/*
utcp.c -- Userspace TCP
- Copyright (C) 2014 Guus Sliepen <guus@tinc-vpn.org>
+ Copyright (C) 2014-2017 Guus Sliepen <guus@tinc-vpn.org>
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
#include <stdbool.h>
#include <string.h>
#include <unistd.h>
-#include <sys/time.h>
-#include <sys/socket.h>
+#include <time.h>
#include "utcp_priv.h"
#undef poll
#endif
-#ifndef timersub
-#define timersub(a, b, r) do {\
- (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;\
-} while (0)
+#ifndef UTCP_CLOCK
+#if defined(CLOCK_MONOTONIC_RAW) && defined(__x86_64__)
+#define UTCP_CLOCK CLOCK_MONOTONIC_RAW
+#else
+#define UTCP_CLOCK CLOCK_MONOTONIC
+#endif
#endif
+static void timespec_sub(const struct timespec *a, const struct timespec *b, struct timespec *r) {
+ r->tv_sec = a->tv_sec - b->tv_sec;
+ r->tv_nsec = a->tv_nsec - b->tv_nsec;
+
+ if(r->tv_nsec < 0) {
+ r->tv_sec--, r->tv_nsec += NSEC_PER_SEC;
+ }
+}
+
+static int32_t timespec_diff_usec(const struct timespec *a, const struct timespec *b) {
+ return (a->tv_sec - b->tv_sec) * 1000000 + (a->tv_nsec - b->tv_nsec) / 1000;
+}
+
+static bool timespec_lt(const struct timespec *a, const struct timespec *b) {
+ if(a->tv_sec == b->tv_sec) {
+ return a->tv_nsec < b->tv_nsec;
+ } else {
+ return a->tv_sec < b->tv_sec;
+ }
+}
+
+static void timespec_clear(struct timespec *a) {
+ a->tv_sec = 0;
+ a->tv_nsec = 0;
+}
+
+static bool timespec_isset(const struct timespec *a) {
+ return a->tv_sec;
+}
+
+static long CLOCK_GRANULARITY; // usec
+
+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;
+}
+
#ifdef UTCP_DEBUG
#include <stdarg.h>
-static void debug(const char *format, ...) {
+#ifndef UTCP_DEBUG_DATALEN
+#define UTCP_DEBUG_DATALEN 20
+#endif
+
+static void debug(struct utcp_connection *c, const char *format, ...) {
+ struct timespec tv;
+ char buf[1024];
+ int len;
+
+ clock_gettime(CLOCK_REALTIME, &tv);
+ len = snprintf(buf, sizeof(buf), "%ld.%06lu %u:%u ", (long)tv.tv_sec, tv.tv_nsec / 1000, c ? c->src : 0, c ? c->dst : 0);
va_list ap;
va_start(ap, format);
- vfprintf(stderr, format, ap);
+ len += vsnprintf(buf + len, sizeof(buf) - len, format, ap);
va_end(ap);
+
+ if(len > 0 && (size_t)len < sizeof(buf)) {
+ fwrite(buf, len, 1, stderr);
+ }
}
-static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, size_t len) {
+static void print_packet(struct utcp_connection *c, 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);
+
+ if(len < sizeof(hdr)) {
+ debug(c, "%s: short packet (%lu bytes)\n", 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);
- if(hdr.ctl & SYN)
- debug("SYN");
- if(hdr.ctl & RST)
- debug("RST");
- if(hdr.ctl & FIN)
- debug("FIN");
- if(hdr.ctl & ACK)
- debug("ACK");
+ memcpy(&hdr, pkt, sizeof(hdr));
- 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;
+
+ if(len > sizeof(hdr)) {
+ datalen = min(len - sizeof(hdr), UTCP_DEBUG_DATALEN);
+ } else {
+ datalen = 0;
+ }
+
+
+ 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];
}
- debug("\n");
+ *p = 0;
+
+ debug(c, "%s: len %lu src %u dst %u seq %u ack %u wnd %u aux %x ctl %s%s%s%s%s data %s\n",
+ dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux,
+ hdr.ctl & SYN ? "SYN" : "",
+ hdr.ctl & RST ? "RST" : "",
+ hdr.ctl & FIN ? "FIN" : "",
+ hdr.ctl & ACK ? "ACK" : "",
+ hdr.ctl & MF ? "MF" : "",
+ str
+ );
+}
+
+static void debug_cwnd(struct utcp_connection *c) {
+ debug(c, "snd.cwnd %u snd.ssthresh %u\n", c->snd.cwnd, ~c->snd.ssthresh ? c->snd.ssthresh : 0);
}
#else
-#define debug(...)
-#define print_packet(...)
+#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) {
c->state = state;
- if(state == ESTABLISHED)
- timerclear(&c->conn_timeout);
- debug("%p new state: %s\n", c->utcp, strstate[state]);
+
+ if(state == ESTABLISHED) {
+ timespec_clear(&c->conn_timeout);
+ }
+
+ debug(c, "state %s\n", strstate[state]);
}
static bool fin_wanted(struct utcp_connection *c, uint32_t seq) {
- if(seq != c->snd.last)
+ if(seq != c->snd.last) {
return false;
+ }
+
switch(c->state) {
case FIN_WAIT_1:
case CLOSING:
case LAST_ACK:
return true;
+
default:
return false;
}
}
-static inline void list_connections(struct utcp *utcp) {
- debug("%p has %d connections:\n", utcp, utcp->nconnections);
- for(int i = 0; i < utcp->nconnections; i++)
- debug(" %u -> %u state %s\n", utcp->connections[i]->src, utcp->connections[i]->dst, strstate[utcp->connections[i]->state]);
+static bool is_reliable(struct utcp_connection *c) {
+ return c->flags & UTCP_RELIABLE;
}
static int32_t seqdiff(uint32_t a, uint32_t b) {
}
// Buffer functions
-// TODO: convert to ringbuffers to avoid memmove() operations.
+static bool buffer_wraps(struct buffer *buf) {
+ return buf->size - buf->offset < buf->used;
+}
+
+static bool buffer_resize(struct buffer *buf, uint32_t newsize) {
+ char *newdata = realloc(buf->data, newsize);
+
+ if(!newdata) {
+ return false;
+ }
+
+ buf->data = newdata;
+
+ if(buffer_wraps(buf)) {
+ // Shift the right part of the buffer until it hits the end of the new buffer.
+ // Old situation:
+ // [345......012]
+ // New situation:
+ // [345.........|........012]
+ uint32_t tailsize = buf->size - buf->offset;
+ uint32_t newoffset = newsize - tailsize;
+ memmove(buf->data + newoffset, buf->data + buf->offset, tailsize);
+ buf->offset = newoffset;
+ }
+
+ buf->size = newsize;
+ return true;
+}
// Store data into the buffer
-static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
- if(buf->maxsize <= buf->used)
- return 0;
- if(len > buf->maxsize - buf->used)
- len = buf->maxsize - buf->used;
- if(len > buf->size - buf->used) {
+static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data, size_t len) {
+ debug(NULL, "buffer_put_at %lu %lu %lu\n", (unsigned long)buf->used, (unsigned long)offset, (unsigned long)len);
+
+ // Ensure we don't store more than maxsize bytes in total
+ size_t required = offset + len;
+
+ if(required > buf->maxsize) {
+ if(offset >= buf->maxsize) {
+ return 0;
+ }
+
+ len = buf->maxsize - offset;
+ required = buf->maxsize;
+ }
+
+ // Check if we need to resize the buffer
+ if(required > buf->size) {
size_t newsize = buf->size;
+
+ if(!newsize) {
+ newsize = 4096;
+ }
+
do {
newsize *= 2;
- } while(newsize < buf->used + len);
- if(newsize > buf->maxsize)
+ } while(newsize < required);
+
+ if(newsize > buf->maxsize) {
newsize = buf->maxsize;
- char *newdata = realloc(buf->data, newsize);
- if(!newdata)
+ }
+
+ if(!buffer_resize(buf, newsize)) {
return -1;
- buf->data = newdata;
- buf->size = newsize;
+ }
+ }
+
+ uint32_t realoffset = buf->offset + offset;
+
+ if(buf->size - buf->offset < offset) {
+ // The offset wrapped
+ realoffset -= buf->size;
}
- memcpy(buf->data + buf->used, data, len);
- buf->used += len;
+
+ if(buf->size - realoffset < len) {
+ // The new chunk of data must be wrapped
+ memcpy(buf->data + realoffset, data, buf->size - realoffset);
+ memcpy(buf->data, (char *)data + buf->size - realoffset, len - (buf->size - realoffset));
+ } else {
+ memcpy(buf->data + realoffset, data, len);
+ }
+
+ if(required > buf->used) {
+ buf->used = required;
+ }
+
return len;
}
-// Get data from the buffer. data can be NULL.
