Page MenuHomeFreeBSD
Differential D15758 Diff 43742 head/sys/netinet/tcp_stacks/rack.c

Changeset View

Standalone View

View Options

head/sys/netinet/tcp_stacks/rack.c

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 2,077 Lines • ▼ Show 20 Linesrack_timer_start(struct tcpcb *tp, struct tcp_rack *rack, uint32_t cts)
if (rack->t_timers_stopped) { if (rack->t_timers_stopped) {
/* All timers have been stopped none are to run */ /* All timers have been stopped none are to run */
return (0); return (0);
} }
if (rack->rc_in_persist) { if (rack->rc_in_persist) {
/* We can't start any timer in persists */ /* We can't start any timer in persists */
return (rack_get_persists_timer_val(tp, rack)); return (rack_get_persists_timer_val(tp, rack));
} }
if (tp->t_state < TCPS_ESTABLISHED)
goto activate_rxt;
rsm = TAILQ_FIRST(&rack->r_ctl.rc_tmap); rsm = TAILQ_FIRST(&rack->r_ctl.rc_tmap);
if (rsm == NULL) { if (rsm == NULL) {
/* Nothing on the send map */ /* Nothing on the send map */
activate_rxt: activate_rxt:
if (SEQ_LT(tp->snd_una, tp->snd_max) || sbavail(&(tp->t_inpcb->inp_socket->so_snd))) { if (SEQ_LT(tp->snd_una, tp->snd_max) || sbavail(&(tp->t_inpcb->inp_socket->so_snd))) {
rack->r_ctl.rc_hpts_flags |= PACE_TMR_RXT; rack->r_ctl.rc_hpts_flags |= PACE_TMR_RXT;
to = TICKS_2_MSEC(tp->t_rxtcur); to = TICKS_2_MSEC(tp->t_rxtcur);
if (to == 0) if (to == 0)
▲ Show 20 Lines • Show All 1,286 Lines • ▼ Show 20 Lines#endif
if (th_flags & TH_FIN) { if (th_flags & TH_FIN) {
rsm->r_flags = RACK_HAS_FIN; rsm->r_flags = RACK_HAS_FIN;
} else { } else {
rsm->r_flags = 0; rsm->r_flags = 0;
} }
rsm->r_tim_lastsent[0] = ts; rsm->r_tim_lastsent[0] = ts;
rsm->r_rtr_cnt = 1; rsm->r_rtr_cnt = 1;
rsm->r_rtr_bytes = 0; rsm->r_rtr_bytes = 0;
if (th_flags & TH_SYN) {
/* The data space is one beyond snd_una */
rsm->r_start = seq_out + 1;
rsm->r_end = rsm->r_start + (len - 1);
} else {
/* Normal case */
rsm->r_start = seq_out; rsm->r_start = seq_out;
rsm->r_end = rsm->r_start + len; rsm->r_end = rsm->r_start + len;
}
rsm->r_sndcnt = 0; rsm->r_sndcnt = 0;
TAILQ_INSERT_TAIL(&rack->r_ctl.rc_map, rsm, r_next); TAILQ_INSERT_TAIL(&rack->r_ctl.rc_map, rsm, r_next);
TAILQ_INSERT_TAIL(&rack->r_ctl.rc_tmap, rsm, r_tnext); TAILQ_INSERT_TAIL(&rack->r_ctl.rc_tmap, rsm, r_tnext);
rsm->r_in_tmap = 1; rsm->r_in_tmap = 1;
return; return;
} }
/* /*
* If we reach here its a retransmission and we need to find it. * If we reach here its a retransmission and we need to find it.
▲ Show 20 Lines • Show All 1,254 Lines • ▼ Show 20 Linesrack_process_data(struct mbuf *m, struct tcphdr *th, struct socket *so,
struct tcpcb *tp, int32_t drop_hdrlen, int32_t tlen, struct tcpcb *tp, int32_t drop_hdrlen, int32_t tlen,
int32_t * ti_locked, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt) int32_t * ti_locked, uint32_t tiwin, int32_t thflags, int32_t nxt_pkt)
{ {
/* /*
* Update window information. Don't look at window if no ACK: TAC's * Update window information. Don't look at window if no ACK: TAC's
* send garbage on first SYN. * send garbage on first SYN.
