(r)->tv_sec = (a)->tv_sec - (b)->tv_sec;\
(r)->tv_usec = (a)->tv_usec - (b)->tv_usec;\
if((r)->tv_usec < 0)\
- (r)->tv_sec--, (r)->tv_usec += 1000000;\
+ (r)->tv_sec--, (r)->tv_usec += USEC_PER_SEC;\
} while (0)
#endif
-#ifndef max
-#define max(a, b) ((a) > (b) ? (a) : (b))
-#endif
+static inline size_t max(size_t a, size_t b) {
+ return a > b ? a : b;
+}
#ifdef UTCP_DEBUG
#include <stdarg.h>
static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
struct hdr hdr;
if(len < sizeof hdr) {
- debug("%p %s: short packet (%zu bytes)\n", utcp, dir, len);
+ debug("%p %s: short packet (%lu bytes)\n", utcp, dir, (unsigned long)len);
return;
}
memcpy(&hdr, pkt, sizeof hdr);
- fprintf (stderr, "%p %s: len=%zu, src=%u dst=%u seq=%u ack=%u wnd=%u ctl=", utcp, dir, len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd);
+ debug("%p %s: len=%lu, src=%u dst=%u seq=%u ack=%u wnd=%u ctl=", utcp, dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd);
if(hdr.ctl & SYN)
debug("SYN");
if(hdr.ctl & RST)
debug("ACK");
if(len > sizeof hdr) {
- debug(" data=");
- for(int i = sizeof hdr; i < len; i++) {
- const char *data = pkt;
- debug("%c", data[i] >= 32 ? data[i] : '.');
+ uint32_t datalen = len - sizeof hdr;
+ const uint8_t *data = (uint8_t *)pkt + sizeof hdr;
+ char str[datalen * 2 + 1];
+ char *p = str;
+
+ for(uint32_t i = 0; i < datalen; i++) {
+ *p++ = "0123456789ABCDEF"[data[i] >> 4];
+ *p++ = "0123456789ABCDEF"[data[i] & 15];
}
+ *p = 0;
+
+ debug(" data=%s", str);
}
debug("\n");
// Store data into the buffer
static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
- if(buf->maxsize <= buf->used)
- return 0;
-
- debug("buffer_put_at %zu %zu %zu\n", buf->used, offset, len);
+ debug("buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
size_t required = offset + len;
if(required > buf->maxsize) {
if(offset >= buf->maxsize)
return 0;
- abort();
len = buf->maxsize - offset;
required = buf->maxsize;
}
} else {
do {
newsize *= 2;
- } while(newsize < buf->used + len);
+ } while(newsize < required);
}
if(newsize > buf->maxsize)
newsize = buf->maxsize;
memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
utcp->nconnections--;
+ buffer_exit(&c->rcvbuf);
buffer_exit(&c->sndbuf);
free(c);
}
}
if(!buffer_init(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
+ buffer_exit(&c->sndbuf);
free(c);
return NULL;
}
c->src = src;
c->dst = dst;
#ifdef UTCP_DEBUG
-#warning debugging
c->snd.iss = 0;
#else
c->snd.iss = rand();
return c;
}
+// Update RTT variables. See RFC 6298.
+static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
+ if(!rtt) {
+ debug("invalid rtt\n");
+ return;
+ }
+
+ struct utcp *utcp = c->utcp;
+
+ if(!utcp->srtt) {
+ utcp->srtt = rtt;
+ utcp->rttvar = rtt / 2;
+ utcp->rto = rtt + max(2 * rtt, CLOCK_GRANULARITY);
+ } else {
+ utcp->rttvar = (utcp->rttvar * 3 + abs(utcp->srtt - rtt)) / 4;
+ utcp->srtt = (utcp->srtt * 7 + rtt) / 8;
+ utcp->rto = utcp->srtt + max(utcp->rttvar, CLOCK_GRANULARITY);
+ }
+
+ if(utcp->rto > MAX_RTO)
+ utcp->rto = MAX_RTO;
+
+ debug("rtt %u srtt %u rttvar %u rto %u\n", rtt, utcp->srtt, utcp->rttvar, utcp->rto);
+}
+
+static void start_retransmit_timer(struct utcp_connection *c) {
+ gettimeofday(&c->rtrx_timeout, NULL);
+ c->rtrx_timeout.tv_usec += c->utcp->rto;
+ while(c->rtrx_timeout.tv_usec >= 1000000) {
+ c->rtrx_timeout.tv_usec -= 1000000;
+ c->rtrx_timeout.tv_sec++;
+ }
+ debug("timeout set to %lu.%06lu (%u)\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_usec, c->utcp->rto);
+}
+
+static void stop_retransmit_timer(struct utcp_connection *c) {
+ timerclear(&c->rtrx_timeout);
+ debug("timeout cleared\n");
+}
+
struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