-static ssize_t buffer_get(struct buffer *buf, void *data, size_t len) {
- if(len > buf->used)
- len = buf->used;
- if(data)
- memcpy(data, buf->data, len);
- if(len < buf->used)
- memmove(buf->data, buf->data + len, buf->used - len);
- buf->used -= len;
- return len;
+static ssize_t buffer_put(struct buffer *buf, const void *data, size_t len) {
+ return buffer_put_at(buf, buf->used, data, len);
}
// Copy data from the buffer without removing it.
static ssize_t buffer_copy(struct buffer *buf, void *data, size_t offset, size_t len) {
- if(offset >= buf->used)
+ // Ensure we don't copy more than is actually stored in the buffer
+ if(offset >= buf->used) {
return 0;
- if(offset + len > buf->used)
+ }
+
+ if(buf->used - offset < len) {
len = buf->used - offset;
- memcpy(data, buf->data + offset, len);
+ }
+
+ uint32_t realoffset = buf->offset + offset;
+
+ if(buf->size - buf->offset < offset) {
+ // The offset wrapped
+ realoffset -= buf->size;
+ }
+
+ if(buf->size - realoffset < len) {
+ // The data is wrapped
+ memcpy(data, buf->data + realoffset, buf->size - realoffset);
+ memcpy((char *)data + buf->size - realoffset, buf->data, len - (buf->size - realoffset));
+ } else {
+ memcpy(data, buf->data + realoffset, len);
+ }
+
return len;
}
-static bool buffer_init(struct buffer *buf, uint32_t len, uint32_t maxlen) {
- memset(buf, 0, sizeof *buf);
- buf->data = malloc(len);
- if(!len)
- return false;
- buf->size = len;
- buf->maxsize = maxlen;
- return true;
+// Copy data from the buffer without removing it.
+static ssize_t buffer_call(struct buffer *buf, utcp_recv_t cb, void *arg, size_t offset, size_t len) {
+ // Ensure we don't copy more than is actually stored in the buffer
+ if(offset >= buf->used) {
+ return 0;
+ }
+
+ if(buf->used - offset < len) {
+ len = buf->used - offset;
+ }
+
+ uint32_t realoffset = buf->offset + offset;
+
+ if(buf->size - buf->offset < offset) {
+ // The offset wrapped
+ realoffset -= buf->size;
+ }
+
+ if(buf->size - realoffset < len) {
+ // The data is wrapped
+ ssize_t rx1 = cb(arg, buf->data + realoffset, buf->size - realoffset);
+
+ if(rx1 < buf->size - realoffset) {
+ return rx1;
+ }
+
+ ssize_t rx2 = cb(arg, buf->data, len - (buf->size - realoffset));
+
+ if(rx2 < 0) {
+ return rx2;
+ } else {
+ return rx1 + rx2;
+ }
+ } else {
+ return cb(arg, buf->data + realoffset, len);
+ }
+}
+
+// Discard data from the buffer.
+static ssize_t buffer_discard(struct buffer *buf, size_t len) {
+ if(buf->used < len) {
+ len = buf->used;
+ }
+
+ if(buf->size - buf->offset < len) {
+ buf->offset -= buf->size;
+ }
+
+ if(buf->used == len) {
+ buf->offset = 0;
+ } else {
+ buf->offset += len;
+ }
+
+ buf->used -= len;
+
+ return len;
+}
+
+static void buffer_clear(struct buffer *buf) {
+ buf->used = 0;
+ buf->offset = 0;
+}
+
+static bool buffer_set_size(struct buffer *buf, uint32_t minsize, uint32_t maxsize) {
+ if(maxsize < minsize) {
+ maxsize = minsize;
+ }
+
+ buf->maxsize = maxsize;
+
+ return buf->size >= minsize || buffer_resize(buf, minsize);
}
static void buffer_exit(struct buffer *buf) {
free(buf->data);
- memset(buf, 0, sizeof *buf);
+ memset(buf, 0, sizeof(*buf));
}
static uint32_t buffer_free(const struct buffer *buf) {
assert(a->src && b->src);
int c = (int)a->src - (int)b->src;
- if(c)
+
+ if(c) {
return c;
+ }
+
c = (int)a->dst - (int)b->dst;
return c;
}
static struct utcp_connection *find_connection(const struct utcp *utcp, uint16_t src, uint16_t dst) {
- if(!utcp->nconnections)
+ if(!utcp->nconnections) {
return NULL;
+ }
+
struct utcp_connection key = {
.src = src,
.dst = dst,
}, *keyp = &key;
- struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
+ struct utcp_connection **match = bsearch(&keyp, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
return match ? *match : NULL;
}
static void free_connection(struct utcp_connection *c) {
struct utcp *utcp = c->utcp;
- struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
+ struct utcp_connection **cp = bsearch(&c, utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
assert(cp);
int i = cp - utcp->connections;
- memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof *cp);
+ memmove(cp, cp + 1, (utcp->nconnections - i - 1) * sizeof(*cp));
utcp->nconnections--;
+ buffer_exit(&c->rcvbuf);
buffer_exit(&c->sndbuf);
free(c);
}
errno = ENOMEM;
return NULL;
}
+
src = rand() | 0x8000;
- while(find_connection(utcp, src, dst))
+
+ while(find_connection(utcp, src, dst)) {
src++;
+ }
}
// Allocate memory for the new connection
if(utcp->nconnections >= utcp->nallocated) {
- if(!utcp->nallocated)
+ if(!utcp->nallocated) {
utcp->nallocated = 4;
- else
+ } else {
utcp->nallocated *= 2;
- struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof *utcp->connections);
- if(!new_array)
+ }
+
+ struct utcp_connection **new_array = realloc(utcp->connections, utcp->nallocated * sizeof(*utcp->connections));
+
+ if(!new_array) {
return NULL;
+ }
+
utcp->connections = new_array;
}
- struct utcp_connection *c = calloc(1, sizeof *c);
- if(!c)
+ struct utcp_connection *c = calloc(1, sizeof(*c));
+
+ if(!c) {
+ return NULL;
+ }
+
+ if(!buffer_set_size(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
+ free(c);
return NULL;
+ }
- if(!buffer_init(&c->sndbuf, DEFAULT_SNDBUFSIZE, DEFAULT_MAXSNDBUFSIZE)) {
+ if(!buffer_set_size(&c->rcvbuf, DEFAULT_RCVBUFSIZE, DEFAULT_MAXRCVBUFSIZE)) {
+ buffer_exit(&c->sndbuf);
free(c);
return NULL;
}
c->src = src;
c->dst = dst;
+#ifdef UTCP_DEBUG
+ c->snd.iss = 0;
+#else
c->snd.iss = rand();
+#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->mss > 2190 ? 2 : utcp->mss > 1095 ? 3 : 4) * utcp->mss;
+ c->snd.ssthresh = ~0;
+ debug_cwnd(c);
+ c->srtt = 0;
+ c->rttvar = 0;
+ c->rto = START_RTO;
c->utcp = utcp;
// Add it to the sorted list of connections
utcp->connections[utcp->nconnections++] = c;
- qsort(utcp->connections, utcp->nconnections, sizeof *utcp->connections, compare);
+ qsort(utcp->connections, utcp->nconnections, sizeof(*utcp->connections), compare);
return c;
}
-struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
+static inline uint32_t absdiff(uint32_t a, uint32_t b) {
+ if(a > b) {
+ return a - b;
+ } else {
+ return b - a;
+ }
+}
+
+// Update RTT variables. See RFC 6298.
