/*
* This function is called from the receive side with the sequence number of
* the last packet received.
*/
void AJ_SerialTx_ReceivedSeq(uint8_t seq)
{
/*
* If we think we have already acked this sequence number we don't adjust
* the ack count.
*/
if (!SEQ_GT(currentTxAck, seq)) {
currentTxAck = (seq + 1) & 0x7;
}
#ifdef ALWAYS_ACK
AJ_SerialTX_EnqueueCtrl(NULL, 0, AJ_SERIAL_ACK);
#else
++pendingAcks;
/*
* If there are no packets to send we are allowed to accumulate a
* backlog of pending ACKs up to a maximum equal to the window size.
* In any case we are required to send an ack within a timeout
* period so if this is the first pending ack we need to prime a timer.
*/
if (pendingAcks == 1) {
AJ_InitTimer(&ackTime);
AJ_TimeAddOffset(&ackTime, AJ_SerialLinkParams.txAckTimeout);
return;
}
/*
* If we have hit our pending ACK limit send a explicit ACK packet immediately.
*/
if (pendingAcks == AJ_SerialLinkParams.windowSize) {
AJ_SerialTX_EnqueueCtrl(NULL, 0, AJ_SERIAL_ACK);
}
#endif
}
static void
dctcp_ack_received(struct cc_var *ccv, uint16_t type)
{
struct dctcp *dctcp_data;
int bytes_acked = 0;
dctcp_data = ccv->cc_data;
if (CCV(ccv, t_flags) & TF_ECN_PERMIT) {
/*
* DCTCP doesn't treat receipt of ECN marked packet as a
* congestion event. Thus, DCTCP always executes the ACK
* processing out of congestion recovery.
*/
if (IN_CONGRECOVERY(CCV(ccv, t_flags))) {
EXIT_CONGRECOVERY(CCV(ccv, t_flags));
newreno_cc_algo.ack_received(ccv, type);
ENTER_CONGRECOVERY(CCV(ccv, t_flags));
} else
newreno_cc_algo.ack_received(ccv, type);
if (type == CC_DUPACK)
bytes_acked = CCV(ccv, t_maxseg);
if (type == CC_ACK)
bytes_acked = ccv->bytes_this_ack;
/* Update total bytes. */
dctcp_data->bytes_total += bytes_acked;
/* Update total marked bytes. */
if (dctcp_data->ece_curr) {
if (!dctcp_data->ece_prev
&& bytes_acked > CCV(ccv, t_maxseg)) {
dctcp_data->bytes_ecn +=
(bytes_acked - CCV(ccv, t_maxseg));
} else
dctcp_data->bytes_ecn += bytes_acked;
dctcp_data->ece_prev = 1;
} else {
if (dctcp_data->ece_prev
&& bytes_acked > CCV(ccv, t_maxseg))
dctcp_data->bytes_ecn += CCV(ccv, t_maxseg);
dctcp_data->ece_prev = 0;
}
dctcp_data->ece_curr = 0;
/*
* Update the fraction of marked bytes at the end of
* current window size.
*/
if ((IN_FASTRECOVERY(CCV(ccv, t_flags)) &&
SEQ_GEQ(ccv->curack, CCV(ccv, snd_recover))) ||
(!IN_FASTRECOVERY(CCV(ccv, t_flags)) &&
SEQ_GT(ccv->curack, dctcp_data->save_sndnxt)))
dctcp_update_alpha(ccv);
} else
newreno_cc_algo.ack_received(ccv, type);
}
int StreamIterator(Flow *f, TcpStream *stream, int close, void *cbdata, uint8_t iflags)
{
SCLogDebug("called with %p, %d, %p, %02x", f, close, cbdata, iflags);
int logged = 0;
/* optimization: don't iterate list if we've logged all,
* so check the last segment's flags */
if (stream->seg_list_tail != NULL &&
(!(stream->seg_list_tail->flags & SEGMENTTCP_FLAG_LOGAPI_PROCESSED)))
{
TcpSegment *seg = stream->seg_list;
while (seg) {
uint8_t flags = iflags;
if (seg->flags & SEGMENTTCP_FLAG_LOGAPI_PROCESSED) {
seg = seg->next;
continue;
}
if (SEQ_GT(seg->seq + seg->payload_len, stream->last_ack)) {
SCLogDebug("seg not (fully) acked yet");
break;
}
if (seg->seq == stream->isn + 1)
flags |= OUTPUT_STREAMING_FLAG_OPEN;
/* if we need to close and we're at the last segment in the list
* we add the 'close' flag so the logger can close up. */
if (close && seg->next == NULL)
flags |= OUTPUT_STREAMING_FLAG_CLOSE;
Streamer(cbdata, f, seg->payload, (uint32_t)seg->payload_len, 0, flags);
seg->flags |= SEGMENTTCP_FLAG_LOGAPI_PROCESSED;
seg = seg->next;
logged = 1;
}
}
/* if we need to close we need to invoke the Streamer for sure. If we
* logged no segments, we call the Streamer with NULL data so it can
* close up. */
if (logged == 0 && close) {
Streamer(cbdata, f, NULL, 0, 0, OUTPUT_STREAMING_FLAG_CLOSE);
}
return 0;
}
static void tcp_state_update(struct tcpup_info *upp, int state)
{
fprintf(stderr, "%x/%-4d %s\t -> %s\n",
upp->t_conv, _tot_tcp, tcpstates[upp->t_state], tcpstates[state]);
if (SEQ_GT(upp->snd_nxt, upp->snd_max)) {
/* update snd max to nxt */
upp->snd_max = upp->snd_nxt;
}
upp->t_state = state;
upp->x_state = state;
return;
}
/**
* \brief Replace (part of) the payload portion of a packet by the data
* in a TCP segment
*
* \param p Packet
* \param seg TCP segment
*
* \todo What about reassembled fragments?
* \todo What about unwrapped tunnel packets?
*/
void StreamTcpInlineSegmentReplacePacket(Packet *p, TcpSegment *seg) {
SCEnter();
uint32_t pseq = TCP_GET_SEQ(p);
uint32_t tseq = seg->seq;
/* check if segment is within the packet */
if (tseq + seg->payload_len < pseq) {
SCReturn;
} else if (pseq + p->payload_len < tseq) {
SCReturn;
} else {
/** \todo review logic */
uint32_t pend = pseq + p->payload_len;
uint32_t tend = tseq + seg->payload_len;
SCLogDebug("pend %u, tend %u", pend, tend);
//SCLogDebug("packet");
//PrintRawDataFp(stdout,p->payload,p->payload_len);
//SCLogDebug("seg");
//PrintRawDataFp(stdout,seg->payload,seg->payload_len);
/* get the minimal seg*_end */
uint32_t end = (SEQ_GT(pend, tend)) ? tend : pend;
/* and the max seq */
uint32_t seq = (SEQ_LT(pseq, tseq)) ? tseq : pseq;
SCLogDebug("seq %u, end %u", seq, end);
uint16_t poff = seq - pseq;
uint16_t toff = seq - tseq;
SCLogDebug("poff %u, toff %u", poff, toff);
uint32_t range = end - seq;
SCLogDebug("range %u", range);
BUG_ON(range > 65536);
if (range) {
/* update the packets payload. As payload is a ptr to either
* p->pkt or p->ext_pkt that is updated as well */
memcpy(p->payload+poff, seg->payload+toff, range);
/* flag as modified so we can reinject / replace after
* recalculating the checksum */
p->flags |= PKT_STREAM_MODIFIED;
}
}
}
/*
* To remove a SACK block.
*
* Parameters:
* sack_blk_t *head: pointer to the array of SACK blks.
* tcp_seq end: to remove all sack blk with seq num less than end.
* int32_t *num: (referenced) total num of SACK blks in the array.
*/
void
tcp_sack_remove(sack_blk_t *head, tcp_seq end, int32_t *num)
{
sack_blk_t tmp[MAX_SACK_BLK];
int32_t i, j, old_num, new_num;
if (*num == 0)
return;
old_num = *num;
new_num = old_num;
j = 0;
/* Walk thru the whole list and copy the new list to tmp[]. */
for (i = 0; i < old_num; i++) {
if (SEQ_GT(end, head[i].begin)) {
/*
* Check to see if the old SACK blk needs to be
* removed or updated. If the old blk is just
* partially covered, update begin and continue.
* If the old blk is completely covered, remove it
* and continue to check.
*/
if (SEQ_GEQ(end, head[i].end)) {
new_num--;
continue;
} else {
tmp[j].begin = end;
tmp[j].end = head[i].end;
}
} else {
tmp[j].begin = head[i].begin;
tmp[j].end = head[i].end;
}
j++;
}
/* Copy tmp[] back to the original list. */
for (i = 0; i < new_num; i++) {
head[i].begin = tmp[i].begin;
head[i].end = tmp[i].end;
}
*num = new_num;
}
/**
* This function is called by the receive layer when a data packet or an explicit ACK
* has been received. The ACK value is one greater (modulo 8) than the seq number of the
* last packet successfully received.
*/
void AJ_SerialTx_ReceivedAck(uint8_t ack)
{
TxPkt volatile* ackedPkt = NULL;
if (txSent == NULL) {
return;
}
/*
* Remove acknowledged packets from sent queue.
*/
while ((txSent != NULL) && SEQ_GT(ack, txSent->seq)) {
ackedPkt = txSent;
txSent = txSent->next;
//AJ_AlwaysPrintf("Releasing seq=%d (acked by %d)\n", ackedPkt->seq, ack);
AJ_ASSERT(ackedPkt->type == AJ_SERIAL_DATA);
/*
* Return pkt to ACL free list.
*/
ackedPkt->next = txFreeList;
txFreeList = ackedPkt;
/*
* If all packet have been ack'd, halt the resend timer and return.
*/
if (txSent == NULL) {
AJ_InitTimer(&resendTime);
AJ_TimeAddOffset(&resendTime, AJ_TIMER_FOREVER);
resendPrimed = FALSE;
return;
}
}
/*
* Reset the resend timer if one or more packets were ack'd.
*/
if (ackedPkt != NULL) {
AJ_InitTimer(&resendTime);
AJ_TimeAddOffset(&resendTime, AJ_SerialLinkParams.txResendTimeout);
resendPrimed = TRUE;
}
}
/*
* Perform any necessary tasks before we exit congestion recovery.
*/
static void
newreno_post_recovery(struct cc_var *ccv)
{
if (IN_FASTRECOVERY(CCV(ccv, t_flags))) {
/*
* Fast recovery will conclude after returning from this
* function. Window inflation should have left us with
* approximately snd_ssthresh outstanding data. But in case we
* would be inclined to send a burst, better to do it via the
* slow start mechanism.
*
* XXXLAS: Find a way to do this without needing curack
*/
if (SEQ_GT(ccv->curack + CCV(ccv, snd_ssthresh),
CCV(ccv, snd_max)))
CCV(ccv, snd_cwnd) = CCV(ccv, snd_max) -
ccv->curack + CCV(ccv, t_maxseg);
else
CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh);
}
}
/*
* USTAT PDU / SOS_READY Processor
*
* Arguments:
* sop pointer to sscop connection block
* m pointer to PDU buffer (without trailer)
* trlr pointer to PDU trailer
*
* Returns:
* none
*
*/
void
sscop_ustat_ready(struct sscop *sop, KBuffer *m, caddr_t trlr)
{
struct ustat_pdu *up = (struct ustat_pdu *)trlr;
struct pdu_hdr *php;
sscop_seq seq1, seq2;
up->ustat_nmr = ntohl(up->ustat_nmr);
up->ustat_nr = ntohl(up->ustat_nr);
/*
* Validate peer's current receive data sequence number
*/
if (SEQ_GT(sop->so_ack, up->ustat_nr, sop->so_ack) ||
SEQ_GEQ(up->ustat_nr, sop->so_send, sop->so_ack)) {
/*
* Bad data sequence number
*/
goto goterr;
}
/*
* Free acknowledged PDUs
*/
for (seq1 = sop->so_ack, SEQ_SET(seq2, up->ustat_nr);
SEQ_LT(seq1, seq2, sop->so_ack);
SEQ_INCR(seq1, 1)) {
sscop_pack_free(sop, seq1);
}
/*
* Update transmit state variables
*/
sop->so_ack = seq2;
SEQ_SET(sop->so_sendmax, up->ustat_nmr);
/*
* Get USTAT list elements
*/
SEQ_SET(seq1, ntohl(up->ustat_le1));
SEQ_SET(seq2, ntohl(up->ustat_le2));
/*
* Validate elements
*/
if (SEQ_GT(sop->so_ack, seq1, sop->so_ack) ||
SEQ_GEQ(seq1, seq2, sop->so_ack) ||
SEQ_GEQ(seq2, sop->so_send, sop->so_ack)) {
/*
* Bad element sequence number
*/
goto goterr;
}
/*
* Process each missing sequence number in this gap
*/
while (SEQ_LT(seq1, seq2, sop->so_ack)) {
/*
* Find corresponding SD PDU on pending ack queue
*/
php = sscop_pack_locate(sop, seq1);
if (php == NULL) {
goto goterr;
}
/*
* Retransmit this SD PDU if it's not
* already scheduled for retranmission.
*/
if ((php->ph_rexmit_lk == NULL) &&
(sop->so_rexmit_tl != php)) {
/*
* Put PDU on retransmit queue and schedule
* transmit servicing
*/
sscop_rexmit_insert(sop, php);
sop->so_flags |= SOF_XMITSRVC;
}
/*
* Bump to next sequence number
*/
SEQ_INCR(seq1, 1);
}
/*
* Report retransmitted PDUs
*/
sscop_maa_error(sop, 'V');
/*
* Free PDU buffer chain
*/
KB_FREEALL(m);
/*
* See if transmit queues need servicing
*/
if (sop->so_flags & SOF_XMITSRVC)
sscop_service_xmit(sop);
return;
goterr:
/*
* Protocol/parameter error encountered
*/
sscop_maa_error(sop, 'T');
/*
* Free PDU buffer chain
*/
KB_FREEALL(m);
if (sop->so_vers == SSCOP_VERS_QSAAL)
/*
* Reestablish a new connection
*/
qsaal1_reestablish(sop);
else
/*
* Initiate error recovery
*/
q2110_error_recovery(sop);
return;
}
/*
* STAT PDU / SOS_READY Processor
*
* Arguments:
* sop pointer to sscop connection block
* m pointer to PDU buffer (without trailer)
* trlr pointer to PDU trailer
*
* Returns:
* none
*
*/
void
sscop_stat_ready(struct sscop *sop, KBuffer *m, caddr_t trlr)
{
struct stat_pdu *sp = (struct stat_pdu *)trlr;
struct pdu_hdr *php;
KBuffer *m0 = m;
sscop_seq seq1, seq2, opa;
int cnt = 0;
sp->stat_nps = ntohl(sp->stat_nps);
sp->stat_nmr = ntohl(sp->stat_nmr);
sp->stat_nr = ntohl(sp->stat_nr);
/*
* Validate peer's received poll sequence number
*/
if (SEQ_GT(sop->so_pollack, sp->stat_nps, sop->so_pollack) ||
SEQ_GT(sp->stat_nps, sop->so_pollsend, sop->so_pollack)) {
/*
* Bad poll sequence number
*/
sscop_maa_error(sop, 'R');
goto goterr;
}
/*
* Validate peer's current receive data sequence number
*/
if (SEQ_GT(sop->so_ack, sp->stat_nr, sop->so_ack) ||
SEQ_GT(sp->stat_nr, sop->so_send, sop->so_ack)) {
/*
* Bad data sequence number
*/
sscop_maa_error(sop, 'S');
goto goterr;
}
/*
* Free acknowledged PDUs
*/
for (seq1 = sop->so_ack, SEQ_SET(seq2, sp->stat_nr);
SEQ_LT(seq1, seq2, sop->so_ack);
SEQ_INCR(seq1, 1)) {
sscop_pack_free(sop, seq1);
}
/*
* Update transmit state variables
*/
opa = sop->so_pollack;
sop->so_ack = seq2;
SEQ_SET(sop->so_pollack, sp->stat_nps);
SEQ_SET(sop->so_sendmax, sp->stat_nmr);
/*
* Get first element in STAT list
*/
while (m && (KB_LEN(m) == 0))
m = KB_NEXT(m);
if (m == NULL)
goto done;
m = sscop_stat_getelem(m, &seq1);
/*
* Make sure there's a second element too
*/
if (m == NULL)
goto done;
/*
* Validate first element (start of missing pdus)
*/
if (SEQ_GT(sop->so_ack, seq1, sop->so_ack) ||
SEQ_GEQ(seq1, sop->so_send, sop->so_ack)) {
/*
* Bad element sequence number
*/
sscop_maa_error(sop, 'S');
goto goterr;
}
/*
* Loop thru all STAT elements in list
*/
while (m) {
/*
* Get next even element (start of received pdus)
*/
m = sscop_stat_getelem(m, &seq2);
/*
* Validate seqence number
*/
if (SEQ_GEQ(seq1, seq2, sop->so_ack) ||
SEQ_GT(seq2, sop->so_send, sop->so_ack)) {
/*
* Bad element sequence number
*/
sscop_maa_error(sop, 'S');
goto goterr;
}
/*
* Process each missing sequence number in this gap
*/
while (SEQ_LT(seq1, seq2, sop->so_ack)) {
/*
* Find corresponding SD PDU on pending ack queue
*/
php = sscop_pack_locate(sop, seq1);
if (php == NULL) {
sscop_maa_error(sop, 'S');
goto goterr;
}
/*
* Retransmit this SD PDU only if it was last sent
* during an earlier poll sequence and it's not
* already scheduled for retranmission.
*/
if (SEQ_LT(php->ph_nps, sp->stat_nps, opa) &&
(php->ph_rexmit_lk == NULL) &&
(sop->so_rexmit_tl != php)) {
/*
* Put PDU on retransmit queue and schedule
* transmit servicing
*/
sscop_rexmit_insert(sop, php);
sop->so_flags |= SOF_XMITSRVC;
cnt++;
}
/*
* Bump to next sequence number
*/
SEQ_INCR(seq1, 1);
}
/*
* Now process series of acknowledged PDUs
*
* Get next odd element (start of missing pdus),
* but make sure there is one and that it's valid
*/
if (m == NULL)
goto done;
m = sscop_stat_getelem(m, &seq2);
if (SEQ_GEQ(seq1, seq2, sop->so_ack) ||
SEQ_GT(seq2, sop->so_send, sop->so_ack)) {
/*
* Bad element sequence number
*/
sscop_maa_error(sop, 'S');
goto goterr;
}
/*
* Process each acked sequence number
*/
while (SEQ_LT(seq1, seq2, sop->so_ack)) {
/*
* Can we clear transmit buffers ??
*/
if ((sop->so_flags & SOF_NOCLRBUF) == 0) {
/*
* Yes, free acked buffers
*/
sscop_pack_free(sop, seq1);
}
/*
* Bump to next sequence number
*/
SEQ_INCR(seq1, 1);
}
}
done:
/*
* Free PDU buffer chain
*/
KB_FREEALL(m0);
/*
* Report retransmitted PDUs
*/
if (cnt)
sscop_maa_error(sop, 'V');
/*
* Record transmit window closed transitions
*/
if (SEQ_LT(sop->so_send, sop->so_sendmax, sop->so_ack)) {
if (sop->so_flags & SOF_NOCREDIT) {
sop->so_flags &= ~SOF_NOCREDIT;
sscop_maa_error(sop, 'X');
}
} else {
if ((sop->so_flags & SOF_NOCREDIT) == 0) {
sop->so_flags |= SOF_NOCREDIT;
sscop_maa_error(sop, 'W');
}
}
if (sop->so_vers == SSCOP_VERS_QSAAL)
/*
* Restart lost poll/stat timer
*/
sop->so_timer[SSCOP_T_NORESP] = sop->so_parm.sp_timeresp;
else {
/*
* Determine new polling phase
*/
if ((sop->so_timer[SSCOP_T_POLL] != 0) &&
((sop->so_flags & SOF_KEEPALIVE) == 0)) {
/*
* Remain in active phase - reset NO-RESPONSE timer
*/
sop->so_timer[SSCOP_T_NORESP] =
sop->so_parm.sp_timeresp;
} else if (sop->so_timer[SSCOP_T_IDLE] == 0) {
/*
* Go from transient to idle phase
*/
sop->so_timer[SSCOP_T_POLL] = 0;
sop->so_flags &= ~SOF_KEEPALIVE;
sop->so_timer[SSCOP_T_NORESP] = 0;
sop->so_timer[SSCOP_T_IDLE] = sop->so_parm.sp_timeidle;
}
}
/*
* See if transmit queues need servicing
*/
if (sop->so_flags & SOF_XMITSRVC)
sscop_service_xmit(sop);
return;
goterr:
/*
* Protocol/parameter error encountered
*/
/*
* Free PDU buffer chain
*/
KB_FREEALL(m0);
if (sop->so_vers == SSCOP_VERS_QSAAL)
/*
* Reestablish a new connection
*/
qsaal1_reestablish(sop);
else
/*
* Initiate error recovery
*/
q2110_error_recovery(sop);
return;
}
/*
* When a new ack with SACK is received, check if it indicates packet
* reordering. If there is packet reordering, the socket is marked and
* the late time offset by which the packet was reordered with
* respect to its closest neighboring packets is computed.
*/
static void
tcp_sack_detect_reordering(struct tcpcb *tp, struct sackhole *s,
tcp_seq sacked_seq, tcp_seq snd_fack)
{
int32_t rext = 0, reordered = 0;
/*
* If the SACK hole is past snd_fack, this is from new SACK
* information, so we can ignore it.
*/
if (SEQ_GT(s->end, snd_fack))
return;
/*
* If there has been a retransmit timeout, then the timestamp on
* the SACK segment will be newer. This might lead to a
* false-positive. Avoid re-ordering detection in this case.
*/
if (tp->t_rxtshift > 0)
return;
/*
* Detect reordering from SACK information by checking
* if recently sacked data was never retransmitted from this hole.
*/
if (SEQ_LT(s->rxmit, sacked_seq)) {
reordered = 1;
tcpstat.tcps_avoid_rxmt++;
}
if (reordered) {
if (tcp_detect_reordering == 1 &&
!(tp->t_flagsext & TF_PKTS_REORDERED)) {
tp->t_flagsext |= TF_PKTS_REORDERED;
tcpstat.tcps_detect_reordering++;
}
tcpstat.tcps_reordered_pkts++;
tp->t_reordered_pkts++;
/*
* If reordering is seen on a connection wth ECN enabled,
* increment the heuristic
*/
if (TCP_ECN_ENABLED(tp)) {
INP_INC_IFNET_STAT(tp->t_inpcb, ecn_fallback_reorder);
tcpstat.tcps_ecn_fallback_reorder++;
tcp_heuristic_ecn_aggressive(tp);
}
VERIFY(SEQ_GEQ(snd_fack, s->rxmit));
if (s->rxmit_start > 0) {
rext = timer_diff(tcp_now, 0, s->rxmit_start, 0);
if (rext < 0)
return;
/*
* We take the maximum reorder window to schedule
* DELAYFR timer as that will take care of jitter
* on the network path.
*
* Computing average and standard deviation seems
* to cause unnecessary retransmissions when there
* is high jitter.
*
* We set a maximum of SRTT/2 and a minimum of
* 10 ms on the reorder window.
*/
tp->t_reorderwin = max(tp->t_reorderwin, rext);
tp->t_reorderwin = min(tp->t_reorderwin,
(tp->t_srtt >> (TCP_RTT_SHIFT - 1)));
tp->t_reorderwin = max(tp->t_reorderwin, 10);
}
}
/*
* Tcp output routine: figure out what should be sent and send it.
*/
int
tcp_output(struct tcpcb *tp)
{
struct inpcb * const inp = tp->t_inpcb;
struct socket *so = inp->inp_socket;
long len, recvwin, sendwin;
int nsacked = 0;
int off, flags, error;
#ifdef TCP_SIGNATURE
int sigoff = 0;
#endif
struct mbuf *m;
struct ip *ip = NULL;
struct ipovly *ipov = NULL;
struct tcphdr *th;
u_char opt[TCP_MAXOLEN];
unsigned int ipoptlen, optlen, hdrlen;
int idle;
boolean_t sendalot;
struct ip6_hdr *ip6 = NULL;
#ifdef INET6
const boolean_t isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
#else
const boolean_t isipv6 = FALSE;
#endif
KKASSERT(so->so_port == &curthread->td_msgport);
/*
* Determine length of data that should be transmitted,
* and flags that will be used.
* If there is some data or critical controls (SYN, RST)
* to send, then transmit; otherwise, investigate further.
*/
/*
* If we have been idle for a while, the send congestion window
* could be no longer representative of the current state of the link.
* So unless we are expecting more acks to come in, slow-start from
* scratch to re-determine the send congestion window.
*/
if (tp->snd_max == tp->snd_una &&
(ticks - tp->t_rcvtime) >= tp->t_rxtcur) {
if (tcp_do_rfc3390) {
int initial_cwnd =
min(4 * tp->t_maxseg, max(2 * tp->t_maxseg, 4380));
tp->snd_cwnd = min(tp->snd_cwnd, initial_cwnd);
} else {
tp->snd_cwnd = tp->t_maxseg;
}
tp->snd_wacked = 0;
}
/*
* Calculate whether the transmit stream was previously idle
* and adjust TF_LASTIDLE for the next time.
*/
idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
if (idle && (tp->t_flags & TF_MORETOCOME))
tp->t_flags |= TF_LASTIDLE;
else
tp->t_flags &= ~TF_LASTIDLE;
if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max &&
!IN_FASTRECOVERY(tp))
nsacked = tcp_sack_bytes_below(&tp->scb, tp->snd_nxt);
again:
/* Make use of SACK information when slow-starting after a RTO. */
if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max &&
!IN_FASTRECOVERY(tp)) {
tcp_seq old_snd_nxt = tp->snd_nxt;
tcp_sack_skip_sacked(&tp->scb, &tp->snd_nxt);
nsacked += tp->snd_nxt - old_snd_nxt;
}
sendalot = FALSE;
off = tp->snd_nxt - tp->snd_una;
sendwin = min(tp->snd_wnd, tp->snd_cwnd + nsacked);
sendwin = min(sendwin, tp->snd_bwnd);
flags = tcp_outflags[tp->t_state];
/*
* Get standard flags, and add SYN or FIN if requested by 'hidden'
* state flags.
*/
if (tp->t_flags & TF_NEEDFIN)
flags |= TH_FIN;
if (tp->t_flags & TF_NEEDSYN)
flags |= TH_SYN;
/*
* If in persist timeout with window of 0, send 1 byte.
* Otherwise, if window is small but nonzero
* and timer expired, we will send what we can
* and go to transmit state.
*/
if (tp->t_flags & TF_FORCE) {
if (sendwin == 0) {
/*
* If we still have some data to send, then
* clear the FIN bit. Usually this would
* happen below when it realizes that we
* aren't sending all the data. However,
* if we have exactly 1 byte of unsent data,
* then it won't clear the FIN bit below,
* and if we are in persist state, we wind
* up sending the packet without recording
* that we sent the FIN bit.
*
* We can't just blindly clear the FIN bit,
* because if we don't have any more data
* to send then the probe will be the FIN
* itself.
