aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorAdam Langley <agl@imperialviolet.org>2008-07-19 03:04:31 -0400
committerDavid S. Miller <davem@davemloft.net>2008-07-19 03:04:31 -0400
commit33ad798c924b4a1afad3593f2796d465040aadd5 (patch)
tree2a0a868849d6e84591b9d4d917722aee2544b77b
parent49a72dfb8814c2d65bd9f8c9c6daf6395a1ec58d (diff)
tcp: options clean up
This should fix the following bugs: * Connections with MD5 signatures produce invalid packets whenever SACK options are included * MD5 signatures are counted twice in the MSS calculations Behaviour changes: * A SYN with MD5 + SACK + TS elicits a SYNACK with MD5 + SACK This is because we can't fit any SACK blocks in a packet with MD5 + TS options. There was discussion about disabling SACK rather than TS in order to fit in better with old, buggy kernels, but that was deemed to be unnecessary. * SYNs with MD5 don't include a TS option See above. Additionally, it removes a bunch of duplicated logic for calculating options, which should help avoid these sort of issues in the future. Signed-off-by: Adam Langley <agl@imperialviolet.org> Signed-off-by: David S. Miller <davem@davemloft.net>
-rw-r--r--include/net/tcp.h2
-rw-r--r--net/ipv4/tcp_output.c432
2 files changed, 238 insertions, 196 deletions
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 31f5bbfc59bc..8983386356a5 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -50,6 +50,7 @@ extern atomic_t tcp_orphan_count;
50extern void tcp_time_wait(struct sock *sk, int state, int timeo); 50extern void tcp_time_wait(struct sock *sk, int state, int timeo);
51 51
52#define MAX_TCP_HEADER (128 + MAX_HEADER) 52#define MAX_TCP_HEADER (128 + MAX_HEADER)
53#define MAX_TCP_OPTION_SPACE 40
53 54
54/* 55/*
55 * Never offer a window over 32767 without using window scaling. Some 56 * Never offer a window over 32767 without using window scaling. Some
@@ -184,6 +185,7 @@ extern void tcp_time_wait(struct sock *sk, int state, int timeo);
184#define TCPOLEN_SACK_BASE_ALIGNED 4 185#define TCPOLEN_SACK_BASE_ALIGNED 4
185#define TCPOLEN_SACK_PERBLOCK 8 186#define TCPOLEN_SACK_PERBLOCK 8
186#define TCPOLEN_MD5SIG_ALIGNED 20 187#define TCPOLEN_MD5SIG_ALIGNED 20
188#define TCPOLEN_MSS_ALIGNED 4
187 189
188/* Flags in tp->nonagle */ 190/* Flags in tp->nonagle */
189#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ 191#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 958ff486165f..1fa683c0ba9b 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -345,28 +345,82 @@ static void tcp_init_nondata_skb(struct sk_buff *skb, u32 seq, u8 flags)
345 TCP_SKB_CB(skb)->end_seq = seq; 345 TCP_SKB_CB(skb)->end_seq = seq;
346} 346}
347 347
348static void tcp_build_and_update_options(__be32 *ptr, struct tcp_sock *tp, 348#define OPTION_SACK_ADVERTISE (1 << 0)
349 __u32 tstamp, __u8 **md5_hash) 349#define OPTION_TS (1 << 1)
350{ 350#define OPTION_MD5 (1 << 2)
351 if (tp->rx_opt.tstamp_ok) { 351
352struct tcp_out_options {
353 u8 options; /* bit field of OPTION_* */
354 u8 ws; /* window scale, 0 to disable */