*/ */
int32_t nsegs; int32_t nsegs;
#ifdef TCP_RFC7413
int32_t tfo_syn; int32_t tfo_syn;
#else
#define tfo_syn (FALSE)
#endif
struct tcp_rack *rack; struct tcp_rack *rack;
rack = (struct tcp_rack *)tp->t_fb_ptr; rack = (struct tcp_rack *)tp->t_fb_ptr;
INP_WLOCK_ASSERT(tp->t_inpcb); INP_WLOCK_ASSERT(tp->t_inpcb);
nsegs = max(1, m->m_pkthdr.lro_nsegs); nsegs = max(1, m->m_pkthdr.lro_nsegs);
if ((thflags & TH_ACK) && if ((thflags & TH_ACK) &&
(SEQ_LT(tp->snd_wl1, th->th_seq) || (SEQ_LT(tp->snd_wl1, th->th_seq) ||
▲ Show 20 Lines • Show All 89 Lines • ▼ Show 20 Linesdodata: /* XXX */
/* /*
* Process the segment text, merging it into the TCP sequencing * Process the segment text, merging it into the TCP sequencing
* queue, and arranging for acknowledgment of receipt if necessary. * queue, and arranging for acknowledgment of receipt if necessary.
* This process logically involves adjusting tp->rcv_wnd as data is * This process logically involves adjusting tp->rcv_wnd as data is
* presented to the user (this happens in tcp_usrreq.c, case * presented to the user (this happens in tcp_usrreq.c, case
* PRU_RCVD). If a FIN has already been received on this connection * PRU_RCVD). If a FIN has already been received on this connection
* then we just ignore the text. * then we just ignore the text.
*/ */
#ifdef TCP_RFC7413
tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
(tp->t_flags & TF_FASTOPEN)); IS_FASTOPEN(tp->t_flags));
#endif
if ((tlen || (thflags & TH_FIN) || tfo_syn) && if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
TCPS_HAVERCVDFIN(tp->t_state) == 0) { TCPS_HAVERCVDFIN(tp->t_state) == 0) {
tcp_seq save_start = th->th_seq; tcp_seq save_start = th->th_seq;
m_adj(m, drop_hdrlen); /* delayed header drop */ m_adj(m, drop_hdrlen); /* delayed header drop */
/* /*
* Insert segment which includes th into TCP reassembly * Insert segment which includes th into TCP reassembly
* queue with control block tp. Set thflags to whether * queue with control block tp. Set thflags to whether
▲ Show 20 Lines • Show All 450 Lines • ▼ Show 20 Linesrack_do_syn_sent(struct mbuf *m, struct tcphdr *th, struct socket *so,
} }
if (!(thflags & TH_SYN)) { if (!(thflags & TH_SYN)) {
rack_do_drop(m, tp, ti_locked); rack_do_drop(m, tp, ti_locked);
return (1); return (1);
} }
tp->irs = th->th_seq; tp->irs = th->th_seq;
tcp_rcvseqinit(tp); tcp_rcvseqinit(tp);
if (thflags & TH_ACK) { if (thflags & TH_ACK) {
int tfo_partial = 0;
TCPSTAT_INC(tcps_connects); TCPSTAT_INC(tcps_connects);
soisconnected(so); soisconnected(so);
#ifdef MAC #ifdef MAC
mac_socketpeer_set_from_mbuf(m, so); mac_socketpeer_set_from_mbuf(m, so);
#endif #endif
/* Do window scaling on this connection? */ /* Do window scaling on this connection? */
if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
(TF_RCVD_SCALE | TF_REQ_SCALE)) { (TF_RCVD_SCALE | TF_REQ_SCALE)) {
tp->rcv_scale = tp->request_r_scale; tp->rcv_scale = tp->request_r_scale;
} }
tp->rcv_adv += min(tp->rcv_wnd, tp->rcv_adv += min(tp->rcv_wnd,
TCP_MAXWIN << tp->rcv_scale); TCP_MAXWIN << tp->rcv_scale);
/* /*
* If not all the data that was sent in the TFO SYN
* has been acked, resend the remainder right away.