*/
if (off < so->so_snd.ssb_cc)
flags &= ~TH_FIN;
sendwin = 1;
} else {
tcp_callout_stop(tp, tp->tt_persist);
tp->t_rxtshift = 0;
}
}
/*
* If snd_nxt == snd_max and we have transmitted a FIN, the
* offset will be > 0 even if so_snd.ssb_cc is 0, resulting in
* a negative length. This can also occur when TCP opens up
* its congestion window while receiving additional duplicate
* acks after fast-retransmit because TCP will reset snd_nxt
* to snd_max after the fast-retransmit.
*
* In the normal retransmit-FIN-only case, however, snd_nxt will
* be set to snd_una, the offset will be 0, and the length may
* wind up 0.
*/
len = (long)ulmin(so->so_snd.ssb_cc, sendwin) - off;
/*
* Lop off SYN bit if it has already been sent. However, if this
* is SYN-SENT state and if segment contains data, suppress sending
* segment (sending the segment would be an option if we still
* did TAO and the remote host supported it).
*/
if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
flags &= ~TH_SYN;
off--, len++;
if (len > 0 && tp->t_state == TCPS_SYN_SENT)
return 0;
}
/*
* Be careful not to send data and/or FIN on SYN segments.
* This measure is needed to prevent interoperability problems
* with not fully conformant TCP implementations.
*/
if (flags & TH_SYN) {
len = 0;
flags &= ~TH_FIN;
}
if (len < 0) {
/*
* If FIN has been sent but not acked,
* but we haven't been called to retransmit,
* len will be < 0. Otherwise, window shrank
* after we sent into it. If window shrank to 0,
* cancel pending retransmit, pull snd_nxt back
* to (closed) window, and set the persist timer
* if it isn't already going. If the window didn't
* close completely, just wait for an ACK.
*/
len = 0;
if (sendwin == 0) {
tcp_callout_stop(tp, tp->tt_rexmt);
tp->t_rxtshift = 0;
tp->snd_nxt = tp->snd_una;
if (!tcp_callout_active(tp, tp->tt_persist))
tcp_setpersist(tp);
}
}
KASSERT(len >= 0, ("%s: len < 0", __func__));
/*
* Automatic sizing of send socket buffer. Often the send buffer
* size is not optimally adjusted to the actual network conditions
* at hand (delay bandwidth product). Setting the buffer size too
* small limits throughput on links with high bandwidth and high
* delay (eg. trans-continental/oceanic links). Setting the
* buffer size too big consumes too much real kernel memory,
* especially with many connections on busy servers.
*
* The criteria to step up the send buffer one notch are:
* 1. receive window of remote host is larger than send buffer
* (with a fudge factor of 5/4th);
* 2. send buffer is filled to 7/8th with data (so we actually
* have data to make use of it);
* 3. send buffer fill has not hit maximal automatic size;
* 4. our send window (slow start and cogestion controlled) is
* larger than sent but unacknowledged data in send buffer.
*
* The remote host receive window scaling factor may limit the
* growing of the send buffer before it reaches its allowed
* maximum.
*
* It scales directly with slow start or congestion window
* and does at most one step per received ACK. This fast
* scaling has the drawback of growing the send buffer beyond
* what is strictly necessary to make full use of a given
* delay*bandwith product. However testing has shown this not
* to be much of an problem. At worst we are trading wasting
* of available bandwith (the non-use of it) for wasting some
* socket buffer memory.
*
* TODO: Shrink send buffer during idle periods together
* with congestion window. Requires another timer. Has to
* wait for upcoming tcp timer rewrite.
*/
if (tcp_do_autosndbuf && so->so_snd.ssb_flags & SSB_AUTOSIZE) {
if ((tp->snd_wnd / 4 * 5) >= so->so_snd.ssb_hiwat &&
so->so_snd.ssb_cc >= (so->so_snd.ssb_hiwat / 8 * 7) &&
so->so_snd.ssb_cc < tcp_autosndbuf_max &&
sendwin >= (so->so_snd.ssb_cc - (tp->snd_nxt - tp->snd_una))) {
u_long newsize;
newsize = ulmin(so->so_snd.ssb_hiwat +
tcp_autosndbuf_inc,
tcp_autosndbuf_max);
if (!ssb_reserve(&so->so_snd, newsize, so, NULL))
atomic_clear_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
if (newsize >= (TCP_MAXWIN << tp->snd_scale))
atomic_clear_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
}
}
/*
* Truncate to the maximum segment length and ensure that FIN is
* removed if the length no longer contains the last data byte.
*/
if (len > tp->t_maxseg) {
len = tp->t_maxseg;
sendalot = TRUE;
}
if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.ssb_cc))
flags &= ~TH_FIN;
recvwin = ssb_space(&so->so_rcv);
/*
* Sender silly window avoidance. We transmit under the following
* conditions when len is non-zero:
*
* - We have a full segment
* - This is the last buffer in a write()/send() and we are
* either idle or running NODELAY
* - we've timed out (e.g. persist timer)
* - we have more then 1/2 the maximum send window's worth of
* data (receiver may be limiting the window size)
* - we need to retransmit
*/
if (len) {
if (len == tp->t_maxseg)
goto send;
/*
* NOTE! on localhost connections an 'ack' from the remote
* end may occur synchronously with the output and cause
* us to flush a buffer queued with moretocome. XXX
*
* note: the len + off check is almost certainly unnecessary.
*/
if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
(idle || (tp->t_flags & TF_NODELAY)) &&
len + off >= so->so_snd.ssb_cc &&
!(tp->t_flags & TF_NOPUSH)) {
goto send;
}
if (tp->t_flags & TF_FORCE) /* typ. timeout case */
goto send;
if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
goto send;
if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
goto send;
}
/*
* Compare available window to amount of window
* known to peer (as advertised window less
* next expected input). If the difference is at least two
* max size segments, or at least 50% of the maximum possible
* window, then want to send a window update to peer.
*/
if (recvwin > 0) {
/*
* "adv" is the amount we can increase the window,
* taking into account that we are limited by
* TCP_MAXWIN << tp->rcv_scale.
*/
long adv = min(recvwin, (long)TCP_MAXWIN << tp->rcv_scale) -
(tp->rcv_adv - tp->rcv_nxt);
long hiwat;
/*
* This ack case typically occurs when the user has drained
* the TCP socket buffer sufficiently to warrent an ack
* containing a 'pure window update'... that is, an ack that
* ONLY updates the tcp window.
*
* It is unclear why we would need to do a pure window update
* past 2 segments if we are going to do one at 1/2 the high
* water mark anyway, especially since under normal conditions
* the user program will drain the socket buffer quickly.
* The 2-segment pure window update will often add a large
* number of extra, unnecessary acks to the stream.
*
* avoid_pure_win_update now defaults to 1.
*/
if (avoid_pure_win_update == 0 ||
(tp->t_flags & TF_RXRESIZED)) {
if (adv >= (long) (2 * tp->t_maxseg)) {
goto send;
}
}
hiwat = (long)(TCP_MAXWIN << tp->rcv_scale);
if (hiwat > (long)so->so_rcv.ssb_hiwat)
hiwat = (long)so->so_rcv.ssb_hiwat;
if (adv >= hiwat / 2)
goto send;
}
/*
* Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
* is also a catch-all for the retransmit timer timeout case.
*/
if (tp->t_flags & TF_ACKNOW)
goto send;
if ((flags & TH_RST) ||
((flags & TH_SYN) && !(tp->t_flags & TF_NEEDSYN)))
goto send;
if (SEQ_GT(tp->snd_up, tp->snd_una))
goto send;
/*
* If our state indicates that FIN should be sent
* and we have not yet done so, then we need to send.
*/
if (flags & TH_FIN &&
(!(tp->t_flags & TF_SENTFIN) || tp->snd_nxt == tp->snd_una))
goto send;
/*
* TCP window updates are not reliable, rather a polling protocol
* using ``persist'' packets is used to insure receipt of window
* updates. The three ``states'' for the output side are:
* idle not doing retransmits or persists
* persisting to move a small or zero window
* (re)transmitting and thereby not persisting
*
* tcp_callout_active(tp, tp->tt_persist)
* is true when we are in persist state.
* The TF_FORCE flag in tp->t_flags
* is set when we are called to send a persist packet.
* tcp_callout_active(tp, tp->tt_rexmt)
* is set when we are retransmitting
* The output side is idle when both timers are zero.
*
* If send window is too small, there is data to transmit, and no
* retransmit or persist is pending, then go to persist state.
* If nothing happens soon, send when timer expires:
* if window is nonzero, transmit what we can,
* otherwise force out a byte.
*/
if (so->so_snd.ssb_cc > 0 &&
!tcp_callout_active(tp, tp->tt_rexmt) &&
!tcp_callout_active(tp, tp->tt_persist)) {
tp->t_rxtshift = 0;
tcp_setpersist(tp);
}
/*
* No reason to send a segment, just return.
*/
return (0);
send:
/*
* Before ESTABLISHED, force sending of initial options
* unless TCP set not to do any options.
* NOTE: we assume that the IP/TCP header plus TCP options
* always fit in a single mbuf, leaving room for a maximum
* link header, i.e.
* max_linkhdr + sizeof(struct tcpiphdr) + optlen <= MCLBYTES
*/
optlen = 0;
if (isipv6)
hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
else
hdrlen = sizeof(struct tcpiphdr);
if (flags & TH_SYN) {
tp->snd_nxt = tp->iss;
if (!(tp->t_flags & TF_NOOPT)) {
u_short mss;
opt[0] = TCPOPT_MAXSEG;
opt[1] = TCPOLEN_MAXSEG;
mss = htons((u_short) tcp_mssopt(tp));
memcpy(opt + 2, &mss, sizeof mss);
optlen = TCPOLEN_MAXSEG;
if ((tp->t_flags & TF_REQ_SCALE) &&
(!(flags & TH_ACK) ||
(tp->t_flags & TF_RCVD_SCALE))) {
*((u_int32_t *)(opt + optlen)) = htonl(
TCPOPT_NOP << 24 |
TCPOPT_WINDOW << 16 |
TCPOLEN_WINDOW << 8 |
tp->request_r_scale);
optlen += 4;
}
if ((tcp_do_sack && !(flags & TH_ACK)) ||
tp->t_flags & TF_SACK_PERMITTED) {
uint32_t *lp = (uint32_t *)(opt + optlen);
*lp = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
optlen += TCPOLEN_SACK_PERMITTED_ALIGNED;
}
}
}
/*
* Send a timestamp and echo-reply if this is a SYN and our side
* wants to use timestamps (TF_REQ_TSTMP is set) or both our side
* and our peer have sent timestamps in our SYN's.
*/
if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
!(flags & TH_RST) &&
(!(flags & TH_ACK) || (tp->t_flags & TF_RCVD_TSTMP))) {
u_int32_t *lp = (u_int32_t *)(opt + optlen);
/* Form timestamp option as shown in appendix A of RFC 1323. */
*lp++ = htonl(TCPOPT_TSTAMP_HDR);
*lp++ = htonl(ticks);
*lp = htonl(tp->ts_recent);
optlen += TCPOLEN_TSTAMP_APPA;
}
/* Set receive buffer autosizing timestamp. */
if (tp->rfbuf_ts == 0 && (so->so_rcv.ssb_flags & SSB_AUTOSIZE))
tp->rfbuf_ts = ticks;
/*
* If this is a SACK connection and we have a block to report,
* fill in the SACK blocks in the TCP options.
*/
if ((tp->t_flags & (TF_SACK_PERMITTED | TF_NOOPT)) ==
TF_SACK_PERMITTED &&
(!LIST_EMPTY(&tp->t_segq) ||
tp->reportblk.rblk_start != tp->reportblk.rblk_end))
tcp_sack_fill_report(tp, opt, &optlen);
#ifdef TCP_SIGNATURE
if (tp->t_flags & TF_SIGNATURE) {
int i;
u_char *bp;
/*
* Initialize TCP-MD5 option (RFC2385)
*/
bp = (u_char *)opt + optlen;
*bp++ = TCPOPT_SIGNATURE;
*bp++ = TCPOLEN_SIGNATURE;
sigoff = optlen + 2;
for (i = 0; i < TCP_SIGLEN; i++)
*bp++ = 0;
optlen += TCPOLEN_SIGNATURE;
/*
* Terminate options list and maintain 32-bit alignment.
*/
*bp++ = TCPOPT_NOP;
*bp++ = TCPOPT_EOL;
optlen += 2;
}
#endif /* TCP_SIGNATURE */
KASSERT(optlen <= TCP_MAXOLEN, ("too many TCP options"));
hdrlen += optlen;
if (isipv6) {
ipoptlen = ip6_optlen(inp);
} else {
if (inp->inp_options) {
ipoptlen = inp->inp_options->m_len -
offsetof(struct ipoption, ipopt_list);
} else {
static int tcpup_state_receive(struct tcpup_info *upp, struct tcpiphdr *tcp, size_t dlen)
{
int xflags = 0;
int snd_una = htonl(tcp->th_ack);
if (tcp->th_flags & TH_RST) {
upp->t_state = TCPS_CLOSED;
return 0;
}
if ((tcp->th_flags & TH_ACK) && SEQ_GT(snd_una, upp->snd_una)) {
/* update snd una from peer */
upp->snd_una = snd_una;
}
switch (upp->t_state) {
case TCPS_SYN_SENT:
xflags = TH_SYN| TH_ACK;
if ((tcp->th_flags & xflags) == TH_SYN) {
assert((tcp->th_flags & TH_FIN) != TH_FIN);
tcp_state_preload(upp, TCPS_SYN_RECEIVED, htonl(tcp->th_seq) + 1);
return 0;
}
if ((tcp->th_flags & xflags) == xflags
&& SEQ_GEQ(htonl(tcp->th_ack), upp->snd_nxt)) {
assert((tcp->th_flags & TH_FIN) != TH_FIN);
tcp_state_preload(upp, TCPS_ESTABLISHED, htonl(tcp->th_seq));
return 0;
}
break;
case TCPS_SYN_RECEIVED:
if ((tcp->th_flags & TH_ACK) == TH_ACK
&& SEQ_GEQ(htonl(tcp->th_ack), upp->snd_nxt)) {
assert((tcp->th_flags & TH_FIN) != TH_FIN);
tcp_state_preload(upp, TCPS_ESTABLISHED, htonl(tcp->th_seq));
return 0;
}
break;
case TCPS_ESTABLISHED:
if ((tcp->th_flags & TH_FIN) == TH_FIN) {
tcp_state_preload(upp, TCPS_CLOSE_WAIT, htonl(tcp->th_seq) + 1);
return 0;
}
break;
case TCPS_FIN_WAIT_1:
xflags = TH_FIN| TH_ACK;
if ((tcp->th_flags & xflags) == xflags
&& SEQ_GEQ(htonl(tcp->th_ack), upp->snd_nxt)) {
tcp_state_preload(upp, TCPS_TIME_WAIT, htonl(tcp->th_seq) + dlen + 1);
return 0;
}
if ((tcp->th_flags & TH_FIN) == TH_FIN) {
tcp_state_preload(upp, TCPS_CLOSING, htonl(tcp->th_seq) + dlen + 1);
return 0;
}
if ((tcp->th_flags & TH_ACK) == TH_ACK
&& SEQ_GEQ(htonl(tcp->th_ack), upp->snd_nxt)) {
tcp_state_preload(upp, TCPS_FIN_WAIT_2, htonl(tcp->th_seq) + dlen);
return 0;
}
break;
case TCPS_FIN_WAIT_2:
if ((tcp->th_flags & TH_FIN) == TH_FIN) {
tcp_state_preload(upp, TCPS_TIME_WAIT, htonl(tcp->th_seq) + dlen + 1);
return 0;
}
break;
case TCPS_CLOSING:
if ((tcp->th_flags & TH_ACK) == TH_ACK
&& SEQ_GEQ(htonl(tcp->th_ack), upp->snd_nxt)) {
tcp_state_preload(upp, TCPS_TIME_WAIT, htonl(tcp->th_seq) + dlen);
return 0;
}
break;
case TCPS_LAST_ACK:
if ((tcp->th_flags & TH_ACK) == TH_ACK
&& SEQ_GEQ(htonl(tcp->th_ack), upp->snd_nxt)) {
tcp_state_preload(upp, TCPS_CLOSED, htonl(tcp->th_seq) + dlen);
return 0;
}
break;
case TCPS_TIME_WAIT:
fprintf(stderr, "before TIME_WAIT -> TIME_WAIT\n");
break;
}
return 0;
}
/*
* This function is called upon receipt of new valid data (while not in
* header prediction mode), and it updates the ordered list of sacks.
*/
void
tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end)
{
/*
* First reported block MUST be the most recent one. Subsequent
* blocks SHOULD be in the order in which they arrived at the
* receiver. These two conditions make the implementation fully
* compliant with RFC 2018.
*/
struct sackblk head_blk, saved_blks[MAX_SACK_BLKS];
int num_head, num_saved, i;
//ScenSim-Port// INP_WLOCK_ASSERT(tp->t_inpcb);
/* Check arguments. */
//ScenSim-Port// KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end"));
/* SACK block for the received segment. */
head_blk.start = rcv_start;
head_blk.end = rcv_end;
/*
* Merge updated SACK blocks into head_blk, and save unchanged SACK
* blocks into saved_blks[]. num_saved will have the number of the
* saved SACK blocks.
*/
num_saved = 0;
for (i = 0; i < tp->rcv_numsacks; i++) {
tcp_seq start = tp->sackblks[i].start;
tcp_seq end = tp->sackblks[i].end;
if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) {
/*
* Discard this SACK block.
*/
} else if (SEQ_LEQ(head_blk.start, end) &&
SEQ_GEQ(head_blk.end, start)) {
/*
* Merge this SACK block into head_blk. This SACK
* block itself will be discarded.
*/
if (SEQ_GT(head_blk.start, start))
head_blk.start = start;
if (SEQ_LT(head_blk.end, end))
head_blk.end = end;
} else {
/*
* Save this SACK block.
*/
saved_blks[num_saved].start = start;
saved_blks[num_saved].end = end;
num_saved++;
}
}
/*
* Update SACK list in tp->sackblks[].
*/
num_head = 0;
if (SEQ_GT(head_blk.start, tp->rcv_nxt)) {
/*
* The received data segment is an out-of-order segment. Put
* head_blk at the top of SACK list.
*/
tp->sackblks[0] = head_blk;
num_head = 1;
/*
* If the number of saved SACK blocks exceeds its limit,
* discard the last SACK block.
*/
if (num_saved >= MAX_SACK_BLKS)
num_saved--;
}
if (num_saved > 0) {
/*
* Copy the saved SACK blocks back.
*/
bcopy(saved_blks, &tp->sackblks[num_head],
sizeof(struct sackblk) * num_saved);
}
/* Save the number of SACK blocks. */
tp->rcv_numsacks = num_head + num_saved;
}
/*
* Process cumulative ACK and the TCP SACK option to update the scoreboard.
* tp->snd_holes is an ordered list of holes (oldest to newest, in terms of
* the sequence space).
*/
void
tcp_sack_doack(struct tcpcb *tp, struct tcpopt *to, tcp_seq th_ack)
{
struct sackhole *cur, *temp;
struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1], *sblkp;
int i, j, num_sack_blks;
//ScenSim-Port// INP_WLOCK_ASSERT(tp->t_inpcb);
num_sack_blks = 0;
/*
* If SND.UNA will be advanced by SEG.ACK, and if SACK holes exist,
* treat [SND.UNA, SEG.ACK) as if it is a SACK block.
*/
if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) {
sack_blocks[num_sack_blks].start = tp->snd_una;
sack_blocks[num_sack_blks++].end = th_ack;
}
/*
* Append received valid SACK blocks to sack_blocks[], but only if we
* received new blocks from the other side.
*/
if (to->to_flags & TOF_SACK) {
for (i = 0; i < to->to_nsacks; i++) {
bcopy((to->to_sacks + i * TCPOLEN_SACK),
&sack, sizeof(sack));
sack.start = ntohl(sack.start);
sack.end = ntohl(sack.end);
if (SEQ_GT(sack.end, sack.start) &&
SEQ_GT(sack.start, tp->snd_una) &&
SEQ_GT(sack.start, th_ack) &&
SEQ_LT(sack.start, tp->snd_max) &&
SEQ_GT(sack.end, tp->snd_una) &&
SEQ_LEQ(sack.end, tp->snd_max))
sack_blocks[num_sack_blks++] = sack;
}
}
/*
* Return if SND.UNA is not advanced and no valid SACK block is
* received.
*/
if (num_sack_blks == 0)
return;
/*
* Sort the SACK blocks so we can update the scoreboard with just one
* pass. The overhead of sorting upto 4+1 elements is less than
* making upto 4+1 passes over the scoreboard.
*/
for (i = 0; i < num_sack_blks; i++) {
for (j = i + 1; j < num_sack_blks; j++) {
if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) {
sack = sack_blocks[i];
sack_blocks[i] = sack_blocks[j];
sack_blocks[j] = sack;
}
}
}
if (TAILQ_EMPTY(&tp->snd_holes))
/*
* Empty scoreboard. Need to initialize snd_fack (it may be
* uninitialized or have a bogus value). Scoreboard holes
* (from the sack blocks received) are created later below
* (in the logic that adds holes to the tail of the
* scoreboard).
*/
tp->snd_fack = SEQ_MAX(tp->snd_una, th_ack);
/*
* In the while-loop below, incoming SACK blocks (sack_blocks[]) and
* SACK holes (snd_holes) are traversed from their tails with just
* one pass in order to reduce the number of compares especially when
* the bandwidth-delay product is large.
*
* Note: Typically, in the first RTT of SACK recovery, the highest
* three or four SACK blocks with the same ack number are received.
* In the second RTT, if retransmitted data segments are not lost,
* the highest three or four SACK blocks with ack number advancing
* are received.
*/
sblkp = &sack_blocks[num_sack_blks - 1]; /* Last SACK block */
tp->sackhint.last_sack_ack = sblkp->end;
if (SEQ_LT(tp->snd_fack, sblkp->start)) {
/*
* The highest SACK block is beyond fack. Append new SACK
* hole at the tail. If the second or later highest SACK
* blocks are also beyond the current fack, they will be
* inserted by way of hole splitting in the while-loop below.
*/
temp = tcp_sackhole_insert(tp, tp->snd_fack,sblkp->start,NULL);
if (temp != NULL) {
tp->snd_fack = sblkp->end;
/* Go to the previous sack block. */
sblkp--;
} else {
/*
* We failed to add a new hole based on the current
* sack block. Skip over all the sack blocks that
* fall completely to the right of snd_fack and
* proceed to trim the scoreboard based on the
* remaining sack blocks. This also trims the
* scoreboard for th_ack (which is sack_blocks[0]).
*/
while (sblkp >= sack_blocks &&
SEQ_LT(tp->snd_fack, sblkp->start))
sblkp--;
if (sblkp >= sack_blocks &&
SEQ_LT(tp->snd_fack, sblkp->end))
tp->snd_fack = sblkp->end;
}
} else if (SEQ_LT(tp->snd_fack, sblkp->end))
/* fack is advanced. */
tp->snd_fack = sblkp->end;
/* We must have at least one SACK hole in scoreboard. */
//ScenSim-Port// KASSERT(!TAILQ_EMPTY(&tp->snd_holes),
//ScenSim-Port// ("SACK scoreboard must not be empty"));
cur = TAILQ_LAST(&tp->snd_holes, sackhole_head); /* Last SACK hole. */
/*
* Since the incoming sack blocks are sorted, we can process them
* making one sweep of the scoreboard.
*/
while (sblkp >= sack_blocks && cur != NULL) {
if (SEQ_GEQ(sblkp->start, cur->end)) {
/*
* SACKs data beyond the current hole. Go to the
* previous sack block.
*/
sblkp--;
continue;
}
if (SEQ_LEQ(sblkp->end, cur->start)) {
/*
* SACKs data before the current hole. Go to the
* previous hole.
*/
cur = TAILQ_PREV(cur, sackhole_head, scblink);
continue;
}
tp->sackhint.sack_bytes_rexmit -= (cur->rxmit - cur->start);
//ScenSim-Port// KASSERT(tp->sackhint.sack_bytes_rexmit >= 0,
//ScenSim-Port// ("sackhint bytes rtx >= 0"));
if (SEQ_LEQ(sblkp->start, cur->start)) {
/* Data acks at least the beginning of hole. */
if (SEQ_GEQ(sblkp->end, cur->end)) {
/* Acks entire hole, so delete hole. */
temp = cur;
cur = TAILQ_PREV(cur, sackhole_head, scblink);
tcp_sackhole_remove(tp, temp);
/*
* The sack block may ack all or part of the
* next hole too, so continue onto the next
* hole.
*/
continue;
} else {
/* Move start of hole forward. */
cur->start = sblkp->end;
cur->rxmit = SEQ_MAX(cur->rxmit, cur->start);
}
} else {
/* Data acks at least the end of hole. */
if (SEQ_GEQ(sblkp->end, cur->end)) {
/* Move end of hole backward. */
cur->end = sblkp->start;
cur->rxmit = SEQ_MIN(cur->rxmit, cur->end);
} else {
/*
* ACKs some data in middle of a hole; need
* to split current hole
*/
temp = tcp_sackhole_insert(tp, sblkp->end,
cur->end, cur);
if (temp != NULL) {
if (SEQ_GT(cur->rxmit, temp->rxmit)) {
temp->rxmit = cur->rxmit;
tp->sackhint.sack_bytes_rexmit
+= (temp->rxmit
- temp->start);
}
cur->end = sblkp->start;
cur->rxmit = SEQ_MIN(cur->rxmit,
cur->end);
}
}
}
tp->sackhint.sack_bytes_rexmit += (cur->rxmit - cur->start);
/*
* Testing sblkp->start against cur->start tells us whether
* we're done with the sack block or the sack hole.
* Accordingly, we advance one or the other.
*/
if (SEQ_LEQ(sblkp->start, cur->start))
cur = TAILQ_PREV(cur, sackhole_head, scblink);
else
sblkp--;
}
}
int
tcp_reass(struct tcpcb *tp,struct tcpiphdr *ti, usn_mbuf_t *m)
{
struct tcpiphdr *q;
struct usn_socket *so = tp->t_inpcb->inp_socket;
int flags;
// Call with ti==0 after become established to
// force pre-ESTABLISHED data up to user socket.
if (ti == 0)
goto present;
// Find a segment which begins after this one does.
for (q = tp->seg_next; q != (struct tcpiphdr *)tp;
q = (struct tcpiphdr *)q->ti_next)
if (SEQ_GT(q->ti_seq, ti->ti_seq))
break;
// If there is a preceding segment, it may provide some of
// our data already. If so, drop the data from the incoming
// segment. If it provides all of our data, drop us.
if ((struct tcpiphdr *)q->ti_prev != (struct tcpiphdr *)tp) {
int i;
q = (struct tcpiphdr *)q->ti_prev;
// conversion to int (in i) handles seq wraparound
i = q->ti_seq + q->ti_len - ti->ti_seq;
if (i > 0) {
if (i >= ti->ti_len) {
g_tcpstat.tcps_rcvduppack++;
g_tcpstat.tcps_rcvdupbyte += ti->ti_len;
usn_free_mbuf(m);
return (0);
}
m_adj(m, i);
ti->ti_len -= i;
ti->ti_seq += i;
}
q = (struct tcpiphdr *)(q->ti_next);
}
g_tcpstat.tcps_rcvoopack++;
g_tcpstat.tcps_rcvoobyte += ti->ti_len;
REASS_MBUF(ti) = m; // XXX: wrong assumtion dst and src port for mbuf pointer
// While we overlap succeeding segments trim them or,
// if they are completely covered, dequeue them.
while (q != (struct tcpiphdr *)tp) {
int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
if (i <= 0)
break;
if (i < q->ti_len) {
q->ti_seq += i;
q->ti_len -= i;
m_adj(REASS_MBUF(q), i);
break;
}
q = (struct tcpiphdr *)q->ti_next;
m = REASS_MBUF((struct tcpiphdr *)q->ti_prev);
// FIXME
//remque(q->ti_prev);
usn_free_mbuf(m);
}
// FIXME: Stick new segment in its place.