355 u8 num_sack_blocks; /* number of SACK blocks to include */
356 u16 mss; /* 0 to disable */
357 __u32 tsval, tsecr; /* need to include OPTION_TS */
358};
359
360static void tcp_options_write(__be32 *ptr, struct tcp_sock *tp,
361 const struct tcp_out_options *opts,
362 __u8 **md5_hash) {
363 if (unlikely(OPTION_MD5 & opts->options)) {
352 *ptr++ = htonl((TCPOPT_NOP << 24) | 364 *ptr++ = htonl((TCPOPT_NOP << 24) |
353 (TCPOPT_NOP << 16) | 365 (TCPOPT_NOP << 16) |
354 (TCPOPT_TIMESTAMP << 8) | 366 (TCPOPT_MD5SIG << 8) |
355 TCPOLEN_TIMESTAMP); 367 TCPOLEN_MD5SIG);
356 *ptr++ = htonl(tstamp); 368 *md5_hash = (__u8 *)ptr;
357 *ptr++ = htonl(tp->rx_opt.ts_recent); 369 ptr += 4;
370 } else {
371 *md5_hash = NULL;
358 } 372 }
359 if (tp->rx_opt.eff_sacks) { 373
360 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks; 374 if (likely(OPTION_TS & opts->options)) {
375 if (unlikely(OPTION_SACK_ADVERTISE & opts->options)) {
376 *ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
377 (TCPOLEN_SACK_PERM << 16) |
378 (TCPOPT_TIMESTAMP << 8) |
379 TCPOLEN_TIMESTAMP);
380 } else {
381 *ptr++ = htonl((TCPOPT_NOP << 24) |
382 (TCPOPT_NOP << 16) |
383 (TCPOPT_TIMESTAMP << 8) |
384 TCPOLEN_TIMESTAMP);
385 }
386 *ptr++ = htonl(opts->tsval);
387 *ptr++ = htonl(opts->tsecr);
388 }
389
390 if (unlikely(opts->mss)) {
391 *ptr++ = htonl((TCPOPT_MSS << 24) |
392 (TCPOLEN_MSS << 16) |
393 opts->mss);
394 }
395
396 if (unlikely(OPTION_SACK_ADVERTISE & opts->options &&
397 !(OPTION_TS & opts->options))) {
398 *ptr++ = htonl((TCPOPT_NOP << 24) |
399 (TCPOPT_NOP << 16) |
400 (TCPOPT_SACK_PERM << 8) |
401 TCPOLEN_SACK_PERM);
402 }
403
404 if (unlikely(opts->ws)) {
405 *ptr++ = htonl((TCPOPT_NOP << 24) |
406 (TCPOPT_WINDOW << 16) |
407 (TCPOLEN_WINDOW << 8) |
408 opts->ws);
409 }
410
411 if (unlikely(opts->num_sack_blocks)) {
412 struct tcp_sack_block *sp = tp->rx_opt.dsack ?
413 tp->duplicate_sack : tp->selective_acks;
361 int this_sack; 414 int this_sack;
362 415
363 *ptr++ = htonl((TCPOPT_NOP << 24) | 416 *ptr++ = htonl((TCPOPT_NOP << 24) |
364 (TCPOPT_NOP << 16) | 417 (TCPOPT_NOP << 16) |
365 (TCPOPT_SACK << 8) | 418 (TCPOPT_SACK << 8) |
366 (TCPOLEN_SACK_BASE + (tp->rx_opt.eff_sacks * 419 (TCPOLEN_SACK_BASE + (opts->num_sack_blocks *
367 TCPOLEN_SACK_PERBLOCK))); 420 TCPOLEN_SACK_PERBLOCK)));
368 421
369 for (this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) { 422 for (this_sack = 0; this_sack < opts->num_sack_blocks;
423 ++this_sack) {
370 *ptr++ = htonl(sp[this_sack].start_seq); 424 *ptr++ = htonl(sp[this_sack].start_seq);
371 *ptr++ = htonl(sp[this_sack].end_seq); 425 *ptr++ = htonl(sp[this_sack].end_seq);
372 } 426 }
@@ -376,81 +430,137 @@ static void tcp_build_and_update_options(__be32 *ptr, struct tcp_sock *tp,
376 tp->rx_opt.eff_sacks--; 430 tp->rx_opt.eff_sacks--;
377 } 431 }
378 } 432 }
433}
434
435static unsigned tcp_syn_options(struct sock *sk, struct sk_buff *skb,
436 struct tcp_out_options *opts,
437 struct tcp_md5sig_key **md5) {
438 struct tcp_sock *tp = tcp_sk(sk);
439 unsigned size = 0;