*/
if (IS_FASTOPEN(tp->t_flags) &&
(tp->snd_una != tp->snd_max)) {
tp->snd_nxt = th->th_ack;
tfo_partial = 1;
}
/*
* If there's data, delay ACK; if there's also a FIN ACKNOW * If there's data, delay ACK; if there's also a FIN ACKNOW
* will be turned on later. * will be turned on later.
*/ */
if (DELAY_ACK(tp, tlen) && tlen != 0) { if (DELAY_ACK(tp, tlen) && tlen != 0 && (tfo_partial == 0)) {
rack_timer_cancel(tp, (struct tcp_rack *)tp->t_fb_ptr, rack_timer_cancel(tp, (struct tcp_rack *)tp->t_fb_ptr,
((struct tcp_rack *)tp->t_fb_ptr)->r_ctl.rc_rcvtime, __LINE__); ((struct tcp_rack *)tp->t_fb_ptr)->r_ctl.rc_rcvtime, __LINE__);
tp->t_flags |= TF_DELACK; tp->t_flags |= TF_DELACK;
} else { } else {
((struct tcp_rack *)tp->t_fb_ptr)->r_wanted_output++; ((struct tcp_rack *)tp->t_fb_ptr)->r_wanted_output++;
tp->t_flags |= TF_ACKNOW; tp->t_flags |= TF_ACKNOW;
} }
if ((thflags & TH_ECE) && V_tcp_do_ecn) { if ((thflags & TH_ECE) && V_tcp_do_ecn) {
tp->t_flags |= TF_ECN_PERMIT; tp->t_flags |= TF_ECN_PERMIT;
TCPSTAT_INC(tcps_ecn_shs); TCPSTAT_INC(tcps_ecn_shs);
} }
if (SEQ_GT(th->th_ack, tp->snd_una)) {
/* /*
* We advance snd_una for the
* fast open case. If th_ack is
* acknowledging data beyond
* snd_una we can't just call
* ack-processing since the
* data stream in our send-map
* will start at snd_una + 1 (one
* beyond the SYN). If its just
* equal we don't need to do that
* and there is no send_map.
*/
tp->snd_una++;
}
/*
* Received <SYN,ACK> in SYN_SENT[*] state. Transitions: * Received <SYN,ACK> in SYN_SENT[*] state. Transitions:
* SYN_SENT --> ESTABLISHED SYN_SENT* --> FIN_WAIT_1 * SYN_SENT --> ESTABLISHED SYN_SENT* --> FIN_WAIT_1
*/ */
tp->t_starttime = ticks; tp->t_starttime = ticks;
if (tp->t_flags & TF_NEEDFIN) { if (tp->t_flags & TF_NEEDFIN) {
tcp_state_change(tp, TCPS_FIN_WAIT_1); tcp_state_change(tp, TCPS_FIN_WAIT_1);
tp->t_flags &= ~TF_NEEDFIN; tp->t_flags &= ~TF_NEEDFIN;
thflags &= ~TH_SYN; thflags &= ~TH_SYN;
▲ Show 20 Lines • Show All 92 Lines • ▼ Show 20 Linesrack_do_syn_recv(struct mbuf *m, struct tcphdr *th, struct socket *so,
rack_calc_rwin(so, tp); rack_calc_rwin(so, tp);
if ((thflags & TH_ACK) && if ((thflags & TH_ACK) &&
(SEQ_LEQ(th->th_ack, tp->snd_una) || (SEQ_LEQ(th->th_ack, tp->snd_una) ||
SEQ_GT(th->th_ack, tp->snd_max))) { SEQ_GT(th->th_ack, tp->snd_max))) {
rack_do_dropwithreset(m, tp, th, ti_locked, BANDLIM_RST_OPENPORT, tlen); rack_do_dropwithreset(m, tp, th, ti_locked, BANDLIM_RST_OPENPORT, tlen);
return (1); return (1);
} }
#ifdef TCP_RFC7413 if (IS_FASTOPEN(tp->t_flags)) {
if (tp->t_flags & TF_FASTOPEN) {
/* /*
* When a TFO connection is in SYN_RECEIVED, the only valid * When a TFO connection is in SYN_RECEIVED, the
* packets are the initial SYN, a retransmit/copy of the * only valid packets are the initial SYN, a
* initial SYN (possibly with a subset of the original * retransmit/copy of the initial SYN (possibly with
* data), a valid ACK, a FIN, or a RST. * a subset of the original data), a valid ACK, a
* FIN, or a RST.
