} while (0)
#endif
+static inline size_t min(size_t a, size_t b) {
+ return a < b ? a : b;
+}
+
static inline size_t max(size_t a, size_t b) {
return a > b ? a : b;
}
debug("\n");
}
+
+static void debug_cwnd(struct utcp_connection *c) {
+ debug("snd.cwnd = %u\n", c->snd.cwnd);
+}
#else
#define debug(...) do {} while(0)
#define print_packet(...) do {} while(0)
+#define debug_cwnd(...) do {} while(0)
#endif
static void set_state(struct utcp_connection *c, enum state state) {
#endif
c->snd.una = c->snd.iss;
c->snd.nxt = c->snd.iss + 1;
- c->rcv.wnd = utcp->mtu;
c->snd.last = c->snd.nxt;
- c->snd.cwnd = utcp->mtu;
+ c->snd.cwnd = (utcp->mtu > 2190 ? 2 : utcp->mtu > 1095 ? 3 : 4) * utcp->mtu;
+ c->snd.ssthresh = ~0;
+ debug_cwnd(c);
c->utcp = utcp;
// Add it to the sorted list of connections
if(!utcp->srtt) {
utcp->srtt = rtt;
utcp->rttvar = rtt / 2;
- utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
} else {
utcp->rttvar = (utcp->rttvar * 3 + absdiff(utcp->srtt, rtt)) / 4;
utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
- utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
}
+ utcp->rto = utcp->srtt + max(4 * utcp->rttvar, CLOCK_GRANULARITY);
+
if(utcp->rto > MAX_RTO) {
utcp->rto = MAX_RTO;
}
return NULL;
}
- assert((flags & ~0xf) == 0);
+ assert((flags & ~0x1f) == 0);
c->flags = flags;
c->recv = recv;
pkt.hdr.dst = c->dst;
pkt.hdr.seq = c->snd.iss;
pkt.hdr.ack = 0;
- pkt.hdr.wnd = c->rcv.wnd;
+ pkt.hdr.wnd = c->rcvbuf.maxsize;
pkt.hdr.ctl = SYN;
pkt.hdr.aux = 0x0101;
pkt.init[0] = 1;
static void ack(struct utcp_connection *c, bool sendatleastone) {
int32_t left = seqdiff(c->snd.last, c->snd.nxt);
- int32_t cwndleft = c->snd.cwnd - seqdiff(c->snd.nxt, c->snd.una);
- debug("cwndleft = %d\n", cwndleft);
+ int32_t cwndleft = min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una);
assert(left >= 0);
if(cwndleft <= 0) {
- cwndleft = 0;
- }
-
- if(cwndleft < left) {
+ left = 0;
+ } else if(cwndleft < left) {
left = cwndleft;
+
+ if(!sendatleastone || cwndleft > c->utcp->mtu) {
+ left -= left % c->utcp->mtu;
+ }
}
+ debug("cwndleft = %d, left = %d\n", cwndleft, left);
+
if(!left && !sendatleastone) {
return;
}
pkt->hdr.src = c->src;
pkt->hdr.dst = c->dst;
pkt->hdr.ack = c->rcv.nxt;
- pkt->hdr.wnd = c->snd.wnd;
+ pkt->hdr.wnd = c->rcvbuf.maxsize;
pkt->hdr.ctl = ACK;
pkt->hdr.aux = 0;
// Add data to send buffer.