//insque(ti, q->ti_prev);
present:
// Present data to user, advancing rcv_nxt through
// completed sequence space.
if (TCPS_HAVERCVDSYN(tp->t_state) == 0)
return (0);
ti = tp->seg_next;
if (ti == (struct tcpiphdr *)tp || ti->ti_seq != tp->rcv_nxt)
return (0);
if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
return (0);
do {
tp->rcv_nxt += ti->ti_len;
flags = ti->ti_flags & TH_FIN;
// FIXME
//remque(ti);
m = REASS_MBUF(ti);
ti = (struct tcpiphdr *)ti->ti_next;
if (so->so_state & USN_CANTRCVMORE) {
usn_free_mbuf(m);
} else {
sbappend(so->so_rcv, m);
}
} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
sorwakeup(so);
usnet_tcpin_rwakeup(so, USN_TCP_IN, USN_TCPEV_READ, m);
return (flags);
}
/*
* TCP input routine, follows pages 65-76 of the
* protocol specification dated September, 1981 very closely.
*/
void
tcp_input(usn_mbuf_t *m, int iphlen)
{
struct tcpiphdr *ti;
struct inpcb *inp;
u_char *optp = NULL;
int optlen;
int len, tlen, off;
struct tcpcb *tp = 0;
int tiflags;
struct usn_socket *so = 0;
int todrop, acked, ourfinisacked;
int needoutput = 0;
short ostate;
struct usn_in_addr laddr;
int dropsocket = 0;
int iss = 0;
u_long tiwin, ts_val, ts_ecr;
int ts_present = 0;
(void)needoutput;
g_tcpstat.tcps_rcvtotal++;
// Get IP and TCP header together in first mbuf.
// Note: IP leaves IP header in first mbuf.
ti = mtod(m, struct tcpiphdr *);
if (iphlen > sizeof (usn_ip_t))
ip_stripoptions(m, (usn_mbuf_t *)0);
if (m->mlen < sizeof (struct tcpiphdr)) {
if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
g_tcpstat.tcps_rcvshort++;
return;
}
ti = mtod(m, struct tcpiphdr *);
}
#ifdef DUMP_PAYLOAD
dump_chain(m,"tcp");
#endif
/*
* Checksum extended TCP header and data.
*/
tlen = ntohs(((usn_ip_t *)ti)->ip_len);
len = sizeof (usn_ip_t) + tlen;
ti->ti_next = ti->ti_prev = 0;
ti->ti_x1 = 0;
ti->ti_len = (u_short)tlen;
HTONS(ti->ti_len);
ti->ti_sum = in_cksum(m, len);
if (ti->ti_sum) {
g_tcpstat.tcps_rcvbadsum++;
goto drop;
}
/*
* Check that TCP offset makes sense,
* pull out TCP options and adjust length. XXX
*/
off = ti->ti_off << 2;
if (off < sizeof (struct tcphdr) || off > tlen) {
g_tcpstat.tcps_rcvbadoff++;
goto drop;
}
tlen -= off;
ti->ti_len = tlen;
if (off > sizeof (struct tcphdr)) {
if (m->mlen < sizeof(usn_ip_t) + off) {
if ((m = m_pullup(m, sizeof (usn_ip_t) + off)) == 0) {
g_tcpstat.tcps_rcvshort++;
return;
}
ti = mtod(m, struct tcpiphdr *);
}
optlen = off - sizeof (struct tcphdr);
optp = mtod(m, u_char *) + sizeof (struct tcpiphdr);
// Do quick retrieval of timestamp options ("options
// prediction?"). If timestamp is the only option and it's
// formatted as recommended in RFC 1323 appendix A, we
// quickly get the values now and not bother calling
// tcp_dooptions(), etc.
if ((optlen == TCPOLEN_TSTAMP_APPA ||
(optlen > TCPOLEN_TSTAMP_APPA &&
optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
*(u_int *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
(ti->ti_flags & TH_SYN) == 0) {
ts_present = 1;
ts_val = ntohl(*(u_long *)(optp + 4));
ts_ecr = ntohl(*(u_long *)(optp + 8));
optp = NULL; // we've parsed the options
}
}
tiflags = ti->ti_flags;
// Convert TCP protocol specific fields to host format.
NTOHL(ti->ti_seq);
NTOHL(ti->ti_ack);
NTOHS(ti->ti_win);
NTOHS(ti->ti_urp);
// Locate pcb for segment.
findpcb:
inp = g_tcp_last_inpcb;
if (inp->inp_lport != ti->ti_dport ||
inp->inp_fport != ti->ti_sport ||
inp->inp_faddr.s_addr != ti->ti_src.s_addr ||
inp->inp_laddr.s_addr != ti->ti_dst.s_addr) {
inp = in_pcblookup(&g_tcb, ti->ti_src, ti->ti_sport,
ti->ti_dst, ti->ti_dport, INPLOOKUP_WILDCARD);
if (inp)
g_tcp_last_inpcb = inp;
++g_tcpstat.tcps_pcbcachemiss;
}
// If the state is CLOSED (i.e., TCB does not exist) then
// all data in the incoming segment is discarded.
// If the TCB exists but is in CLOSED state, it is embryonic,
// but should either do a listen or a connect soon.
if (inp == 0)
goto dropwithreset;
tp = intotcpcb(inp);
DEBUG("found inp cb, laddr=%x, lport=%d, faddr=%x,"
" fport=%d, tp_state=%d, tp_flags=%d",
inp->inp_laddr.s_addr,
inp->inp_lport,
inp->inp_faddr.s_addr,
inp->inp_fport, tp->t_state, tp->t_flags);
if (tp == 0)
goto dropwithreset;
if (tp->t_state == TCPS_CLOSED)
goto drop;
// Unscale the window into a 32-bit value.
if ((tiflags & TH_SYN) == 0)
tiwin = ti->ti_win << tp->snd_scale;
else
tiwin = ti->ti_win;
so = inp->inp_socket;
DEBUG("socket info, options=%x", so->so_options);
if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
if (so->so_options & SO_DEBUG) {
ostate = tp->t_state;
g_tcp_saveti = *ti;
}
if (so->so_options & SO_ACCEPTCONN) {
if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
// Note: dropwithreset makes sure we don't
// send a reset in response to a RST.
if (tiflags & TH_ACK) {
g_tcpstat.tcps_badsyn++;
goto dropwithreset;
}
DEBUG("SYN is expected, tiflags=%d", tiflags);
goto drop;
}
so = sonewconn(so, 0);
if (so == 0) {
DEBUG("failed to create new connection, tiflags=%d", tiflags);
goto drop;
}
// Mark socket as temporary until we're
// committed to keeping it. The code at
// ``drop'' and ``dropwithreset'' check the
// flag dropsocket to see if the temporary
// socket created here should be discarded.
// We mark the socket as discardable until
// we're committed to it below in TCPS_LISTEN.
dropsocket++;
inp = (struct inpcb *)so->so_pcb;
inp->inp_laddr = ti->ti_dst;
inp->inp_lport = ti->ti_dport;
// BSD >= 4.3
inp->inp_options = ip_srcroute();
tp = intotcpcb(inp);
tp->t_state = TCPS_LISTEN;
// Compute proper scaling value from buffer space
while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
TCP_MAXWIN << tp->request_r_scale < so->so_rcv->sb_hiwat)
tp->request_r_scale++;
}
}
// Segment received on connection.
// Reset idle time and keep-alive timer.
tp->t_idle = 0;
tp->t_timer[TCPT_KEEP] = g_tcp_keepidle;
// Process options if not in LISTEN state,
// else do it below (after getting remote address).
if (optp && tp->t_state != TCPS_LISTEN)
tcp_dooptions(tp, optp, optlen, ti,
&ts_present, &ts_val, &ts_ecr);
// Header prediction: check for the two common cases
// of a uni-directional data xfer. If the packet has
// no control flags, is in-sequence, the window didn't
// change and we're not retransmitting, it's a
// candidate. If the length is zero and the ack moved
// forward, we're the sender side of the xfer. Just
// free the data acked & wake any higher level process
// that was blocked waiting for space. If the length
// is non-zero and the ack didn't move, we're the
// receiver side. If we're getting packets in-order
// (the reassembly queue is empty), add the data to
// the socket buffer and note that we need a delayed ack.
if (tp->t_state == TCPS_ESTABLISHED &&
(tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
(!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) &&
ti->ti_seq == tp->rcv_nxt &&
tiwin && tiwin == tp->snd_wnd &&
tp->snd_nxt == tp->snd_max) {
// If last ACK falls within this segment's sequence numbers,
// record the timestamp.
if ( ts_present && TSTMP_GEQ(ts_val, tp->ts_recent) &&
SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) ){
tp->ts_recent_age = g_tcp_now;
tp->ts_recent = ts_val;
}
if (ti->ti_len == 0) {
if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
tp->snd_cwnd >= tp->snd_wnd) {
// this is a pure ack for outstanding data.
++g_tcpstat.tcps_predack;
if (ts_present)
tcp_xmit_timer(tp, g_tcp_now-ts_ecr+1);
else if (tp->t_rtt &&
SEQ_GT(ti->ti_ack, tp->t_rtseq))
tcp_xmit_timer(tp, tp->t_rtt);
acked = ti->ti_ack - tp->snd_una;
g_tcpstat.tcps_rcvackpack++;
g_tcpstat.tcps_rcvackbyte += acked;
TRACE("drop so_snd buffer, drop_bytes=%d, len=%d",
acked, so->so_snd.sb_cc);
sbdrop(so->so_snd, acked);
tp->snd_una = ti->ti_ack;
usn_free_cmbuf(m);
// If all outstanding data are acked, stop
// retransmit timer, otherwise restart timer
// using current (possibly backed-off) value.
// If process is waiting for space,
// wakeup/selwakeup/signal. If data
// are ready to send, let tcp_output
// decide between more output or persist.
if (tp->snd_una == tp->snd_max)
tp->t_timer[TCPT_REXMT] = 0;
else if (tp->t_timer[TCPT_PERSIST] == 0)
tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
if (so->so_options & SO_DEBUG)
tcp_trace(TA_INPUT, ostate, tp, &g_tcp_saveti, 0);
//if (so->so_snd->sb_flags & SB_NOTIFY) {
// usnet_tcpin_wwakeup(so, USN_TCP_IN, usn_tcpev_sbnotify, 0);
// sowwakeup(so);
//}
// send buffer is available for app thread.
usnet_tcpin_wwakeup(so, USN_TCP_IN, USN_TCPEV_WRITE, 0);
if (so->so_snd->sb_cc)
tcp_output(tp);
return;
}
} else if (ti->ti_ack == tp->snd_una &&
tp->seg_next == (struct tcpiphdr *)tp &&
ti->ti_len <= sbspace(so->so_rcv)) {
// this is a pure, in-sequence data packet
// with nothing on the reassembly queue and
// we have enough buffer space to take it.
++g_tcpstat.tcps_preddat;
tp->rcv_nxt += ti->ti_len;
g_tcpstat.tcps_rcvpack++;
g_tcpstat.tcps_rcvbyte += ti->ti_len;
// Drop TCP, IP headers and TCP options then add data
// to socket buffer.
m->head += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
m->mlen -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
TRACE("add data to rcv buf");
sbappend(so->so_rcv, m);
sorwakeup(so);
// new data is available for app threads.
usnet_tcpin_rwakeup(so, USN_TCP_IN, USN_TCPEV_READ, m);
if (so->so_options & SO_DEBUG) {
TRACE("tcp trace, so_options=%d", so->so_options);
tcp_trace(TA_INPUT, ostate, tp, &g_tcp_saveti, 0);
}
tp->t_flags |= TF_DELACK;
return;
}
}
// Drop TCP, IP headers and TCP options.
m->head += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
m->mlen -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
// Calculate amount of space in receive window,
// and then do TCP input processing.
// Receive window is amount of space in rcv queue,
// but not less than advertised window.
{
int win;
win = sbspace(so->so_rcv);
if (win < 0)
win = 0;
tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
}
switch (tp->t_state) {
// If the state is LISTEN then ignore segment if it contains an RST.
// If the segment contains an ACK then it is bad and send a RST.
// If it does not contain a SYN then it is not interesting; drop it.
// Don't bother responding if the destination was a broadcast.
// Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
// tp->iss, and send a segment:
// <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
// Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
// Fill in remote peer address fields if not previously specified.
// Enter SYN_RECEIVED state, and process any other fields of this
// segment in this state.
case TCPS_LISTEN: {
usn_mbuf_t *am;
struct usn_sockaddr_in *sin;
if (tiflags & TH_RST)
goto drop;
if (tiflags & TH_ACK)
goto dropwithreset;
if ((tiflags & TH_SYN) == 0)
goto drop;
// RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
// in_broadcast() should never return true on a received
// packet with M_BCAST not set.
//if (m->m_flags & (M_BCAST|M_MCAST) ||
// IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
// goto drop;
am = usn_get_mbuf(0, BUF_MSIZE, 0); // XXX: the size!
if (am == NULL)
goto drop;
am->mlen = sizeof (struct usn_sockaddr_in);
sin = mtod(am, struct usn_sockaddr_in *);
sin->sin_family = AF_INET;
sin->sin_len = sizeof(*sin);
sin->sin_addr = ti->ti_src;
sin->sin_port = ti->ti_sport;
bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
laddr = inp->inp_laddr;
if (inp->inp_laddr.s_addr == USN_INADDR_ANY)
inp->inp_laddr = ti->ti_dst;
if (in_pcbconnect(inp, am)) {
inp->inp_laddr = laddr;
usn_free_mbuf(am);
goto drop;
}
usn_free_mbuf(am);
tp->t_template = tcp_template(tp);
if (tp->t_template == 0) {
tp = tcp_drop(tp, ENOBUFS);
dropsocket = 0; // socket is already gone
goto drop;
}
if (optp)
tcp_dooptions(tp, optp, optlen, ti,
&ts_present, &ts_val, &ts_ecr);
if (iss)
tp->iss = iss;
else
tp->iss = g_tcp_iss;
g_tcp_iss += TCP_ISSINCR/4;
tp->irs = ti->ti_seq;
tcp_sendseqinit(tp);
tcp_rcvseqinit(tp);
tp->t_flags |= TF_ACKNOW;
TRACE("change tcp state to TCPS_SYN_RECEIVED, state=%d, tp_flags=%d",
tp->t_state, tp->t_flags);
tp->t_state = TCPS_SYN_RECEIVED;
// tcp event
usnet_tcpin_ewakeup(so, USN_TCP_IN, USN_TCPST_SYN_RECEIVED, 0);
tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
dropsocket = 0; // committed to socket
g_tcpstat.tcps_accepts++;
goto trimthenstep6;
}
// If the state is SYN_SENT:
// if seg contains an ACK, but not for our SYN, drop the input.
// if seg contains a RST, then drop the connection.
// if seg does not contain SYN, then drop it.
// Otherwise this is an acceptable SYN segment
// initialize tp->rcv_nxt and tp->irs
// if seg contains ack then advance tp->snd_una
// if SYN has been acked change to ESTABLISHED else SYN_RCVD state
// arrange for segment to be acked (eventually)
// continue processing rest of data/controls, beginning with URG
case TCPS_SYN_SENT:
if ((tiflags & TH_ACK) &&
(SEQ_LEQ(ti->ti_ack, tp->iss) ||
SEQ_GT(ti->ti_ack, tp->snd_max)))
goto dropwithreset;
if (tiflags & TH_RST) {
if (tiflags & TH_ACK)
tp = tcp_drop(tp, ECONNREFUSED);
goto drop;
}
if ((tiflags & TH_SYN) == 0)
goto drop;
if (tiflags & TH_ACK) {
tp->snd_una = ti->ti_ack;
if (SEQ_LT(tp->snd_nxt, tp->snd_una))
tp->snd_nxt = tp->snd_una;
tp->t_timer[TCPT_REXMT] = 0;
}
tp->irs = ti->ti_seq;
tcp_rcvseqinit(tp);
tp->t_flags |= TF_ACKNOW;
TRACE("ack now, tp flags=%d", tp->t_flags);
// XXX: remove second test.
if (tiflags & TH_ACK /*&& SEQ_GT(tp->snd_una, tp->iss)*/) {
g_tcpstat.tcps_connects++;
soisconnected(so);
TRACE("change tcp state to TCPS_ESTABLISHED,"
" state=%d, tp_flags=%d", tp->t_state, tp->t_flags);
tp->t_state = TCPS_ESTABLISHED;
// Do window scaling on this connection?
if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
(TF_RCVD_SCALE|TF_REQ_SCALE)) {
tp->snd_scale = tp->requested_s_scale;
tp->rcv_scale = tp->request_r_scale;
}
tcp_reass(tp, (struct tcpiphdr *)0, (usn_mbuf_t *)0);
// if we didn't have to retransmit the SYN,
// use its rtt as our initial srtt & rtt var.
if (tp->t_rtt)
tcp_xmit_timer(tp, tp->t_rtt);
} else {
TRACE("change tcp state to TCPS_SYN_RECEIVED, state=%d, tp_flags=%d",
tp->t_state, tp->t_flags);
tp->t_state = TCPS_SYN_RECEIVED;
// tcp event
usnet_tcpin_ewakeup(so, USN_TCP_IN, USN_TCPST_SYN_RECEIVED, 0);
}
trimthenstep6:
// Advance ti->ti_seq to correspond to first data byte.
// If data, trim to stay within window,
// dropping FIN if necessary.
ti->ti_seq++;
if (ti->ti_len > tp->rcv_wnd) {
todrop = ti->ti_len - tp->rcv_wnd;
m_adj(m, -todrop);
ti->ti_len = tp->rcv_wnd;
tiflags &= ~TH_FIN;
g_tcpstat.tcps_rcvpackafterwin++;
g_tcpstat.tcps_rcvbyteafterwin += todrop;
}
tp->snd_wl1 = ti->ti_seq - 1;
tp->rcv_up = ti->ti_seq;
goto step6;
}
// States other than LISTEN or SYN_SENT.
// First check timestamp, if present.
// Then check that at least some bytes of segment are within
// receive window. If segment begins before rcv_nxt,
// drop leading data (and SYN); if nothing left, just ack.
//
// RFC 1323 PAWS: If we have a timestamp reply on this segment
// and it's less than ts_recent, drop it.
if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent &&
TSTMP_LT(ts_val, tp->ts_recent)) {
// Check to see if ts_recent is over 24 days old.
if ((int)(g_tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
// Invalidate ts_recent. If this segment updates
// ts_recent, the age will be reset later and ts_recent
// will get a valid value. If it does not, setting
// ts_recent to zero will at least satisfy the
// requirement that zero be placed in the timestamp
// echo reply when ts_recent isn't valid. The
// age isn't reset until we get a valid ts_recent
// because we don't want out-of-order segments to be
// dropped when ts_recent is old.
tp->ts_recent = 0;
} else {
g_tcpstat.tcps_rcvduppack++;
g_tcpstat.tcps_rcvdupbyte += ti->ti_len;
g_tcpstat.tcps_pawsdrop++;
goto dropafterack;
}
}
todrop = tp->rcv_nxt - ti->ti_seq;
if (todrop > 0) {
if (tiflags & TH_SYN) {
tiflags &= ~TH_SYN;
ti->ti_seq++;
if (ti->ti_urp > 1)
ti->ti_urp--;
else
tiflags &= ~TH_URG;
todrop--;
}
if ( todrop >= ti->ti_len ||
( todrop == ti->ti_len && (tiflags & TH_FIN ) == 0 ) ) {
// Any valid FIN must be to the left of the window.
// At this point the FIN must be a duplicate or
// out of sequence; drop it.
tiflags &= ~TH_FIN;
// Send an ACK to resynchronize and drop any data
// But keep on processing for RST or ACK.
tp->t_flags |= TF_ACKNOW;
TRACE("send ack now to resync, tp_flags=%d", tp->t_flags);
todrop = ti->ti_len;
g_tcpstat.tcps_rcvdupbyte += ti->ti_len;
g_tcpstat.tcps_rcvduppack++;
} else {
g_tcpstat.tcps_rcvpartduppack++;
g_tcpstat.tcps_rcvpartdupbyte += ti->ti_len;
}
m_adj(m, todrop);
ti->ti_seq += todrop;
ti->ti_len -= todrop;
if (ti->ti_urp > todrop)
ti->ti_urp -= todrop;
else {
tiflags &= ~TH_URG;
ti->ti_urp = 0;
}
}
// If new data are received on a connection after the
// user processes are gone, then RST the other end.
if ((so->so_state & USN_NOFDREF) &&
tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
tp = tcp_close(tp);
g_tcpstat.tcps_rcvafterclose++;
goto dropwithreset;
}
// If segment ends after window, drop trailing data
// (and PUSH and FIN); if nothing left, just ACK.
todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
if (todrop > 0) {
g_tcpstat.tcps_rcvpackafterwin++;
if (todrop >= ti->ti_len) {
g_tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
// If a new connection request is received
// while in TIME_WAIT, drop the old connection
// and start over if the sequence numbers
// are above the previous ones.
if (tiflags & TH_SYN &&
tp->t_state == TCPS_TIME_WAIT &&
SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
iss = tp->snd_nxt + TCP_ISSINCR;
tp = tcp_close(tp);
goto findpcb;
}
// If window is closed can only take segments at
// window edge, and have to drop data and PUSH from
// incoming segments. Continue processing, but
// remember to ack. Otherwise, drop segment
// and ack.
if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
tp->t_flags |= TF_ACKNOW;
g_tcpstat.tcps_rcvwinprobe++;
} else
goto dropafterack;
} else
g_tcpstat.tcps_rcvbyteafterwin += todrop;
m_adj(m, -todrop);
ti->ti_len -= todrop;
tiflags &= ~(TH_PUSH|TH_FIN);
}
// check valid timestamp. Replace code above.
if (ts_present && TSTMP_GEQ(ts_val, tp->ts_recent) &&
SEQ_LEQ(ti->ti_seq, tp->last_ack_sent) ) {
tp->ts_recent_age = g_tcp_now;
tp->ts_recent = ts_val;
}
// If the RST bit is set examine the state:
// SYN_RECEIVED STATE:
// If passive open, return to LISTEN state.
// If active open, inform user that connection was refused.
// ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
// Inform user that connection was reset, and close tcb.
// CLOSING, LAST_ACK, TIME_WAIT STATES
// Close the tcb.
if (tiflags&TH_RST) switch (tp->t_state) {
case TCPS_SYN_RECEIVED:
so->so_error = ECONNREFUSED;
goto close;
case TCPS_ESTABLISHED:
case TCPS_FIN_WAIT_1:
case TCPS_FIN_WAIT_2:
case TCPS_CLOSE_WAIT:
so->so_error = ECONNRESET;
close:
DEBUG("change tcp state to TCPS_CLOSED, state=%d", tp->t_state);
tp->t_state = TCPS_CLOSED;
// tcp event
usnet_tcpin_ewakeup(so, USN_TCP_IN, USN_TCPST_CLOSED, 0);
g_tcpstat.tcps_drops++;
tp = tcp_close(tp);
goto drop;
case TCPS_CLOSING:
case TCPS_LAST_ACK:
case TCPS_TIME_WAIT:
tp = tcp_close(tp);
goto drop;
}
// If a SYN is in the window, then this is an
// error and we send an RST and drop the connection.
if (tiflags & TH_SYN) {
tp = tcp_drop(tp, ECONNRESET);
goto dropwithreset;
}
// If the ACK bit is off we drop the segment and return.
if ((tiflags & TH_ACK) == 0)
goto drop;
// Ack processing.
switch (tp->t_state) {
// In SYN_RECEIVED state if the ack ACKs our SYN then enter
// ESTABLISHED state and continue processing, otherwise
// send an RST.
case TCPS_SYN_RECEIVED:
if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
SEQ_GT(ti->ti_ack, tp->snd_max))
goto dropwithreset;
g_tcpstat.tcps_connects++;
DEBUG("change tcp state to TCPS_ESTABLISHED, state=%d", tp->t_state);
tp->t_state = TCPS_ESTABLISHED;
soisconnected(so);
// Do window scaling?
if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
(TF_RCVD_SCALE|TF_REQ_SCALE)) {
tp->snd_scale = tp->requested_s_scale;
tp->rcv_scale = tp->request_r_scale;
}
tcp_reass(tp, (struct tcpiphdr *)0, (usn_mbuf_t *)0);
tp->snd_wl1 = ti->ti_seq - 1;
// fall into ...
// In ESTABLISHED state: drop duplicate ACKs; ACK out of range
// ACKs. If the ack is in the range
// tp->snd_una < ti->ti_ack <= tp->snd_max
// then advance tp->snd_una to ti->ti_ack and drop
// data from the retransmission queue. If this ACK reflects
// more up to date window information we update our window information.
case TCPS_ESTABLISHED:
case TCPS_FIN_WAIT_1:
case TCPS_FIN_WAIT_2:
case TCPS_CLOSE_WAIT:
case TCPS_CLOSING:
case TCPS_LAST_ACK:
case TCPS_TIME_WAIT:
if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
g_tcpstat.tcps_rcvdupack++;
// If we have outstanding data (other than
// a window probe), this is a completely
// duplicate ack (ie, window info didn't
// change), the ack is the biggest we've
// seen and we've seen exactly our rexmt
// threshhold of them, assume a packet
// has been dropped and retransmit it.
// Kludge snd_nxt & the congestion
// window so we send only this one
// packet.
//
// We know we're losing at the current
// window size so do congestion avoidance
// (set ssthresh to half the current window
// and pull our congestion window back to
// the new ssthresh).
//
// Dup acks mean that packets have left the
// network (they're now cached at the receiver)
// so bump cwnd by the amount in the receiver
// to keep a constant cwnd packets in the
// network.
if (tp->t_timer[TCPT_REXMT] == 0 ||
ti->ti_ack != tp->snd_una)
tp->t_dupacks = 0;
else if (++tp->t_dupacks == g_tcprexmtthresh) {
// congestion avoidance
tcp_seq onxt = tp->snd_nxt;
u_int win =
min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
if (win < 2)
win = 2;
tp->snd_ssthresh = win * tp->t_maxseg;
tp->t_timer[TCPT_REXMT] = 0;
tp->t_rtt = 0;
tp->snd_nxt = ti->ti_ack;
tp->snd_cwnd = tp->t_maxseg;
tcp_output(tp);
tp->snd_cwnd = tp->snd_ssthresh +
tp->t_maxseg * tp->t_dupacks;
if (SEQ_GT(onxt, tp->snd_nxt))
tp->snd_nxt = onxt;
goto drop;
} else if (tp->t_dupacks > g_tcprexmtthresh) {
tp->snd_cwnd += tp->t_maxseg;
tcp_output(tp);
goto drop;
}
} else
tp->t_dupacks = 0;
break;
}
// If the congestion window was inflated to account
// for the other side's cached packets, retract it.
if (tp->t_dupacks > g_tcprexmtthresh &&
tp->snd_cwnd > tp->snd_ssthresh)
tp->snd_cwnd = tp->snd_ssthresh;
tp->t_dupacks = 0;
if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
g_tcpstat.tcps_rcvacktoomuch++;
goto dropafterack;
}
acked = ti->ti_ack - tp->snd_una;
g_tcpstat.tcps_rcvackpack++;
g_tcpstat.tcps_rcvackbyte += acked;
// If we have a timestamp reply, update smoothed
// round trip time. If no timestamp is present but
// transmit timer is running and timed sequence
// number was acked, update smoothed round trip time.