+static void update_rtt(struct utcp_connection *c, uint32_t rtt) {
+ if(!rtt) {
+ debug(c, "invalid rtt\n");
+ return;
+ }
+
+ if(!c->srtt) {
+ c->srtt = rtt;
+ c->rttvar = rtt / 2;
+ } else {
+ c->rttvar = (c->rttvar * 3 + absdiff(c->srtt, rtt)) / 4;
+ c->srtt = (c->srtt * 7 + rtt) / 8;
+ }
+
+ c->rto = c->srtt + max(4 * c->rttvar, CLOCK_GRANULARITY);
+
+ if(c->rto > MAX_RTO) {
+ c->rto = MAX_RTO;
+ }
+
+ debug(c, "rtt %u srtt %u rttvar %u rto %u\n", rtt, c->srtt, c->rttvar, c->rto);
+}
+
+static void start_retransmit_timer(struct utcp_connection *c) {
+ clock_gettime(UTCP_CLOCK, &c->rtrx_timeout);
+
+ uint32_t rto = c->rto;
+
+ while(rto > USEC_PER_SEC) {
+ c->rtrx_timeout.tv_sec++;
+ rto -= USEC_PER_SEC;
+ }
+
+ c->rtrx_timeout.tv_nsec += rto * 1000;
+
+ if(c->rtrx_timeout.tv_nsec >= NSEC_PER_SEC) {
+ c->rtrx_timeout.tv_nsec -= NSEC_PER_SEC;
+ c->rtrx_timeout.tv_sec++;
+ }
+
+ debug(c, "rtrx_timeout %ld.%06lu\n", c->rtrx_timeout.tv_sec, c->rtrx_timeout.tv_nsec);
+}
+
+static void stop_retransmit_timer(struct utcp_connection *c) {
+ timespec_clear(&c->rtrx_timeout);
+ debug(c, "rtrx_timeout cleared\n");
+}
+
+struct utcp_connection *utcp_connect_ex(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv, uint32_t flags) {
struct utcp_connection *c = allocate_connection(utcp, 0, dst);
- if(!c)
+
+ if(!c) {
return NULL;
+ }
+ assert((flags & ~0x1f) == 0);
+
+ c->flags = flags;
c->recv = recv;
c->priv = priv;
- struct hdr hdr;
-
- hdr.src = c->src;
- hdr.dst = c->dst;
- hdr.seq = c->snd.iss;
- hdr.ack = 0;
- hdr.wnd = c->rcv.wnd;
- hdr.ctl = SYN;
- hdr.aux = 0;
+ struct {
+ struct hdr hdr;
+ uint8_t init[4];
+ } pkt;
+
+ pkt.hdr.src = c->src;
+ pkt.hdr.dst = c->dst;
+ pkt.hdr.seq = c->snd.iss;
+ pkt.hdr.ack = 0;
+ pkt.hdr.wnd = c->rcvbuf.maxsize;
+ pkt.hdr.ctl = SYN;
+ pkt.hdr.aux = 0x0101;
+ pkt.init[0] = 1;
+ pkt.init[1] = 0;
+ pkt.init[2] = 0;
+ pkt.init[3] = flags & 0x7;
set_state(c, SYN_SENT);
- print_packet(utcp, "send", &hdr, sizeof hdr);
- utcp->send(utcp, &hdr, sizeof hdr);
+ print_packet(c, "send", &pkt, sizeof(pkt));
+ utcp->send(utcp, &pkt, sizeof(pkt));
- gettimeofday(&c->conn_timeout, NULL);
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
c->conn_timeout.tv_sec += utcp->timeout;
+ start_retransmit_timer(c);
+
return c;
}
+struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) {
+ return utcp_connect_ex(utcp, dst, recv, priv, UTCP_TCP);
+}
+
void utcp_accept(struct utcp_connection *c, utcp_recv_t recv, void *priv) {
if(c->reapable || c->state != SYN_RECEIVED) {
- debug("Error: accept() called on invalid connection %p in state %s\n", c, strstate[c->state]);
+ debug(c, "accept() called on invalid connection in state %s\n", c, strstate[c->state]);
return;
}
- debug("%p accepted, %p %p\n", c, recv, priv);
+ debug(c, "accepted %p %p\n", c, recv, priv);
c->recv = recv;
c->priv = priv;
set_state(c, ESTABLISHED);
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);
+ int32_t cwndleft = is_reliable(c) ? min(c->snd.cwnd, c->snd.wnd) - seqdiff(c->snd.nxt, c->snd.una) : MAX_UNRELIABLE_SIZE;
assert(left >= 0);
- if(cwndleft <= 0)
- cwndleft = 0;
-
- if(cwndleft < left)
+ if(cwndleft <= 0) {
+ left = 0;
+ } else if(cwndleft < left) {
left = cwndleft;
- if(!left && !sendatleastone)
+ if(!sendatleastone || cwndleft > c->utcp->mss) {
+ left -= left % c->utcp->mss;
+ }
+ }
+
+ debug(c, "cwndleft %d left %d\n", cwndleft, left);
+
+ if(!left && !sendatleastone) {
return;
+ }
struct {
struct hdr hdr;
- char data[];
- } *pkt;
-
- pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
- if(!pkt->data)
- return;
+ uint8_t data[];
+ } *pkt = c->utcp->pkt;
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 = is_reliable(c) ? c->rcvbuf.maxsize : 0;
pkt->hdr.ctl = ACK;
pkt->hdr.aux = 0;
do {
- uint32_t seglen = left > c->utcp->mtu ? c->utcp->mtu : left;
+ uint32_t seglen = left > c->utcp->mss ? c->utcp->mss : left;
pkt->hdr.seq = c->snd.nxt;
buffer_copy(&c->sndbuf, pkt->data, seqdiff(c->snd.nxt, c->snd.una), seglen);
c->snd.nxt += seglen;
left -= seglen;
+ if(!is_reliable(c)) {
+ if(left) {
+ pkt->hdr.ctl |= MF;
+ } else {
+ pkt->hdr.ctl &= ~MF;
+ }
+ }
+
if(seglen && fin_wanted(c, c->snd.nxt)) {
seglen--;
pkt->hdr.ctl |= FIN;
}
- print_packet(c->utcp, "send", pkt, sizeof pkt->hdr + seglen);
- c->utcp->send(c->utcp, pkt, sizeof pkt->hdr + seglen);
- } while(left);
+ if(!c->rtt_start.tv_sec) {
+ // Start RTT measurement
+ clock_gettime(UTCP_CLOCK, &c->rtt_start);
+ c->rtt_seq = pkt->hdr.seq + seglen;
+ debug(c, "starting RTT measurement, expecting ack %u\n", c->rtt_seq);
+ }
- free(pkt);
+ print_packet(c, "send", pkt, sizeof(pkt->hdr) + seglen);
+ c->utcp->send(c->utcp, pkt, sizeof(pkt->hdr) + seglen);
+
+ if(left && !is_reliable(c)) {
+ pkt->hdr.wnd += seglen;
+ }
+ } while(left);
}
ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) {
if(c->reapable) {
- debug("Error: send() called on closed connection %p\n", c);
+ debug(c, "send() called on closed connection\n");
errno = EBADF;
return -1;
}
switch(c->state) {
case CLOSED:
case LISTEN:
- case SYN_SENT:
- case SYN_RECEIVED:
- debug("Error: send() called on unconnected connection %p\n", c);
+ debug(c, "send() called on unconnected connection\n");
errno = ENOTCONN;
return -1;
+
+ case SYN_SENT:
+ case SYN_RECEIVED:
case ESTABLISHED:
case CLOSE_WAIT:
break;
+
case FIN_WAIT_1:
case FIN_WAIT_2:
case CLOSING:
case LAST_ACK:
case TIME_WAIT:
- debug("Error: send() called on closing connection %p\n", c);
+ debug(c, "send() called on closed connection\n");
errno = EPIPE;
return -1;
}
- // Add data to send buffer
+ // Exit early if we have nothing to send.
- if(!len)
+ if(!len) {
return 0;
+ }
if(!data) {
errno = EFAULT;
return -1;
}
- len = buffer_put(&c->sndbuf, data, len);
- if(len <= 0) {
- errno = EWOULDBLOCK;
+ // Check if we need to be able to buffer all data
+
+ if(c->flags & UTCP_NO_PARTIAL) {
+ if(len > buffer_free(&c->sndbuf)) {
+ if(len > c->sndbuf.maxsize) {
+ errno = EMSGSIZE;
+ return -1;
+ } else {
+ errno = EWOULDBLOCK;
+ return 0;
+ }
+ }
+ }
+
+ // Add data to send buffer.