- len = buffer_put(&c->sndbuf, data, len);
+ if(is_reliable(c) || (c->state != SYN_SENT && c->state != SYN_RECEIVED)) {
+ len = buffer_put(&c->sndbuf, data, len);
+ } else {
+ return 0;
+ }
if(len <= 0) {
- errno = EWOULDBLOCK;
- return 0;
+ if(is_reliable(c)) {
+ errno = EWOULDBLOCK;
+ return 0;
+ } else {
+ return len;
+ }
}
c->snd.last += len;
hdr->dst = tmp;
}
+static void fast_retransmit(struct utcp_connection *c) {
+ if(c->state == CLOSED || c->snd.last == c->snd.una) {
+ debug("fast_retransmit() called but nothing to retransmit!\n");
+ return;
+ }
+
+ struct utcp *utcp = c->utcp;
+
+ struct {
+ struct hdr hdr;
+ uint8_t data[];
+ } *pkt;
+
+ pkt = malloc(sizeof(pkt->hdr) + c->utcp->mtu);
+
+ if(!pkt) {
+ return;
+ }
+
+ pkt->hdr.src = c->src;
+ pkt->hdr.dst = c->dst;
+ pkt->hdr.wnd = c->rcvbuf.maxsize;
+ pkt->hdr.aux = 0;
+
+ switch(c->state) {
+ case ESTABLISHED:
+ case FIN_WAIT_1:
+ case CLOSE_WAIT:
+ case CLOSING:
+ case LAST_ACK:
+ // Send unacked data again.
+ pkt->hdr.seq = c->snd.una;
+ pkt->hdr.ack = c->rcv.nxt;
+ pkt->hdr.ctl = ACK;
+ uint32_t len = min(seqdiff(c->snd.last, c->snd.una), utcp->mtu);
+
+ if(fin_wanted(c, c->snd.una + len)) {
+ len--;
+ pkt->hdr.ctl |= FIN;
+ }
+
+ buffer_copy(&c->sndbuf, pkt->data, 0, len);
+ print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
+ utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
+ break;
+
+ default:
+ break;
+ }
+
+ free(pkt);
+}
+
static void retransmit(struct utcp_connection *c) {
if(c->state == CLOSED || c->snd.last == c->snd.una) {
debug("Retransmit() called but nothing to retransmit!\n");
pkt->hdr.src = c->src;
pkt->hdr.dst = c->dst;
- pkt->hdr.wnd = c->rcv.wnd;
+ pkt->hdr.wnd = c->rcvbuf.maxsize;
pkt->hdr.aux = 0;
switch(c->state) {
}
c->snd.nxt = c->snd.una + len;
- c->snd.cwnd = utcp->mtu; // reduce cwnd on retransmit
+
+ // RFC 5681 slow start after timeout
+ c->snd.ssthresh = max(c->snd.cwnd / 2, utcp->mtu * 2); // eq. 4
+ c->snd.cwnd = utcp->mtu;
+ debug_cwnd(c);
+
buffer_copy(&c->sndbuf, pkt->data, 0, len);
print_packet(c->utcp, "rtrx", pkt, sizeof(pkt->hdr) + len);
utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
ptr += 2;
}
+ bool has_data = len || (hdr.ctl & (SYN | FIN));
+
// Try to match the packet to an existing connection
struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
c->flags = UTCP_TCP;
}
+synack:
// Return SYN+ACK, go to SYN_RECEIVED state
c->snd.wnd = hdr.wnd;
c->rcv.irs = hdr.seq;
pkt.hdr.dst = c->dst;
pkt.hdr.ack = c->rcv.irs + 1;
pkt.hdr.seq = c->snd.iss;
- pkt.hdr.wnd = c->rcv.wnd;
+ pkt.hdr.wnd = c->rcvbuf.maxsize;
pkt.hdr.ctl = SYN | ACK;
if(init) {
// It is for an existing connection.
- uint32_t prevrcvnxt = c->rcv.nxt;
-
// 1. Drop invalid packets.
// 1a. Drop packets that should not happen in our current state.
break;
}
- // 1b. Drop packets with a sequence number not in our receive window.
+ // 1b. Discard data that is not in our receive window.
- bool acceptable;
+ if(is_reliable(c)) {
+ bool acceptable;
- if(c->state == SYN_SENT) {
- acceptable = true;
- } else if(len == 0) {
- acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
- } else {
- int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
+ if(c->state == SYN_SENT) {
+ acceptable = true;
+ } else if(len == 0) {
+ acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
+ } else {
+ int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
- // cut already accepted front overlapping
- if(rcv_offset < 0) {
- acceptable = len > (size_t) - rcv_offset;
+ // cut already accepted front overlapping
+ if(rcv_offset < 0) {
+ acceptable = len > (size_t) - rcv_offset;
- if(acceptable) {
- ptr -= rcv_offset;
- len += rcv_offset;
- hdr.seq -= rcv_offset;
+ if(acceptable) {
+ ptr -= rcv_offset;
+ len += rcv_offset;
+ hdr.seq -= rcv_offset;
+ }
+ } else {
+ acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
}
- } else {
- acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
}
- }
- if(!acceptable) {
- debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
+ if(!acceptable) {
+ debug("Packet not acceptable, %u <= %u + %lu < %u\n", c->rcv.nxt, hdr.seq, (unsigned long)len, c->rcv.nxt + c->rcvbuf.maxsize);
- // Ignore unacceptable RST packets.