// Since we now have an rtt measurement, cancel the
// timer backoff (cf., Phil Karn's retransmit alg.).
// Recompute the initial retransmit timer.
if (ts_present)
tcp_xmit_timer(tp, g_tcp_now-ts_ecr+1);
else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
tcp_xmit_timer(tp,tp->t_rtt);
// If all outstanding data is acked, stop retransmit
// timer and remember to restart (more output or persist).
// If there is more data to be acked, restart retransmit
// timer, using current (possibly backed-off) value.
if (ti->ti_ack == tp->snd_max) {
tp->t_timer[TCPT_REXMT] = 0;
DEBUG("change needoutput to 1");
needoutput = 1;
tp->t_flags |= TF_NEEDOUTPUT;
} else if (tp->t_timer[TCPT_PERSIST] == 0)
tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
// When new data is acked, open the congestion window.
// If the window gives us less than ssthresh packets
// in flight, open exponentially (maxseg per packet).
// Otherwise open linearly: maxseg per window
// (maxseg * (maxseg / cwnd) per packet).
{
u_int cw = tp->snd_cwnd;
u_int incr = tp->t_maxseg;
if (cw > tp->snd_ssthresh)
incr = incr * incr / cw;
tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
}
if (acked > so->so_snd->sb_cc) {
tp->snd_wnd -= so->so_snd->sb_cc;
DEBUG("drop all so_snd buffer, drop_bytes=%d, acked=%d",
so->so_snd->sb_cc, acked);
sbdrop(so->so_snd, (int)so->so_snd->sb_cc);
ourfinisacked = 1;
} else {
DEBUG("drop so_snd buffer, drop_bytes=%d, len=%d", acked, so->so_snd->sb_cc);
sbdrop(so->so_snd, acked);
tp->snd_wnd -= acked;
ourfinisacked = 0;
}
//if (so->so_snd->sb_flags & SB_NOTIFY) {
sowwakeup(so);
usnet_tcpin_wwakeup(so, USN_TCP_IN, USN_TCPEV_WRITE, 0);
//}
tp->snd_una = ti->ti_ack;
if (SEQ_LT(tp->snd_nxt, tp->snd_una))
tp->snd_nxt = tp->snd_una;
switch (tp->t_state) {
// In FIN_WAIT_1 STATE in addition to the processing
// for the ESTABLISHED state if our FIN is now acknowledged
// then enter FIN_WAIT_2.
case TCPS_FIN_WAIT_1:
if (ourfinisacked) {
// If we can't receive any more
// data, then closing user can proceed.
// Starting the timer is contrary to the
// specification, but if we don't get a FIN
// we'll hang forever.
if (so->so_state & USN_CANTRCVMORE) {
soisdisconnected(so);
tp->t_timer[TCPT_2MSL] = g_tcp_maxidle;
}
DEBUG("change tcp state to TCPS_FIN_WAIT_2, state=%d", tp->t_state);
tp->t_state = TCPS_FIN_WAIT_2;
usnet_tcpin_ewakeup(so, USN_TCP_IN, USN_TCPST_FIN_WAIT2, 0);
}
break;
// In CLOSING STATE in addition to the processing for
// the ESTABLISHED state if the ACK acknowledges our FIN
// then enter the TIME-WAIT state, otherwise ignore
// the segment.
case TCPS_CLOSING:
if (ourfinisacked) {
DEBUG("change tcp state to TCPS_TIME_WAIT, state=%d", tp->t_state);
tp->t_state = TCPS_TIME_WAIT;
usnet_tcpin_ewakeup(so, USN_TCP_IN, USN_TCPST_TIME_WAIT, 0);
tcp_canceltimers(tp);
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
soisdisconnected(so);
}
break;
// In LAST_ACK, we may still be waiting for data to drain
// and/or to be acked, as well as for the ack of our FIN.
// If our FIN is now acknowledged, delete the TCB,
// enter the closed state and return.
case TCPS_LAST_ACK:
if (ourfinisacked) {
tp = tcp_close(tp);
goto drop;
}
break;
// In TIME_WAIT state the only thing that should arrive
// is a retransmission of the remote FIN. Acknowledge
// it and restart the finack timer.
case TCPS_TIME_WAIT:
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
goto dropafterack;
}
}
step6:
// Update window information.
// Don't look at window if no ACK: TAC's send garbage on first SYN.
if ((tiflags & TH_ACK) &&
(SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
(tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
(tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd) )) )) {
// keep track of pure window updates
if (ti->ti_len == 0 &&
tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
g_tcpstat.tcps_rcvwinupd++;
tp->snd_wnd = tiwin;
tp->snd_wl1 = ti->ti_seq;
tp->snd_wl2 = ti->ti_ack;
if (tp->snd_wnd > tp->max_sndwnd)
tp->max_sndwnd = tp->snd_wnd;
DEBUG("change needoutput to 1");
tp->t_flags |= TF_NEEDOUTPUT;
needoutput = 1;
}
// Process segments with URG.
if ((tiflags & TH_URG) && ti->ti_urp &&
TCPS_HAVERCVDFIN(tp->t_state) == 0) {
// This is a kludge, but if we receive and accept
// random urgent pointers, we'll crash in
// soreceive. It's hard to imagine someone
// actually wanting to send this much urgent data.
if (ti->ti_urp + so->so_rcv->sb_cc > g_sb_max) {
ti->ti_urp = 0; // XXX
tiflags &= ~TH_URG; // XXX
goto dodata; // XXX
}
// If this segment advances the known urgent pointer,
// then mark the data stream. This should not happen
// in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
// a FIN has been received from the remote side.
// In these states we ignore the URG.
//
// According to RFC961 (Assigned Protocols),
// the urgent pointer points to the last octet
// of urgent data. We continue, however,
// to consider it to indicate the first octet
// of data past the urgent section as the original
// spec states (in one of two places).
if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
tp->rcv_up = ti->ti_seq + ti->ti_urp;
so->so_oobmark = so->so_rcv->sb_cc +
(tp->rcv_up - tp->rcv_nxt) - 1;
if (so->so_oobmark == 0)
so->so_state |= USN_RCVATMARK;
sohasoutofband(so);
// send async event to app threads.
usnet_tcpin_ewakeup(so, USN_TCP_IN, USN_TCPEV_OUTOFBOUND, 0);
tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
}
// Remove out of band data so doesn't get presented to user.
// This can happen independent of advancing the URG pointer,
// but if two URG's are pending at once, some out-of-band
// data may creep in... ick.
if (ti->ti_urp <= ti->ti_len
#ifdef SO_OOBINLINE
&& (so->so_options & SO_OOBINLINE) == 0
#endif
)
tcp_pulloutofband(so, ti, m);
} else
// If no out of band data is expected,
// pull receive urgent pointer along
// with the receive window.
if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
tp->rcv_up = tp->rcv_nxt;
dodata: // XXX
#ifdef DUMP_PAYLOAD
DEBUG("Handle data");
dump_chain(m,"tcp");
#endif
// Process the segment text, merging it into the TCP sequencing queue,
// and arranging for acknowledgment of receipt if necessary.
// This process logically involves adjusting tp->rcv_wnd as data
// is presented to the user (this happens in tcp_usrreq.c,
// case PRU_RCVD). If a FIN has already been received on this
// connection then we just ignore the text.
if ((ti->ti_len || (tiflags&TH_FIN)) &&
TCPS_HAVERCVDFIN(tp->t_state) == 0) {
TCP_REASS(tp, ti, m, so, tiflags);
// Note the amount of data that peer has sent into
// our window, in order to estimate the sender's
// buffer size.
len = so->so_rcv->sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
} else {
usn_free_cmbuf(m);
tiflags &= ~TH_FIN;
}
// If FIN is received ACK the FIN and let the user know
// that the connection is closing.
if (tiflags & TH_FIN) {
if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
socantrcvmore(so);
tp->t_flags |= TF_ACKNOW;
TRACE("ack FIN now, tp flags=%d", tp->t_flags);
tp->rcv_nxt++;
}
switch (tp->t_state) {
// In SYN_RECEIVED and ESTABLISHED STATES
// enter the CLOSE_WAIT state.
case TCPS_SYN_RECEIVED:
case TCPS_ESTABLISHED:
TRACE("change tcp state to TCPS_CLOSE_WAIT, state=%d", tp->t_state);
tp->t_state = TCPS_CLOSE_WAIT;
soewakeup(so, 0);
usnet_tcpin_ewakeup(so, USN_TCP_IN, USN_TCPST_CLOSE_WAIT, 0);
break;
// If still in FIN_WAIT_1 STATE FIN has not been acked so
// enter the CLOSING state.
case TCPS_FIN_WAIT_1:
TRACE("change tcp state to TCPS_CLOSING, state=%d", tp->t_state);
tp->t_state = TCPS_CLOSING;
usnet_tcpin_ewakeup(so, USN_TCP_IN, USN_TCPST_CLOSING, 0);
break;
// In FIN_WAIT_2 state enter the TIME_WAIT state,
// starting the time-wait timer, turning off the other
// standard timers.
case TCPS_FIN_WAIT_2:
TRACE("change tcp state to TCPS_TIME_WAIT, state=%d", tp->t_state);
tp->t_state = TCPS_TIME_WAIT;
tcp_canceltimers(tp);
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
soisdisconnected(so);
usnet_tcpin_ewakeup(so, USN_TCP_IN, USN_TCPST_TIME_WAIT, 0);
break;
// In TIME_WAIT state restart the 2 MSL time_wait timer.
case TCPS_TIME_WAIT:
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
break;
}
}
if (so->so_options & SO_DEBUG) {
TRACE("tcp trace, so_options=%d", so->so_options);
tcp_trace(TA_INPUT, ostate, tp, &g_tcp_saveti, 0);
}
// Return any desired output.
//if (needoutput || (tp->t_flags & TF_ACKNOW)){
if (tp->t_flags & TF_NEEDOUTPUT || (tp->t_flags & TF_ACKNOW)){
TRACE("ack now or need to ouput, tp->t_flags=%d", tp->t_flags);
tcp_output(tp);
}
return;
dropafterack:
TRACE("dropafterack");
// Generate an ACK dropping incoming segment if it occupies
// sequence space, where the ACK reflects our state.
if (tiflags & TH_RST)
goto drop;
usn_free_cmbuf(m);
tp->t_flags |= TF_ACKNOW;
TRACE("ack now, tp flags=%d", tp->t_flags);
tcp_output(tp);
return;
dropwithreset:
TRACE("dropwithreset");
// Generate a RST, dropping incoming segment.
// Make ACK acceptable to originator of segment.
// Don't bother to respond if destination was broadcast/multicast.
#define USN_MULTICAST(i) (((u_int)(i) & 0xf0000000) == 0xe0000000)
if ((tiflags & TH_RST) || m->flags & (BUF_BCAST|BUF_MCAST) ||
USN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
goto drop;
if (tiflags & TH_ACK)
tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
else {
if (tiflags & TH_SYN)
ti->ti_len++;
tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
TH_RST|TH_ACK);
}
// destroy temporarily created socket
if (dropsocket)
soabort(so);
return;
drop:
TRACE("drop");
// Drop space held by incoming segment and return.
if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) {
TRACE("tcp trace: drop a socket");
tcp_trace(TA_DROP, ostate, tp, &g_tcp_saveti, 0);
}
usn_free_cmbuf(m);
// destroy temporarily created socket
if (dropsocket)
soabort(so);
return;
}
/*
* To insert a new blk to the array of SACK blk in receiver.
*
* Parameters:
* sack_blk_t *head: pointer to the array of SACK blks.
* tcp_seq begin: starting seq num of the new blk.
* tcp_seq end: ending seq num of the new blk.
* int32_t *num: (referenced) total num of SACK blks on the list.
*/
void
tcp_sack_insert(sack_blk_t *head, tcp_seq begin, tcp_seq end, int32_t *num)
{
int32_t i, j, old_num, new_num;
sack_blk_t tmp[MAX_SACK_BLK - 1];
/* The array is empty, just add the new one. */
if (*num == 0) {
head[0].begin = begin;
head[0].end = end;
*num = 1;
return;
}
/*
* Check for overlap. There are five cases.
*
* 1. there is no overlap with any other SACK blks.
* 2. new SACK blk is completely contained in another blk.
* 3. tail part of new SACK blk overlaps with another blk.
* 4. head part of new SACK blk overlaps with another blk.
* 5. new SACK blk completely contains another blk.
*
* Use tmp to hold old SACK blks. After the loop, copy them back
* to head.
*/
old_num = *num;
if (old_num > MAX_SACK_BLK - 1) {
old_num = MAX_SACK_BLK - 1;
}
new_num = old_num;
j = 0;
for (i = 0; i < old_num; i++) {
if (SEQ_LT(end, head[i].begin) || SEQ_GT(begin, head[i].end)) {
/* Case 1: continue to check. */
tmp[j].begin = head[i].begin;
tmp[j].end = head[i].end;
j++;
continue;
} else if (SEQ_GEQ(begin, head[i].begin) &&
SEQ_LEQ(end, head[i].end)) {
/* Case 2: re-insert the old blk to the head. */
begin = head[i].begin;
end = head[i].end;
} else if (SEQ_LEQ(end, head[i].end) &&
SEQ_GEQ(end, head[i].begin)) {
/*
* Case 3: Extend the new blk, remove the old one
* and continue to check.
*/
end = head[i].end;
} else if (SEQ_GEQ(begin, head[i].begin) &&
SEQ_LEQ(begin, head[i].end)) {
/* Case 4 */
begin = head[i].begin;
}
/*
* Common code for all cases except the first one, which
* copies the original SACK blk into the tmp storage. Other
* cases remove the original SACK blk by not copying into
* tmp storage.
*/
new_num--;
}
head[0].begin = begin;
head[0].end = end;
for (i = 0; i < new_num; i++) {
head[i+1].begin = tmp[i].begin;
head[i+1].end = tmp[i].end;
}
*num = new_num + 1;
}
int
tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
{
struct socket *so = tp->t_inpcb->inp_socket;
struct mbuf *mq, *mp;
int flags, wakeup;
INP_WLOCK_ASSERT(tp->t_inpcb);
/*
* XXX: tcp_reass() is rather inefficient with its data structures
* and should be rewritten (see NetBSD for optimizations).
*/
/*
* Call with th==NULL after become established to
* force pre-ESTABLISHED data up to user socket.
*/
if (th == NULL)
goto present;
M_ASSERTPKTHDR(m);
KASSERT(*tlenp == m->m_pkthdr.len, ("%s: tlenp %u len %u", __func__,
*tlenp, m->m_pkthdr.len));
/*
* Limit the number of segments that can be queued to reduce the
* potential for mbuf exhaustion. For best performance, we want to be
* able to queue a full window's worth of segments. The size of the
* socket receive buffer determines our advertised window and grows
* automatically when socket buffer autotuning is enabled. Use it as the
* basis for our queue limit.
* Always let the missing segment through which caused this queue.
* NB: Access to the socket buffer is left intentionally unlocked as we
* can tolerate stale information here.
*/
if ((th->th_seq != tp->rcv_nxt || !TCPS_HAVEESTABLISHED(tp->t_state)) &&
tp->t_segqlen + m->m_pkthdr.len >= sbspace(&so->so_rcv)) {
char *s;
TCPSTAT_INC(tcps_rcvreassfull);
*tlenp = 0;
if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL,
NULL))) {
log(LOG_DEBUG, "%s; %s: queue limit reached, "
"segment dropped\n", s, __func__);
free(s, M_TCPLOG);
}
m_freem(m);
return (0);
}
/*
* Find a segment which begins after this one does.
*/
mp = NULL;
for (mq = tp->t_segq; mq != NULL; mq = mq->m_nextpkt) {
if (SEQ_GT(M_TCPHDR(mq)->th_seq, th->th_seq))
break;
mp = mq;
}
/*
* If there is a preceding segment, it may provide some of
* our data already. If so, drop the data from the incoming
* segment. If it provides all of our data, drop us.
*/
if (mp != NULL) {
int i;
/* conversion to int (in i) handles seq wraparound */
i = M_TCPHDR(mp)->th_seq + mp->m_pkthdr.len - th->th_seq;
if (i > 0) {
if (i >= *tlenp) {
TCPSTAT_INC(tcps_rcvduppack);
TCPSTAT_ADD(tcps_rcvdupbyte, *tlenp);
m_freem(m);
/*
* Try to present any queued data
* at the left window edge to the user.
* This is needed after the 3-WHS
* completes.
*/
goto present; /* ??? */
}
m_adj(m, i);
*tlenp -= i;
th->th_seq += i;
}
}
tp->t_rcvoopack++;
TCPSTAT_INC(tcps_rcvoopack);
TCPSTAT_ADD(tcps_rcvoobyte, *tlenp);
/*
* While we overlap succeeding segments trim them or,
* if they are completely covered, dequeue them.
*/
while (mq) {
struct mbuf *nq;
int i;
i = (th->th_seq + *tlenp) - M_TCPHDR(mq)->th_seq;
if (i <= 0)
break;
if (i < mq->m_pkthdr.len) {
M_TCPHDR(mq)->th_seq += i;
m_adj(mq, i);
tp->t_segqlen -= i;
break;
}
nq = mq->m_nextpkt;
tp->t_segqlen -= mq->m_pkthdr.len;
m_freem(mq);
if (mp)
mp->m_nextpkt = nq;
else
tp->t_segq = nq;
mq = nq;
}
/*
* Insert the new segment queue entry into place. Try to collapse
* mbuf chains if segments are adjacent.
*/
if (mp) {
if (M_TCPHDR(mp)->th_seq + mp->m_pkthdr.len == th->th_seq)
m_catpkt(mp, m);
else {
m->m_nextpkt = mp->m_nextpkt;
mp->m_nextpkt = m;
m->m_pkthdr.pkt_tcphdr = th;
}
} else {
mq = tp->t_segq;
tp->t_segq = m;
if (mq && th->th_seq + *tlenp == M_TCPHDR(mq)->th_seq) {
m->m_nextpkt = mq->m_nextpkt;
mq->m_nextpkt = NULL;
m_catpkt(m, mq);
} else
m->m_nextpkt = mq;
m->m_pkthdr.pkt_tcphdr = th;
}
tp->t_segqlen += *tlenp;
present:
/*
* Present data to user, advancing rcv_nxt through
* completed sequence space.
*/
if (!TCPS_HAVEESTABLISHED(tp->t_state))
return (0);
flags = 0;
wakeup = 0;
SOCKBUF_LOCK(&so->so_rcv);
while ((mq = tp->t_segq) != NULL &&
M_TCPHDR(mq)->th_seq == tp->rcv_nxt) {
tp->t_segq = mq->m_nextpkt;
tp->rcv_nxt += mq->m_pkthdr.len;
tp->t_segqlen -= mq->m_pkthdr.len;
flags = M_TCPHDR(mq)->th_flags & TH_FIN;
if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
m_freem(mq);
else {
mq->m_nextpkt = NULL;
sbappendstream_locked(&so->so_rcv, mq, 0);
wakeup = 1;
}
}
ND6_HINT(tp);
if (wakeup)
sorwakeup_locked(so);
else
SOCKBUF_UNLOCK(&so->so_rcv);
return (flags);
}
static int
tcp_reass(register struct tcpcb *tp, register struct tcpiphdr *ti,
struct mbuf *m)
{
register struct tcpiphdr *q;
struct socket *so = tp->t_socket;
int flags;
/*
* Call with ti==NULL after become established to
* force pre-ESTABLISHED data up to user socket.
*/
if (ti == NULL)
goto present;
/*
* Find a segment which begins after this one does.
*/
for (q = tcpfrag_list_first(tp); !tcpfrag_list_end(q, tp);
q = tcpiphdr_next(q))
if (SEQ_GT(q->ti_seq, ti->ti_seq))
break;
/*
* If there is a preceding segment, it may provide some of
* our data already. If so, drop the data from the incoming
* segment. If it provides all of our data, drop us.
*/
if (!tcpfrag_list_end(tcpiphdr_prev(q), tp)) {
register int i;
q = tcpiphdr_prev(q);
/* conversion to int (in i) handles seq wraparound */
i = q->ti_seq + q->ti_len - ti->ti_seq;
if (i > 0) {
if (i >= ti->ti_len) {
m_free(m);
/*
* Try to present any queued data
* at the left window edge to the user.
* This is needed after the 3-WHS
* completes.
*/
goto present; /* ??? */
}
m_adj(m, i);
ti->ti_len -= i;
ti->ti_seq += i;
}
q = tcpiphdr_next(q);
}
ti->ti_mbuf = m;
/*
* While we overlap succeeding segments trim them or,
* if they are completely covered, dequeue them.
*/
while (!tcpfrag_list_end(q, tp)) {
register int i = (ti->ti_seq + ti->ti_len) - q->ti_seq;
if (i <= 0)
break;
if (i < q->ti_len) {
q->ti_seq += i;
q->ti_len -= i;
m_adj(q->ti_mbuf, i);
break;
}
q = tcpiphdr_next(q);
m = tcpiphdr_prev(q)->ti_mbuf;
remque(tcpiphdr2qlink(tcpiphdr_prev(q)));
m_free(m);
}
/*
* Stick new segment in its place.
*/
insque(tcpiphdr2qlink(ti), tcpiphdr2qlink(tcpiphdr_prev(q)));
present:
/*
* Present data to user, advancing rcv_nxt through
* completed sequence space.