+
+ if(is_reliable(c)) {
+ len = buffer_put(&c->sndbuf, data, len);
+ } else if(c->state != SYN_SENT && c->state != SYN_RECEIVED) {
+ if(len > MAX_UNRELIABLE_SIZE || buffer_put(&c->sndbuf, data, len) != (ssize_t)len) {
+ errno = EMSGSIZE;
+ return -1;
+ }
+ } else {
return 0;
}
+ if(len <= 0) {
+ if(is_reliable(c)) {
+ errno = EWOULDBLOCK;
+ return 0;
+ } else {
+ return len;
+ }
+ }
+
c->snd.last += len;
+
+ // Don't send anything yet if the connection has not fully established yet
+
+ if(c->state == SYN_SENT || c->state == SYN_RECEIVED) {
+ return len;
+ }
+
ack(c, false);
+
+ if(!is_reliable(c)) {
+ c->snd.una = c->snd.nxt = c->snd.last;
+ buffer_discard(&c->sndbuf, c->sndbuf.used);
+ c->do_poll = true;
+ }
+
+ if(is_reliable(c) && !timespec_isset(&c->rtrx_timeout)) {
+ start_retransmit_timer(c);
+ }
+
+ if(is_reliable(c) && !timespec_isset(&c->conn_timeout)) {
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
+ c->conn_timeout.tv_sec += c->utcp->timeout;
+ }
+
return len;
}
hdr->dst = tmp;
}
-static void retransmit(struct utcp_connection *c) {
- if(c->state == CLOSED || c->snd.nxt == c->snd.una)
+static void fast_retransmit(struct utcp_connection *c) {
+ if(c->state == CLOSED || c->snd.last == c->snd.una) {
+ debug(c, "fast_retransmit() called but nothing to retransmit!\n");
return;
+ }
struct utcp *utcp = c->utcp;
struct {
struct hdr hdr;
- char data[];
+ uint8_t data[];
} *pkt;
- pkt = malloc(sizeof pkt->hdr + c->utcp->mtu);
- if(!pkt)
+ pkt = malloc(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 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);
- break;
+ 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->mss);
+
+ if(fin_wanted(c, c->snd.una + len)) {
+ len--;
+ pkt->hdr.ctl |= FIN;
+ }
- case SYN_RECEIVED:
- // Send SYNACK again
- pkt->hdr.seq = c->snd.nxt;
- pkt->hdr.ack = c->rcv.nxt;
- pkt->hdr.ctl = SYN | ACK;
- print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr);
- utcp->send(utcp, pkt, sizeof pkt->hdr);
+ buffer_copy(&c->sndbuf, pkt->data, 0, len);
+ print_packet(c, "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(c, "retransmit() called but nothing to retransmit!\n");
+ stop_retransmit_timer(c);
+ return;
+ }
+
+ struct utcp *utcp = c->utcp;
+
+ struct {
+ struct hdr hdr;
+ uint8_t data[];
+ } *pkt = c->utcp->pkt;
+
+ pkt->hdr.src = c->src;
+ pkt->hdr.dst = c->dst;
+ pkt->hdr.wnd = c->rcvbuf.maxsize;
+ 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.ctl = SYN;
+ pkt->hdr.aux = 0x0101;
+ pkt->data[0] = 1;
+ pkt->data[1] = 0;
+ pkt->data[2] = 0;
+ pkt->data[3] = c->flags & 0x7;
+ print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + 4);
+ utcp->send(utcp, pkt, sizeof(pkt->hdr) + 4);
+ break;
+
+ case SYN_RECEIVED:
+ // Send SYNACK again
+ pkt->hdr.seq = c->snd.nxt;
+ pkt->hdr.ack = c->rcv.nxt;
+ pkt->hdr.ctl = SYN | ACK;
+ print_packet(c, "rtrx", pkt, sizeof(pkt->hdr));
+ utcp->send(utcp, pkt, sizeof(pkt->hdr));
+ break;
+
+ 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->mss);
+
+ if(fin_wanted(c, c->snd.una + len)) {
+ len--;
+ pkt->hdr.ctl |= FIN;
+ }
+
+ // RFC 5681 slow start after timeout
+ uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
+ c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
+ c->snd.cwnd = utcp->mss;
+ debug_cwnd(c);
+
+ buffer_copy(&c->sndbuf, pkt->data, 0, len);
+ print_packet(c, "rtrx", pkt, sizeof(pkt->hdr) + len);
+ utcp->send(utcp, pkt, sizeof(pkt->hdr) + len);
+
+ c->snd.nxt = c->snd.una + len;
+ break;
+
+ case CLOSED:
+ case LISTEN:
+ case TIME_WAIT:
+ case FIN_WAIT_2:
+ // We shouldn't need to retransmit anything in this state.
+#ifdef UTCP_DEBUG
+ abort();
+#endif
+ stop_retransmit_timer(c);
+ goto cleanup;
+ }
+
+ start_retransmit_timer(c);
+ c->rto *= 2;
+
+ if(c->rto > MAX_RTO) {
+ c->rto = MAX_RTO;
+ }
+
+ c->rtt_start.tv_sec = 0; // invalidate RTT timer
+ c->dupack = 0; // cancel any ongoing fast recovery
+
+cleanup:
+ return;
+}
+
+/* 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(c, "sack_consume %lu\n", (unsigned long)len);
+
+ if(len > c->rcvbuf.used) {
+ debug(c, "all SACK entries consumed\n");
+ c->sacks[0].len = 0;
+ return;
+ }
+
+ buffer_discard(&c->rcvbuf, len);
+
+ for(int i = 0; i < NSACKS && c->sacks[i].len;) {
+ if(len < c->sacks[i].offset) {
+ c->sacks[i].offset -= len;
+ i++;
+ } else if(len < c->sacks[i].offset + c->sacks[i].len) {
+ 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[NSACKS - 1].len = 0;
+ } else {
+ c->sacks[i].len = 0;
+ break;
+ }
+ }
+ }
+
+ for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
+ debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
+ }
+}
+
+static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, const void *data, size_t len) {
+ debug(c, "out of order packet, offset %u\n", offset);
+ // Packet loss or reordering occured. Store the data in the buffer.
+ ssize_t rxd = buffer_put_at(&c->rcvbuf, offset, data, len);
+
+ if(rxd < 0 || (size_t)rxd < len) {
+ abort();
+ }
+
+ // Make note of where we put it.
+ for(int i = 0; i < NSACKS; i++) {
+ if(!c->sacks[i].len) { // nothing to merge, add new entry
+ debug(c, "new SACK entry %d\n", i);
+ c->sacks[i].offset = offset;
+ c->sacks[i].len = rxd;
break;
+ } else if(offset < c->sacks[i].offset) {
+ if(offset + rxd < c->sacks[i].offset) { // insert before
+ if(!c->sacks[NSACKS - 1].len) { // only if room left
+ debug(c, "insert SACK entry at %d\n", i);
+ 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(c, "SACK entries full, dropping packet\n");
+ }
- 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 = seqdiff(c->snd.last, c->snd.una);
- if(len > utcp->mtu)
- len = utcp->mtu;
- if(fin_wanted(c, c->snd.una + len)) {
- len--;
- pkt->hdr.ctl |= FIN;
+ break;
+ } else { // merge
+ debug(c, "merge with start of SACK entry at %d\n", i);
+ c->sacks[i].offset = offset;
+ break;
}
- 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);
+ } else if(offset <= c->sacks[i].offset + c->sacks[i].len) {
+ if(offset + rxd > c->sacks[i].offset + c->sacks[i].len) { // merge
+ debug(c, "merge with end of SACK entry at %d\n", i);
+ c->sacks[i].len = offset + rxd - c->sacks[i].offset;
+ // TODO: handle potential merge with next entry
+ }
+
break;
+ }
+ }
- case CLOSED:
- case LISTEN:
- case TIME_WAIT:
- case FIN_WAIT_2:
- // We shouldn't need to retransmit anything in this state.
-#ifdef UTCP_DEBUG
+ for(int i = 0; i < NSACKS && c->sacks[i].len; i++) {
+ debug(c, "SACK[%d] offset %u len %u\n", i, c->sacks[i].offset, c->sacks[i].len);
+ }
+}
+
+static void handle_in_order(struct utcp_connection *c, const void *data, size_t len) {
+ if(c->recv) {
+ ssize_t rxd = c->recv(c, data, len);
+
+ if(rxd != (ssize_t)len) {
+ // TODO: handle the application not accepting all data.
abort();
-#endif
- timerclear(&c->rtrx_timeout);
- break;
+ }
}
- free(pkt);
+ // Check if we can process out-of-order data now.
+ if(c->sacks[0].len && len >= c->sacks[0].offset) {
+ debug(c, "incoming packet len %lu connected with SACK at %u\n", (unsigned long)len, c->sacks[0].offset);
+
+ if(len < c->sacks[0].offset + c->sacks[0].len) {
+ size_t offset = len;
+ len = c->sacks[0].offset + c->sacks[0].len;
+ size_t remainder = len - offset;
+ ssize_t rxd = buffer_call(&c->rcvbuf, c->recv, c, offset, remainder);
+
+ if(rxd != (ssize_t)remainder) {
+ // TODO: handle the application not accepting all data.