*/ */
if ((thflags & (TH_SYN | TH_ACK)) == (TH_SYN | TH_ACK)) { if ((thflags & (TH_SYN | TH_ACK)) == (TH_SYN | TH_ACK)) {
rack_do_dropwithreset(m, tp, th, ti_locked, BANDLIM_RST_OPENPORT, tlen); rack_do_dropwithreset(m, tp, th, ti_locked, BANDLIM_RST_OPENPORT, tlen);
return (1); return (1);
} else if (thflags & TH_SYN) { } else if (thflags & TH_SYN) {
/* non-initial SYN is ignored */ /* non-initial SYN is ignored */
struct tcp_rack *rack; struct tcp_rack *rack;
rack = (struct tcp_rack *)tp->t_fb_ptr; rack = (struct tcp_rack *)tp->t_fb_ptr;
if ((rack->r_ctl.rc_hpts_flags & PACE_TMR_RXT) || if ((rack->r_ctl.rc_hpts_flags & PACE_TMR_RXT) ||
(rack->r_ctl.rc_hpts_flags & PACE_TMR_TLP) || (rack->r_ctl.rc_hpts_flags & PACE_TMR_TLP) ||
(rack->r_ctl.rc_hpts_flags & PACE_TMR_RACK)) { (rack->r_ctl.rc_hpts_flags & PACE_TMR_RACK)) {
rack_do_drop(m, NULL, ti_locked); rack_do_drop(m, NULL, ti_locked);
return (0); return (0);
} }
} else if (!(thflags & (TH_ACK | TH_FIN | TH_RST))) { } else if (!(thflags & (TH_ACK | TH_FIN | TH_RST))) {
rack_do_drop(m, NULL, ti_locked); rack_do_drop(m, NULL, ti_locked);
return (0); return (0);
} }
} }
#endif
if (thflags & TH_RST) if (thflags & TH_RST)
return (rack_process_rst(m, th, so, tp, ti_locked)); return (rack_process_rst(m, th, so, tp, ti_locked));
/* /*
* RFC5961 Section 4.2 Send challenge ACK for any SYN in * RFC5961 Section 4.2 Send challenge ACK for any SYN in
* synchronized state. * synchronized state.
*/ */
if (thflags & TH_SYN) { if (thflags & TH_SYN) {
rack_challenge_ack(m, th, tp, ti_locked, &ret_val); rack_challenge_ack(m, th, tp, ti_locked, &ret_val);
▲ Show 20 Lines • Show All 44 Lines • ▼ Show 20 Lines if ((to->to_flags & TOF_TS) != 0 &&
tp->ts_recent = to->to_tsval; tp->ts_recent = to->to_tsval;
} }
/* /*
* If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN flag * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN flag
* is on (half-synchronized state), then queue data for later * is on (half-synchronized state), then queue data for later
* processing; else drop segment and return. * processing; else drop segment and return.