440
379#ifdef CONFIG_TCP_MD5SIG 441#ifdef CONFIG_TCP_MD5SIG
380 if (md5_hash) { 442 *md5 = tp->af_specific->md5_lookup(sk, sk);
381 *ptr++ = htonl((TCPOPT_NOP << 24) | 443 if (*md5) {
382 (TCPOPT_NOP << 16) | 444 opts->options |= OPTION_MD5;
383 (TCPOPT_MD5SIG << 8) | 445 size += TCPOLEN_MD5SIG_ALIGNED;
384 TCPOLEN_MD5SIG);
385 *md5_hash = (__u8 *)ptr;
386 } 446 }
447#else
448 *md5 = NULL;
387#endif 449#endif
450
451 /* We always get an MSS option. The option bytes which will be seen in
452 * normal data packets should timestamps be used, must be in the MSS
453 * advertised. But we subtract them from tp->mss_cache so that
454 * calculations in tcp_sendmsg are simpler etc. So account for this
455 * fact here if necessary. If we don't do this correctly, as a
456 * receiver we won't recognize data packets as being full sized when we
457 * should, and thus we won't abide by the delayed ACK rules correctly.
458 * SACKs don't matter, we never delay an ACK when we have any of those
459 * going out. */
460 opts->mss = tcp_advertise_mss(sk);
461 size += TCPOLEN_MSS_ALIGNED;
462
463 if (likely(sysctl_tcp_timestamps && *md5 == NULL)) {
464 opts->options |= OPTION_TS;
465 opts->tsval = TCP_SKB_CB(skb)->when;
466 opts->tsecr = tp->rx_opt.ts_recent;
467 size += TCPOLEN_TSTAMP_ALIGNED;
468 }
469 if (likely(sysctl_tcp_window_scaling)) {
470 opts->ws = tp->rx_opt.rcv_wscale;
471 size += TCPOLEN_WSCALE_ALIGNED;
472 }
473 if (likely(sysctl_tcp_sack)) {
474 opts->options |= OPTION_SACK_ADVERTISE;
475 if (unlikely(!OPTION_TS & opts->options))
476 size += TCPOLEN_SACKPERM_ALIGNED;
477 }
478
479 return size;
388} 480}
389 481
390/* Construct a tcp options header for a SYN or SYN_ACK packet. 482static unsigned tcp_synack_options(struct sock *sk,
391 * If this is every changed make sure to change the definition of 483 struct request_sock *req,
392 * MAX_SYN_SIZE to match the new maximum number of options that you 484 unsigned mss, struct sk_buff *skb,
393 * can generate. 485 struct tcp_out_options *opts,
394 * 486 struct tcp_md5sig_key **md5) {
395 * Note - that with the RFC2385 TCP option, we make room for the 487 unsigned size = 0;
396 * 16 byte MD5 hash. This will be filled in later, so the pointer for the 488 struct inet_request_sock *ireq = inet_rsk(req);
397 * location to be filled is passed back up. 489 char doing_ts;
398 */ 490
399static void tcp_syn_build_options(__be32 *ptr, int mss, int ts, int sack,
400 int offer_wscale, int wscale, __u32 tstamp,
401 __u32 ts_recent, __u8 **md5_hash)
402{
403 /* We always get an MSS option.
404 * The option bytes which will be seen in normal data
405 * packets should timestamps be used, must be in the MSS
406 * advertised. But we subtract them from tp->mss_cache so
407 * that calculations in tcp_sendmsg are simpler etc.
408 * So account for this fact here if necessary. If we
409 * don't do this correctly, as a receiver we won't
410 * recognize data packets as being full sized when we
411 * should, and thus we won't abide by the delayed ACK
412 * rules correctly.