+ abort();
+ }
+ }
+ }
+
+ if(c->rcvbuf.used) {
+ sack_consume(c, len);
+ }
+
+ c->rcv.nxt += len;
+}
+
+static void handle_unreliable(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
+ // Fast path for unfragmented packets
+ if(!hdr->wnd && !(hdr->ctl & MF)) {
+ c->recv(c, data, len);
+ c->rcv.nxt = hdr->seq + len;
+ return;
+ }
+
+ // Ensure reassembled packet are not larger than 64 kiB
+ if(hdr->wnd >= MAX_UNRELIABLE_SIZE || hdr->wnd + len > MAX_UNRELIABLE_SIZE) {
+ return;
+ }
+
+ // Don't accept out of order fragments
+ if(hdr->wnd && hdr->seq != c->rcv.nxt) {
+ return;
+ }
+
+ // Reset the receive buffer for the first fragment
+ if(!hdr->wnd) {
+ buffer_clear(&c->rcvbuf);
+ }
+
+ ssize_t rxd = buffer_put_at(&c->rcvbuf, hdr->wnd, data, len);
+
+ if(rxd != (ssize_t)len) {
+ return;
+ }
+
+ // Send the packet if it's the final fragment
+ if(!(hdr->ctl & MF)) {
+ buffer_call(&c->rcvbuf, c->recv, c, 0, hdr->wnd + len);
+ }
+
+ c->rcv.nxt = hdr->seq + len;
+}
+
+static void handle_incoming_data(struct utcp_connection *c, const struct hdr *hdr, const void *data, size_t len) {
+ if(!is_reliable(c)) {
+ handle_unreliable(c, hdr, data, len);
+ return;
+ }
+
+ uint32_t offset = seqdiff(hdr->seq, c->rcv.nxt);
+
+ if(offset + len > c->rcvbuf.maxsize) {
+ abort();
+ }
+
+ if(offset) {
+ handle_out_of_order(c, offset, data, len);
+ } else {
+ handle_in_order(c, data, len);
+ }
}
ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) {
+ const uint8_t *ptr = data;
+
if(!utcp) {
errno = EFAULT;
return -1;
}
- if(!len)
+ if(!len) {
return 0;
+ }
if(!data) {
errno = EFAULT;
return -1;
}
- print_packet(utcp, "recv", data, len);
-
// Drop packets smaller than the header
struct hdr hdr;
- if(len < sizeof hdr) {
+
+ if(len < sizeof(hdr)) {
+ print_packet(NULL, "recv", data, len);
errno = EBADMSG;
return -1;
}
// Make a copy from the potentially unaligned data to a struct hdr
- memcpy(&hdr, data, sizeof hdr);
- data += sizeof hdr;
- len -= sizeof hdr;
+ memcpy(&hdr, ptr, sizeof(hdr));
+
+ // Try to match the packet to an existing connection
+
+ struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
+ print_packet(c, "recv", data, len);
+
+ // Process the header
+
+ ptr += sizeof(hdr);
+ len -= sizeof(hdr);
// Drop packets with an unknown CTL flag
- if(hdr.ctl & ~(SYN | ACK | RST | FIN)) {
+ if(hdr.ctl & ~(SYN | ACK | RST | FIN | MF)) {
+ print_packet(NULL, "recv", data, len);
errno = EBADMSG;
return -1;
}
- // Try to match the packet to an existing connection
+ // Check for auxiliary headers
- struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src);
+ const uint8_t *init = NULL;
+
+ uint16_t aux = hdr.aux;
+
+ while(aux) {
+ size_t auxlen = 4 * (aux >> 8) & 0xf;
+ uint8_t auxtype = aux & 0xff;
+
+ if(len < auxlen) {
+ errno = EBADMSG;
+ return -1;
+ }
+
+ switch(auxtype) {
+ case AUX_INIT:
+ if(!(hdr.ctl & SYN) || auxlen != 4) {
+ errno = EBADMSG;
+ return -1;
+ }
+
+ init = ptr;
+ break;
+
+ default:
+ errno = EBADMSG;
+ return -1;
+ }
+
+ len -= auxlen;
+ ptr += auxlen;
+
+ if(!(aux & 0x800)) {
+ break;
+ }
+
+ if(len < 2) {
+ errno = EBADMSG;
+ return -1;
+ }
+
+ memcpy(&aux, ptr, 2);
+ len -= 2;
+ ptr += 2;
+ }
+
+ bool has_data = len || (hdr.ctl & (SYN | FIN));
// Is it for a new connection?
if(!c) {
// Ignore RST packets
- if(hdr.ctl & RST)
+ if(hdr.ctl & RST) {
return 0;
+ }
// Is it a SYN packet and are we LISTENing?
// Try to allocate memory, otherwise send a RST back
c = allocate_connection(utcp, hdr.dst, hdr.src);
+
if(!c) {
len = 1;
goto reset;
}
+ // Parse auxilliary information
+ if(init) {
+ if(init[0] < 1) {
+ len = 1;
+ goto reset;
+ }
+
+ c->flags = init[3] & 0x7;
+ } else {
+ c->flags = UTCP_TCP;
+ }
+
+synack:
// Return SYN+ACK, go to SYN_RECEIVED state
c->snd.wnd = hdr.wnd;
c->rcv.irs = hdr.seq;
c->rcv.nxt = c->rcv.irs + 1;
set_state(c, SYN_RECEIVED);
- hdr.dst = c->dst;
- hdr.src = c->src;
- hdr.ack = c->rcv.irs + 1;
- hdr.seq = c->snd.iss;
- hdr.ctl = SYN | ACK;
- print_packet(c->utcp, "send", &hdr, sizeof hdr);
- utcp->send(utcp, &hdr, sizeof hdr);
+ struct {
+ struct hdr hdr;
+ uint8_t data[4];
+ } pkt;
+
+ pkt.hdr.src = c->src;
+ pkt.hdr.dst = c->dst;
+ pkt.hdr.ack = c->rcv.irs + 1;
+ pkt.hdr.seq = c->snd.iss;
+ pkt.hdr.wnd = c->rcvbuf.maxsize;
+ pkt.hdr.ctl = SYN | ACK;
+
+ if(init) {
+ pkt.hdr.aux = 0x0101;
+ pkt.data[0] = 1;
+ pkt.data[1] = 0;
+ pkt.data[2] = 0;
+ pkt.data[3] = c->flags & 0x7;
+ print_packet(c, "send", &pkt, sizeof(hdr) + 4);
+ utcp->send(utcp, &pkt, sizeof(hdr) + 4);
+ } else {
+ pkt.hdr.aux = 0;
+ print_packet(c, "send", &pkt, sizeof(hdr));
+ utcp->send(utcp, &pkt, sizeof(hdr));
+ }
} else {
// No, we don't want your packets, send a RST back
len = 1;
return 0;
}
- debug("%p state %s\n", c->utcp, strstate[c->state]);
+ debug(c, "state %s\n", strstate[c->state]);
// 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(c, "got packet for closed connection\n");
return 0;
+ }
// 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.
case LAST_ACK:
case TIME_WAIT:
break;
+
default:
#ifdef UTCP_DEBUG
abort();
break;
}
- // 1b. Drop packets with a sequence number not in our receive window.
-
- bool acceptable;
-
- if(c->state == SYN_SENT)
- acceptable = true;
-
- // TODO: handle packets overlapping c->rcv.nxt.
-#if 0
- // Only use this when accepting out-of-order packets.
- else if(len == 0)
- if(c->rcv.wnd == 0)
- acceptable = hdr.seq == c->rcv.nxt;
- else
- acceptable = (seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt + c->rcv.wnd) < 0);
- else
- if(c->rcv.wnd == 0)
- // We don't accept data when the receive window is zero.
- acceptable = false;
- else
- // Both start and end of packet must be within the receive window
- acceptable = (seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt + c->rcv.wnd) < 0)
- || (seqdiff(hdr.seq + len + 1, c->rcv.nxt) >= 0 && seqdiff(hdr.seq + len - 1, c->rcv.nxt + c->rcv.wnd) < 0);
-#else
- if(c->state != SYN_SENT)
- acceptable = hdr.seq == c->rcv.nxt;
-#endif
+ // 1b. Discard data that is not in our receive window.
- if(!acceptable) {
- debug("Packet not acceptable, %u <= %u + %zu < %u\n", c->rcv.nxt, hdr.seq, len, c->rcv.nxt + c->rcv.wnd);
- // Ignore unacceptable RST packets.
- if(hdr.ctl & RST)
- return 0;
- // Otherwise, send an ACK back in the hope things improve.
- ack(c, true);
- return 0;
+ 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);
+
+ // 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;
+ }
+ } else {
+ acceptable = seqdiff(hdr.seq, c->rcv.nxt) >= 0 && seqdiff(hdr.seq, c->rcv.nxt) + len <= c->rcvbuf.maxsize;
+ }
+ }
+
+ if(!acceptable) {
+ debug(c, "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;
+ }
+
+ // Otherwise, continue processing.