- if(hdr.ctl & RST) {
- return 0;
- }
+ // Ignore unacceptable RST packets.
+ if(hdr.ctl & RST) {
+ return 0;
+ }
- // Otherwise, continue processing.
- len = 0;
+ // Otherwise, continue processing.
+ len = 0;
+ }
}
c->snd.wnd = hdr.wnd; // TODO: move below
// ackno should not roll back, and it should also not be bigger than what we ever could have sent
// (= snd.una + c->sndbuf.used).
+ if(!is_reliable(c)) {
+ if(hdr.ack != c->snd.last && c->state >= ESTABLISHED) {
+ hdr.ack = c->snd.una;
+ }
+ }
+
if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.last) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
debug("Packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
c->recv(c, NULL, 0);
}
+ if(c->poll && !c->reapable) {
+ c->poll(c, 0);
+ }
+
return 0;
case SYN_RECEIVED:
c->recv(c, NULL, 0);
}
+ if(c->poll && !c->reapable) {
+ c->poll(c, 0);
+ }
+
return 0;
case CLOSING:
// 3. Advance snd.una
advanced = seqdiff(hdr.ack, c->snd.una);
- prevrcvnxt = c->rcv.nxt;
if(advanced) {
// RTT measurement
assert(data_acked >= 0);
+#ifndef NDEBUG
int32_t bufused = seqdiff(c->snd.last, c->snd.una);
assert(data_acked <= bufused);
+#endif
if(data_acked) {
buffer_get(&c->sndbuf, NULL, data_acked);
c->snd.una = hdr.ack;
- c->dupack = 0;
- c->snd.cwnd += utcp->mtu;
+ if(c->dupack) {
+ if(c->dupack >= 3) {
+ c->snd.cwnd = c->snd.ssthresh;
+ }
+
+ c->dupack = 0;
+ }
+
+ // Increase the congestion window according to RFC 5681
+ if(c->snd.cwnd < c->snd.ssthresh) {
+ c->snd.cwnd += min(advanced, utcp->mtu); // eq. 2
+ } else {
+ c->snd.cwnd += max(1, (utcp->mtu * utcp->mtu) / c->snd.cwnd); // eq. 3
+ }
if(c->snd.cwnd > c->sndbuf.maxsize) {
c->snd.cwnd = c->sndbuf.maxsize;
}
+ debug_cwnd(c);
+
// Check if we have sent a FIN that is now ACKed.
switch(c->state) {
case FIN_WAIT_1:
case CLOSING:
if(c->snd.una == c->snd.last) {
gettimeofday(&c->conn_timeout, NULL);
- c->conn_timeout.tv_sec += 60;
+ c->conn_timeout.tv_sec += utcp->timeout;
set_state(c, TIME_WAIT);
}
if(c->dupack == 3) {
debug("Triplicate ACK\n");
- //TODO: Resend one packet and go to fast recovery mode. See RFC 6582.
- //We do a very simple variant here; reset the nxt pointer to the last acknowledged packet from the peer.
- //Reset the congestion window so we wait for ACKs.