*/
if (!TCPS_HAVEESTABLISHED(tp->t_state))
return (0);
ti = tcpfrag_list_first(tp);
if (tcpfrag_list_end(ti, tp) || ti->ti_seq != tp->rcv_nxt)
return (0);
if (tp->t_state == TCPS_SYN_RECEIVED && ti->ti_len)
return (0);
do {
tp->rcv_nxt += ti->ti_len;
flags = ti->ti_flags & TH_FIN;
remque(tcpiphdr2qlink(ti));
m = ti->ti_mbuf;
ti = tcpiphdr_next(ti);
if (so->so_state & SS_FCANTSENDMORE)
m_free(m);
else {
if (so->so_emu) {
if (tcp_emu(so,m)) sbappend(so, m);
} else
sbappend(so, m);
}
} while (ti != (struct tcpiphdr *)tp && ti->ti_seq == tp->rcv_nxt);
return (flags);
}
static int tcpup_state_send(struct tcpup_info *upp, struct tcpiphdr *tcp, size_t dlen)
{
int xflags = 0;
if (tcp->th_flags & TH_RST) {
upp->t_state = TCPS_CLOSED;
return 0;
}
if (upp->x_state != upp->t_state
&& (tcp->th_flags & TH_ACK)
&& SEQ_GEQ(htonl(tcp->th_ack), upp->rcv_una)) {
tcp_state_update(upp, upp->x_state);
upp->t_state = upp->x_state;
}
switch (upp->t_state) {
case TCPS_CLOSED:
xflags = TH_SYN| TH_ACK;
if ((tcp->th_flags & xflags) == TH_SYN) {
upp->snd_nxt = htonl(tcp->th_seq) + 1;
upp->snd_max = htonl(tcp->th_seq) + 1;
upp->snd_una = htonl(tcp->th_seq) + 1;
tcp_state_update(upp, TCPS_SYN_SENT);
return 0;
}
break;
case TCPS_SYN_RECEIVED:
assert((tcp->th_flags & TH_FIN) != TH_FIN);
xflags = TH_SYN| TH_ACK;
if ((tcp->th_flags & xflags) == TH_ACK
&& SEQ_GT(htonl(tcp->th_seq), upp->snd_nxt)) {
tcp_state_update(upp, upp->x_state);
return 0;
}
break;
case TCPS_ESTABLISHED:
if ((tcp->th_flags & TH_FIN) == TH_FIN) {
upp->snd_nxt = htonl(tcp->th_seq) + dlen + 1;
tcp_state_update(upp, TCPS_FIN_WAIT_1);
return 0;
}
break;
case TCPS_CLOSE_WAIT:
if ((tcp->th_flags & TH_FIN) == TH_FIN) {
upp->snd_nxt = htonl(tcp->th_seq) + dlen + 1;
tcp_state_update(upp, TCPS_LAST_ACK);
return 0;
}
break;
case TCPS_FIN_WAIT_1:
xflags = TH_FIN| TH_ACK;
if ((tcp->th_flags & xflags) == TH_ACK) {
tcp_state_update(upp, upp->x_state);
return 0;
}
break;
}
if (dlen > 0) {
upp->snd_nxt = htonl(tcp->th_seq) + dlen;
if (SEQ_GT(upp->snd_nxt, upp->snd_max)) {
/* update snd max to nxt */
upp->snd_max = upp->snd_nxt;
}
}
return 0;
}
enum CT_UPDATE_RES
OvsConntrackUpdateTcpEntry(OVS_CT_ENTRY* conn_,
const TCPHdr *tcp,
PNET_BUFFER_LIST nbl,
BOOLEAN reply,
UINT64 now)
{
struct conn_tcp *conn = OvsCastConntrackEntryToTcpEntry(conn_);
/* The peer that sent 'pkt' */
struct tcp_peer *src = &conn->peer[reply ? 1 : 0];
/* The peer that should receive 'pkt' */
struct tcp_peer *dst = &conn->peer[reply ? 0 : 1];
uint8_t sws = 0, dws = 0;
UINT16 tcp_flags = ntohs(tcp->flags);
uint16_t win = ntohs(tcp->window);
uint32_t ack, end, seq, orig_seq;
uint32_t p_len = OvsGetTcpPayloadLength(nbl);
int ackskew;
if (OvsCtInvalidTcpFlags(tcp_flags)) {
return CT_UPDATE_INVALID;
}
if (((tcp_flags & (TCP_SYN|TCP_ACK)) == TCP_SYN)
&& dst->state >= CT_DPIF_TCPS_FIN_WAIT_2
&& src->state >= CT_DPIF_TCPS_FIN_WAIT_2) {
src->state = dst->state = CT_DPIF_TCPS_CLOSED;
return CT_UPDATE_NEW;
}
if (src->wscale & CT_WSCALE_FLAG
&& dst->wscale & CT_WSCALE_FLAG
&& !(tcp_flags & TCP_SYN)) {
sws = src->wscale & CT_WSCALE_MASK;
dws = dst->wscale & CT_WSCALE_MASK;
} else if (src->wscale & CT_WSCALE_UNKNOWN
&& dst->wscale & CT_WSCALE_UNKNOWN
&& !(tcp_flags & TCP_SYN)) {
sws = TCP_MAX_WSCALE;
dws = TCP_MAX_WSCALE;
}
/*
* Sequence tracking algorithm from Guido van Rooij's paper:
* http://www.madison-gurkha.com/publications/tcp_filtering/
* tcp_filtering.ps
*/
orig_seq = seq = ntohl(tcp->seq);
if (src->state < CT_DPIF_TCPS_SYN_SENT) {
/* First packet from this end. Set its state */
ack = ntohl(tcp->ack_seq);
end = seq + p_len;
if (tcp_flags & TCP_SYN) {
end++;
if (dst->wscale & CT_WSCALE_FLAG) {
src->wscale = OvsTcpGetWscale(tcp);
if (src->wscale & CT_WSCALE_FLAG) {
/* Remove scale factor from initial window */
sws = src->wscale & CT_WSCALE_MASK;
win = DIV_ROUND_UP((uint32_t) win, 1 << sws);
dws = dst->wscale & CT_WSCALE_MASK;
} else {
/* fixup other window */
dst->max_win <<= dst->wscale & CT_WSCALE_MASK;
/* in case of a retrans SYN|ACK */
dst->wscale = 0;
}
}
}
if (tcp_flags & TCP_FIN) {
end++;
}
src->seqlo = seq;
src->state = CT_DPIF_TCPS_SYN_SENT;
/*
* May need to slide the window (seqhi may have been set by
* the crappy stack check or if we picked up the connection
* after establishment)
*/
if (src->seqhi == 1 ||
SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) {
src->seqhi = end + MAX(1, dst->max_win << dws);
}
if (win > src->max_win) {
src->max_win = win;
}
} else {
ack = ntohl(tcp->ack_seq);
end = seq + p_len;
if (tcp_flags & TCP_SYN) {
end++;
}
if (tcp_flags & TCP_FIN) {
end++;
}
}
if ((tcp_flags & TCP_ACK) == 0) {
/* Let it pass through the ack skew check */
ack = dst->seqlo;
} else if ((ack == 0
&& (tcp_flags & (TCP_ACK|TCP_RST)) == (TCP_ACK|TCP_RST))
/* broken tcp stacks do not set ack */) {
/* Many stacks (ours included) will set the ACK number in an
* FIN|ACK if the SYN times out -- no sequence to ACK. */
ack = dst->seqlo;
}
if (seq == end) {
/* Ease sequencing restrictions on no data packets */
seq = src->seqlo;
end = seq;
}
ackskew = dst->seqlo - ack;
#define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */
if (SEQ_GEQ(src->seqhi, end)
/* Last octet inside other's window space */
&& SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws))
/* Retrans: not more than one window back */
&& (ackskew >= -MAXACKWINDOW)
/* Acking not more than one reassembled fragment backwards */
&& (ackskew <= (MAXACKWINDOW << sws))
/* Acking not more than one window forward */
&& ((tcp_flags & TCP_RST) == 0 || orig_seq == src->seqlo
|| (orig_seq == src->seqlo + 1)
|| (orig_seq + 1 == src->seqlo))) {
/* Require an exact/+1 sequence match on resets when possible */
/* update max window */
if (src->max_win < win) {
src->max_win = win;
}
/* synchronize sequencing */
if (SEQ_GT(end, src->seqlo)) {
src->seqlo = end;
}
/* slide the window of what the other end can send */
if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) {
dst->seqhi = ack + MAX((win << sws), 1);
}
/* update states */
if (tcp_flags & TCP_SYN && src->state < CT_DPIF_TCPS_SYN_SENT) {
src->state = CT_DPIF_TCPS_SYN_SENT;
}
if (tcp_flags & TCP_FIN && src->state < CT_DPIF_TCPS_CLOSING) {
src->state = CT_DPIF_TCPS_CLOSING;
}
if (tcp_flags & TCP_ACK) {
if (dst->state == CT_DPIF_TCPS_SYN_SENT) {
dst->state = CT_DPIF_TCPS_ESTABLISHED;
} else if (dst->state == CT_DPIF_TCPS_CLOSING) {
dst->state = CT_DPIF_TCPS_FIN_WAIT_2;
}
}
if (tcp_flags & TCP_RST) {
src->state = dst->state = CT_DPIF_TCPS_TIME_WAIT;
}
if (src->state >= CT_DPIF_TCPS_FIN_WAIT_2
&& dst->state >= CT_DPIF_TCPS_FIN_WAIT_2) {
OvsConntrackUpdateExpiration(conn, now, 30 * CT_INTERVAL_SEC);
} else if (src->state >= CT_DPIF_TCPS_CLOSING
&& dst->state >= CT_DPIF_TCPS_CLOSING) {
OvsConntrackUpdateExpiration(conn, now, 45 * CT_INTERVAL_SEC);
} else if (src->state < CT_DPIF_TCPS_ESTABLISHED
|| dst->state < CT_DPIF_TCPS_ESTABLISHED) {
OvsConntrackUpdateExpiration(conn, now, 30 * CT_INTERVAL_SEC);
} else if (src->state >= CT_DPIF_TCPS_CLOSING
|| dst->state >= CT_DPIF_TCPS_CLOSING) {
OvsConntrackUpdateExpiration(conn, now, 15 * 60 * CT_INTERVAL_SEC);
} else {
OvsConntrackUpdateExpiration(conn, now, 24 * 60 * 60 * CT_INTERVAL_SEC);
}
} else if ((dst->state < CT_DPIF_TCPS_SYN_SENT
|| dst->state >= CT_DPIF_TCPS_FIN_WAIT_2
|| src->state >= CT_DPIF_TCPS_FIN_WAIT_2)
&& SEQ_GEQ(src->seqhi + MAXACKWINDOW, end)
/* Within a window forward of the originating packet */
&& SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) {
/* Within a window backward of the originating packet */
/*
* This currently handles three situations:
* 1) Stupid stacks will shotgun SYNs before their peer
* replies.
* 2) When PF catches an already established stream (the
* firewall rebooted, the state table was flushed, routes
* changed...)
* 3) Packets get funky immediately after the connection
* closes (this should catch Solaris spurious ACK|FINs
* that web servers like to spew after a close)
*
* This must be a little more careful than the above code
* since packet floods will also be caught here. We don't
* update the TTL here to mitigate the damage of a packet
* flood and so the same code can handle awkward establishment
* and a loosened connection close.
* In the establishment case, a correct peer response will
* validate the connection, go through the normal state code
* and keep updating the state TTL.
*/
/* update max window */
if (src->max_win < win) {
src->max_win = win;
}
/* synchronize sequencing */
if (SEQ_GT(end, src->seqlo)) {
src->seqlo = end;
}
/* slide the window of what the other end can send */
if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) {
dst->seqhi = ack + MAX((win << sws), 1);
}
/*
* Cannot set dst->seqhi here since this could be a shotgunned
* SYN and not an already established connection.
*/
if (tcp_flags & TCP_FIN && src->state < CT_DPIF_TCPS_CLOSING) {
src->state = CT_DPIF_TCPS_CLOSING;
}
if (tcp_flags & TCP_RST) {
src->state = dst->state = CT_DPIF_TCPS_TIME_WAIT;
}
} else {
return CT_UPDATE_INVALID;
}
return CT_UPDATE_VALID;
}
/*
* Returns 1 if the TIME_WAIT state was killed and we should start over,
* looking for a pcb in the listen state. Returns 0 otherwise.
*/
int
tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
struct mbuf *m, int tlen)
{
struct tcptw *tw;
int thflags;
tcp_seq seq;
/* tcbinfo lock required for tcp_twclose(), tcp_tw_2msl_reset(). */
//ScenSim-Port// INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
//ScenSim-Port// INP_WLOCK_ASSERT(inp);
/*
* XXXRW: Time wait state for inpcb has been recycled, but inpcb is
* still present. This is undesirable, but temporarily necessary
* until we work out how to handle inpcb's who's timewait state has
* been removed.
*/
tw = intotw(inp);
if (tw == NULL)
goto drop;
thflags = th->th_flags;
/*
* NOTE: for FIN_WAIT_2 (to be added later),
* must validate sequence number before accepting RST
*/
/*
* If the segment contains RST:
* Drop the segment - see Stevens, vol. 2, p. 964 and
* RFC 1337.
*/
if (thflags & TH_RST)
goto drop;
//ScenSim-Port//#if 0
//ScenSim-Port///* PAWS not needed at the moment */
//ScenSim-Port// /*
//ScenSim-Port// * RFC 1323 PAWS: If we have a timestamp reply on this segment
//ScenSim-Port// * and it's less than ts_recent, drop it.
//ScenSim-Port// */
//ScenSim-Port// if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
//ScenSim-Port// TSTMP_LT(to.to_tsval, tp->ts_recent)) {
//ScenSim-Port// if ((thflags & TH_ACK) == 0)
//ScenSim-Port// goto drop;
//ScenSim-Port// goto ack;
//ScenSim-Port// }
//ScenSim-Port// /*
//ScenSim-Port// * ts_recent is never updated because we never accept new segments.
//ScenSim-Port// */
//ScenSim-Port//#endif
/*
* If a new connection request is received
* while in TIME_WAIT, drop the old connection
* and start over if the sequence numbers
* are above the previous ones.
*/
if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
tcp_twclose(tw, 0);
return (1);
}
/*
* Drop the segment if it does not contain an ACK.
*/
if ((thflags & TH_ACK) == 0)
goto drop;
/*
* Reset the 2MSL timer if this is a duplicate FIN.
*/
if (thflags & TH_FIN) {
seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
if (seq + 1 == tw->rcv_nxt)
tcp_tw_2msl_reset(tw, 1);
}
/*
* Acknowledge the segment if it has data or is not a duplicate ACK.
*/
if (thflags != TH_ACK || tlen != 0 ||
th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
tcp_twrespond(tw, TH_ACK);
drop:
//ScenSim-Port// INP_WUNLOCK(inp);
m_freem(m);
return (0);
}
/* Receive packet */
void
l2tp_ctrl_input(l2tpd *_this, int listener_index, struct sockaddr *peer,
struct sockaddr *sock, void *nat_t_ctx, u_char *pkt, int pktlen)
{
int i, len, offsiz, reqlen, is_ctrl;
uint16_t mestype;
struct l2tp_avp *avp, *avp0;
l2tp_ctrl *ctrl;
l2tp_call *call;
char buf[L2TP_AVP_MAXSIZ], errmsg[256];
time_t curr_time;
u_char *pkt0;
struct l2tp_header hdr;
char hbuf[NI_MAXHOST + NI_MAXSERV + 16];
ctrl = NULL;
curr_time = get_monosec();
pkt0 = pkt;
L2TP_CTRL_ASSERT(peer->sa_family == sock->sa_family);
L2TP_CTRL_ASSERT(peer->sa_family == AF_INET ||
peer->sa_family == AF_INET6)
/*
* Parse L2TP Header
*/
memset(&hdr, 0, sizeof(hdr));
if (pktlen < 2) {
snprintf(errmsg, sizeof(errmsg), "a short packet. "
"length=%d", pktlen);
goto bad_packet;
}
memcpy(&hdr, pkt, 2);
pkt += 2;
if (hdr.ver != L2TP_HEADER_VERSION_RFC2661) {
/* XXX: only RFC2661 is supported */
snprintf(errmsg, sizeof(errmsg),
"Unsupported version at header = %d", hdr.ver);
goto bad_packet;
}
is_ctrl = (hdr.t != 0)? 1 : 0;
/* calc required length */
reqlen = 6; /* for Flags, Tunnel-Id, Session-Id field */
if (hdr.l) reqlen += 2; /* for Length field (opt) */
if (hdr.s) reqlen += 4; /* for Ns, Nr field (opt) */
if (hdr.o) reqlen += 2; /* for Offset Size field (opt) */
if (reqlen > pktlen) {
snprintf(errmsg, sizeof(errmsg),
"a short packet. length=%d", pktlen);
goto bad_packet;
}
if (hdr.l != 0) {
GETSHORT(hdr.length, pkt);
if (hdr.length > pktlen) {
snprintf(errmsg, sizeof(errmsg),
"Actual packet size is smaller than the length "
"field %d < %d", pktlen, hdr.length);
goto bad_packet;
}
pktlen = hdr.length; /* remove trailing trash */
}
GETSHORT(hdr.tunnel_id, pkt);
GETSHORT(hdr.session_id, pkt);
if (hdr.s != 0) {
GETSHORT(hdr.ns, pkt);
GETSHORT(hdr.nr, pkt);
}
if (hdr.o != 0) {
GETSHORT(offsiz, pkt);
if (pktlen < offsiz) {
snprintf(errmsg, sizeof(errmsg),
"offset field is bigger than remaining packet "
"length %d > %d", offsiz, pktlen);
goto bad_packet;
}
pkt += offsiz;
}
L2TP_CTRL_ASSERT(pkt - pkt0 == reqlen);
pktlen -= (pkt - pkt0); /* cut down the length of header */
ctrl = NULL;
memset(buf, 0, sizeof(buf));
mestype = 0;
avp = NULL;
if (is_ctrl) {
avp0 = (struct l2tp_avp *)buf;
avp = avp_find_message_type_avp(avp0, pkt, pktlen);
if (avp != NULL)
mestype = avp->attr_value[0] << 8 | avp->attr_value[1];
}
ctrl = l2tpd_get_ctrl(_this, hdr.tunnel_id);
if (ctrl == NULL) {
/* new control */
if (!is_ctrl) {
snprintf(errmsg, sizeof(errmsg),
"bad data message: tunnelId=%d is not "
"found.", hdr.tunnel_id);
goto bad_packet;
}
if (mestype != L2TP_AVP_MESSAGE_TYPE_SCCRQ) {
snprintf(errmsg, sizeof(errmsg),
"bad control message: tunnelId=%d is not "
"found. mestype=%s", hdr.tunnel_id,
avp_mes_type_string(mestype));
goto bad_packet;
}
if ((ctrl = l2tp_ctrl_create()) == NULL) {
l2tp_ctrl_log(ctrl, LOG_ERR,
"l2tp_ctrl_create() failed: %m");
goto fail;
}
if (l2tp_ctrl_init(ctrl, _this, peer, sock, nat_t_ctx) != 0) {
l2tp_ctrl_log(ctrl, LOG_ERR,
"l2tp_ctrl_start() failed: %m");
goto fail;
}
ctrl->listener_index = listener_index;
l2tp_ctrl_reload(ctrl);
} else {
/*
* treat as an error if src address and port is not
* match. (because it is potentially DoS attach)
*/
int notmatch = 0;
if (ctrl->peer.ss_family != peer->sa_family)
notmatch = 1;
else if (peer->sa_family == AF_INET) {
if (SIN(peer)->sin_addr.s_addr !=
SIN(&ctrl->peer)->sin_addr.s_addr ||
SIN(peer)->sin_port != SIN(&ctrl->peer)->sin_port)
notmatch = 1;
} else if (peer->sa_family == AF_INET6) {
if (!IN6_ARE_ADDR_EQUAL(&(SIN6(peer)->sin6_addr),
&(SIN6(&ctrl->peer)->sin6_addr)) ||
SIN6(peer)->sin6_port !=
SIN6(&ctrl->peer)->sin6_port)
notmatch = 1;
}
if (notmatch) {
snprintf(errmsg, sizeof(errmsg),
"tunnelId=%u is already assigned for %s",
hdr.tunnel_id, addrport_tostring(
(struct sockaddr *)&ctrl->peer,
ctrl->peer.ss_len, hbuf, sizeof(hbuf)));
goto bad_packet;
}
}
ctrl->last_rcv = curr_time;
call = NULL;
if (hdr.session_id != 0) {
/* search l2tp_call by Session ID */
/* linear search is enough for this purpose */
len = slist_length(&ctrl->call_list);
for (i = 0; i < len; i++) {
call = slist_get(&ctrl->call_list, i);
if (call->session_id == hdr.session_id)
break;
call = NULL;
}
}
if (!is_ctrl) {
int delayed = 0;
/* L2TP data */
if (ctrl->state != L2TP_CTRL_STATE_ESTABLISHED) {
l2tp_ctrl_log(ctrl, LOG_WARNING,
"Received Data packet in '%s'",
l2tp_ctrl_state_string(ctrl));
goto fail;
}
if (call == NULL) {
l2tp_ctrl_log(ctrl, LOG_WARNING,
"Received a data packet but it has no call. "
"session_id=%u", hdr.session_id);
goto fail;
}
L2TP_CTRL_DBG((ctrl, DEBUG_LEVEL_2,
"call=%u RECV ns=%u nr=%u snd_nxt=%u rcv_nxt=%u len=%d",
call->id, hdr.ns, hdr.nr, call->snd_nxt, call->rcv_nxt,
pktlen));
if (call->state != L2TP_CALL_STATE_ESTABLISHED){
l2tp_ctrl_log(ctrl, LOG_WARNING,
"Received a data packet but call is not "
"established");
goto fail;
}
if (hdr.s != 0) {
if (SEQ_LT(hdr.ns, call->rcv_nxt)) {
if (SEQ_LT(hdr.ns,
call->rcv_nxt - L2TP_CALL_DELAY_LIMIT)) {
/* sequence number seems to be delayed */
/* XXX: need to log? */
L2TP_CTRL_DBG((ctrl, LOG_DEBUG,
"receive a out of sequence "
"data packet: %u < %u.",
hdr.ns, call->rcv_nxt));
return;
}
delayed = 1;
} else {
call->rcv_nxt = hdr.ns + 1;
}
}
l2tp_call_ppp_input(call, pkt, pktlen, delayed);
return;
}
if (hdr.s != 0) {
L2TP_CTRL_DBG((ctrl, DEBUG_LEVEL_2,
"RECV %s ns=%u nr=%u snd_nxt=%u snd_una=%u rcv_nxt=%u "
"len=%d", (is_ctrl)? "C" : "", hdr.ns, hdr.nr,
ctrl->snd_nxt, ctrl->snd_una, ctrl->rcv_nxt, pktlen));
if (pktlen <= 0)
l2tp_ctrl_log(ctrl, LOG_INFO, "RecvZLB");
if (SEQ_GT(hdr.nr, ctrl->snd_una)) {
if (hdr.nr == ctrl->snd_nxt ||
SEQ_LT(hdr.nr, ctrl->snd_nxt))
ctrl->snd_una = hdr.nr;
else {
l2tp_ctrl_log(ctrl, LOG_INFO,
"Received message has bad Nr field: "
"%u < %u.", hdr.ns, ctrl->snd_nxt);
/* XXX Drop with ZLB? */
goto fail;
}
}
if (l2tp_ctrl_txwin_size(ctrl) <= 0) {
/* no waiting ack */
if (ctrl->hello_wait_ack != 0) {
/*
* Reset Hello state, as an ack for the Hello
* is recived.
*/
ctrl->hello_wait_ack = 0;
ctrl->hello_io_time = curr_time;
}
switch (ctrl->state) {
case L2TP_CTRL_STATE_CLEANUP_WAIT:
l2tp_ctrl_stop(ctrl, 0);
return;
}
}
if (hdr.ns != ctrl->rcv_nxt) {
/* there are remaining packet */
if (l2tp_ctrl_resend_una_packets(ctrl) <= 0) {
/* resend or sent ZLB */
l2tp_ctrl_send_ZLB(ctrl);
}
#ifdef L2TP_CTRL_DEBUG
if (pktlen != 0) { /* not ZLB */
L2TP_CTRL_DBG((ctrl, LOG_DEBUG,
"receive out of sequence %u must be %u. "
"mestype=%s", hdr.ns, ctrl->rcv_nxt,
avp_mes_type_string(mestype)));
}
#endif
return;
}
if (pktlen <= 0)
return; /* ZLB */
if (l2tp_ctrl_txwin_is_full(ctrl)) {
L2TP_CTRL_DBG((ctrl, LOG_DEBUG,
"Received message cannot be handled. "
"Transmission window is full."));
l2tp_ctrl_send_ZLB(ctrl);
return;
}
ctrl->rcv_nxt++;
if (avp == NULL) {
l2tpd_log(_this, LOG_WARNING,
"bad control message: no message-type AVP.");
goto fail;
}
}
/*
* state machine (RFC2661 pp. 56-57)
*/
switch (ctrl->state) {
case L2TP_CTRL_STATE_IDLE:
switch (mestype) {
case L2TP_AVP_MESSAGE_TYPE_SCCRQ:
if (l2tp_ctrl_recv_SCCRQ(ctrl, pkt, pktlen, _this,
peer) == 0) {
/* acceptable */
l2tp_ctrl_send_SCCRP(ctrl);
ctrl->state = L2TP_CTRL_STATE_WAIT_CTL_CONN;
return;
}
/*
* in case un-acceptable, it was already processed
* at l2tcp_ctrl_recv_SCCRQ
*/
return;
case L2TP_AVP_MESSAGE_TYPE_SCCRP:
/*
* RFC specifies that sent of StopCCN in the state,
* However as this implementation only support Passive
* open, this packet will not received.
*/
/* FALLTHROUGH */
case L2TP_AVP_MESSAGE_TYPE_SCCCN:
default:
break;
}
goto fsm_fail;
case L2TP_CTRL_STATE_WAIT_CTL_CONN:
/* Wait-Ctl-Conn */
switch (mestype) {
case L2TP_AVP_MESSAGE_TYPE_SCCCN:
l2tp_ctrl_log(ctrl, LOG_INFO, "RecvSCCN");
if (l2tp_ctrl_send_ZLB(ctrl) == 0) {
ctrl->state = L2TP_CTRL_STATE_ESTABLISHED;
}
return;
case L2TP_AVP_MESSAGE_TYPE_StopCCN:
goto receive_stop_ccn;
case L2TP_AVP_MESSAGE_TYPE_SCCRQ:
case L2TP_AVP_MESSAGE_TYPE_SCCRP:
default:
break;
}
break; /* fsm_fail */
case L2TP_CTRL_STATE_ESTABLISHED:
/* Established */
switch (mestype) {
case L2TP_AVP_MESSAGE_TYPE_SCCCN:
case L2TP_AVP_MESSAGE_TYPE_SCCRQ:
case L2TP_AVP_MESSAGE_TYPE_SCCRP:
break;
receive_stop_ccn:
case L2TP_AVP_MESSAGE_TYPE_StopCCN:
if (l2tp_ctrl_recv_StopCCN(ctrl, pkt, pktlen) == 0) {
if (l2tp_ctrl_resend_una_packets(ctrl) <= 0)
l2tp_ctrl_send_ZLB(ctrl);
l2tp_ctrl_stop(ctrl, 0);
return;
}
l2tp_ctrl_log(ctrl, LOG_ERR, "Received bad StopCCN");
l2tp_ctrl_send_ZLB(ctrl);
l2tp_ctrl_stop(ctrl, 0);
return;
case L2TP_AVP_MESSAGE_TYPE_HELLO:
if (l2tp_ctrl_resend_una_packets(ctrl) <= 0)
l2tp_ctrl_send_ZLB(ctrl);
return;
case L2TP_AVP_MESSAGE_TYPE_CDN:
case L2TP_AVP_MESSAGE_TYPE_ICRP:
case L2TP_AVP_MESSAGE_TYPE_ICCN:
if (call == NULL) {
l2tp_ctrl_log(ctrl, LOG_INFO,
"Unknown call message: %s",
avp_mes_type_string(mestype));
goto fail;
}
/* FALLTHROUGH */
case L2TP_AVP_MESSAGE_TYPE_ICRQ:
l2tp_call_recv_packet(ctrl, call, mestype, pkt,
pktlen);
return;
default:
break;
}
break; /* fsm_fail */
case L2TP_CTRL_STATE_CLEANUP_WAIT:
if (mestype == L2TP_AVP_MESSAGE_TYPE_StopCCN) {
/*
* We left ESTABLISHED state, but the peer sent StopCCN.
*/
goto receive_stop_ccn;
}
break; /* fsm_fail */
}
fsm_fail:
/* state machine error */
l2tp_ctrl_log(ctrl, LOG_WARNING, "Received %s in '%s' state",
avp_mes_type_string(mestype), l2tp_ctrl_state_string(ctrl));
l2tp_ctrl_stop(ctrl, L2TP_STOP_CCN_RCODE_FSM_ERROR);
return;
fail:
if (ctrl != NULL && mestype != 0) {
l2tp_ctrl_log(ctrl, LOG_WARNING, "Received %s in '%s' state",
avp_mes_type_string(mestype), l2tp_ctrl_state_string(ctrl));
l2tp_ctrl_stop(ctrl, L2TP_STOP_CCN_RCODE_GENERAL_ERROR);
}
return;
bad_packet:
l2tpd_log(_this, LOG_INFO, "Received from=%s: %s",
addrport_tostring(peer, peer->sa_len, hbuf, sizeof(hbuf)), errmsg);
return;
}
int
tcp_output(struct tcpcb * tp)
{
struct socket * so = tp->t_inpcb->inp_socket;
int len;
long win;
int off, flags, error;
struct mbuf * m;
struct tcpiphdr * ti;
unsigned optlen = 0;
int idle, sendalot;
struct mbuf * sendm; /* mbuf which contains data to send */
struct mbuf * tcp_mbuf; /* mbuf containing TCP header */
int bufoff; /* offset of data in sendm->m_data */
#ifdef TCP_SACK
int sack_resend;
int sack_hole = 0; /* next sack hole to fill */
if(tp->t_flags & TF_SACKREPLY)
{
/* we are resending based on a received SACK header */
sack_resend = TRUE;
tp->t_flags &= ~TF_SACKREPLY; /* clear flag */
}
else
sack_resend = FALSE;
#endif /* TCP_SACK */
/*
* Determine length of data that should be transmitted,
* and flags that will be used.
* If there is some data or critical controls (SYN, RST)
* to send, then transmit; otherwise, investigate further.
*/
idle = (tp->snd_max == tp->snd_una);
again:
sendalot = 0;
off = (int)(tp->snd_nxt - tp->snd_una);
win = (long)tp->snd_wnd; /* set basic send window */
if (win > (long)tp->snd_cwnd) /* see if we need congestion control */
{
win = (int)(tp->snd_cwnd & ~(ALIGN_TYPE-1)); /* keep data aligned */
}
/*
* If in persist timeout with window of 0, send 1 byte.
* Otherwise, if window is small but nonzero
* and timer expired, we will send what we can
* and go to transmit state.
*/
if (tp->t_force)
{
if (win == 0)
win = 1;
else
{
tp->t_timer[TCPT_PERSIST] = 0;
tp->t_rxtshift = 0;
}
}
#ifdef TCP_SACK
/* See if we need to adjust the offset for a sack resend */
if(sack_resend)
{
off = (int)(tp->sack_hole_start[sack_hole] - tp->snd_una);
/* if this hole's already been acked then punt and move to next hole */
if(off < 0)
{
/* clear out the acked hole */
tp->sack_hole_start[sack_hole] = tp->sack_hole_end[sack_hole] = 0;
/* see if we're done with SACK hole list (2 tests) */
if(++sack_hole >= SACK_BLOCKS)
return 0;
if(tp->sack_hole_start[sack_hole] == tp->sack_hole_end[sack_hole])
return 0;
goto again;
}
tp->snd_nxt = tp->sack_hole_start[sack_hole];
len = (int)(tp->sack_hole_end[sack_hole] - tp->sack_hole_start[sack_hole]);
len = (int)MIN(len, (int)win);
}
else
#endif /* TCP_SACK */
{
/* set length of packets which are not sack resends */
len = (int)MIN(so->so_snd.sb_cc, (unsigned)win) - off;
}
flags = tcp_outflags[tp->t_state];
/* See if we need to build TCP options field. This test should be fast. */
#if (defined(TCP_TIMESTAMP) | defined(TCP_SACK))
if((flags & TH_SYN) ||
/* !!!??? (so->so_options & SO_TIMESTAMP) || */
(tp->t_flags & TF_SACKNOW)
)
{
optlen = bld_options(tp, &tcp_optionbuf[optlen], flags, so);
}
#else
/* If other options not defined this build then don't bother to call bld_options() except
* on SYN packets
*/
if(flags & TH_SYN)
{
optlen = bld_options(tp, &tcp_optionbuf[optlen], flags, so);
}
#endif
if (len < 0)
{
/*
* If FIN has been sent but not acked,
* but we haven't been called to retransmit,
* len will be -1. Otherwise, window shrank
* after we sent into it. If window shrank to 0,
* cancel pending retransmit and pull snd_nxt
* back to (closed) window. We will enter persist
* state below. If the window didn't close completely,
* just wait for an ACK.