+ len = 0;
+ }
+ } else {
+#if UTCP_DEBUG
+ int32_t rcv_offset = seqdiff(hdr.seq, c->rcv.nxt);
+
+ if(rcv_offset) {
+ debug(c, "packet out of order, offset %u bytes", rcv_offset);
+ }
+
+#endif
}
c->snd.wnd = hdr.wnd; // TODO: move below
// 1c. Drop packets with an invalid ACK.
- // ackno should not roll back, and it should also not be bigger than snd.nxt.
+ // 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(c, "packet ack seqno out of range, %u <= %u < %u\n", c->snd.una, hdr.ack, c->snd.una + c->sndbuf.used);
- if(hdr.ctl & ACK && (seqdiff(hdr.ack, c->snd.nxt) > 0 || seqdiff(hdr.ack, c->snd.una) < 0)) {
- debug("Packet ack seqno out of range, %u %u %u\n", hdr.ack, c->snd.una, c->snd.nxt);
// Ignore unacceptable RST packets.
- if(hdr.ctl & RST)
+ if(hdr.ctl & RST) {
return 0;
+ }
+
goto reset;
}
if(hdr.ctl & RST) {
switch(c->state) {
case SYN_SENT:
- if(!(hdr.ctl & ACK))
+ if(!(hdr.ctl & ACK)) {
return 0;
+ }
+
// The peer has refused our connection.
set_state(c, CLOSED);
errno = ECONNREFUSED;
- if(c->recv)
+
+ if(c->recv) {
c->recv(c, NULL, 0);
+ }
+
+ if(c->poll && !c->reapable) {
+ c->poll(c, 0);
+ }
+
return 0;
+
case SYN_RECEIVED:
- if(hdr.ctl & ACK)
+ if(hdr.ctl & ACK) {
return 0;
+ }
+
// We haven't told the application about this connection yet. Silently delete.
free_connection(c);
return 0;
+
case ESTABLISHED:
case FIN_WAIT_1:
case FIN_WAIT_2:
case CLOSE_WAIT:
- if(hdr.ctl & ACK)
+ if(hdr.ctl & ACK) {
return 0;
+ }
+
// The peer has aborted our connection.
set_state(c, CLOSED);
errno = ECONNRESET;
- if(c->recv)
+
+ if(c->recv) {
c->recv(c, NULL, 0);
+ }
+
+ if(c->poll && !c->reapable) {
+ c->poll(c, 0);
+ }
+
return 0;
+
case CLOSING:
case LAST_ACK:
case TIME_WAIT:
- if(hdr.ctl & ACK)
+ if(hdr.ctl & ACK) {
return 0;
+ }
+
// As far as the application is concerned, the connection has already been closed.
// If it has called utcp_close() already, we can immediately free this connection.
if(c->reapable) {
free_connection(c);
return 0;
}
+
// Otherwise, immediately move to the CLOSED state.
set_state(c, CLOSED);
return 0;
+
default:
#ifdef UTCP_DEBUG
abort();
}
}
+ uint32_t advanced;
+
+ if(!(hdr.ctl & ACK)) {
+ advanced = 0;
+ goto skip_ack;
+ }
+
// 3. Advance snd.una
- uint32_t advanced = seqdiff(hdr.ack, c->snd.una);
- prevrcvnxt = c->rcv.nxt;
+ advanced = seqdiff(hdr.ack, c->snd.una);
if(advanced) {
+ // RTT measurement
+ if(c->rtt_start.tv_sec) {
+ if(c->rtt_seq == hdr.ack) {
+ struct timespec now;
+ clock_gettime(UTCP_CLOCK, &now);
+ int32_t diff = timespec_diff_usec(&now, &c->rtt_start);
+ update_rtt(c, diff);
+ c->rtt_start.tv_sec = 0;
+ } else if(c->rtt_seq < hdr.ack) {
+ debug(c, "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) {
- case SYN_SENT:
- case SYN_RECEIVED:
- data_acked--;
- break;
- // TODO: handle FIN as well.
- default:
- break;
+ case SYN_SENT:
+ case SYN_RECEIVED:
+ data_acked--;
+ break;
+
+ // TODO: handle FIN as well.
+ default:
+ break;
}
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);
+ if(data_acked) {
+ buffer_discard(&c->sndbuf, data_acked);
+ c->do_poll = true;
+ }
+
+ // Also advance snd.nxt if possible
+ if(seqdiff(c->snd.nxt, hdr.ack) < 0) {
+ c->snd.nxt = hdr.ack;
+ }
c->snd.una = hdr.ack;
- c->dupack = 0;
- c->snd.cwnd += utcp->mtu;
- if(c->snd.cwnd > c->sndbuf.maxsize)
+ if(c->dupack) {
+ if(c->dupack >= 3) {
+ debug(c, "fast recovery ended\n");
+ 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->mss); // eq. 2
+ } else {
+ c->snd.cwnd += max(1, (utcp->mss * utcp->mss) / 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:
- if(c->snd.una == c->snd.last)
+ if(c->snd.una == c->snd.last) {
set_state(c, FIN_WAIT_2);
+ }
+
break;
+
case CLOSING:
if(c->snd.una == c->snd.last) {
- gettimeofday(&c->conn_timeout, NULL);
- c->conn_timeout.tv_sec += 60;
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
+ c->conn_timeout.tv_sec += utcp->timeout;
set_state(c, TIME_WAIT);
}
+
break;
+
default:
break;
}
} else {
- if(!len) {
+ if(!len && is_reliable(c) && c->snd.una != c->snd.last) {
c->dupack++;
+ debug(c, "duplicate ACK %d\n", c->dupack);
+
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.
- //This will cause us to start retransmitting, but at the same speed as the incoming ACKs arrive,
- //thus preventing a drop in speed.
- c->snd.nxt = c->snd.una;
+ // RFC 5681 fast recovery
+ debug(c, "fast recovery started\n", c->dupack);
+ uint32_t flightsize = seqdiff(c->snd.nxt, c->snd.una);
+ c->snd.ssthresh = max(flightsize / 2, utcp->mss * 2); // eq. 4
+ c->snd.cwnd = min(c->snd.ssthresh + 3 * utcp->mss, 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->mss;
+
+ if(c->snd.cwnd > c->sndbuf.maxsize) {
+ c->snd.cwnd = c->sndbuf.maxsize;
+ }
+
+ debug_cwnd(c);
}
+
+ // We got an ACK which indicates the other side did get one of our packets.
+ // Reset the retransmission timer to avoid going to slow start,
+ // but don't touch the connection timeout.
+ start_retransmit_timer(c);
}
}
// 4. Update timers
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);
+ timespec_clear(&c->conn_timeout);
+ } else if(is_reliable(c)) {
+ start_retransmit_timer(c);
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
+ c->conn_timeout.tv_sec += utcp->timeout;
+ }
}
+skip_ack:
// 5. Process SYN stuff
if(hdr.ctl & SYN) {
switch(c->state) {
case SYN_SENT:
+
// This is a SYNACK. It should always have ACKed the SYN.
- if(!advanced)
+ if(!advanced) {
goto reset;
+ }
+
c->rcv.irs = hdr.seq;
c->rcv.nxt = hdr.seq;
- set_state(c, ESTABLISHED);
+
+ if(c->shut_wr) {
+ c->snd.last++;
+ set_state(c, FIN_WAIT_1);
+ } else {
+ set_state(c, ESTABLISHED);
+ }
+
// TODO: notify application of this somehow.
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:
case LAST_ACK:
case TIME_WAIT:
// Ehm, no. We should never receive a second SYN.
- goto reset;
+ return 0;
+
default:
#ifdef UTCP_DEBUG
abort();
if(c->state == SYN_RECEIVED) {
// This is the ACK after the SYNACK. It should always have ACKed the SYNACK.
- if(!advanced)
+ if(!advanced) {
goto reset;
+ }
// Are we still LISTENing?
- if(utcp->accept)
+ if(utcp->accept) {
utcp->accept(c, c->src);
+ }
if(c->state != ESTABLISHED) {
set_state(c, CLOSED);
abort();
#endif
return 0;
+
case ESTABLISHED:
case FIN_WAIT_1:
case FIN_WAIT_2:
break;
+
case CLOSE_WAIT:
case CLOSING:
case LAST_ACK:
case TIME_WAIT:
// Ehm no, We should never receive more data after a FIN.
goto reset;
+
default:
#ifdef UTCP_DEBUG
abort();
return 0;
}
- ssize_t rxd;
-
- if(c->recv) {
- rxd = c->recv(c, data, len);
- if(rxd != len) {
- // TODO: once we have a receive buffer, handle the application not accepting all data.