413 * SACKs don't matter, we never delay an ACK when we
414 * have any of those going out.
415 */
416 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
417 if (ts) {
418 if (sack)
419 *ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
420 (TCPOLEN_SACK_PERM << 16) |
421 (TCPOPT_TIMESTAMP << 8) |
422 TCPOLEN_TIMESTAMP);
423 else
424 *ptr++ = htonl((TCPOPT_NOP << 24) |
425 (TCPOPT_NOP << 16) |
426 (TCPOPT_TIMESTAMP << 8) |
427 TCPOLEN_TIMESTAMP);
428 *ptr++ = htonl(tstamp); /* TSVAL */
429 *ptr++ = htonl(ts_recent); /* TSECR */
430 } else if (sack)
431 *ptr++ = htonl((TCPOPT_NOP << 24) |
432 (TCPOPT_NOP << 16) |
433 (TCPOPT_SACK_PERM << 8) |
434 TCPOLEN_SACK_PERM);
435 if (offer_wscale)
436 *ptr++ = htonl((TCPOPT_NOP << 24) |
437 (TCPOPT_WINDOW << 16) |
438 (TCPOLEN_WINDOW << 8) |
439 (wscale));
440#ifdef CONFIG_TCP_MD5SIG 491#ifdef CONFIG_TCP_MD5SIG
441 /* 492 *md5 = tcp_rsk(req)->af_specific->md5_lookup(sk, req);
442 * If MD5 is enabled, then we set the option, and include the size 493 if (*md5) {
443 * (always 18). The actual MD5 hash is added just before the 494 opts->options |= OPTION_MD5;
444 * packet is sent. 495 size += TCPOLEN_MD5SIG_ALIGNED;
445 */
446 if (md5_hash) {
447 *ptr++ = htonl((TCPOPT_NOP << 24) |
448 (TCPOPT_NOP << 16) |
449 (TCPOPT_MD5SIG << 8) |
450 TCPOLEN_MD5SIG);
451 *md5_hash = (__u8 *)ptr;
452 } 496 }
497#else
498 *md5 = NULL;
453#endif 499#endif
500
501 /* we can't fit any SACK blocks in a packet with MD5 + TS
502 options. There was discussion about disabling SACK rather than TS in
503 order to fit in better with old, buggy kernels, but that was deemed
504 to be unnecessary. */
505 doing_ts = ireq->tstamp_ok && !(*md5 && ireq->sack_ok);
506
507 opts->mss = mss;
508 size += TCPOLEN_MSS_ALIGNED;
509
510 if (likely(ireq->wscale_ok)) {
511 opts->ws = ireq->rcv_wscale;
512 size += TCPOLEN_WSCALE_ALIGNED;
513 }
514 if (likely(doing_ts)) {
515 opts->options |= OPTION_TS;
516 opts->tsval = TCP_SKB_CB(skb)->when;
517 opts->tsecr = req->ts_recent;
518 size += TCPOLEN_TSTAMP_ALIGNED;
519 }
520 if (likely(ireq->sack_ok)) {
521 opts->options |= OPTION_SACK_ADVERTISE;
522 if (unlikely(!doing_ts))
523 size += TCPOLEN_SACKPERM_ALIGNED;
524 }
525
526 return size;
527}
528
529static unsigned tcp_established_options(struct sock *sk, struct sk_buff *skb,
530 struct tcp_out_options *opts,
531 struct tcp_md5sig_key **md5) {
532 struct tcp_skb_cb *tcb = skb ? TCP_SKB_CB(skb) : NULL;
533 struct tcp_sock *tp = tcp_sk(sk);
534 unsigned size = 0;
535
536#ifdef CONFIG_TCP_MD5SIG
537 *md5 = tp->af_specific->md5_lookup(sk, sk);
538 if (unlikely(*md5)) {
539 opts->options |= OPTION_MD5;
540 size += TCPOLEN_MD5SIG_ALIGNED;
541 }
542#else
543 *md5 = NULL;
544#endif
545
546 if (likely(tp->rx_opt.tstamp_ok)) {