*/
len = 0;
if (win == 0)
{
tp->t_timer[TCPT_REXMT] = 0;
tp->snd_nxt = tp->snd_una;
}
}
if (len > (int)tp->t_maxseg)
{
len = tp->t_maxseg;
sendalot = 1;
}
#ifdef IP_V4
#ifdef IP_PMTU
{
int pmtu = tp->t_inpcb->inp_pmtu - 40;
if (len > pmtu)
{
len = pmtu - 40;
sendalot = 1;
}
}
#endif /* IP_PMTU */
/* We don't need a pmtu test for IPv6. V6 code limits t_maxseg to
* the Path MTU, so the test above the v4 ifdef above covers us.
*/
#endif /* IP_V4 */
if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.sb_cc))
flags &= ~TH_FIN;
win = (long)(sbspace(&so->so_rcv));
/*
* If our state indicates that FIN should be sent
* and we have not yet done so, or we're retransmitting the FIN,
* then we need to send.
*/
if ((flags & TH_FIN) &&
(so->so_snd.sb_cc == 0) &&
((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una))
{
goto send;
}
/*
* Send if we owe peer an ACK.
*/
if (tp->t_flags & TF_ACKNOW)
goto send;
if (flags & (TH_SYN|TH_RST))
goto send;
if (SEQ_GT(tp->snd_up, tp->snd_una))
goto send;
/*
* Sender silly window avoidance. If connection is idle
* and can send all data, a maximum segment,
* at least a maximum default-size segment do it,
* or are forced, do it; otherwise don't bother.
* If peer's buffer is tiny, then send
* when window is at least half open.
* If retransmitting (possibly after persist timer forced us
* to send into a small window), then must resend.
*/
if (len)
{
if (len == (int)tp->t_maxseg)
goto send;
if ((idle || tp->t_flags & TF_NODELAY) &&
len + off >= (int)so->so_snd.sb_cc)
{
goto send;
}
if (tp->t_force)
goto send;
if (len >= (int)(tp->max_sndwnd / 2))
goto send;
if (SEQ_LT(tp->snd_nxt, tp->snd_max))
goto send;
}
/*
* Compare available window to amount of window
* known to peer (as advertised window less
* next expected input). If the difference is at least two
* max size segments or at least 35% of the maximum possible
* window, then want to send a window update to peer.
*/
if (win > 0)
{
int adv = (int)win - (int)(tp->rcv_adv - tp->rcv_nxt);
if (so->so_rcv.sb_cc == 0 && adv >= (int)(tp->t_maxseg * 2))
goto send;
if (100 * (u_int)adv / so->so_rcv.sb_hiwat >= 35)
goto send;
}
/*
* TCP window updates are not reliable, rather a polling protocol
* using ``persist'' packets is used to insure receipt of window
* updates. The three ``states'' for the output side are:
* idle not doing retransmits or persists
* persisting to move a small or zero window
* (re)transmitting and thereby not persisting
*
* tp->t_timer[TCPT_PERSIST]
* is set when we are in persist state.
* tp->t_force
* is set when we are called to send a persist packet.
* tp->t_timer[TCPT_REXMT]
* is set when we are retransmitting
* The output side is idle when both timers are zero.
*
* If send window is too small, there is data to transmit, and no
* retransmit or persist is pending, then go to persist state.
* If nothing happens soon, send when timer expires:
* if window is nonzero, transmit what we can,
* otherwise force out a byte.
*/
if (so->so_snd.sb_cc && tp->t_timer[TCPT_REXMT] == 0 &&
tp->t_timer[TCPT_PERSIST] == 0)
{
tp->t_rxtshift = 0;
tcp_setpersist(tp);
}
/*
* No reason to send a segment, just return.
*/
return (0);
send:
ENTER_CRIT_SECTION(tp);
/* Limit send length to the current buffer so as to
* avoid doing the "mbuf shuffle" in m_copy().
*/
bufoff = off;
sendm = so->so_snd.sb_mb;
if (len)
{
/* find mbuf containing data to send (at "off") */
while (sendm) /* loop through socket send list */
{
bufoff -= sendm->m_len;
if (bufoff < 0) /* if off is in this buffer, break */
break;
sendm = sendm->m_next;
}
if (!sendm) { dtrap(); /* shouldn't happen */ }
bufoff += sendm->m_len; /* index to next data to send in msend */
/* if socket has multiple unsent mbufs, set flag for send to loop */
if ((sendm->m_next) && (len > (int)sendm->m_len))
{
flags &= ~TH_FIN; /* don't FIN on segment prior to last */
sendalot = 1; /* set to send more segments */
}
if((flags & TH_FIN) && (so->so_snd.sb_cc > (unsigned)len))
{
/* This can happen on slow links (PPP) which retry the last
* segment - the one with the FIN bit attached to data.
*/
flags &= ~TH_FIN; /* don't FIN on segment prior to last */
}
/* only send the rest of msend */
len = min(len, (int)sendm->m_len);
/* if we're not sending starting at sendm->m_data (in which
* case bufoff != 0), then we will copy the data; else we would
* write IP/TCP headers over sent but un-ack'ed data in sendm.
* Similarly, if sendm->m_data is not aligned with respect to
* sendm->m_base and ALIGN_TYPE, we will copy the data to
* ensure that it (and the then-prepended IP/TCP headers) will
* be aligned according to ALIGN_TYPE.
*/
if ((bufoff != 0) || /* data not front aligned in send mbuf? */
(((sendm->m_data - sendm->m_base) & (ALIGN_TYPE - 1)) != 0))
{
len = min(len, (int)(sendm->m_len - bufoff)); /* limit len again */
/* One more test - if this data is not aligned with the front
* of the m_data buffer then we can't use it in place, else we
* might write the IP/TCP header over data that has not yet
* been acked. In this case we must make sure our send
* fits into a little buffer and send what we can.
*/
if ((len > (int)(lilbufsiz - HDRSLEN)) && /* length is bigger the small buffer? */
(bigfreeq.q_len < 2)) /* and we are low on big buffers */
{
len = lilbufsiz - HDRSLEN;
}
}
}
/* if send data is sufficiently aligned in packet, prepend TCP/IP header
* in the space provided.
*/
if (len && (bufoff == 0) &&
(sendm->pkt->inuse == 1) &&
(((sendm->m_data - sendm->m_base) & (ALIGN_TYPE - 1)) == 0) &&
(optlen == 0))
{
/* get an empty mbuf to "clone" the data */
m = m_getnbuf(MT_TXDATA, 0);
if (!m)
{
EXIT_CRIT_SECTION(tp);
return (ENOBUFS);
}
m->pkt = sendm->pkt; /* copy packet location in new mbuf */
m->pkt->inuse++; /* bump packet's use count */
m->m_base = sendm->m_base; /* clone mbuf members */
m->m_memsz = sendm->m_memsz;
m->m_len = len + TCPIPHDRSZ; /* adjust clone for header */
m->m_data = sendm->m_data - TCPIPHDRSZ;
}
else /* either no data or data is not front aligned in mbuf */
{
/* Grab a header mbuf, attaching a copy of data to be
* transmitted, and initialize the header from
* the template for sends on this connection.
*/
m = m_getwithdata (MT_HEADER, IFNETHDR_SIZE + TCPIPHDRSZ);
if (m ==(struct mbuf *)NULL)
{
EXIT_CRIT_SECTION(tp);
return ENOBUFS;
}
m->m_len = TCPIPHDRSZ;
m->m_data += IFNETHDR_SIZE;/* Move this to sizeof tcpip hdr leave*/
/* 14 bytes for ethernet header */
if (len) /* attach any data to send */
{
m->m_next = m_copy(so->so_snd.sb_mb, off, (int) len);
if (m->m_next == 0)
{
m_freem(m);
EXIT_CRIT_SECTION(tp);
return ENOBUFS;
}
}
}
EXIT_CRIT_SECTION(tp);
if (len)
{
if (tp->t_force && len == 1)
tcpstat.tcps_sndprobe++;
else if (SEQ_LT(tp->snd_nxt, tp->snd_max))
{
tcpstat.tcps_sndrexmitpack++;
tcpstat.tcps_sndrexmitbyte += len;
#ifdef TCP_SACK
if(sack_resend)
tcpstat.tcps_sackresend++;
#endif
}
else
{
tcpstat.tcps_sndpack++;
tcpstat.tcps_sndbyte += len;
}
}
else if (tp->t_flags & TF_ACKNOW)
{
tcpstat.tcps_sndacks++;
}
else if (flags & (TH_SYN|TH_FIN|TH_RST))
tcpstat.tcps_sndctrl++;
else if (SEQ_GT(tp->snd_up, tp->snd_una))
tcpstat.tcps_sndurg++;
else
tcpstat.tcps_sndwinup++;
ti = (struct tcpiphdr *)(m->m_data+sizeof(struct ip)-sizeof(struct ipovly));
if ((char *)ti < m->pkt->nb_buff)
{
panic("tcp_out- packet ptr underflow\n");
}
tcp_mbuf = m; /* flag TCP header mbuf */
#ifdef IP_V6 /* Dual mode code */
if(so->so_domain == AF_INET6)
{
m = mbuf_prepend(m, sizeof(struct ipv6));
if(m == NULL)
{
/* this can happen when we run out of mbufs or pkt buffers
* That is, mfreeq is empty or (lilfreeq, bigfreeq) are empty.
* One solution is to find out which one is getting full and
* then increase them.
*/
dtrap(); /* This is really rare... */
m_freem(tcp_mbuf); /* Free TCP/data chain */
return ENOBUFS;
}
/* strip overlay from front of TCP header */
tcp_mbuf->m_data += sizeof(struct ipovly);
tcp_mbuf->m_len -= sizeof(struct ipovly);
}
#endif /* end IP_V6 */
if (tp->t_template == 0)
panic("tcp_output");
MEMCPY((char*)ti, (char*)tp->t_template, sizeof(struct tcpiphdr));
/*
* Fill in fields, remembering maximum advertised
* window for use in delaying messages about window sizes.
* If resending a FIN, be sure not to use a new sequence number.
*/
if (flags & TH_FIN && tp->t_flags & TF_SENTFIN &&
tp->snd_nxt == tp->snd_max)
{
tp->snd_nxt--;
}
ti->ti_seq = htonl(tp->snd_nxt);
ti->ti_ack = htonl(tp->rcv_nxt);
/*
* If we're sending a SYN, check the IP address of the interface
* that we will (likely) use to send the IP datagram -- if it's
* changed from what is in the template (as it might if this is
* a retransmission, and the original SYN caused PPP to start
* bringing the interface up, and PPP has got a new IP address
* via IPCP), update the template and the inpcb with the new
* address.
*/
if (flags & TH_SYN)
{
struct inpcb * inp;
inp = (struct inpcb *)so->so_pcb;
switch(so->so_domain)
{
#ifdef IP_V4
case AF_INET:
{
ip_addr src;
#ifdef INCLUDE_PPP
if(((flags & TH_ACK) == 0) && /* SYN only, not SYN/ACK */
(inp->ifp) && /* Make sure we have iface */
(inp->ifp->mib.ifType == PPP)) /* only PPP type */
{
dtrap(); /* remove after confirmed to work in PPP */
src = ip_mymach(ti->ti_dst.s_addr);
if (src != ti->ti_src.s_addr)
{
ti->ti_src.s_addr = src;
tp->t_template->ti_src.s_addr = src;
tp->t_inpcb->inp_laddr.s_addr = src;
}
}
#endif /* INCLUDE_PPP */
/* If this is a SYN (not a SYN/ACK) then set the pmtu */
if((flags & TH_ACK) == 0)
{
#ifdef IP_PMTU
inp->inp_pmtu = pmtucache_get(inp->inp_faddr.s_addr);
#else /* not compiled for pathmtu, guess based on iface */
{
NET ifp;
/* find iface for route. Pass "src" as nexthop return */
ifp = iproute(ti->ti_dst.s_addr, &src);
if(ifp)
inp->inp_pmtu = ifp->n_mtu - (ifp->n_lnh + 40);
else
inp->inp_pmtu = 580; /* Ugh. */
}
#endif /* IP_PMTU */
}
break;
}
#endif /* IP_V4 */
#ifdef IP_V6
case AF_INET6:
{
struct ip6_inaddr * local;
local = ip6_myaddr(&tp->t_inpcb->ip6_faddr, inp->ifp);
/* If we got a local address & it's not the one in the pcb, then
* we assume it changed at the iface and fix it in the pcb. Unlike
* v4, we don't have an IP header yet, not do we have a template
* to worry about.
*/
if((local) &&
(!IP6EQ(&local->addr, &tp->t_inpcb->ip6_laddr)))
{
IP6CPY(&tp->t_inpcb->ip6_laddr, &local->addr);
}
/* If this is a SYN (not a SYN/ACK) then set the pmtu */
if((flags & TH_ACK) == 0)
{
inp->inp_pmtu = ip6_pmtulookup(&inp->ip6_laddr, inp->ifp);
}
break;
}
#endif /* IP_V6 */
default:
dtrap(); /* bad domain setting */
}
}
/* fill in options if any are set */
if (optlen)
{
struct mbuf * mopt;
mopt = m_getwithdata(MT_TXDATA, MAXOPTLEN);
if (mopt == NULL)
{
m_freem(m);
return (ENOBUFS);
}
/* insert options mbuf after after tmp_mbuf */
mopt->m_next = tcp_mbuf->m_next;
tcp_mbuf->m_next = mopt;
/* extend options to aligned address */
while(optlen & 0x03)
tcp_optionbuf[optlen++] = TCPOPT_EOL;
MEMCPY(mtod(mopt, char *), tcp_optionbuf, optlen);
mopt->m_len = optlen;
/* use portable macro to set tcp data offset bits */
SET_TH_OFF(ti->ti_t, ((sizeof (struct tcphdr) + optlen) >> 2));
}
ti->ti_flags = (u_char)flags;
/*
* Calculate receive window. Don't shrink window,
* but avoid silly window syndrome.
*/
if (win < (long)(so->so_rcv.sb_hiwat / 4) && win < (long)tp->t_maxseg)
win = 0;
if (win < (long)(tp->rcv_adv - tp->rcv_nxt))
win = (long)(tp->rcv_adv - tp->rcv_nxt);
/* do check for Iniche buffer limits -JB- */
if (bigfreeq.q_len == 0) /* If queue length is 0, set window to 0 */
{
win = 0;
}
else if(win > (((long)bigfreeq.q_len - 1) * (long)bigbufsiz))
{
win = ((long)bigfreeq.q_len - 1) * bigbufsiz;
}
#ifdef TCP_WIN_SCALE
if(tp->t_flags & TF_WINSCALE)
{
ti->ti_win = htons((u_short)(win >> tp->rcv_wind_scale)); /* apply scale */
}
/*
* Tcp output routine: figure out what should be sent and send it.
*/
int
tcp_output(struct tcpcb *tp)
{
struct inpcb * const inp = tp->t_inpcb;
struct socket *so = inp->inp_socket;
long len, recvwin, sendwin;
int nsacked = 0;
int off, flags, error = 0;
#ifdef TCP_SIGNATURE
int sigoff = 0;
#endif
struct mbuf *m;
struct ip *ip;
struct tcphdr *th;
u_char opt[TCP_MAXOLEN];
unsigned int ipoptlen, optlen, hdrlen;
int idle;
boolean_t sendalot;
struct ip6_hdr *ip6;
#ifdef INET6
const boolean_t isipv6 = INP_ISIPV6(inp);
#else
const boolean_t isipv6 = FALSE;
#endif
boolean_t can_tso = FALSE, use_tso;
boolean_t report_sack, idle_cwv = FALSE;
u_int segsz, tso_hlen, tso_lenmax = 0;
int segcnt = 0;
boolean_t need_sched = FALSE;
KKASSERT(so->so_port == &curthread->td_msgport);
/*
* Determine length of data that should be transmitted,
* and flags that will be used.
* If there is some data or critical controls (SYN, RST)
* to send, then transmit; otherwise, investigate further.
*/
/*
* If we have been idle for a while, the send congestion window
* could be no longer representative of the current state of the
* link; need to validate congestion window. However, we should
* not perform congestion window validation here, since we could
* be asked to send pure ACK.
*/
if (tp->snd_max == tp->snd_una &&
(ticks - tp->snd_last) >= tp->t_rxtcur && tcp_idle_restart)
idle_cwv = TRUE;
/*
* Calculate whether the transmit stream was previously idle
* and adjust TF_LASTIDLE for the next time.
*/
idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una);
if (idle && (tp->t_flags & TF_MORETOCOME))
tp->t_flags |= TF_LASTIDLE;
else
tp->t_flags &= ~TF_LASTIDLE;
if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max &&
!IN_FASTRECOVERY(tp))
nsacked = tcp_sack_bytes_below(&tp->scb, tp->snd_nxt);
/*
* Find out whether TSO could be used or not
*
* For TSO capable devices, the following assumptions apply to
* the processing of TCP flags:
* - If FIN is set on the large TCP segment, the device must set
* FIN on the last segment that it creates from the large TCP
* segment.
* - If PUSH is set on the large TCP segment, the device must set
* PUSH on the last segment that it creates from the large TCP
* segment.
*/
#if !defined(IPSEC) && !defined(FAST_IPSEC)
if (tcp_do_tso
#ifdef TCP_SIGNATURE
&& (tp->t_flags & TF_SIGNATURE) == 0
#endif
) {
if (!isipv6) {
struct rtentry *rt = inp->inp_route.ro_rt;
if (rt != NULL && (rt->rt_flags & RTF_UP) &&
(rt->rt_ifp->if_hwassist & CSUM_TSO)) {
can_tso = TRUE;
tso_lenmax = rt->rt_ifp->if_tsolen;
}
}
}
#endif /* !IPSEC && !FAST_IPSEC */
again:
m = NULL;
ip = NULL;
th = NULL;
ip6 = NULL;
if ((tp->t_flags & (TF_SACK_PERMITTED | TF_NOOPT)) ==
TF_SACK_PERMITTED &&
(!TAILQ_EMPTY(&tp->t_segq) ||
tp->reportblk.rblk_start != tp->reportblk.rblk_end))
report_sack = TRUE;
else
report_sack = FALSE;
/* Make use of SACK information when slow-starting after a RTO. */
if (TCP_DO_SACK(tp) && tp->snd_nxt != tp->snd_max &&
!IN_FASTRECOVERY(tp)) {
tcp_seq old_snd_nxt = tp->snd_nxt;
tcp_sack_skip_sacked(&tp->scb, &tp->snd_nxt);
nsacked += tp->snd_nxt - old_snd_nxt;
}
sendalot = FALSE;
off = tp->snd_nxt - tp->snd_una;
sendwin = min(tp->snd_wnd, tp->snd_cwnd + nsacked);
sendwin = min(sendwin, tp->snd_bwnd);
flags = tcp_outflags[tp->t_state];
/*
* Get standard flags, and add SYN or FIN if requested by 'hidden'
* state flags.
*/
if (tp->t_flags & TF_NEEDFIN)
flags |= TH_FIN;
if (tp->t_flags & TF_NEEDSYN)
flags |= TH_SYN;
/*
* If in persist timeout with window of 0, send 1 byte.
* Otherwise, if window is small but nonzero
* and timer expired, we will send what we can
* and go to transmit state.
*/
if (tp->t_flags & TF_FORCE) {
if (sendwin == 0) {
/*
* If we still have some data to send, then
* clear the FIN bit. Usually this would
* happen below when it realizes that we
* aren't sending all the data. However,
* if we have exactly 1 byte of unsent data,
* then it won't clear the FIN bit below,
* and if we are in persist state, we wind
* up sending the packet without recording
* that we sent the FIN bit.
*
* We can't just blindly clear the FIN bit,
* because if we don't have any more data
* to send then the probe will be the FIN
* itself.
*/
if (off < so->so_snd.ssb_cc)
flags &= ~TH_FIN;
sendwin = 1;
} else {
tcp_callout_stop(tp, tp->tt_persist);
tp->t_rxtshift = 0;
}
}
/*
* If snd_nxt == snd_max and we have transmitted a FIN, the
* offset will be > 0 even if so_snd.ssb_cc is 0, resulting in
* a negative length. This can also occur when TCP opens up
* its congestion window while receiving additional duplicate
* acks after fast-retransmit because TCP will reset snd_nxt
* to snd_max after the fast-retransmit.
*
* A negative length can also occur when we are in the
* TCPS_SYN_RECEIVED state due to a simultanious connect where
* our SYN has not been acked yet.
*
* In the normal retransmit-FIN-only case, however, snd_nxt will
* be set to snd_una, the offset will be 0, and the length may
* wind up 0.
*/
len = (long)ulmin(so->so_snd.ssb_cc, sendwin) - off;
/*
* Lop off SYN bit if it has already been sent. However, if this
* is SYN-SENT state and if segment contains data, suppress sending
* segment (sending the segment would be an option if we still
* did TAO and the remote host supported it).
*/
if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) {
flags &= ~TH_SYN;
off--, len++;
if (len > 0 && tp->t_state == TCPS_SYN_SENT) {
tp->t_flags &= ~(TF_ACKNOW | TF_XMITNOW);
return 0;
}
}
/*
* Be careful not to send data and/or FIN on SYN segments.
* This measure is needed to prevent interoperability problems
* with not fully conformant TCP implementations.
*/
if (flags & TH_SYN) {
len = 0;
flags &= ~TH_FIN;
}
if (len < 0) {
/*
* A negative len can occur if our FIN has been sent but not
* acked, or if we are in a simultanious connect in the
* TCPS_SYN_RECEIVED state with our SYN sent but not yet
* acked.
*
* If our window has contracted to 0 in the FIN case
* (which can only occur if we have NOT been called to
* retransmit as per code a few paragraphs up) then we
* want to shift the retransmit timer over to the
* persist timer.
*
* However, if we are in the TCPS_SYN_RECEIVED state
* (the SYN case) we will be in a simultanious connect and
* the window may be zero degeneratively. In this case we
* do not want to shift to the persist timer after the SYN
* or the SYN+ACK transmission.
*/
len = 0;
if (sendwin == 0 && tp->t_state != TCPS_SYN_RECEIVED) {
tcp_callout_stop(tp, tp->tt_rexmt);
tp->t_rxtshift = 0;
tp->snd_nxt = tp->snd_una;
if (!tcp_callout_active(tp, tp->tt_persist))
tcp_setpersist(tp);
}
}
KASSERT(len >= 0, ("%s: len < 0", __func__));
/*
* Automatic sizing of send socket buffer. Often the send buffer
* size is not optimally adjusted to the actual network conditions
* at hand (delay bandwidth product). Setting the buffer size too
* small limits throughput on links with high bandwidth and high
* delay (eg. trans-continental/oceanic links). Setting the
* buffer size too big consumes too much real kernel memory,
* especially with many connections on busy servers.
*
* The criteria to step up the send buffer one notch are:
* 1. receive window of remote host is larger than send buffer
* (with a fudge factor of 5/4th);
* 2. hiwat has not significantly exceeded bwnd (inflight)
* (bwnd is a maximal value if inflight is disabled).
* 3. send buffer is filled to 7/8th with data (so we actually
* have data to make use of it);
* 4. hiwat has not hit maximal automatic size;
* 5. our send window (slow start and cogestion controlled) is
* larger than sent but unacknowledged data in send buffer.
*
* The remote host receive window scaling factor may limit the
* growing of the send buffer before it reaches its allowed
* maximum.
*
* It scales directly with slow start or congestion window
* and does at most one step per received ACK. This fast
* scaling has the drawback of growing the send buffer beyond
* what is strictly necessary to make full use of a given
* delay*bandwith product. However testing has shown this not
* to be much of an problem. At worst we are trading wasting
* of available bandwith (the non-use of it) for wasting some
* socket buffer memory.
*
* The criteria for shrinking the buffer is based solely on
* the inflight code (snd_bwnd). If inflight is disabled,
* the buffer will not be shrinked. Note that snd_bwnd already
* has a fudge factor. Our test adds a little hysteresis.
*/
if (tcp_do_autosndbuf && (so->so_snd.ssb_flags & SSB_AUTOSIZE)) {
const int asbinc = tcp_autosndbuf_inc;
const int hiwat = so->so_snd.ssb_hiwat;
const int lowat = so->so_snd.ssb_lowat;
u_long newsize;
if ((tp->snd_wnd / 4 * 5) >= hiwat &&
so->so_snd.ssb_cc >= (hiwat / 8 * 7) &&
hiwat < tp->snd_bwnd + hiwat / 10 &&
hiwat + asbinc < tcp_autosndbuf_max &&
hiwat < (TCP_MAXWIN << tp->snd_scale) &&
sendwin >= (so->so_snd.ssb_cc -
(tp->snd_nxt - tp->snd_una))) {
newsize = ulmin(hiwat + asbinc, tcp_autosndbuf_max);
if (!ssb_reserve(&so->so_snd, newsize, so, NULL))
atomic_clear_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
#if 0
if (newsize >= (TCP_MAXWIN << tp->snd_scale))
atomic_clear_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
#endif
} else if ((long)tp->snd_bwnd <
(long)(hiwat * 3 / 4 - lowat - asbinc) &&
hiwat > tp->t_maxseg * 2 + asbinc &&
hiwat + asbinc >= tcp_autosndbuf_min &&
tcp_do_autosndbuf == 1) {
newsize = ulmax(hiwat - asbinc, tp->t_maxseg * 2);
ssb_reserve(&so->so_snd, newsize, so, NULL);
}
}
/*
* Don't use TSO, if:
* - Congestion window needs validation
* - There are SACK blocks to report
* - RST or SYN flags is set
* - URG will be set
*
* XXX
* Checking for SYN|RST looks overkill, just to be safe than sorry
*/
use_tso = can_tso;
if (report_sack || idle_cwv || (flags & (TH_RST | TH_SYN)))
use_tso = FALSE;
if (use_tso) {
tcp_seq ugr_nxt = tp->snd_nxt;
if ((flags & TH_FIN) && (tp->t_flags & TF_SENTFIN) &&
tp->snd_nxt == tp->snd_max)
--ugr_nxt;
if (SEQ_GT(tp->snd_up, ugr_nxt))
use_tso = FALSE;
}
if (use_tso) {
/*
* Find out segment size and header length for TSO
*/
error = tcp_tso_getsize(tp, &segsz, &tso_hlen);
if (error)
use_tso = FALSE;
}
if (!use_tso) {
segsz = tp->t_maxseg;
tso_hlen = 0; /* not used */
}
/*
* Truncate to the maximum segment length if not TSO, and ensure that
* FIN is removed if the length no longer contains the last data byte.