- abort();
- }
- if(rxd < 0)
- rxd = 0;
- else if(rxd > len)
- rxd = len; // Bad application, bad!
- } else {
- rxd = len;
- }
-
- c->rcv.nxt += len;
+ handle_incoming_data(c, &hdr, ptr, len);
}
// 7. Process FIN stuff
- if(hdr.ctl & FIN) {
+ if((hdr.ctl & FIN) && (!is_reliable(c) || hdr.seq + len == c->rcv.nxt)) {
switch(c->state) {
case SYN_SENT:
case SYN_RECEIVED:
abort();
#endif
break;
+
case ESTABLISHED:
set_state(c, CLOSE_WAIT);
break;
+
case FIN_WAIT_1:
set_state(c, CLOSING);
break;
+
case FIN_WAIT_2:
- gettimeofday(&c->conn_timeout, NULL);
- c->conn_timeout.tv_sec += 60;
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
+ c->conn_timeout.tv_sec += utcp->timeout;
set_state(c, TIME_WAIT);
break;
+
case CLOSE_WAIT:
case CLOSING:
case LAST_ACK:
case TIME_WAIT:
// Ehm, no. We should never receive a second FIN.
goto reset;
+
default:
#ifdef UTCP_DEBUG
abort();
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:
- ack(c, prevrcvnxt != c->rcv.nxt);
+ if(is_reliable(c) || hdr.ctl & SYN || hdr.ctl & FIN) {
+ ack(c, has_data);
+ }
+
return 0;
reset:
swap_ports(&hdr);
hdr.wnd = 0;
+ hdr.aux = 0;
+
if(hdr.ctl & ACK) {
hdr.seq = hdr.ack;
hdr.ctl = RST;
hdr.seq = 0;
hdr.ctl = RST | ACK;
}
- print_packet(utcp, "send", &hdr, sizeof hdr);
- utcp->send(utcp, &hdr, sizeof hdr);
+
+ print_packet(c, "send", &hdr, sizeof(hdr));
+ utcp->send(utcp, &hdr, sizeof(hdr));
return 0;
}
int utcp_shutdown(struct utcp_connection *c, int dir) {
- debug("%p shutdown %d at %u\n", c ? c->utcp : NULL, dir, c->snd.last);
+ debug(c, "shutdown %d at %u\n", dir, c ? c->snd.last : 0);
+
if(!c) {
errno = EFAULT;
return -1;
}
if(c->reapable) {
- debug("Error: shutdown() called on closed connection %p\n", c);
+ debug(c, "shutdown() called on closed connection\n");
errno = EBADF;
return -1;
}
// TCP does not have a provision for stopping incoming packets.
// The best we can do is to just ignore them.
- if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR)
+ if(dir == UTCP_SHUT_RD || dir == UTCP_SHUT_RDWR) {
c->recv = NULL;
+ }
// The rest of the code deals with shutting down writes.
- if(dir == UTCP_SHUT_RD)
+ if(dir == UTCP_SHUT_RD) {
return 0;
+ }
+
+ // Only process shutting down writes once.
+ if(c->shut_wr) {
+ return 0;
+ }
+
+ c->shut_wr = true;
switch(c->state) {
case CLOSED:
return -1;
case SYN_SENT:
- set_state(c, CLOSED);
return 0;
case SYN_RECEIVED:
case ESTABLISHED:
set_state(c, FIN_WAIT_1);
break;
+
case FIN_WAIT_1:
case FIN_WAIT_2:
return 0;
+
case CLOSE_WAIT:
set_state(c, CLOSING);
break;
c->snd.last++;
ack(c, false);
- return 0;
-}
-int utcp_close(struct utcp_connection *c) {
- if(utcp_shutdown(c, SHUT_RDWR))
- return -1;
- c->recv = NULL;
- c->poll = NULL;
- c->reapable = true;
+ if(!timespec_isset(&c->rtrx_timeout)) {
+ start_retransmit_timer(c);
+ }
+
return 0;
}
-int utcp_abort(struct utcp_connection *c) {
+static bool reset_connection(struct utcp_connection *c) {
if(!c) {
errno = EFAULT;
- return -1;
+ return false;
}
if(c->reapable) {
- debug("Error: abort() called on closed connection %p\n", c);
+ debug(c, "abort() called on closed connection\n");
errno = EBADF;
- return -1;
+ return false;
}
c->recv = NULL;
c->poll = NULL;
- c->reapable = true;
switch(c->state) {
case CLOSED:
- return 0;
+ return true;
+
case LISTEN:
case SYN_SENT:
case CLOSING:
case LAST_ACK:
case TIME_WAIT:
set_state(c, CLOSED);
- return 0;
+ return true;
case SYN_RECEIVED:
case ESTABLISHED:
hdr.wnd = 0;
hdr.ctl = RST;
- print_packet(c->utcp, "send", &hdr, sizeof hdr);
- c->utcp->send(c->utcp, &hdr, sizeof hdr);
+ print_packet(c, "send", &hdr, sizeof(hdr));
+ c->utcp->send(c->utcp, &hdr, sizeof(hdr));
+ return true;
+}
+
+// Closes all the opened connections
+void utcp_abort_all_connections(struct utcp *utcp) {
+ if(!utcp) {
+ errno = EINVAL;
+ return;
+ }
+
+ for(int i = 0; i < utcp->nconnections; i++) {
+ struct utcp_connection *c = utcp->connections[i];
+
+ if(c->reapable || c->state == CLOSED) {
+ continue;
+ }
+
+ utcp_recv_t old_recv = c->recv;
+ utcp_poll_t old_poll = c->poll;
+
+ reset_connection(c);
+
+ if(old_recv) {
+ errno = 0;
+ old_recv(c, NULL, 0);
+ }
+
+ if(old_poll && !c->reapable) {
+ errno = 0;
+ old_poll(c, 0);
+ }
+ }
+
+ return;
+}
+
+int utcp_close(struct utcp_connection *c) {
+ if(utcp_shutdown(c, SHUT_RDWR) && errno != ENOTCONN) {
+ return -1;
+ }
+
+ c->recv = NULL;
+ c->poll = NULL;
+ c->reapable = true;
+ return 0;
+}
+
+int utcp_abort(struct utcp_connection *c) {
+ if(!reset_connection(c)) {
+ return -1;
+ }
+
+ c->reapable = true;
return 0;
}
* The return value is the time to the next timeout in milliseconds,
* or maybe a negative value if the timeout is infinite.
*/
-struct timeval utcp_timeout(struct utcp *utcp) {
- struct timeval now;
- gettimeofday(&now, NULL);
- struct timeval next = {now.tv_sec + 3600, now.tv_usec};
+struct timespec utcp_timeout(struct utcp *utcp) {
+ struct timespec now;
+ clock_gettime(UTCP_CLOCK, &now);
+ struct timespec next = {now.tv_sec + 3600, now.tv_nsec};
for(int i = 0; i < utcp->nconnections; i++) {
struct utcp_connection *c = utcp->connections[i];
- if(!c)
+
+ if(!c) {
continue;
+ }
+ // delete connections that have been utcp_close()d.