547 opts->options |= OPTION_TS;
548 opts->tsval = tcb ? tcb->when : 0;
549 opts->tsecr = tp->rx_opt.ts_recent;
550 size += TCPOLEN_TSTAMP_ALIGNED;
551 }
552
553 if (unlikely(tp->rx_opt.eff_sacks)) {
554 const unsigned remaining = MAX_TCP_OPTION_SPACE - size;
555 opts->num_sack_blocks =
556 min_t(unsigned, tp->rx_opt.eff_sacks,
557 (remaining - TCPOLEN_SACK_BASE_ALIGNED) /
558 TCPOLEN_SACK_PERBLOCK);
559 size += TCPOLEN_SACK_BASE_ALIGNED +
560 opts->num_sack_blocks * TCPOLEN_SACK_PERBLOCK;
561 }
562
563 return size;
454} 564}
455 565
456/* This routine actually transmits TCP packets queued in by 566/* This routine actually transmits TCP packets queued in by
@@ -471,13 +581,11 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
471 struct inet_sock *inet; 581 struct inet_sock *inet;
472 struct tcp_sock *tp; 582 struct tcp_sock *tp;
473 struct tcp_skb_cb *tcb; 583 struct tcp_skb_cb *tcb;
474 int tcp_header_size; 584 struct tcp_out_options opts;
475#ifdef CONFIG_TCP_MD5SIG 585 unsigned tcp_options_size, tcp_header_size;
476 struct tcp_md5sig_key *md5; 586 struct tcp_md5sig_key *md5;
477 __u8 *md5_hash_location; 587 __u8 *md5_hash_location;
478#endif
479 struct tcphdr *th; 588 struct tcphdr *th;
480 int sysctl_flags;
481 int err; 589 int err;
482 590
483 BUG_ON(!skb || !tcp_skb_pcount(skb)); 591 BUG_ON(!skb || !tcp_skb_pcount(skb));
@@ -500,52 +608,18 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
500 inet = inet_sk(sk); 608 inet = inet_sk(sk);
501 tp = tcp_sk(sk); 609 tp = tcp_sk(sk);
502 tcb = TCP_SKB_CB(skb); 610 tcb = TCP_SKB_CB(skb);
503 tcp_header_size = tp->tcp_header_len; 611 memset(&opts, 0, sizeof(opts));
504
505#define SYSCTL_FLAG_TSTAMPS 0x1
506#define SYSCTL_FLAG_WSCALE 0x2
507#define SYSCTL_FLAG_SACK 0x4
508 612
509 sysctl_flags = 0; 613 if (unlikely(tcb->flags & TCPCB_FLAG_SYN))
510 if (unlikely(tcb->flags & TCPCB_FLAG_SYN)) { 614 tcp_options_size = tcp_syn_options(sk, skb, &opts, &md5);
511 tcp_header_size = sizeof(struct tcphdr) + TCPOLEN_MSS; 615 else
512 if (sysctl_tcp_timestamps) { 616 tcp_options_size = tcp_established_options(sk, skb, &opts,
513 tcp_header_size += TCPOLEN_TSTAMP_ALIGNED; 617 &md5);
514 sysctl_flags |= SYSCTL_FLAG_TSTAMPS; 618 tcp_header_size = tcp_options_size + sizeof(struct tcphdr);
515 }
516 if (sysctl_tcp_window_scaling) {
517 tcp_header_size += TCPOLEN_WSCALE_ALIGNED;
518 sysctl_flags |= SYSCTL_FLAG_WSCALE;
519 }
520 if (sysctl_tcp_sack) {
521 sysctl_flags |= SYSCTL_FLAG_SACK;
522 if (!(sysctl_flags & SYSCTL_FLAG_TSTAMPS))
523 tcp_header_size += TCPOLEN_SACKPERM_ALIGNED;
524 }
525 } else if (unlikely(tp->rx_opt.eff_sacks)) {
526 /* A SACK is 2 pad bytes, a 2 byte header, plus
527 * 2 32-bit sequence numbers for each SACK block.
