same as us discovering intelligent life on another plant tomorrow.
 625
 626           But generally, we should (RFC lies!) to accept ACK
 627           from SYNACK both here and in tcp_rcv_state_process().
 628           tcp_rcv_state_process() does not, hence, we do not too.
 629
 630           Note that the case is absolutely generic:
 631           we cannot optimize anything here without
 632           violating protocol. All the checks must be made
 633           before attempt to create socket.
 634         */
 635
 636        /* RFC793 page 36: "If the connection is in any non-synchronized state ...
 637         *                  and the incoming segment acknowledges something not yet
 638         *                  sent (the segment carries an unacceptable ACK) ...
 639         *                  a reset is sent."
 640         *
 641         * Invalid ACK: reset will be sent by listening socket.
 642         * Note that the ACK validity check for a Fast Open socket is done
 643         * elsewhere and is checked directly against the child socket rather
 644         * than req because user data may have been sent out.
 645         */
 646        if ((flg & TCP_FLAG_ACK) && !fastopen &&
 647            (TCP_SKB_CB(skb)->ack_seq !=
 648             tcp_rsk(req)->snt_isn + 1 + tcp_s_data_size(tcp_sk(sk))))
 649                return sk;
 650
 651        /* Also, it would be not so bad idea to check rcv_tsecr, which
 652         * is essentially ACK extension and too early or too late values
 653         * should cause reset in unsynchronized states.
 654         */
 655
 656        /* RFC793: "first check sequence number". */
 657
 658        if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
 659                                          tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rcv_wnd)) {
 660                /* Out of window: send ACK and drop. */
 661                if (!(flg & TCP_FLAG_RST))
 662                        req->rsk_ops->send_ack(sk, skb, req);
 663                if (paws_reject)
 664                        NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
 665                return NULL;
 666        }
 667
 668        /* In sequence, PAWS is OK. */
 669
 670        if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
 671                req->ts_recent = tmp_opt.rcv_tsval;
 672
 673        if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
 674                /* Truncate SYN, it is out of window starting
 675                   at tcp_rsk(req)->rcv_isn + 1. */
 676                flg &= ~TCP_FLAG_SYN;
 677        }
 678
 679        /* RFC793: "second check the RST bit" and
 680         *         "fourth, check the SYN bit"
 681         */
 682        if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
 683                TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
 684                goto embryonic_reset;
 685        }
 686
 687        /* ACK sequence verified above, just make sure ACK is
 688         * set.  If ACK not set, just silently drop the packet.
 689         *
 690         * XXX (TFO) - if we ever allow "data after SYN", the
 691         * following check needs to be removed.
 692         */
 693        if (!(flg & TCP_FLAG_ACK))
 694                return NULL;
 695
 696        /* Got ACK for our SYNACK, so update baseline for SYNACK RTT sample. */
 697        if (tmp_opt.saw_tstamp && tmp_opt.rcv_tsecr)
 698                tcp_rsk(req)->snt_synack = tmp_opt.rcv_tsecr;
 699        else if (req->retrans) /* don't take RTT sample if retrans && ~TS */
 700                tcp_rsk(req)->snt_synack = 0;
 701
 702        /* For Fast Open no more processing is needed (sk is the
 703         * child socket).
 704         */
 705        if (fastopen)
 706                return sk;
 707
 708        /* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
 709        if (req->retrans < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
 710            TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
 711                inet_rsk(req)->acked = 1;
 712                NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
 713                return NULL;
 714        }
 715
 716        /* OK, ACK is valid, create big socket and
 717         * feed this segment to it. It will repeat all
 718         * the tests. THIS SEGMENT MUST MOVE SOCKET TO
 719         * ESTABLISHED STATE. If it will be dropped after
 720         * socket is created, wait for troubles.
 721         */
 722        child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
 723        if (child == NULL)
 724                goto listen_overflow;
 725
 726        inet_csk_reqsk_queue_unlink(sk, req, prev);
 727        inet_csk_reqsk_queue_removed(sk, req);
 728
 729        inet_csk_reqsk_queue_add(sk, req, child);
 730        return child;
 731
 732listen_overflow:
 733        if (!sysctl_tcp_abort_on_overflow) {
 734                inet_rsk(req)->acked = 1;
 735                return NULL;
 736        }
 737
 738embryonic_reset:
 739        if (!(flg & TCP_FLAG_RST)) {
 740                /* Received a bad SYN pkt - for TFO We try not to reset
 741                 * the local connection unless it's really necessary to
 742                 * avoid becoming vulnerable to outside attack aiming at
 743                 * resetting legit local connections.
 744                 */
 745                req->rsk_ops->send_reset(sk, skb);
 746        } else if (fastopen) { /* received a valid RST pkt */
 747                reqsk_fastopen_remove(sk, req, true);
 748                tcp_reset(sk);
 749        }
 750        if (!fastopen) {
 751                inet_csk_reqsk_queue_drop(sk, req, prev);
 752                NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
 753        }
 754        return NULL;
 755}
 756EXPORT_SYMBOL(tcp_check_req);
 757
 758/*
 759 * Queue segment on the new socket if the new socket is active,
 760 * otherwise we just shortcircuit this and continue with
 761 * the new socket.
 762 *
 763 * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
 764 * when entering. But other states are possible due to a race condition
 765 * where after __inet_lookup_established() fails but before the listener
 766 * locked is obtained, other packets cause the same connection to
 767 * be created.
 768 */
 769
 770int tcp_child_process(struct sock *parent, struct sock *child,
 771                      struct sk_buff *skb)
 772{
 773        int ret = 0;
 774        int state = child->sk_state;
 775
 776        if (!sock_owned_by_user(child)) {
 777                ret = tcp_rcv_state_process(child, skb, tcp_hdr(skb),
 778                                            skb->len);
 779                /* Wakeup parent, send SIGIO */
 780                if (state == TCP_SYN_RECV && child->sk_state != state)
 781                        parent->sk_data_ready(parent, 0);
 782        } else {
 783                /* Alas, it is possible again, because we do lookup
 784                 * in main socket hash table and lock on listening
 785                 * socket does not protect us more.
 786                 */
 787                __sk_add_backlog(child, skb);
 788        }
 789
 790        bh_unlock_sock(child);
 791        sock_put(child);
 792        return ret;
 793}
 794EXPORT_SYMBOL(tcp_child_process);
 795
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.