*/ */
if ((thflags & TH_ACK) == 0) { if ((thflags & TH_ACK) == 0) {
#ifdef TCP_RFC7413 if (IS_FASTOPEN(tp->t_flags)) {
if (tp->t_flags & TF_FASTOPEN) {
tp->snd_wnd = tiwin; tp->snd_wnd = tiwin;
cc_conn_init(tp); cc_conn_init(tp);
} }
#endif
return (rack_process_data(m, th, so, tp, drop_hdrlen, tlen, return (rack_process_data(m, th, so, tp, drop_hdrlen, tlen,
ti_locked, tiwin, thflags, nxt_pkt)); ti_locked, tiwin, thflags, nxt_pkt));
} }
TCPSTAT_INC(tcps_connects); TCPSTAT_INC(tcps_connects);
soisconnected(so); soisconnected(so);
/* Do window scaling? */ /* Do window scaling? */
if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
(TF_RCVD_SCALE | TF_REQ_SCALE)) { (TF_RCVD_SCALE | TF_REQ_SCALE)) {
tp->rcv_scale = tp->request_r_scale; tp->rcv_scale = tp->request_r_scale;
tp->snd_wnd = tiwin; tp->snd_wnd = tiwin;
} }
/* /*
* Make transitions: SYN-RECEIVED -> ESTABLISHED SYN-RECEIVED* -> * Make transitions: SYN-RECEIVED -> ESTABLISHED SYN-RECEIVED* ->
* FIN-WAIT-1 * FIN-WAIT-1
*/ */
tp->t_starttime = ticks; tp->t_starttime = ticks;
if (tp->t_flags & TF_NEEDFIN) { if (tp->t_flags & TF_NEEDFIN) {
tcp_state_change(tp, TCPS_FIN_WAIT_1); tcp_state_change(tp, TCPS_FIN_WAIT_1);
tp->t_flags &= ~TF_NEEDFIN; tp->t_flags &= ~TF_NEEDFIN;
} else { } else {
tcp_state_change(tp, TCPS_ESTABLISHED); tcp_state_change(tp, TCPS_ESTABLISHED);
TCP_PROBE5(accept__established, NULL, tp, TCP_PROBE5(accept__established, NULL, tp,
mtod(m, const char *), tp, th); mtod(m, const char *), tp, th);
#ifdef TCP_RFC7413 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) {
if (tp->t_tfo_pending) {
tcp_fastopen_decrement_counter(tp->t_tfo_pending); tcp_fastopen_decrement_counter(tp->t_tfo_pending);
tp->t_tfo_pending = NULL; tp->t_tfo_pending = NULL;
/* /*
* Account for the ACK of our SYN prior to regular * Account for the ACK of our SYN prior to regular
* ACK processing below. * ACK processing below.
*/ */
tp->snd_una++; tp->snd_una++;
} }
/* /*
* TFO connections call cc_conn_init() during SYN * TFO connections call cc_conn_init() during SYN
* processing. Calling it again here for such connections * processing. Calling it again here for such connections
* is not harmless as it would undo the snd_cwnd reduction * is not harmless as it would undo the snd_cwnd reduction
* that occurs when a TFO SYN|ACK is retransmitted. * that occurs when a TFO SYN|ACK is retransmitted.
*/ */
if (!(tp->t_flags & TF_FASTOPEN)) if (!IS_FASTOPEN(tp->t_flags))
#endif
cc_conn_init(tp); cc_conn_init(tp);
} }
/* /*
* If segment contains data or ACK, will call tcp_reass() later; if * If segment contains data or ACK, will call tcp_reass() later; if
* not, do so now to pass queued data to user. * not, do so now to pass queued data to user.