*/
if (len > segsz) {
if (!use_tso) {
len = segsz;
++segcnt;
} else {
int nsegs;
if (__predict_false(tso_lenmax < segsz))
tso_lenmax = segsz << 1;
/*
* Truncate TSO transfers to (IP_MAXPACKET - iphlen -
* thoff), and make sure that we send equal size
* transfers down the stack (rather than big-small-
* big-small-...).
*/
len = min(len, tso_lenmax);
nsegs = min(len, (IP_MAXPACKET - tso_hlen)) / segsz;
KKASSERT(nsegs > 0);
len = nsegs * segsz;
if (len <= segsz) {
use_tso = FALSE;
++segcnt;
} else {
segcnt += nsegs;
}
}
sendalot = TRUE;
} else {
use_tso = FALSE;
if (len > 0)
++segcnt;
}
if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + so->so_snd.ssb_cc))
flags &= ~TH_FIN;
recvwin = ssb_space(&so->so_rcv);
/*
* Sender silly window avoidance. We transmit under the following
* conditions when len is non-zero:
*
* - We have a full segment
* - This is the last buffer in a write()/send() and we are
* either idle or running NODELAY
* - we've timed out (e.g. persist timer)
* - we have more then 1/2 the maximum send window's worth of
* data (receiver may be limiting the window size)
* - we need to retransmit
*/
if (len) {
if (len >= segsz)
goto send;
/*
* NOTE! on localhost connections an 'ack' from the remote
* end may occur synchronously with the output and cause
* us to flush a buffer queued with moretocome. XXX
*
* note: the len + off check is almost certainly unnecessary.
*/
if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */
(idle || (tp->t_flags & TF_NODELAY)) &&
len + off >= so->so_snd.ssb_cc &&
!(tp->t_flags & TF_NOPUSH)) {
goto send;
}
if (tp->t_flags & TF_FORCE) /* typ. timeout case */
goto send;
if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0)
goto send;
if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */
goto send;
if (tp->t_flags & TF_XMITNOW)
goto send;
}
/*
* Compare available window to amount of window
* known to peer (as advertised window less
* next expected input). If the difference is at least two
* max size segments, or at least 50% of the maximum possible
* window, then want to send a window update to peer.
*/
if (recvwin > 0) {
/*
* "adv" is the amount we can increase the window,
* taking into account that we are limited by
* TCP_MAXWIN << tp->rcv_scale.
*/
long adv = min(recvwin, (long)TCP_MAXWIN << tp->rcv_scale) -
(tp->rcv_adv - tp->rcv_nxt);
long hiwat;
/*
* This ack case typically occurs when the user has drained
* the TCP socket buffer sufficiently to warrent an ack
* containing a 'pure window update'... that is, an ack that
* ONLY updates the tcp window.
*
* It is unclear why we would need to do a pure window update
* past 2 segments if we are going to do one at 1/2 the high
* water mark anyway, especially since under normal conditions
* the user program will drain the socket buffer quickly.
* The 2-segment pure window update will often add a large
* number of extra, unnecessary acks to the stream.
*
* avoid_pure_win_update now defaults to 1.
*/
if (avoid_pure_win_update == 0 ||
(tp->t_flags & TF_RXRESIZED)) {
if (adv >= (long) (2 * segsz)) {
goto send;
}
}
hiwat = (long)(TCP_MAXWIN << tp->rcv_scale);
if (hiwat > (long)so->so_rcv.ssb_hiwat)
hiwat = (long)so->so_rcv.ssb_hiwat;
if (adv >= hiwat / 2)
goto send;
}
/*
* Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW
* is also a catch-all for the retransmit timer timeout case.
*/
if (tp->t_flags & TF_ACKNOW)
goto send;
if ((flags & TH_RST) ||
((flags & TH_SYN) && !(tp->t_flags & TF_NEEDSYN)))
goto send;
if (SEQ_GT(tp->snd_up, tp->snd_una))
goto send;
/*
* If our state indicates that FIN should be sent
* and we have not yet done so, then we need to send.
*/
if ((flags & TH_FIN) &&
(!(tp->t_flags & TF_SENTFIN) || tp->snd_nxt == tp->snd_una))
goto send;
/*
* TCP window updates are not reliable, rather a polling protocol
* using ``persist'' packets is used to insure receipt of window
* updates. The three ``states'' for the output side are:
* idle not doing retransmits or persists
* persisting to move a small or zero window
* (re)transmitting and thereby not persisting
*
* tcp_callout_active(tp, tp->tt_persist)
* is true when we are in persist state.
* The TF_FORCE flag in tp->t_flags
* is set when we are called to send a persist packet.
* tcp_callout_active(tp, tp->tt_rexmt)
* is set when we are retransmitting
* The output side is idle when both timers are zero.
*
* If send window is too small, there is data to transmit, and no
* retransmit or persist is pending, then go to persist state.
*
* If nothing happens soon, send when timer expires:
* if window is nonzero, transmit what we can, otherwise force out
* a byte.
*
* Don't try to set the persist state if we are in TCPS_SYN_RECEIVED
* with data pending. This situation can occur during a
* simultanious connect.
*/
if (so->so_snd.ssb_cc > 0 &&
tp->t_state != TCPS_SYN_RECEIVED &&
!tcp_callout_active(tp, tp->tt_rexmt) &&
!tcp_callout_active(tp, tp->tt_persist)) {
tp->t_rxtshift = 0;
tcp_setpersist(tp);
}
/*
* No reason to send a segment, just return.
*/
tp->t_flags &= ~TF_XMITNOW;
return (0);
send:
if (need_sched && len > 0) {
tcp_output_sched(tp);
return 0;
}
/*
* Before ESTABLISHED, force sending of initial options
* unless TCP set not to do any options.
* NOTE: we assume that the IP/TCP header plus TCP options
* always fit in a single mbuf, leaving room for a maximum
* link header, i.e.
* max_linkhdr + sizeof(struct tcpiphdr) + optlen <= MCLBYTES
*/
optlen = 0;
if (isipv6)
hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
else
hdrlen = sizeof(struct tcpiphdr);
if (flags & TH_SYN) {
tp->snd_nxt = tp->iss;
if (!(tp->t_flags & TF_NOOPT)) {
u_short mss;
opt[0] = TCPOPT_MAXSEG;
opt[1] = TCPOLEN_MAXSEG;
mss = htons((u_short) tcp_mssopt(tp));
memcpy(opt + 2, &mss, sizeof mss);
optlen = TCPOLEN_MAXSEG;
if ((tp->t_flags & TF_REQ_SCALE) &&
(!(flags & TH_ACK) ||
(tp->t_flags & TF_RCVD_SCALE))) {
*((u_int32_t *)(opt + optlen)) = htonl(
TCPOPT_NOP << 24 |
TCPOPT_WINDOW << 16 |
TCPOLEN_WINDOW << 8 |
tp->request_r_scale);
optlen += 4;
}
if ((tcp_do_sack && !(flags & TH_ACK)) ||
tp->t_flags & TF_SACK_PERMITTED) {
uint32_t *lp = (uint32_t *)(opt + optlen);
*lp = htonl(TCPOPT_SACK_PERMITTED_ALIGNED);
optlen += TCPOLEN_SACK_PERMITTED_ALIGNED;
}
}
}
/*
* Send a timestamp and echo-reply if this is a SYN and our side
* wants to use timestamps (TF_REQ_TSTMP is set) or both our side
* and our peer have sent timestamps in our SYN's.
*/
if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
!(flags & TH_RST) &&
(!(flags & TH_ACK) || (tp->t_flags & TF_RCVD_TSTMP))) {
u_int32_t *lp = (u_int32_t *)(opt + optlen);
/* Form timestamp option as shown in appendix A of RFC 1323. */
*lp++ = htonl(TCPOPT_TSTAMP_HDR);
*lp++ = htonl(ticks);
*lp = htonl(tp->ts_recent);
optlen += TCPOLEN_TSTAMP_APPA;
}
/* Set receive buffer autosizing timestamp. */
if (tp->rfbuf_ts == 0 && (so->so_rcv.ssb_flags & SSB_AUTOSIZE))
tp->rfbuf_ts = ticks;
/*
* If this is a SACK connection and we have a block to report,
* fill in the SACK blocks in the TCP options.
*/
if (report_sack)
tcp_sack_fill_report(tp, opt, &optlen);
#ifdef TCP_SIGNATURE
if (tp->t_flags & TF_SIGNATURE) {
int i;
u_char *bp;
/*
* Initialize TCP-MD5 option (RFC2385)
*/
bp = (u_char *)opt + optlen;
*bp++ = TCPOPT_SIGNATURE;
*bp++ = TCPOLEN_SIGNATURE;
sigoff = optlen + 2;
for (i = 0; i < TCP_SIGLEN; i++)
*bp++ = 0;
optlen += TCPOLEN_SIGNATURE;
/*
* Terminate options list and maintain 32-bit alignment.
*/
*bp++ = TCPOPT_NOP;
*bp++ = TCPOPT_EOL;
optlen += 2;
}
#endif /* TCP_SIGNATURE */
KASSERT(optlen <= TCP_MAXOLEN, ("too many TCP options"));
hdrlen += optlen;
if (isipv6) {
ipoptlen = ip6_optlen(inp);
} else {
if (inp->inp_options) {
ipoptlen = inp->inp_options->m_len -
offsetof(struct ipoption, ipopt_list);
} else {
/*
* TCP input routine, follows pages 65-76 of the
* protocol specification dated September, 1981 very closely.
*/
void
tcp_input(struct mbuf *m, int iphlen, struct socket *inso)
{
struct ip save_ip, *ip;
register struct tcpiphdr *ti;
caddr_t optp = NULL;
int optlen = 0;
int len, tlen, off;
register struct tcpcb *tp = NULL;
register int tiflags;
struct socket *so = NULL;
int todrop, acked, ourfinisacked, needoutput = 0;
int iss = 0;
u_long tiwin;
int ret;
struct ex_list *ex_ptr;
Slirp *slirp;
DEBUG_CALL("tcp_input");
DEBUG_ARGS((dfd, " m = %8lx iphlen = %2d inso = %lx\n",
(long )m, iphlen, (long )inso ));
/*
* If called with m == 0, then we're continuing the connect
*/
if (m == NULL) {
so = inso;
slirp = so->slirp;
/* Re-set a few variables */
tp = sototcpcb(so);
m = so->so_m;
so->so_m = NULL;
ti = so->so_ti;
tiwin = ti->ti_win;
tiflags = ti->ti_flags;
goto cont_conn;
}
slirp = m->slirp;
/*
* Get IP and TCP header together in first mbuf.
* Note: IP leaves IP header in first mbuf.
*/
ti = mtod(m, struct tcpiphdr *);
if (iphlen > sizeof(struct ip )) {
ip_stripoptions(m, (struct mbuf *)0);
iphlen=sizeof(struct ip );
}
/* XXX Check if too short */
/*
* Save a copy of the IP header in case we want restore it
* for sending an ICMP error message in response.
*/
ip=mtod(m, struct ip *);
save_ip = *ip;
save_ip.ip_len+= iphlen;
/*
* Checksum extended TCP header and data.
*/
tlen = ((struct ip *)ti)->ip_len;
tcpiphdr2qlink(ti)->next = tcpiphdr2qlink(ti)->prev = NULL;
memset(&ti->ti_i.ih_mbuf, 0 , sizeof(struct mbuf_ptr));
ti->ti_x1 = 0;
ti->ti_len = htons((uint16_t)tlen);
len = sizeof(struct ip ) + tlen;
if(cksum(m, len)) {
goto drop;
}
/*
* Check that TCP offset makes sense,
* pull out TCP options and adjust length. XXX
*/
off = ti->ti_off << 2;
if (off < sizeof (struct tcphdr) || off > tlen) {
goto drop;
}
tlen -= off;
ti->ti_len = tlen;
if (off > sizeof (struct tcphdr)) {
optlen = off - sizeof (struct tcphdr);
optp = mtod(m, caddr_t) + sizeof (struct tcpiphdr);
}
tiflags = ti->ti_flags;
/*
* Convert TCP protocol specific fields to host format.
*/
NTOHL(ti->ti_seq);
NTOHL(ti->ti_ack);
NTOHS(ti->ti_win);
NTOHS(ti->ti_urp);
/*
* Drop TCP, IP headers and TCP options.
*/
m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
/*
* Locate pcb for segment.
*/
findso:
so = slirp->tcp_last_so;
if (so->so_fport != ti->ti_dport ||
so->so_lport != ti->ti_sport ||
so->so_laddr.s_addr != ti->ti_src.s_addr ||
so->so_faddr.s_addr != ti->ti_dst.s_addr) {
so = solookup(&slirp->tcb, ti->ti_src, ti->ti_sport,
ti->ti_dst, ti->ti_dport);
if (so)
slirp->tcp_last_so = so;
}
/*
* If the state is CLOSED (i.e., TCB does not exist) then
* all data in the incoming segment is discarded.
* If the TCB exists but is in CLOSED state, it is embryonic,
* but should either do a listen or a connect soon.
*
* state == CLOSED means we've done socreate() but haven't
* attached it to a protocol yet...
*
* XXX If a TCB does not exist, and the TH_SYN flag is
* the only flag set, then create a session, mark it
* as if it was LISTENING, and continue...
*/
if (so == NULL) {
if (slirp->restricted) {
/* Any hostfwds will have an existing socket, so we only get here
* for non-hostfwd connections. These should be dropped, unless it
* happens to be a guestfwd.
*/
for (ex_ptr = slirp->exec_list; ex_ptr; ex_ptr = ex_ptr->ex_next) {
if (ex_ptr->ex_fport == ti->ti_dport &&
ti->ti_dst.s_addr == ex_ptr->ex_addr.s_addr) {
break;
}
}
if (!ex_ptr) {
goto dropwithreset;
}
}
if ((tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) != TH_SYN)
goto dropwithreset;
if ((so = socreate(slirp)) == NULL)
goto dropwithreset;
if (tcp_attach(so) < 0) {
free(so); /* Not sofree (if it failed, it's not insqued) */
goto dropwithreset;
}
sbreserve(&so->so_snd, TCP_SNDSPACE);
sbreserve(&so->so_rcv, TCP_RCVSPACE);
so->so_laddr = ti->ti_src;
so->so_lport = ti->ti_sport;
so->so_faddr = ti->ti_dst;
so->so_fport = ti->ti_dport;
if ((so->so_iptos = tcp_tos(so)) == 0)
so->so_iptos = ((struct ip *)ti)->ip_tos;
tp = sototcpcb(so);
tp->t_state = TCPS_LISTEN;
}
/*
* If this is a still-connecting socket, this probably
* a retransmit of the SYN. Whether it's a retransmit SYN
* or something else, we nuke it.
*/
if (so->so_state & SS_ISFCONNECTING)
goto drop;
tp = sototcpcb(so);
/* XXX Should never fail */
if (tp == NULL)
goto dropwithreset;
if (tp->t_state == TCPS_CLOSED)
goto drop;
tiwin = ti->ti_win;
/*
* Segment received on connection.
* Reset idle time and keep-alive timer.
*/
tp->t_idle = 0;
if (SO_OPTIONS)
tp->t_timer[TCPT_KEEP] = TCPTV_KEEPINTVL;
else
tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_IDLE;
/*
* Process options if not in LISTEN state,
* else do it below (after getting remote address).
*/
if (optp && tp->t_state != TCPS_LISTEN)
tcp_dooptions(tp, (u_char *)optp, optlen, ti);
/*
* Header prediction: check for the two common cases
* of a uni-directional data xfer. If the packet has
* no control flags, is in-sequence, the window didn't
* change and we're not retransmitting, it's a
* candidate. If the length is zero and the ack moved
* forward, we're the sender side of the xfer. Just
* free the data acked & wake any higher level process
* that was blocked waiting for space. If the length
* is non-zero and the ack didn't move, we're the
* receiver side. If we're getting packets in-order
* (the reassembly queue is empty), add the data to
* the socket buffer and note that we need a delayed ack.
*
* XXX Some of these tests are not needed
* eg: the tiwin == tp->snd_wnd prevents many more
* predictions.. with no *real* advantage..
*/
if (tp->t_state == TCPS_ESTABLISHED &&
(tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
ti->ti_seq == tp->rcv_nxt &&
tiwin && tiwin == tp->snd_wnd &&
tp->snd_nxt == tp->snd_max) {
if (ti->ti_len == 0) {
if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
tp->snd_cwnd >= tp->snd_wnd) {
/*
* this is a pure ack for outstanding data.
*/
if (tp->t_rtt &&
SEQ_GT(ti->ti_ack, tp->t_rtseq))
tcp_xmit_timer(tp, tp->t_rtt);
acked = ti->ti_ack - tp->snd_una;
sbdrop(&so->so_snd, acked);
tp->snd_una = ti->ti_ack;
m_free(m);
/*
* If all outstanding data are acked, stop
* retransmit timer, otherwise restart timer
* using current (possibly backed-off) value.
* If process is waiting for space,
* wakeup/selwakeup/signal. If data
* are ready to send, let tcp_output
* decide between more output or persist.
*/
if (tp->snd_una == tp->snd_max)
tp->t_timer[TCPT_REXMT] = 0;
else if (tp->t_timer[TCPT_PERSIST] == 0)
tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
/*
* This is called because sowwakeup might have
* put data into so_snd. Since we don't so sowwakeup,
* we don't need this.. XXX???
*/
if (so->so_snd.sb_cc)
(void) tcp_output(tp);
return;
}
} else if (ti->ti_ack == tp->snd_una &&
tcpfrag_list_empty(tp) &&
ti->ti_len <= sbspace(&so->so_rcv)) {
/*
* this is a pure, in-sequence data packet
* with nothing on the reassembly queue and
* we have enough buffer space to take it.
*/
tp->rcv_nxt += ti->ti_len;
/*
* Add data to socket buffer.
*/
if (so->so_emu) {
if (tcp_emu(so,m)) sbappend(so, m);
} else
sbappend(so, m);
/*
* If this is a short packet, then ACK now - with Nagel
* congestion avoidance sender won't send more until
* he gets an ACK.
*
* It is better to not delay acks at all to maximize
* TCP throughput. See RFC 2581.
*/
tp->t_flags |= TF_ACKNOW;
tcp_output(tp);
return;
}
} /* header prediction */
/*
* Calculate amount of space in receive window,
* and then do TCP input processing.
* Receive window is amount of space in rcv queue,
* but not less than advertised window.
*/
{ int win;
win = sbspace(&so->so_rcv);
if (win < 0)
win = 0;
tp->rcv_wnd = max(win, (int)(tp->rcv_adv - tp->rcv_nxt));
}
switch (tp->t_state) {
/*
* If the state is LISTEN then ignore segment if it contains an RST.
* If the segment contains an ACK then it is bad and send a RST.
* If it does not contain a SYN then it is not interesting; drop it.
* Don't bother responding if the destination was a broadcast.
* Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
* tp->iss, and send a segment:
* <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
* Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
* Fill in remote peer address fields if not previously specified.
* Enter SYN_RECEIVED state, and process any other fields of this
* segment in this state.
*/
case TCPS_LISTEN: {
if (tiflags & TH_RST)
goto drop;
if (tiflags & TH_ACK)
goto dropwithreset;
if ((tiflags & TH_SYN) == 0)
goto drop;
/*
* This has way too many gotos...
* But a bit of spaghetti code never hurt anybody :)
*/
/*
* If this is destined for the control address, then flag to
* tcp_ctl once connected, otherwise connect
*/
if ((so->so_faddr.s_addr & slirp->vnetwork_mask.s_addr) ==
slirp->vnetwork_addr.s_addr) {
if (so->so_faddr.s_addr != slirp->vhost_addr.s_addr &&
so->so_faddr.s_addr != slirp->vnameserver_addr.s_addr) {
/* May be an add exec */
for (ex_ptr = slirp->exec_list; ex_ptr;
ex_ptr = ex_ptr->ex_next) {
if(ex_ptr->ex_fport == so->so_fport &&
so->so_faddr.s_addr == ex_ptr->ex_addr.s_addr) {
so->so_state |= SS_CTL;
break;
}
}
if (so->so_state & SS_CTL) {
goto cont_input;
}
}
/* CTL_ALIAS: Do nothing, tcp_fconnect will be called on it */
}
if (so->so_emu & EMU_NOCONNECT) {
so->so_emu &= ~EMU_NOCONNECT;
goto cont_input;
}
if ((tcp_fconnect(so) == -1) &&
#if defined(_WIN32)
socket_error() != WSAEWOULDBLOCK
#else
(errno != EINPROGRESS) && (errno != EWOULDBLOCK)
#endif
) {
u_char code=ICMP_UNREACH_NET;
DEBUG_MISC((dfd, " tcp fconnect errno = %d-%s\n",
errno,strerror(errno)));
if(errno == ECONNREFUSED) {
/* ACK the SYN, send RST to refuse the connection */
tcp_respond(tp, ti, m, ti->ti_seq+1, (tcp_seq)0,
TH_RST|TH_ACK);
} else {
if(errno == EHOSTUNREACH) code=ICMP_UNREACH_HOST;
HTONL(ti->ti_seq); /* restore tcp header */
HTONL(ti->ti_ack);
HTONS(ti->ti_win);
HTONS(ti->ti_urp);
m->m_data -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
m->m_len += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
*ip=save_ip;
icmp_error(m, ICMP_UNREACH,code, 0,strerror(errno));
}
tcp_close(tp);
m_free(m);
} else {
/*
* Haven't connected yet, save the current mbuf
* and ti, and return
* XXX Some OS's don't tell us whether the connect()
* succeeded or not. So we must time it out.
*/
so->so_m = m;
so->so_ti = ti;
tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
tp->t_state = TCPS_SYN_RECEIVED;
tcp_template(tp);
}
return;
cont_conn:
/* m==NULL
* Check if the connect succeeded
*/
if (so->so_state & SS_NOFDREF) {
tp = tcp_close(tp);
goto dropwithreset;
}
cont_input:
tcp_template(tp);
if (optp)
tcp_dooptions(tp, (u_char *)optp, optlen, ti);
if (iss)
tp->iss = iss;
else
tp->iss = slirp->tcp_iss;
slirp->tcp_iss += TCP_ISSINCR/2;
tp->irs = ti->ti_seq;
tcp_sendseqinit(tp);
tcp_rcvseqinit(tp);
tp->t_flags |= TF_ACKNOW;
tp->t_state = TCPS_SYN_RECEIVED;
tp->t_timer[TCPT_KEEP] = TCPTV_KEEP_INIT;
goto trimthenstep6;
} /* case TCPS_LISTEN */
/*
* If the state is SYN_SENT:
* if seg contains an ACK, but not for our SYN, drop the input.
* if seg contains a RST, then drop the connection.
* if seg does not contain SYN, then drop it.
* Otherwise this is an acceptable SYN segment
* initialize tp->rcv_nxt and tp->irs
* if seg contains ack then advance tp->snd_una
* if SYN has been acked change to ESTABLISHED else SYN_RCVD state
* arrange for segment to be acked (eventually)
* continue processing rest of data/controls, beginning with URG
*/
case TCPS_SYN_SENT:
if ((tiflags & TH_ACK) &&
(SEQ_LEQ(ti->ti_ack, tp->iss) ||
SEQ_GT(ti->ti_ack, tp->snd_max)))
goto dropwithreset;
if (tiflags & TH_RST) {
if (tiflags & TH_ACK) {
tcp_drop(tp, 0); /* XXX Check t_softerror! */
}
goto drop;
}
if ((tiflags & TH_SYN) == 0)
goto drop;
if (tiflags & TH_ACK) {
tp->snd_una = ti->ti_ack;
if (SEQ_LT(tp->snd_nxt, tp->snd_una))
tp->snd_nxt = tp->snd_una;
}
tp->t_timer[TCPT_REXMT] = 0;
tp->irs = ti->ti_seq;
tcp_rcvseqinit(tp);
tp->t_flags |= TF_ACKNOW;
if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) {
soisfconnected(so);
tp->t_state = TCPS_ESTABLISHED;
(void) tcp_reass(tp, (struct tcpiphdr *)0,
(struct mbuf *)0);
/*
* if we didn't have to retransmit the SYN,
* use its rtt as our initial srtt & rtt var.
*/
if (tp->t_rtt)
tcp_xmit_timer(tp, tp->t_rtt);
} else
tp->t_state = TCPS_SYN_RECEIVED;
trimthenstep6:
/*
* Advance ti->ti_seq to correspond to first data byte.
* If data, trim to stay within window,
* dropping FIN if necessary.
*/
ti->ti_seq++;
if (ti->ti_len > tp->rcv_wnd) {
todrop = ti->ti_len - tp->rcv_wnd;
m_adj(m, -todrop);
ti->ti_len = tp->rcv_wnd;
tiflags &= ~TH_FIN;
}
tp->snd_wl1 = ti->ti_seq - 1;
tp->rcv_up = ti->ti_seq;
goto step6;
} /* switch tp->t_state */
/*
* States other than LISTEN or SYN_SENT.
* Check that at least some bytes of segment are within
* receive window. If segment begins before rcv_nxt,
* drop leading data (and SYN); if nothing left, just ack.
*/
todrop = tp->rcv_nxt - ti->ti_seq;
if (todrop > 0) {
if (tiflags & TH_SYN) {
tiflags &= ~TH_SYN;
ti->ti_seq++;
if (ti->ti_urp > 1)
ti->ti_urp--;
else
tiflags &= ~TH_URG;
todrop--;
}
/*
* Following if statement from Stevens, vol. 2, p. 960.
*/
if (todrop > ti->ti_len
|| (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
/*
* Any valid FIN must be to the left of the window.
* At this point the FIN must be a duplicate or out
* of sequence; drop it.
*/
tiflags &= ~TH_FIN;
/*
* Send an ACK to resynchronize and drop any data.
* But keep on processing for RST or ACK.
*/
tp->t_flags |= TF_ACKNOW;
todrop = ti->ti_len;
}
m_adj(m, todrop);
ti->ti_seq += todrop;
ti->ti_len -= todrop;
if (ti->ti_urp > todrop)
ti->ti_urp -= todrop;
else {
tiflags &= ~TH_URG;
ti->ti_urp = 0;
}
}
/*
* If new data are received on a connection after the
* user processes are gone, then RST the other end.
*/
if ((so->so_state & SS_NOFDREF) &&
tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
tp = tcp_close(tp);
goto dropwithreset;
}
/*
* If segment ends after window, drop trailing data
* (and PUSH and FIN); if nothing left, just ACK.
*/
todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
if (todrop > 0) {
if (todrop >= ti->ti_len) {
/*
* If a new connection request is received
* while in TIME_WAIT, drop the old connection
* and start over if the sequence numbers
* are above the previous ones.
*/
if (tiflags & TH_SYN &&
tp->t_state == TCPS_TIME_WAIT &&
SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
iss = tp->rcv_nxt + TCP_ISSINCR;
tp = tcp_close(tp);
goto findso;
}
/*
* If window is closed can only take segments at
* window edge, and have to drop data and PUSH from
* incoming segments. Continue processing, but
* remember to ack. Otherwise, drop segment
* and ack.