528 */
529 tcp_header_size += (TCPOLEN_SACK_BASE_ALIGNED +
530 (tp->rx_opt.eff_sacks *
531 TCPOLEN_SACK_PERBLOCK));
532 }
533 619
534 if (tcp_packets_in_flight(tp) == 0) 620 if (tcp_packets_in_flight(tp) == 0)
535 tcp_ca_event(sk, CA_EVENT_TX_START); 621 tcp_ca_event(sk, CA_EVENT_TX_START);
536 622
537#ifdef CONFIG_TCP_MD5SIG
538 /*
539 * Are we doing MD5 on this segment? If so - make
540 * room for it.
541 */
542 md5 = tp->af_specific->md5_lookup(sk, sk);
543 if (md5) {
544 tcp_header_size += TCPOLEN_MD5SIG_ALIGNED;
545 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
546 }
547#endif
548
549 skb_push(skb, tcp_header_size); 623 skb_push(skb, tcp_header_size);
550 skb_reset_transport_header(skb); 624 skb_reset_transport_header(skb);
551 skb_set_owner_w(skb, sk); 625 skb_set_owner_w(skb, sk);
@@ -576,33 +650,14 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
576 th->urg = 1; 650 th->urg = 1;
577 } 651 }
578 652
579 if (unlikely(tcb->flags & TCPCB_FLAG_SYN)) { 653 tcp_options_write((__be32 *)(th + 1), tp, &opts, &md5_hash_location);
580 tcp_syn_build_options((__be32 *)(th + 1), 654 if (likely((tcb->flags & TCPCB_FLAG_SYN) == 0))
581 tcp_advertise_mss(sk),
582 (sysctl_flags & SYSCTL_FLAG_TSTAMPS),
583 (sysctl_flags & SYSCTL_FLAG_SACK),
584 (sysctl_flags & SYSCTL_FLAG_WSCALE),
585 tp->rx_opt.rcv_wscale,
586 tcb->when,
587 tp->rx_opt.ts_recent,
588
589#ifdef CONFIG_TCP_MD5SIG
590 md5 ? &md5_hash_location :
591#endif
592 NULL);
593 } else {
594 tcp_build_and_update_options((__be32 *)(th + 1),
595 tp, tcb->when,
596#ifdef CONFIG_TCP_MD5SIG
597 md5 ? &md5_hash_location :
598#endif
599 NULL);
600 TCP_ECN_send(sk, skb, tcp_header_size); 655 TCP_ECN_send(sk, skb, tcp_header_size);
601 }
602 656
603#ifdef CONFIG_TCP_MD5SIG 657#ifdef CONFIG_TCP_MD5SIG
604 /* Calculate the MD5 hash, as we have all we need now */ 658 /* Calculate the MD5 hash, as we have all we need now */
605 if (md5) { 659 if (md5) {
660 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
606 tp->af_specific->calc_md5_hash(md5_hash_location, 661 tp->af_specific->calc_md5_hash(md5_hash_location,
607 md5, sk, NULL, skb); 662 md5, sk, NULL, skb);
608 } 663 }
@@ -626,10 +681,6 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
626 tcp_enter_cwr(sk, 1); 681 tcp_enter_cwr(sk, 1);
627 682
628 return net_xmit_eval(err); 683 return net_xmit_eval(err);
629
630#undef SYSCTL_FLAG_TSTAMPS
631#undef SYSCTL_FLAG_WSCALE
632#undef SYSCTL_FLAG_SACK
633} 684}
634 685
635/* This routine just queue's the buffer 686/* This routine just queue's the buffer
@@ -970,6 +1021,9 @@ unsigned int tcp_current_mss(struct sock *sk, int large_allowed)
970 u32 mss_now; 1021 u32 mss_now;
971 u16 xmit_size_goal; 1022 u16 xmit_size_goal;
972 int doing_tso = 0; 1023 int doing_tso = 0;
1024 unsigned header_len;