*/ */
if (tlen == 0 && (thflags & TH_FIN) == 0) if (tlen == 0 && (thflags & TH_FIN) == 0)
(void)tcp_reass(tp, (struct tcphdr *)0, 0, (void)tcp_reass(tp, (struct tcphdr *)0, 0,
▲ Show 20 Lines • Show All 1,399 Lines • ▼ Show 20 Linesrack_output(struct tcpcb *tp)
struct ip *ip = NULL; struct ip *ip = NULL;
#ifdef TCPDEBUG #ifdef TCPDEBUG
struct ipovly *ipov = NULL; struct ipovly *ipov = NULL;
#endif #endif
struct udphdr *udp = NULL; struct udphdr *udp = NULL;
struct tcp_rack *rack; struct tcp_rack *rack;
struct tcphdr *th; struct tcphdr *th;
uint8_t pass = 0; uint8_t pass = 0;
uint8_t wanted_cookie = 0;
u_char opt[TCP_MAXOLEN]; u_char opt[TCP_MAXOLEN];
unsigned ipoptlen, optlen, hdrlen, ulen=0; unsigned ipoptlen, optlen, hdrlen, ulen=0;
uint32_t rack_seq; uint32_t rack_seq;
#if defined(IPSEC) || defined(IPSEC_SUPPORT) #if defined(IPSEC) || defined(IPSEC_SUPPORT)
unsigned ipsec_optlen = 0; unsigned ipsec_optlen = 0;
#endif #endif
Show All 25 Lines#endif
do_a_prefetch = 1; do_a_prefetch = 1;
INP_WLOCK_ASSERT(inp); INP_WLOCK_ASSERT(inp);
#ifdef TCP_OFFLOAD #ifdef TCP_OFFLOAD
if (tp->t_flags & TF_TOE) if (tp->t_flags & TF_TOE)
return (tcp_offload_output(tp)); return (tcp_offload_output(tp));
#endif #endif
#ifdef TCP_RFC7413
/* /*
* For TFO connections in SYN_RECEIVED, only allow the initial * For TFO connections in SYN_RECEIVED, only allow the initial
* SYN|ACK and those sent by the retransmit timer. * SYN|ACK and those sent by the retransmit timer.
*/ */
if ((tp->t_flags & TF_FASTOPEN) && if (IS_FASTOPEN(tp->t_flags) &&
(tp->t_state == TCPS_SYN_RECEIVED) && (tp->t_state == TCPS_SYN_RECEIVED) &&
SEQ_GT(tp->snd_max, tp->snd_una) && /* inital SYN|ACK sent */ SEQ_GT(tp->snd_max, tp->snd_una) && /* initial SYN|ACK sent */
(tp->snd_nxt != tp->snd_una)) /* not a retransmit */ (rack->r_ctl.rc_resend == NULL)) /* not a retransmit */
return (0); return (0);
#endif
#ifdef INET6 #ifdef INET6
if (rack->r_state) { if (rack->r_state) {
/* Use the cache line loaded if possible */ /* Use the cache line loaded if possible */
isipv6 = rack->r_is_v6; isipv6 = rack->r_is_v6;
} else { } else {
isipv6 = (inp->inp_vflag & INP_IPV6) != 0; isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
} }
#endif #endif
▲ Show 20 Lines • Show All 273 Lines • ▼ Show 20 Lines#endif
* *
* If sack_rxmit is true we are retransmitting from the scoreboard * If sack_rxmit is true we are retransmitting from the scoreboard
* in which case len is already set. * in which case len is already set.
*/ */
if (sack_rxmit == 0) { if (sack_rxmit == 0) {
uint32_t avail; uint32_t avail;
avail = sbavail(sb); avail = sbavail(sb);
if (SEQ_GT(tp->snd_nxt, tp->snd_una)) if (SEQ_GT(tp->snd_nxt, tp->snd_una) && avail)
sb_offset = tp->snd_nxt - tp->snd_una; sb_offset = tp->snd_nxt - tp->snd_una;
else else
sb_offset = 0; sb_offset = 0;
if (IN_RECOVERY(tp->t_flags) == 0) { if (IN_RECOVERY(tp->t_flags) == 0) {
if (rack->r_ctl.rc_tlp_new_data) { if (rack->r_ctl.rc_tlp_new_data) {
/* TLP is forcing out new data */ /* TLP is forcing out new data */
if (rack->r_ctl.rc_tlp_new_data > (uint32_t) (avail - sb_offset)) { if (rack->r_ctl.rc_tlp_new_data > (uint32_t) (avail - sb_offset)) {
rack->r_ctl.rc_tlp_new_data = (uint32_t) (avail - sb_offset); rack->r_ctl.rc_tlp_new_data = (uint32_t) (avail - sb_offset);
▲ Show 20 Lines • Show All 75 Lines • ▼ Show 20 Lines if (prefetch_so_done == 0) {
kern_prefetch(so, &prefetch_so_done); kern_prefetch(so, &prefetch_so_done);
prefetch_so_done = 1; prefetch_so_done = 1;
} }
/* /*
* Lop off SYN bit if it has already been sent. However, if this is * Lop off SYN bit if it has already been sent. However, if this is
* SYN-SENT state and if segment contains data and if we don't know * SYN-SENT state and if segment contains data and if we don't know
* that foreign host supports TAO, suppress sending segment. * that foreign host supports TAO, suppress sending segment.