if(c->state == CLOSED) {
if(c->reapable) {
- debug("Reaping %p\n", c);
+ debug(c, "reaping\n");
free_connection(c);
i--;
}
+
continue;
}
- if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &now, <)) {
+ if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &now)) {
errno = ETIMEDOUT;
c->state = CLOSED;
- if(c->recv)
+
+ if(c->recv) {
c->recv(c, NULL, 0);
+ }
+
+ if(c->poll && !c->reapable) {
+ c->poll(c, 0);
+ }
+
continue;
}
- if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &now, <)) {
+ if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &now)) {
+ debug(c, "retransmitting after timeout\n");
retransmit(c);
}
- if(c->poll && buffer_free(&c->sndbuf) && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
- c->poll(c, buffer_free(&c->sndbuf));
+ if(c->poll) {
+ if((c->state == ESTABLISHED || c->state == CLOSE_WAIT) && c->do_poll) {
+ c->do_poll = false;
+ uint32_t len = buffer_free(&c->sndbuf);
- if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <))
- next = c->conn_timeout;
+ if(len) {
+ c->poll(c, len);
+ }
+ } else if(c->state == CLOSED) {
+ c->poll(c, 0);
+ }
+ }
- if(c->snd.nxt != c->snd.una) {
- c->rtrx_timeout = now;
- c->rtrx_timeout.tv_sec++;
- } else {
- timerclear(&c->rtrx_timeout);
+ if(timespec_isset(&c->conn_timeout) && timespec_lt(&c->conn_timeout, &next)) {
+ next = c->conn_timeout;
}
- if(timerisset(&c->rtrx_timeout) && timercmp(&c->rtrx_timeout, &next, <))
+ if(timespec_isset(&c->rtrx_timeout) && timespec_lt(&c->rtrx_timeout, &next)) {
next = c->rtrx_timeout;
+ }
}
- struct timeval diff;
- timersub(&next, &now, &diff);
+ struct timespec diff;
+
+ timespec_sub(&next, &now, &diff);
+
return diff;
}
bool utcp_is_active(struct utcp *utcp) {
- if(!utcp)
+ if(!utcp) {
return false;
+ }
for(int i = 0; i < utcp->nconnections; i++)
- if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT)
+ if(utcp->connections[i]->state != CLOSED && utcp->connections[i]->state != TIME_WAIT) {
return true;
+ }
return false;
}
struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_send_t send, void *priv) {
- struct utcp *utcp = calloc(1, sizeof *utcp);
- if(!utcp)
- return NULL;
-
if(!send) {
errno = EFAULT;
return NULL;
}
+ struct utcp *utcp = calloc(1, sizeof(*utcp));
+
+ if(!utcp) {
+ return NULL;
+ }
+
+ if(!CLOCK_GRANULARITY) {
+ struct timespec res;
+ clock_getres(UTCP_CLOCK, &res);
+ CLOCK_GRANULARITY = res.tv_sec * USEC_PER_SEC + res.tv_nsec / 1000;
+ }
+
utcp->accept = accept;
utcp->pre_accept = pre_accept;
utcp->send = send;
utcp->priv = priv;
- utcp->mtu = 1000;
- utcp->timeout = 60;
+ utcp_set_mtu(utcp, DEFAULT_MTU);
+ utcp->timeout = DEFAULT_USER_TIMEOUT; // sec
return utcp;
}
void utcp_exit(struct utcp *utcp) {
- if(!utcp)
+ if(!utcp) {
return;
+ }
+
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]->sndbuf);
- free(utcp->connections[i]);
+ struct utcp_connection *c = utcp->connections[i];
+
+ 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);
}
+
free(utcp->connections);
+ free(utcp->pkt);
free(utcp);
}
return utcp ? utcp->mtu : 0;
}
+uint16_t utcp_get_mss(struct utcp *utcp) {
+ return utcp ? utcp->mss : 0;
+}
+
void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) {
- // TODO: handle overhead of the header
- if(utcp)
- utcp->mtu = mtu;
+ if(!utcp) {
+ return;
+ }
+
+ if(mtu <= sizeof(struct hdr)) {
+ return;
+ }
+
+ if(mtu > utcp->mtu) {
+ char *new = realloc(utcp->pkt, mtu + sizeof(struct hdr));
+
+ if(!new) {
+ return;
+ }
+
+ utcp->pkt = new;
+ }
+
+ utcp->mtu = mtu;
+ utcp->mss = mtu - sizeof(struct hdr);
+}
+
+void utcp_reset_timers(struct utcp *utcp) {
+ if(!utcp) {
+ return;
+ }
+
+ struct timespec now, then;
+
+ clock_gettime(UTCP_CLOCK, &now);
+
+ then = now;
+
+ then.tv_sec += utcp->timeout;
+
+ for(int i = 0; i < utcp->nconnections; i++) {
+ struct utcp_connection *c = utcp->connections[i];
+
+ if(c->reapable) {
+ continue;
+ }
+
+ if(timespec_isset(&c->rtrx_timeout)) {
+ c->rtrx_timeout = now;
+ }
+
+ if(timespec_isset(&c->conn_timeout)) {
+ c->conn_timeout = then;
+ }
+
+ c->rtt_start.tv_sec = 0;
+
+ if(c->rto > START_RTO) {
+ c->rto = START_RTO;
+ }
+ }
}
int utcp_get_user_timeout(struct utcp *u) {
}
void utcp_set_user_timeout(struct utcp *u, int timeout) {
- if(u)
+ if(u) {
u->timeout = timeout;
+ }
}
size_t utcp_get_sndbuf(struct utcp_connection *c) {
}
size_t utcp_get_sndbuf_free(struct utcp_connection *c) {
- if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT))
+ if(!c) {
+ return 0;
+ }
+
+ switch(c->state) {
+ case SYN_SENT:
+ case SYN_RECEIVED:
+ case ESTABLISHED:
+ case CLOSE_WAIT:
return buffer_free(&c->sndbuf);
- else
+
+ default:
return 0;
+ }
}
void utcp_set_sndbuf(struct utcp_connection *c, size_t size) {
- if(!c)
+ if(!c) {
return;
+ }
+
c->sndbuf.maxsize = size;
- if(c->sndbuf.maxsize != size)
+
+ if(c->sndbuf.maxsize != size) {
c->sndbuf.maxsize = -1;
+ }
+
+ c->do_poll = buffer_free(&c->sndbuf);
+}
+
+size_t utcp_get_rcvbuf(struct utcp_connection *c) {
+ return c ? c->rcvbuf.maxsize : 0;
+}
+
+size_t utcp_get_rcvbuf_free(struct utcp_connection *c) {
+ if(c && (c->state == ESTABLISHED || c->state == CLOSE_WAIT)) {
+ return buffer_free(&c->rcvbuf);
+ } else {
+ return 0;
+ }
+}
+
+void utcp_set_rcvbuf(struct utcp_connection *c, size_t size) {
+ if(!c) {
+ return;
+ }
+
+ c->rcvbuf.maxsize = size;
+
+ if(c->rcvbuf.maxsize != size) {
+ c->rcvbuf.maxsize = -1;
+ }
+}
+
+size_t utcp_get_sendq(struct utcp_connection *c) {
+ return c->sndbuf.used;
+}
+
+size_t utcp_get_recvq(struct utcp_connection *c) {
+ return c->rcvbuf.used;
}
bool utcp_get_nodelay(struct utcp_connection *c) {
}
void utcp_set_nodelay(struct utcp_connection *c, bool nodelay) {
- if(c)
+ if(c) {
c->nodelay = nodelay;
+ }
}
bool utcp_get_keepalive(struct utcp_connection *c) {
}
void utcp_set_keepalive(struct utcp_connection *c, bool keepalive) {
- if(c)
+ if(c) {
c->keepalive = keepalive;
+ }
}
size_t utcp_get_outq(struct utcp_connection *c) {
}
void utcp_set_recv_cb(struct utcp_connection *c, utcp_recv_t recv) {
- if(c)
+ if(c) {
c->recv = recv;
+ }
}
void utcp_set_poll_cb(struct utcp_connection *c, utcp_poll_t poll) {
- if(c)
+ if(c) {
c->poll = poll;
+ c->do_poll = buffer_free(&c->sndbuf);
+ }
}
void utcp_set_accept_cb(struct utcp *utcp, utcp_accept_t accept, utcp_pre_accept_t pre_accept) {
utcp->pre_accept = pre_accept;
}
}
+
+void utcp_expect_data(struct utcp_connection *c, bool expect) {
+ if(!c || c->reapable) {
+ return;
+ }
+
+ if(!(c->state == ESTABLISHED || c->state == FIN_WAIT_1 || c->state == FIN_WAIT_2)) {
+ return;
+ }
+
+ if(expect) {
+ // If we expect data, start the connection timer.
+ if(!timespec_isset(&c->conn_timeout)) {
+ clock_gettime(UTCP_CLOCK, &c->conn_timeout);
+ c->conn_timeout.tv_sec += c->utcp->timeout;
+ }
+ } else {
+ // If we want to cancel expecting data, only clear the timer when there is no unACKed data.
+ if(c->snd.una == c->snd.last) {
+ timespec_clear(&c->conn_timeout);
+ }
+ }
+}
+
+void utcp_offline(struct utcp *utcp, bool offline) {
+ struct timespec now;
+ clock_gettime(UTCP_CLOCK, &now);
+
+ for(int i = 0; i < utcp->nconnections; i++) {
+ struct utcp_connection *c = utcp->connections[i];
+
+ if(c->reapable) {
+ continue;
+ }
+
+ utcp_expect_data(c, offline);
+
+ if(!offline) {
+ if(timespec_isset(&c->rtrx_timeout)) {
+ c->rtrx_timeout = now;
+ }
+
+ utcp->connections[i]->rtt_start.tv_sec = 0;
+
+ if(c->rto > START_RTO) {
+ c->rto = START_RTO;
+ }
+ }
+ }
+}
+
+void utcp_set_clock_granularity(long granularity) {
+ CLOCK_GRANULARITY = granularity;
+}