1025 struct tcp_out_options opts;
1026 struct tcp_md5sig_key *md5;
973 1027
974 mss_now = tp->mss_cache; 1028 mss_now = tp->mss_cache;
975 1029
@@ -982,14 +1036,16 @@ unsigned int tcp_current_mss(struct sock *sk, int large_allowed)
982 mss_now = tcp_sync_mss(sk, mtu); 1036 mss_now = tcp_sync_mss(sk, mtu);
983 } 1037 }
984 1038
985 if (tp->rx_opt.eff_sacks) 1039 header_len = tcp_established_options(sk, NULL, &opts, &md5) +
986 mss_now -= (TCPOLEN_SACK_BASE_ALIGNED + 1040 sizeof(struct tcphdr);
987 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)); 1041 /* The mss_cache is sized based on tp->tcp_header_len, which assumes
988 1042 * some common options. If this is an odd packet (because we have SACK
989#ifdef CONFIG_TCP_MD5SIG 1043 * blocks etc) then our calculated header_len will be different, and
990 if (tp->af_specific->md5_lookup(sk, sk)) 1044 * we have to adjust mss_now correspondingly */
991 mss_now -= TCPOLEN_MD5SIG_ALIGNED; 1045 if (header_len != tp->tcp_header_len) {
992#endif 1046 int delta = (int) header_len - tp->tcp_header_len;
1047 mss_now -= delta;
1048 }
993 1049
994 xmit_size_goal = mss_now; 1050 xmit_size_goal = mss_now;
995 1051
@@ -2179,11 +2235,10 @@ struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
2179 struct tcp_sock *tp = tcp_sk(sk); 2235 struct tcp_sock *tp = tcp_sk(sk);
2180 struct tcphdr *th; 2236 struct tcphdr *th;
2181 int tcp_header_size; 2237 int tcp_header_size;
2238 struct tcp_out_options opts;
2182 struct sk_buff *skb; 2239 struct sk_buff *skb;
2183#ifdef CONFIG_TCP_MD5SIG
2184 struct tcp_md5sig_key *md5; 2240 struct tcp_md5sig_key *md5;
2185 __u8 *md5_hash_location; 2241 __u8 *md5_hash_location;
2186#endif
2187 2242
2188 skb = sock_wmalloc(sk, MAX_TCP_HEADER + 15, 1, GFP_ATOMIC); 2243 skb = sock_wmalloc(sk, MAX_TCP_HEADER + 15, 1, GFP_ATOMIC);
2189 if (skb == NULL) 2244 if (skb == NULL)
@@ -2194,18 +2249,27 @@ struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
2194 2249
2195 skb->dst = dst_clone(dst); 2250 skb->dst = dst_clone(dst);
2196 2251
2197 tcp_header_size = (sizeof(struct tcphdr) + TCPOLEN_MSS + 2252 if (req->rcv_wnd == 0) { /* ignored for retransmitted syns */
2198 (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0) + 2253 __u8 rcv_wscale;
2199 (ireq->wscale_ok ? TCPOLEN_WSCALE_ALIGNED : 0) + 2254 /* Set this up on the first call only */
2200 /* SACK_PERM is in the place of NOP NOP of TS */ 2255 req->window_clamp = tp->window_clamp ? : dst_metric(dst, RTAX_WINDOW);
2201 ((ireq->sack_ok && !ireq->tstamp_ok) ? TCPOLEN_SACKPERM_ALIGNED : 0)); 2256 /* tcp_full_space because it is guaranteed to be the first packet */
2257 tcp_select_initial_window(tcp_full_space(sk),
2258 dst_metric(dst, RTAX_ADVMSS) - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