*/ */
if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una) &&
if ((tp->t_state != TCPS_SYN_RECEIVED) && ((sack_rxmit == 0) && (tp->t_rxtshift == 0))) {
(tp->t_state != TCPS_SYN_SENT)) if (tp->t_state != TCPS_SYN_RECEIVED)
flags &= ~TH_SYN; flags &= ~TH_SYN;
#ifdef TCP_RFC7413
/* /*
* When sending additional segments following a TFO SYN|ACK, * When sending additional segments following a TFO SYN|ACK,
* do not include the SYN bit. * do not include the SYN bit.
*/ */
if ((tp->t_flags & TF_FASTOPEN) && if (IS_FASTOPEN(tp->t_flags) &&
(tp->t_state == TCPS_SYN_RECEIVED)) (tp->t_state == TCPS_SYN_RECEIVED))
flags &= ~TH_SYN; flags &= ~TH_SYN;
#endif
sb_offset--, len++; sb_offset--, len++;
if (sbavail(sb) == 0)
len = 0;
} }
/* /*
* Be careful not to send data and/or FIN on SYN segments. This * Be careful not to send data and/or FIN on SYN segments. This
* measure is needed to prevent interoperability problems with not * measure is needed to prevent interoperability problems with not
* fully conformant TCP implementations. * fully conformant TCP implementations.
*/ */
if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) { if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) {
len = 0; len = 0;
flags &= ~TH_FIN; flags &= ~TH_FIN;
} }
#ifdef TCP_RFC7413
/* /*
* When retransmitting SYN|ACK on a passively-created TFO socket, * On TFO sockets, ensure no data is sent in the following cases:
* don't include data, as the presence of data may have caused the *
* original SYN|ACK to have been dropped by a middlebox. * - When retransmitting SYN|ACK on a passively-created socket
*
* - When retransmitting SYN on an actively created socket
*
* - When sending a zero-length cookie (cookie request) on an
* actively created socket
*
* - When the socket is in the CLOSED state (RST is being sent)
*/ */
if ((tp->t_flags & TF_FASTOPEN) && if (IS_FASTOPEN(tp->t_flags) &&
((tp->t_state == TCPS_SYN_RECEIVED) && (tp->t_rxtshift > 0))) (((flags & TH_SYN) && (tp->t_rxtshift > 0)) ||
((tp->t_state == TCPS_SYN_SENT) &&
(tp->t_tfo_client_cookie_len == 0)) ||
(flags & TH_RST)))
len = 0; len = 0;
#endif /* Without fast-open there should never be data sent on a SYN */
if ((flags & TH_SYN) && (!IS_FASTOPEN(tp->t_flags)))
len = 0;
if (len <= 0) { if (len <= 0) {
/* /*
* If FIN has been sent but not acked, but we haven't been * If FIN has been sent but not acked, but we haven't been
* called to retransmit, len will be < 0. Otherwise, window * called to retransmit, len will be < 0. Otherwise, window
* shrank after we sent into it. If window shrank to 0, * shrank after we sent into it. If window shrank to 0,
* cancel pending retransmit, pull snd_nxt back to (closed) * cancel pending retransmit, pull snd_nxt back to (closed)
* window, and set the persist timer if it isn't already * window, and set the persist timer if it isn't already
* going. If the window didn't close completely, just wait * going. If the window didn't close completely, just wait
▲ Show 20 Lines • Show All 299 Lines • ▼ Show 20 Lines if ((tp->t_flags & TF_NOOPT) == 0) {
if (flags & TH_SYN) { if (flags & TH_SYN) {
tp->snd_nxt = tp->iss; tp->snd_nxt = tp->iss;
to.to_mss = tcp_mssopt(&inp->inp_inc); to.to_mss = tcp_mssopt(&inp->inp_inc);
#ifdef NETFLIX_TCPOUDP #ifdef NETFLIX_TCPOUDP
if (tp->t_port) if (tp->t_port)
to.to_mss -= V_tcp_udp_tunneling_overhead; to.to_mss -= V_tcp_udp_tunneling_overhead;
#endif #endif
to.to_flags |= TOF_MSS; to.to_flags |= TOF_MSS;
#ifdef TCP_RFC7413
/* /*
* Only include the TFO option on the first * On SYN or SYN|ACK transmits on TFO connections,
* transmission of the SYN|ACK on a * only include the TFO option if it is not a
* passively-created TFO socket, as the presence of * retransmit, as the presence of the TFO option may
* the TFO option may have caused the original * have caused the original SYN or SYN|ACK to have
* SYN|ACK to have been dropped by a middlebox. * been dropped by a middlebox.