struct utcp_connection *c = allocate_connection(utcp, 0, dst);
if(!c)
pkt->hdr.ctl |= FIN;
}
+ if(!c->rtt_start.tv_sec) {
+ // Start RTT measurement
+ gettimeofday(&c->rtt_start, NULL);
+ c->rtt_seq = pkt->hdr.seq + seglen;
+ debug("Starting RTT measurement, expecting ack %u\n", c->rtt_seq);
+ }
+
print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
} while(left);
c->snd.last += len;
ack(c, false);
+ if(!timerisset(&c->rtrx_timeout))
+ start_retransmit_timer(c);
return len;
}
}
static void retransmit(struct utcp_connection *c) {
- if(c->state == CLOSED || c->snd.nxt == c->snd.una)
+ if(c->state == CLOSED || c->snd.last == c->snd.una) {
+ debug("Retransmit() called but nothing to retransmit!\n");
+ stop_retransmit_timer(c);
return;
+ }
struct utcp *utcp = c->utcp;
pkt->hdr.src = c->src;
pkt->hdr.dst = c->dst;
+ pkt->hdr.wnd = c->rcv.wnd;
+ pkt->hdr.aux = 0;
switch(c->state) {
case SYN_SENT:
// Send our SYN again
pkt->hdr.seq = c->snd.iss;
pkt->hdr.ack = 0;
- pkt->hdr.wnd = c->rcv.wnd;
pkt->hdr.ctl = SYN;
print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
utcp->send(utcp, pkt, sizeof pkt->hdr);
#ifdef UTCP_DEBUG
abort();
#endif
- timerclear(&c->rtrx_timeout);
- break;
+ stop_retransmit_timer(c);
+ goto cleanup;
}
+ start_retransmit_timer(c);
+ utcp->rto *= 2;
+ if(utcp->rto > MAX_RTO)
+ utcp->rto = MAX_RTO;
+ c->rtt_start.tv_sec = 0; // invalidate RTT timer
+
+cleanup:
free(pkt);
}
-// Update receive buffer and SACK entries after consuming data.
+/* Update receive buffer and SACK entries after consuming data.
+ *
+ * Situation:
+ *
+ * |.....0000..1111111111.....22222......3333|
+ * |---------------^
+ *
+ * 0..3 represent the SACK entries. The ^ indicates up to which point we want
+ * to remove data from the receive buffer. The idea is to substract "len"
+ * from the offset of all the SACK entries, and then remove/cut down entries
+ * that are shifted to before the start of the receive buffer.
+ *
+ * There are three cases:
+ * - the SACK entry is after ^, in that case just change the offset.
+ * - the SACK entry starts before and ends after ^, so we have to
+ * change both its offset and size.
+ * - the SACK entry is completely before ^, in that case delete it.
+ */
static void sack_consume(struct utcp_connection *c, size_t len) {
- debug("sack_consume %zu\n", len);
- if(len > c->rcvbuf.used)
- abort();
+ debug("sack_consume %lu\n", (unsigned long)len);
+ if(len > c->rcvbuf.used) {
+ debug("All SACK entries consumed");
+ c->sacks[0].len = 0;
+ return;
+ }
buffer_get(&c->rcvbuf, NULL, len);
c->sacks[i].offset -= len;
i++;
} else if(len < c->sacks[i].offset + c->sacks[i].len) {
- c->sacks[i].offset = 0;
c->sacks[i].len -= len - c->sacks[i].offset;
+ c->sacks[i].offset = 0;
i++;
} else {
if(i < NSACKS - 1) {
memmove(&c->sacks[i], &c->sacks[i + 1], (NSACKS - 1 - i) * sizeof c->sacks[i]);
- c->sacks[i + 1].len = 0;
+ c->sacks[NSACKS - 1].len = 0;
} else {
c->sacks[i].len = 0;
break;
memmove(&c->sacks[i + 1], &c->sacks[i], (NSACKS - i - 1) * sizeof c->sacks[i]);
c->sacks[i].offset = offset;
c->sacks[i].len = rxd;
+ } else {
+ debug("SACK entries full, dropping packet\n");
}
break;
} else { // merge
static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
// Check if we can process out-of-order data now.
if(c->sacks[0].len && len >= c->sacks[0].offset) { // TODO: handle overlap with second SACK
- debug("incoming packet len %zu connected with SACK at %u\n", len, c->sacks[0].offset);
+ debug("incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
buffer_put_at(&c->rcvbuf, 0, data, len); // TODO: handle return value
len = max(len, c->sacks[0].offset + c->sacks[0].len);
data = c->rcvbuf.data;
// In case this is for a CLOSED connection, ignore the packet.