*/
if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
tp->t_flags |= TF_ACKNOW;
} else {
goto dropafterack;
}
}
m_adj(m, -todrop);
ti->ti_len -= todrop;
tiflags &= ~(TH_PUSH|TH_FIN);
}
/*
* If the RST bit is set examine the state:
* SYN_RECEIVED STATE:
* If passive open, return to LISTEN state.
* If active open, inform user that connection was refused.
* ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
* Inform user that connection was reset, and close tcb.
* CLOSING, LAST_ACK, TIME_WAIT STATES
* Close the tcb.
*/
if (tiflags&TH_RST) switch (tp->t_state) {
case TCPS_SYN_RECEIVED:
case TCPS_ESTABLISHED:
case TCPS_FIN_WAIT_1:
case TCPS_FIN_WAIT_2:
case TCPS_CLOSE_WAIT:
tp->t_state = TCPS_CLOSED;
tcp_close(tp);
goto drop;
case TCPS_CLOSING:
case TCPS_LAST_ACK:
case TCPS_TIME_WAIT:
tcp_close(tp);
goto drop;
}
/*
* If a SYN is in the window, then this is an
* error and we send an RST and drop the connection.
*/
if (tiflags & TH_SYN) {
tp = tcp_drop(tp,0);
goto dropwithreset;
}
/*
* If the ACK bit is off we drop the segment and return.
*/
if ((tiflags & TH_ACK) == 0) goto drop;
/*
* Ack processing.
*/
switch (tp->t_state) {
/*
* In SYN_RECEIVED state if the ack ACKs our SYN then enter
* ESTABLISHED state and continue processing, otherwise
* send an RST. una<=ack<=max
*/
case TCPS_SYN_RECEIVED:
if (SEQ_GT(tp->snd_una, ti->ti_ack) ||
SEQ_GT(ti->ti_ack, tp->snd_max))
goto dropwithreset;
tp->t_state = TCPS_ESTABLISHED;
/*
* The sent SYN is ack'ed with our sequence number +1
* The first data byte already in the buffer will get
* lost if no correction is made. This is only needed for
* SS_CTL since the buffer is empty otherwise.
* tp->snd_una++; or:
*/
tp->snd_una=ti->ti_ack;
if (so->so_state & SS_CTL) {
/* So tcp_ctl reports the right state */
ret = tcp_ctl(so);
if (ret == 1) {
soisfconnected(so);
so->so_state &= ~SS_CTL; /* success XXX */
} else if (ret == 2) {
so->so_state &= SS_PERSISTENT_MASK;
so->so_state |= SS_NOFDREF; /* CTL_CMD */
} else {
needoutput = 1;
tp->t_state = TCPS_FIN_WAIT_1;
}
} else {
soisfconnected(so);
}
(void) tcp_reass(tp, (struct tcpiphdr *)0, (struct mbuf *)0);
tp->snd_wl1 = ti->ti_seq - 1;
/* Avoid ack processing; snd_una==ti_ack => dup ack */
goto synrx_to_est;
/* fall into ... */
/*
* In ESTABLISHED state: drop duplicate ACKs; ACK out of range
* ACKs. If the ack is in the range
* tp->snd_una < ti->ti_ack <= tp->snd_max
* then advance tp->snd_una to ti->ti_ack and drop
* data from the retransmission queue. If this ACK reflects
* more up to date window information we update our window information.
*/
case TCPS_ESTABLISHED:
case TCPS_FIN_WAIT_1:
case TCPS_FIN_WAIT_2:
case TCPS_CLOSE_WAIT:
case TCPS_CLOSING:
case TCPS_LAST_ACK:
case TCPS_TIME_WAIT:
if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
DEBUG_MISC((dfd, " dup ack m = %lx so = %lx\n",
(long )m, (long )so));
/*
* If we have outstanding data (other than
* a window probe), this is a completely
* duplicate ack (ie, window info didn't
* change), the ack is the biggest we've
* seen and we've seen exactly our rexmt
* threshold of them, assume a packet
* has been dropped and retransmit it.
* Kludge snd_nxt & the congestion
* window so we send only this one
* packet.
*
* We know we're losing at the current
* window size so do congestion avoidance
* (set ssthresh to half the current window
* and pull our congestion window back to
* the new ssthresh).
*
* Dup acks mean that packets have left the
* network (they're now cached at the receiver)
* so bump cwnd by the amount in the receiver
* to keep a constant cwnd packets in the
* network.
*/
if (tp->t_timer[TCPT_REXMT] == 0 ||
ti->ti_ack != tp->snd_una)
tp->t_dupacks = 0;
else if (++tp->t_dupacks == TCPREXMTTHRESH) {
tcp_seq onxt = tp->snd_nxt;
u_int win =
min(tp->snd_wnd, tp->snd_cwnd) / 2 /
tp->t_maxseg;
if (win < 2)
win = 2;
tp->snd_ssthresh = win * tp->t_maxseg;
tp->t_timer[TCPT_REXMT] = 0;
tp->t_rtt = 0;
tp->snd_nxt = ti->ti_ack;
tp->snd_cwnd = tp->t_maxseg;
(void) tcp_output(tp);
tp->snd_cwnd = tp->snd_ssthresh +
tp->t_maxseg * tp->t_dupacks;
if (SEQ_GT(onxt, tp->snd_nxt))
tp->snd_nxt = onxt;
goto drop;
} else if (tp->t_dupacks > TCPREXMTTHRESH) {
tp->snd_cwnd += tp->t_maxseg;
(void) tcp_output(tp);
goto drop;
}
} else
tp->t_dupacks = 0;
break;
}
synrx_to_est:
/*
* If the congestion window was inflated to account
* for the other side's cached packets, retract it.
*/
if (tp->t_dupacks > TCPREXMTTHRESH &&
tp->snd_cwnd > tp->snd_ssthresh)
tp->snd_cwnd = tp->snd_ssthresh;
tp->t_dupacks = 0;
if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
goto dropafterack;
}
acked = ti->ti_ack - tp->snd_una;
/*
* If transmit timer is running and timed sequence
* number was acked, update smoothed round trip time.
* Since we now have an rtt measurement, cancel the
* timer backoff (cf., Phil Karn's retransmit alg.).
* Recompute the initial retransmit timer.
*/
if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
tcp_xmit_timer(tp,tp->t_rtt);
/*
* If all outstanding data is acked, stop retransmit
* timer and remember to restart (more output or persist).
* If there is more data to be acked, restart retransmit
* timer, using current (possibly backed-off) value.
*/
if (ti->ti_ack == tp->snd_max) {
tp->t_timer[TCPT_REXMT] = 0;
needoutput = 1;
} else if (tp->t_timer[TCPT_PERSIST] == 0)
tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
/*
* When new data is acked, open the congestion window.
* If the window gives us less than ssthresh packets
* in flight, open exponentially (maxseg per packet).
* Otherwise open linearly: maxseg per window
* (maxseg^2 / cwnd per packet).
*/
{
register u_int cw = tp->snd_cwnd;
register u_int incr = tp->t_maxseg;
if (cw > tp->snd_ssthresh)
incr = incr * incr / cw;
tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
}
if (acked > so->so_snd.sb_cc) {
tp->snd_wnd -= so->so_snd.sb_cc;
sbdrop(&so->so_snd, (int )so->so_snd.sb_cc);
ourfinisacked = 1;
} else {
sbdrop(&so->so_snd, acked);
tp->snd_wnd -= acked;
ourfinisacked = 0;
}
tp->snd_una = ti->ti_ack;
if (SEQ_LT(tp->snd_nxt, tp->snd_una))
tp->snd_nxt = tp->snd_una;
switch (tp->t_state) {
/*
* In FIN_WAIT_1 STATE in addition to the processing
* for the ESTABLISHED state if our FIN is now acknowledged
* then enter FIN_WAIT_2.
*/
case TCPS_FIN_WAIT_1:
if (ourfinisacked) {
/*
* If we can't receive any more
* data, then closing user can proceed.
* Starting the timer is contrary to the
* specification, but if we don't get a FIN
* we'll hang forever.
*/
if (so->so_state & SS_FCANTRCVMORE) {
tp->t_timer[TCPT_2MSL] = TCP_MAXIDLE;
}
tp->t_state = TCPS_FIN_WAIT_2;
}
break;
/*
* In CLOSING STATE in addition to the processing for
* the ESTABLISHED state if the ACK acknowledges our FIN
* then enter the TIME-WAIT state, otherwise ignore
* the segment.
*/
case TCPS_CLOSING:
if (ourfinisacked) {
tp->t_state = TCPS_TIME_WAIT;
tcp_canceltimers(tp);
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
}
break;
/*
* In LAST_ACK, we may still be waiting for data to drain
* and/or to be acked, as well as for the ack of our FIN.
* If our FIN is now acknowledged, delete the TCB,
* enter the closed state and return.
*/
case TCPS_LAST_ACK:
if (ourfinisacked) {
tcp_close(tp);
goto drop;
}
break;
/*
* In TIME_WAIT state the only thing that should arrive
* is a retransmission of the remote FIN. Acknowledge
* it and restart the finack timer.
*/
case TCPS_TIME_WAIT:
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
goto dropafterack;
}
} /* switch(tp->t_state) */
step6:
/*
* Update window information.
* Don't look at window if no ACK: TAC's send garbage on first SYN.
*/
if ((tiflags & TH_ACK) &&
(SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
(tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
(tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
tp->snd_wnd = tiwin;
tp->snd_wl1 = ti->ti_seq;
tp->snd_wl2 = ti->ti_ack;
if (tp->snd_wnd > tp->max_sndwnd)
tp->max_sndwnd = tp->snd_wnd;
needoutput = 1;
}
/*
* Process segments with URG.
*/
if ((tiflags & TH_URG) && ti->ti_urp &&
TCPS_HAVERCVDFIN(tp->t_state) == 0) {
/*
* This is a kludge, but if we receive and accept
* random urgent pointers, we'll crash in
* soreceive. It's hard to imagine someone
* actually wanting to send this much urgent data.
*/
if (ti->ti_urp + so->so_rcv.sb_cc > so->so_rcv.sb_datalen) {
ti->ti_urp = 0;
tiflags &= ~TH_URG;
goto dodata;
}
/*
* If this segment advances the known urgent pointer,
* then mark the data stream. This should not happen
* in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
* a FIN has been received from the remote side.
* In these states we ignore the URG.
*
* According to RFC961 (Assigned Protocols),
* the urgent pointer points to the last octet
* of urgent data. We continue, however,
* to consider it to indicate the first octet
* of data past the urgent section as the original
* spec states (in one of two places).
*/
if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
tp->rcv_up = ti->ti_seq + ti->ti_urp;
so->so_urgc = so->so_rcv.sb_cc +
(tp->rcv_up - tp->rcv_nxt); /* -1; */
tp->rcv_up = ti->ti_seq + ti->ti_urp;
}
} else
/*
* If no out of band data is expected,
* pull receive urgent pointer along
* with the receive window.
*/
if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
tp->rcv_up = tp->rcv_nxt;
dodata:
/*
* If this is a small packet, then ACK now - with Nagel
* congestion avoidance sender won't send more until
* he gets an ACK.
*/
if (ti->ti_len && (unsigned)ti->ti_len <= 5 &&
((struct tcpiphdr_2 *)ti)->first_char == (char)27) {
tp->t_flags |= TF_ACKNOW;
}
/*
* Process the segment text, merging it into the TCP sequencing queue,
* and arranging for acknowledgment of receipt if necessary.
* This process logically involves adjusting tp->rcv_wnd as data
* is presented to the user (this happens in tcp_usrreq.c,
* case PRU_RCVD). If a FIN has already been received on this
* connection then we just ignore the text.
*/
if ((ti->ti_len || (tiflags&TH_FIN)) &&
TCPS_HAVERCVDFIN(tp->t_state) == 0) {
TCP_REASS(tp, ti, m, so, tiflags);
} else {
m_free(m);
tiflags &= ~TH_FIN;
}
/*
* If FIN is received ACK the FIN and let the user know
* that the connection is closing.
*/
if (tiflags & TH_FIN) {
if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
/*
* If we receive a FIN we can't send more data,
* set it SS_FDRAIN
* Shutdown the socket if there is no rx data in the
* buffer.
* soread() is called on completion of shutdown() and
* will got to TCPS_LAST_ACK, and use tcp_output()
* to send the FIN.
*/
sofwdrain(so);
tp->t_flags |= TF_ACKNOW;
tp->rcv_nxt++;
}
switch (tp->t_state) {
/*
* In SYN_RECEIVED and ESTABLISHED STATES
* enter the CLOSE_WAIT state.
*/
case TCPS_SYN_RECEIVED:
case TCPS_ESTABLISHED:
if(so->so_emu == EMU_CTL) /* no shutdown on socket */
tp->t_state = TCPS_LAST_ACK;
else
tp->t_state = TCPS_CLOSE_WAIT;
break;
/*
* If still in FIN_WAIT_1 STATE FIN has not been acked so
* enter the CLOSING state.
*/
case TCPS_FIN_WAIT_1:
tp->t_state = TCPS_CLOSING;
break;
/*
* In FIN_WAIT_2 state enter the TIME_WAIT state,
* starting the time-wait timer, turning off the other
* standard timers.
*/
case TCPS_FIN_WAIT_2:
tp->t_state = TCPS_TIME_WAIT;
tcp_canceltimers(tp);
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
break;
/*
* In TIME_WAIT state restart the 2 MSL time_wait timer.
*/
case TCPS_TIME_WAIT:
tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
break;
}
}
/*
* Return any desired output.
*/
if (needoutput || (tp->t_flags & TF_ACKNOW)) {
(void) tcp_output(tp);
}
return;
dropafterack:
/*
* Generate an ACK dropping incoming segment if it occupies
* sequence space, where the ACK reflects our state.
*/
if (tiflags & TH_RST)
goto drop;
m_free(m);
tp->t_flags |= TF_ACKNOW;
(void) tcp_output(tp);
return;
dropwithreset:
/* reuses m if m!=NULL, m_free() unnecessary */
if (tiflags & TH_ACK)
tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
else {
if (tiflags & TH_SYN) ti->ti_len++;
tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
TH_RST|TH_ACK);
}
return;
drop:
/*
* Drop space held by incoming segment and return.
*/
m_free(m);
}
static int
tcp_lro_matching_tuple(struct ip* ip_hdr, struct tcphdr *tcp_hdr, int *hash,
int *flow_id )
{
struct lro_flow *flow;
tcp_seq seqnum;
unsigned int off = 0;
int payload_len = 0;
*hash = LRO_HASH(ip_hdr->ip_src.s_addr, ip_hdr->ip_dst.s_addr,
tcp_hdr->th_sport, tcp_hdr->th_dport, (TCP_LRO_FLOW_MAP - 1));
*flow_id = lro_flow_map[*hash];
if (*flow_id == TCP_LRO_FLOW_NOTFOUND) {
return TCP_LRO_NAN;
}
seqnum = tcp_hdr->th_seq;
off = tcp_hdr->th_off << 2;
payload_len = ip_hdr->ip_len - off;
flow = &lro_flow_list[*flow_id];
if ((flow->lr_faddr.s_addr == ip_hdr->ip_src.s_addr) &&
(flow->lr_laddr.s_addr == ip_hdr->ip_dst.s_addr) &&
(flow->lr_fport == tcp_hdr->th_sport) &&
(flow->lr_lport == tcp_hdr->th_dport)) {
if (flow->lr_tcphdr == NULL) {
if (ntohl(seqnum) == flow->lr_seq) {
return TCP_LRO_COALESCE;
}
if (lrodebug >= 4) {
printf("%s: seqnum = %x, lr_seq = %x\n",
__func__, ntohl(seqnum), flow->lr_seq);
}
lro_seq_mismatch++;
if (SEQ_GT(ntohl(seqnum), flow->lr_seq)) {
lro_seq_outoforder++;
/*
* Whenever we receive out of order packets it
* signals loss and recovery and LRO doesn't
* let flows recover quickly. So eject.
*/
flow->lr_flags |= LRO_EJECT_REQ;
}
return TCP_LRO_NAN;
}
if (flow->lr_flags & LRO_EJECT_REQ) {
if (lrodebug)
printf("%s: eject. \n", __func__);
return TCP_LRO_EJECT_FLOW;
}
if (SEQ_GT(tcp_hdr->th_ack, flow->lr_tcphdr->th_ack)) {
if (lrodebug) {
printf("%s: th_ack = %x flow_ack = %x \n",
__func__, tcp_hdr->th_ack,
flow->lr_tcphdr->th_ack);
}
return TCP_LRO_EJECT_FLOW;
}
if (ntohl(seqnum) == (ntohl(lro_flow_list[*flow_id].lr_tcphdr->th_seq) + lro_flow_list[*flow_id].lr_len)) {
return TCP_LRO_COALESCE;
} else {
/* LRO does not handle loss recovery well, eject */
flow->lr_flags |= LRO_EJECT_REQ;
return TCP_LRO_EJECT_FLOW;
}
}
if (lrodebug) printf("tcp_lro_matching_tuple: collision \n");
return TCP_LRO_COLLISION;
}
/*
* Ertt_packet_measurements uses a small amount of state kept on each packet
* sent to match incoming acknowledgements. This enables more accurate and
* secure round trip time measurements. The resulting measurement is used for
* congestion control algorithms which require a more accurate time.
* Ertt_packet_measurements is called via the helper hook in tcp_input.c
*/
static int
ertt_packet_measurement_hook(int hhook_type, int hhook_id, void *udata,
void *ctx_data, void *hdata, struct osd *hosd)
{
struct ertt *e_t;
struct tcpcb *tp;
struct tcphdr *th;
struct tcpopt *to;
struct tcp_hhook_data *thdp;
struct txseginfo *txsi;
int acked, measurenext_len, multiack, new_sacked_bytes, rtt_bytes_adjust;
uint32_t measurenext, rts;
tcp_seq ack;
KASSERT(ctx_data != NULL, ("%s: ctx_data is NULL!", __func__));
KASSERT(hdata != NULL, ("%s: hdata is NULL!", __func__));
e_t = (struct ertt *)hdata;
thdp = ctx_data;
tp = thdp->tp;
th = thdp->th;
to = thdp->to;
new_sacked_bytes = (tp->sackhint.last_sack_ack != 0);
measurenext = measurenext_len = multiack = rts = rtt_bytes_adjust = 0;
acked = th->th_ack - tp->snd_una;
INP_WLOCK_ASSERT(tp->t_inpcb);
/* Packet has provided new acknowledgements. */
if (acked > 0 || new_sacked_bytes) {
if (acked == 0 && new_sacked_bytes) {
/* Use last sacked data. */
ack = tp->sackhint.last_sack_ack;
} else
ack = th->th_ack;
txsi = TAILQ_FIRST(&e_t->txsegi_q);
while (txsi != NULL) {
rts = 0;
/* Acknowledgement is acking more than this txsi. */
if (SEQ_GT(ack, txsi->seq + txsi->len)) {
if (txsi->flags & TXSI_RTT_MEASURE_START ||
measurenext) {
marked_packet_rtt(txsi, e_t, tp,
&measurenext, &measurenext_len,
&rtt_bytes_adjust, MULTI_ACK);
}
TAILQ_REMOVE(&e_t->txsegi_q, txsi, txsegi_lnk);
uma_zfree(txseginfo_zone, txsi);
txsi = TAILQ_FIRST(&e_t->txsegi_q);
continue;
}
/*
* Guess if delayed acks are being used by the receiver.
*
* XXXDH: A simple heuristic that could be improved
*/
if (!new_sacked_bytes) {
if (acked > tp->t_maxseg) {
e_t->dlyack_rx +=
(e_t->dlyack_rx < DLYACK_SMOOTH) ?
1 : 0;
multiack = 1;
} else if (acked > txsi->len) {
multiack = 1;
e_t->dlyack_rx +=
(e_t->dlyack_rx < DLYACK_SMOOTH) ?
1 : 0;
} else if (acked == tp->t_maxseg ||
acked == txsi->len) {
e_t->dlyack_rx -=
(e_t->dlyack_rx > 0) ? 1 : 0;
}
/* Otherwise leave dlyack_rx the way it was. */
}
/*
* Time stamps are only to help match the txsi with the
* received acknowledgements.
*/
if (e_t->timestamp_errors < MAX_TS_ERR &&
(to->to_flags & TOF_TS) != 0 && to->to_tsecr) {
/*
* Note: All packets sent with the offload will
* have the same time stamp. If we are sending
* on a fast interface and the t_maxseg is much
* smaller than one tick, this will be fine. The
* time stamp would be the same whether we were
* using tso or not. However, if the interface
* is slow, this will cause problems with the
* calculations. If the interface is slow, there
* is not reason to be using tso, and it should
* be turned off.
*/
/*
* If there are too many time stamp errors, time
* stamps won't be trusted
*/
rts = to->to_tsecr;
/* Before this packet. */
if (!e_t->dlyack_rx && TSTMP_LT(rts, txsi->tx_ts))
/* When delayed acking is used, the
* reflected time stamp is of the first
* packet and thus may be before
* txsi->tx_ts.
*/
break;
if (TSTMP_GT(rts, txsi->tx_ts)) {
/*
* If reflected time stamp is later than
* tx_tsi, then this txsi is old.
*/
if (txsi->flags & TXSI_RTT_MEASURE_START
|| measurenext) {
marked_packet_rtt(txsi, e_t, tp,
&measurenext, &measurenext_len,
&rtt_bytes_adjust, OLD_TXSI);
}
TAILQ_REMOVE(&e_t->txsegi_q, txsi,
txsegi_lnk);
uma_zfree(txseginfo_zone, txsi);
txsi = TAILQ_FIRST(&e_t->txsegi_q);
continue;
}
if (rts == txsi->tx_ts &&
TSTMP_LT(to->to_tsval, txsi->rx_ts)) {
/*
* Segment received before sent!
* Something is wrong with the received
* timestamps so increment errors. If
* this keeps up we will ignore
* timestamps.
*/
e_t->timestamp_errors++;
}
}
/*
* Acknowledging a sequence number before this txsi.
* If it is an old txsi that may have had the same seq
* numbers, it should have been removed if time stamps
* are being used.
*/
if (SEQ_LEQ(ack, txsi->seq))
break; /* Before first packet in txsi. */
/*
* Only ack > txsi->seq and ack <= txsi->seq+txsi->len
* past this point.
*
* If delayed acks are being used, an acknowledgement
* for a single segment will have been delayed by the
* receiver and will yield an inaccurate measurement. In
* this case, we only make the measurement if more than
* one segment is being acknowledged or sack is
* currently being used.
*/
if (!e_t->dlyack_rx || multiack || new_sacked_bytes) {
/* Make an accurate new measurement. */
e_t->rtt = tcp_ts_getticks() - txsi->tx_ts + 1;
if (e_t->rtt < e_t->minrtt || e_t->minrtt == 0)
e_t->minrtt = e_t->rtt;
if (e_t->rtt > e_t->maxrtt || e_t->maxrtt == 0)
e_t->maxrtt = e_t->rtt;
}
if (txsi->flags & TXSI_RTT_MEASURE_START || measurenext)
marked_packet_rtt(txsi, e_t, tp,
&measurenext, &measurenext_len,
&rtt_bytes_adjust, CORRECT_ACK);
if (txsi->flags & TXSI_TSO) {
if (txsi->len > acked) {
txsi->len -= acked;
/*
* This presumes ack for first bytes in
* txsi, this may not be true but it
* shouldn't cause problems for the
* timing.
*
* We remeasure RTT even though we only
* have a single txsi. The rationale
* behind this is that it is better to
* have a slightly inaccurate
* measurement than no additional
* measurement for the rest of the bulk
* transfer. Since TSO is only used on
* high speed interface cards, so the
* packets should be transmitted at line
* rate back to back with little
* difference in transmission times (in
* ticks).
*/
txsi->seq += acked;
/*
* Reset txsi measure flag so we don't
* use it for another RTT measurement.
*/
txsi->flags &= ~TXSI_RTT_MEASURE_START;
/*
* There is still more data to be acked
* from tso bulk transmission, so we
* won't remove it from the TAILQ yet.
*/
break;
}
txsi->len = 0;
}
TAILQ_REMOVE(&e_t->txsegi_q, txsi, txsegi_lnk);
uma_zfree(txseginfo_zone, txsi);
break;
}
if (measurenext) {
/*
* We need to do a RTT measurement. It won't be the best
* if we do it here.
*/
marked_packet_rtt(txsi, e_t, tp,
&measurenext, &measurenext_len,
&rtt_bytes_adjust, FORCED_MEASUREMENT);
}
}
return (0);
}
/*
* This function is called upon receipt of new valid data (while not in header
* prediction mode), and it updates the ordered list of sacks.
*/
void
tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end)
{
/*
* First reported block MUST be the most recent one. Subsequent
* blocks SHOULD be in the order in which they arrived at the
* receiver. These two conditions make the implementation fully
* compliant with RFC 2018.
*/
struct sackblk head_blk, saved_blks[MAX_SACK_BLKS];
int num_head, num_saved, i;
/* SACK block for the received segment. */
head_blk.start = rcv_start;
head_blk.end = rcv_end;
/*
* Merge updated SACK blocks into head_blk, and
* save unchanged SACK blocks into saved_blks[].
* num_saved will have the number of the saved SACK blocks.
*/
num_saved = 0;
for (i = 0; i < tp->rcv_numsacks; i++) {
tcp_seq start = tp->sackblks[i].start;
tcp_seq end = tp->sackblks[i].end;
if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) {
/*
* Discard this SACK block.
*/
} else if (SEQ_LEQ(head_blk.start, end) &&
SEQ_GEQ(head_blk.end, start)) {
/*
* Merge this SACK block into head_blk.
* This SACK block itself will be discarded.
*/
if (SEQ_GT(head_blk.start, start))
head_blk.start = start;
if (SEQ_LT(head_blk.end, end))
head_blk.end = end;
} else {
/*
* Save this SACK block.
*/
saved_blks[num_saved].start = start;
saved_blks[num_saved].end = end;
num_saved++;
}
}
/*
* Update SACK list in tp->sackblks[].
*/
num_head = 0;
if (SEQ_GT(head_blk.start, tp->rcv_nxt)) {
/*
* The received data segment is an out-of-order segment.
* Put head_blk at the top of SACK list.
*/
tp->sackblks[0] = head_blk;
num_head = 1;
/*
* If the number of saved SACK blocks exceeds its limit,
* discard the last SACK block.
*/
if (num_saved >= MAX_SACK_BLKS)
num_saved--;
}
if (num_saved > 0) {
/*
* Copy the saved SACK blocks back.
*/
bcopy(saved_blks, &tp->sackblks[num_head],
sizeof(struct sackblk) * num_saved);
}
/* Save the number of SACK blocks. */
tp->rcv_numsacks = num_head + num_saved;
/* If we are requesting SACK recovery, reset the stretch-ack state
* so that connection will generate more acks after recovery and
* sender's cwnd will open.
*/
if ((tp->t_flags & TF_STRETCHACK) != 0 && tp->rcv_numsacks > 0)
tcp_reset_stretch_ack(tp);
#if TRAFFIC_MGT
if (tp->acc_iaj > 0 && tp->rcv_numsacks > 0)
reset_acc_iaj(tp);
#endif /* TRAFFIC_MGT */
}