*/ */
if ((tp->t_flags & TF_FASTOPEN) && if (IS_FASTOPEN(tp->t_flags) &&
(tp->t_state == TCPS_SYN_RECEIVED) &&
(tp->t_rxtshift == 0)) { (tp->t_rxtshift == 0)) {
to.to_tfo_len = TCP_FASTOPEN_MAX_COOKIE_LEN; if (tp->t_state == TCPS_SYN_RECEIVED) {
to.to_tfo_cookie = (u_char *)&tp->t_tfo_cookie; to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN;
to.to_tfo_cookie =
(u_int8_t *)&tp->t_tfo_cookie.server;
to.to_flags |= TOF_FASTOPEN; to.to_flags |= TOF_FASTOPEN;
wanted_cookie = 1;
} else if (tp->t_state == TCPS_SYN_SENT) {
to.to_tfo_len =
tp->t_tfo_client_cookie_len;
to.to_tfo_cookie =
tp->t_tfo_cookie.client;
to.to_flags |= TOF_FASTOPEN;
wanted_cookie = 1;
/*
* If we wind up having more data to
* send with the SYN than can fit in
* one segment, don't send any more
* until the SYN|ACK comes back from
* the other end.
*/
sendalot = 0;
} }
#endif
} }
}
/* Window scaling. */ /* Window scaling. */
if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) { if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) {
to.to_wscale = tp->request_r_scale; to.to_wscale = tp->request_r_scale;
to.to_flags |= TOF_SCALE; to.to_flags |= TOF_SCALE;
} }
/* Timestamps. */ /* Timestamps. */
if ((tp->t_flags & TF_RCVD_TSTMP) || if ((tp->t_flags & TF_RCVD_TSTMP) ||
((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) { ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) {
Show All 17 Lines
#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
/* TCP-MD5 (RFC2385). */ /* TCP-MD5 (RFC2385). */
if (tp->t_flags & TF_SIGNATURE) if (tp->t_flags & TF_SIGNATURE)
to.to_flags |= TOF_SIGNATURE; to.to_flags |= TOF_SIGNATURE;
#endif /* TCP_SIGNATURE */ #endif /* TCP_SIGNATURE */
/* Processing the options. */ /* Processing the options. */
hdrlen += optlen = tcp_addoptions(&to, opt); hdrlen += optlen = tcp_addoptions(&to, opt);
/*
* If we wanted a TFO option to be added, but it was unable
* to fit, ensure no data is sent.
*/
if (IS_FASTOPEN(tp->t_flags) && wanted_cookie &&
!(to.to_flags & TOF_FASTOPEN))
len = 0;
} }
#ifdef NETFLIX_TCPOUDP #ifdef NETFLIX_TCPOUDP
if (tp->t_port) { if (tp->t_port) {
if (V_tcp_udp_tunneling_port == 0) { if (V_tcp_udp_tunneling_port == 0) {
/* The port was removed?? */ /* The port was removed?? */
SOCKBUF_UNLOCK(&so->so_snd); SOCKBUF_UNLOCK(&so->so_snd);
return (EHOSTUNREACH); return (EHOSTUNREACH);
} }
▲ Show 20 Lines • Show All 1,391 Lines • Show Last 20 Lines