2259 &req->rcv_wnd,
2260 &req->window_clamp,
2261 ireq->wscale_ok,
2262 &rcv_wscale);
2263 ireq->rcv_wscale = rcv_wscale;
2264 }
2265
2266 memset(&opts, 0, sizeof(opts));
2267 TCP_SKB_CB(skb)->when = tcp_time_stamp;
2268 tcp_header_size = tcp_synack_options(sk, req,
2269 dst_metric(dst, RTAX_ADVMSS),
2270 skb, &opts, &md5) +
2271 sizeof(struct tcphdr);
2202 2272
2203#ifdef CONFIG_TCP_MD5SIG
2204 /* Are we doing MD5 on this segment? If so - make room for it */
2205 md5 = tcp_rsk(req)->af_specific->md5_lookup(sk, req);
2206 if (md5)
2207 tcp_header_size += TCPOLEN_MD5SIG_ALIGNED;
2208#endif
2209 skb_push(skb, tcp_header_size); 2273 skb_push(skb, tcp_header_size);
2210 skb_reset_transport_header(skb); 2274 skb_reset_transport_header(skb);
2211 2275
@@ -2223,19 +2287,6 @@ struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
2223 TCPCB_FLAG_SYN | TCPCB_FLAG_ACK); 2287 TCPCB_FLAG_SYN | TCPCB_FLAG_ACK);
2224 th->seq = htonl(TCP_SKB_CB(skb)->seq); 2288 th->seq = htonl(TCP_SKB_CB(skb)->seq);
2225 th->ack_seq = htonl(tcp_rsk(req)->rcv_isn + 1); 2289 th->ack_seq = htonl(tcp_rsk(req)->rcv_isn + 1);
2226 if (req->rcv_wnd == 0) { /* ignored for retransmitted syns */
2227 __u8 rcv_wscale;
2228 /* Set this up on the first call only */
2229 req->window_clamp = tp->window_clamp ? : dst_metric(dst, RTAX_WINDOW);
2230 /* tcp_full_space because it is guaranteed to be the first packet */
2231 tcp_select_initial_window(tcp_full_space(sk),
2232 dst_metric(dst, RTAX_ADVMSS) - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
2233 &req->rcv_wnd,
2234 &req->window_clamp,
2235 ireq->wscale_ok,
2236 &rcv_wscale);
2237 ireq->rcv_wscale = rcv_wscale;
2238 }
2239 2290
2240 /* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */ 2291 /* RFC1323: The window in SYN & SYN/ACK segments is never scaled. */
2241 th->window = htons(min(req->rcv_wnd, 65535U)); 2292 th->window = htons(min(req->rcv_wnd, 65535U));
@@ -2244,18 +2295,7 @@ struct sk_buff *tcp_make_synack(struct sock *sk, struct dst_entry *dst,
2244 TCP_SKB_CB(skb)->when = cookie_init_timestamp(req); 2295 TCP_SKB_CB(skb)->when = cookie_init_timestamp(req);
2245 else 2296 else
2246#endif 2297#endif
2247 TCP_SKB_CB(skb)->when = tcp_time_stamp; 2298 tcp_options_write((__be32 *)(th + 1), tp, &opts, &md5_hash_location);
2248 tcp_syn_build_options((__be32 *)(th + 1), dst_metric(dst, RTAX_ADVMSS), ireq->tstamp_ok,
2249 ireq->sack_ok, ireq->wscale_ok, ireq->rcv_wscale,
2250 TCP_SKB_CB(skb)->when,
2251 req->ts_recent,
2252 (
2253#ifdef CONFIG_TCP_MD5SIG
2254 md5 ? &md5_hash_location :
2255#endif
2256 NULL)
2257 );
2258
2259 th->doff = (tcp_header_size >> 2); 2299 th->doff = (tcp_header_size >> 2);
2260 TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS); 2300 TCP_INC_STATS(sock_net(sk), TCP_MIB_OUTSEGS);
2261 2301