- c->snd.nxt = c->snd.una;
- c->snd.cwnd = utcp->mtu;
- start_retransmit_timer(c);
+
+ // RFC 5681 fast recovery
+ c->snd.ssthresh = max(c->snd.cwnd / 2, utcp->mtu * 2); // eq. 4
+ c->snd.cwnd = max(c->snd.ssthresh + 3 * utcp->mtu, c->sndbuf.maxsize);
+
+ if(c->snd.cwnd > c->sndbuf.maxsize) {
+ c->snd.cwnd = c->sndbuf.maxsize;
+ }
+
+ debug_cwnd(c);
+
+ fast_retransmit(c);
+ } else if(c->dupack > 3) {
+ c->snd.cwnd += utcp->mtu;
+
+ if(c->snd.cwnd > c->sndbuf.maxsize) {
+ c->snd.cwnd = c->sndbuf.maxsize;
+ }
+
+ debug_cwnd(c);
}
}
}
break;
case SYN_RECEIVED:
+ // This is a retransmit of a SYN, send back the SYNACK.
+ goto synack;
+
case ESTABLISHED:
case FIN_WAIT_1:
case FIN_WAIT_2:
// 7. Process FIN stuff
- if((hdr.ctl & FIN) && hdr.seq + len == c->rcv.nxt) {
+ if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
switch(c->state) {
case SYN_SENT:
case SYN_RECEIVED:
case FIN_WAIT_2:
gettimeofday(&c->conn_timeout, NULL);
- c->conn_timeout.tv_sec += 60;
+ c->conn_timeout.tv_sec += utcp->timeout;
set_state(c, TIME_WAIT);
break;
c->rcv.nxt++;
len++;
- // Inform the application that the peer closed the connection.
+ // Inform the application that the peer closed its end of the connection.
if(c->recv) {
errno = 0;
c->recv(c, NULL, 0);
}
// Now we send something back if:
- // - we advanced rcv.nxt (ie, we got some data that needs to be ACKed)
+ // - we received data, so we have to send back an ACK
// -> sendatleastone = true
// - or we got an ack, so we should maybe send a bit more data
// -> sendatleastone = false
- ack(c, len || prevrcvnxt != c->rcv.nxt);
+ if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
+ ack(c, has_data);
+ }
+
return 0;
reset:
}
utcp_recv_t old_recv = c->recv;
+ utcp_poll_t old_poll = c->poll;
reset_connection(c);
errno = 0;
old_recv(c, NULL, 0);
}
+
+ if(old_poll && !c->reapable) {
+ errno = 0;
+ old_poll(c, 0);
+ }
}
return;
c->recv(c, NULL, 0);
}
- if(c->poll) {
+ if(c->poll && !c->reapable) {
c->poll(c, 0);
}
for(int i = 0; i < utcp->nconnections; i++) {
struct utcp_connection *c = utcp->connections[i];
- if(!c->reapable)
+ if(!c->reapable) {
if(c->recv) {
c->recv(c, NULL, 0);
}
+ if(c->poll && !c->reapable) {
+ c->poll(c, 0);
+ }
+ }
+
buffer_exit(&c->rcvbuf);
buffer_exit(&c->sndbuf);
free(c);
continue;
}
- c->rtrx_timeout = now;
- c->conn_timeout = then;
+ if(timerisset(&c->rtrx_timeout)) {
+ c->rtrx_timeout = now;
+ }
+
+ if(timerisset(&c->conn_timeout)) {
+ c->conn_timeout = then;
+ }
+
c->rtt_start.tv_sec = 0;
}
}
void utcp_offline(struct utcp *utcp, bool offline) {
+ struct timeval now;
+ gettimeofday(&now, NULL);
+
for(int i = 0; i < utcp->nconnections; i++) {
struct utcp_connection *c = utcp->connections[i];
- if(!c->reapable) {
- utcp_expect_data(c, offline);
-
- // If we are online again, reset the retransmission timers, but keep the connection timeout as it is,
- // to prevent peers toggling online/offline state frequently from keeping connections alive
- // if there is no progress in sending actual data.
- if(!offline) {
- gettimeofday(&utcp->connections[i]->rtrx_timeout, NULL);
- utcp->connections[i]->rtt_start.tv_sec = 0;
+ if(c->reapable) {
+ continue;
+ }
+
+ utcp_expect_data(c, offline);
+
+ if(!offline) {
+ if(timerisset(&c->rtrx_timeout)) {
+ c->rtrx_timeout = now;
}
+
+ utcp->connections[i]->rtt_start.tv_sec = 0;
}
}