// TODO: make it so incoming packets can never match a CLOSED connection.
- if(c->state == CLOSED)
+ if(c->state == CLOSED) {
+ debug("Got packet for closed connection\n");
return 0;
+ }
// It is for an existing connection.
if(c->state == SYN_SENT)
acceptable = true;
-
- // TODO: handle packets overlapping c->rcv.nxt.
-#if 1
- // Only use this when accepting out-of-order packets.
else if(len == 0)
acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0;
- else
+ else {
+ int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
+
+ // cut already accepted front overlapping
+ if(rcv_offset < 0) {
+ acceptable = rcv_offset + len >= 0;
+ if(acceptable) {
+ data -= rcv_offset;
+ len += rcv_offset;
+ }
+ }
+
acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
-#else
- if(c->state != SYN_SENT)
- acceptable = hdr.seq == c->rcv.nxt;
-#endif
+ }
if(!acceptable) {
- debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcvbuf.maxsize);
+ 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;
// 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(hdr.ctl & ACK &&
- ((seqdiff(hdr.ack, c->snd.una + c->sndbuf.used) > 0 &&
- seqdiff(hdr.ack, c->snd.nxt) > 0) // TODO: simplify this if
- || seqdiff(hdr.ack, c->snd.una) < 0)) {
+ 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);
// Ignore unacceptable RST packets.
if(hdr.ctl & RST)
prevrcvnxt = c->rcv.nxt;
if(advanced) {
+ // RTT measurement
+ if(c->rtt_start.tv_sec) {
+ if(c->rtt_seq == hdr.ack) {
+ struct timeval now, diff;
+ gettimeofday(&now, NULL);
+ timersub(&now, &c->rtt_start, &diff);
+ update_rtt(c, diff.tv_sec * 1000000 + diff.tv_usec);
+ c->rtt_start.tv_sec = 0;
+ } else if(c->rtt_seq < hdr.ack) {
+ debug("Cancelling RTT measurement: %u < %u\n", c->rtt_seq, hdr.ack);
+ c->rtt_start.tv_sec = 0;
+ }
+ }
+
int32_t data_acked = advanced;
switch(c->state) {
//Reset the congestion window so we wait for ACKs.
c->snd.nxt = c->snd.una;
c->snd.cwnd = utcp->mtu;
+ start_retransmit_timer(c);
}
}
}
if(advanced) {
timerclear(&c->conn_timeout); // It will be set anew in utcp_timeout() if c->snd.una != c->snd.nxt.
- if(c->snd.una == c->snd.nxt)
- timerclear(&c->rtrx_timeout);
+ if(c->snd.una == c->snd.last)
+ stop_retransmit_timer(c);
+ else
+ start_retransmit_timer(c);
}
// 5. Process SYN stuff
// - or we got an ack, so we should maybe send a bit more data
// -> sendatleastone = false
-ack:
- ack(c, prevrcvnxt != c->rcv.nxt);
+ ack(c, len || prevrcvnxt != c->rcv.nxt);
return 0;
reset:
c->snd.last++;
ack(c, false);
+ if(!timerisset(&c->rtrx_timeout))
+ start_retransmit_timer(c);
return 0;
}
int utcp_close(struct utcp_connection *c) {
- if(utcp_shutdown(c, SHUT_RDWR))
+ if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN)
return -1;
c->recv = NULL;
c->poll = NULL;
if(!c)
continue;
+ // delete connections that have been utcp_close()d.
if(c->state == CLOSED) {
if(c->reapable) {
debug("Reaping %p\n", c);
}
if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
+ debug("retransmit()\n");
retransmit(c);
}
if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
next = c->conn_timeout;
- if(c->snd.nxt != c->snd.una) {
- c->rtrx_timeout = now;
- c->rtrx_timeout.tv_sec++;
- } else {
- timerclear(&c->rtrx_timeout);
- }
-
if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
next = c->rtrx_timeout;
}
utcp->pre_accept = pre_accept;
utcp->send = send;
utcp->priv = priv;
- utcp->mtu = 1000;
- utcp->timeout = 60;
+ utcp->mtu = DEFAULT_MTU;
+ utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
+ utcp->rto = START_RTO; // usec
return utcp;
}
for(int i = 0; i < utcp->nconnections; i++) {
if(!utcp->connections[i]->reapable)
debug("Warning, freeing unclosed connection %p\n", utcp->connections[i]);
+ buffer_exit(&utcp->connections[i]->rcvbuf);
buffer_exit(&utcp->connections[i]->sndbuf);
free(utcp->connections[i]);
}
projects / utcp / blobdiff