X-Git-Url: http://git.meshlink.io/?a=blobdiff_plain;f=utcp.c;h=d990081a30cef77b858c2c5e300122f910a6c49f;hb=558c6d183e2a580b811c5603dd48c33b01a3dc7e;hp=607af2d149ee9242646d3f1a97638ecb64e01078;hpb=027abc36cfb0a8247625f5f6af65b7ce63065a33;p=utcp diff --git a/utcp.c b/utcp.c index 607af2d..d990081 100644 --- a/utcp.c +++ b/utcp.c @@ -1,6 +1,6 @@ /* utcp.c -- Userspace TCP - Copyright (C) 2014 Guus Sliepen + Copyright (C) 2014-2017 Guus Sliepen 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 @@ -49,13 +49,13 @@ (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 @@ -70,12 +70,12 @@ static void debug(const char *format, ...) { 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 aux=%x ctl=", utcp, dir, (unsigned long)len, hdr.src, hdr.dst, hdr.seq, hdr.ack, hdr.wnd, hdr.aux); if(hdr.ctl & SYN) debug("SYN"); if(hdr.ctl & RST) @@ -87,30 +87,24 @@ static void print_packet(struct utcp *utcp, const char *dir, const void *pkt, si if(len > sizeof hdr) { uint32_t datalen = len - sizeof hdr; - uint8_t *str = malloc((datalen << 1) + 7); - memcpy(str, " data=", 6); - uint8_t *strptr = str + 6; - const uint8_t *data = pkt; - const uint8_t *dataend = data + datalen; - - while(data != dataend) { - *strptr = (*data >> 4) > 9? (*data >> 4) + 55 : (*data >> 4) + 48; - ++strptr; - *strptr = (*data & 0xf) > 9? (*data & 0xf) + 55 : (*data & 0xf) + 48; - ++strptr; - ++data; + 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]; } - *strptr = 0; + *p = 0; - debug(str); - free(str); + debug(" data=%s", str); } debug("\n"); } #else -#define debug(...) -#define print_packet(...) +#define debug(...) do {} while(0) +#define print_packet(...) do {} while(0) #endif static void set_state(struct utcp_connection *c, enum state state) { @@ -133,10 +127,8 @@ static bool fin_wanted(struct utcp_connection *c, uint32_t seq) { } } -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) { @@ -148,16 +140,12 @@ static int32_t seqdiff(uint32_t a, uint32_t b) { // 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; } @@ -169,7 +157,7 @@ static ssize_t buffer_put_at(struct buffer *buf, size_t offset, const void *data } else { do { newsize *= 2; - } while(newsize < buf->used + len); + } while(newsize < required); } if(newsize > buf->maxsize) newsize = buf->maxsize; @@ -348,6 +336,13 @@ static struct utcp_connection *allocate_connection(struct utcp *utcp, uint16_t s return c; } +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) { @@ -362,7 +357,7 @@ static void update_rtt(struct utcp_connection *c, uint32_t 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->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); } @@ -388,35 +383,51 @@ static void stop_retransmit_timer(struct utcp_connection *c) { debug("timeout cleared\n"); } -struct utcp_connection *utcp_connect(struct utcp *utcp, uint16_t dst, utcp_recv_t recv, void *priv) { +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) return NULL; + assert((flags & ~0xf) == 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->rcv.wnd; + 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(utcp, "send", &pkt, sizeof pkt); + utcp->send(utcp, &pkt, sizeof pkt); gettimeofday(&c->conn_timeout, NULL); 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]); @@ -447,7 +458,7 @@ static void ack(struct utcp_connection *c, bool sendatleastone) { struct { struct hdr hdr; - char data[]; + uint8_t data[]; } *pkt; pkt = malloc(sizeof pkt->hdr + c->utcp->mtu); @@ -517,7 +528,7 @@ ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) { return -1; } - // Add data to send buffer + // Exit early if we have nothing to send. if(!len) return 0; @@ -527,6 +538,8 @@ ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) { return -1; } + // Add data to send buffer. + len = buffer_put(&c->sndbuf, data, len); if(len <= 0) { errno = EWOULDBLOCK; @@ -535,7 +548,11 @@ ssize_t utcp_send(struct utcp_connection *c, const void *data, size_t len) { c->snd.last += len; ack(c, false); - if(!timerisset(&c->rtrx_timeout)) + if(!is_reliable(c)) { + c->snd.una = c->snd.nxt = c->snd.last; + buffer_get(&c->sndbuf, NULL, c->sndbuf.used); + } + if(is_reliable(c) && !timerisset(&c->rtrx_timeout)) start_retransmit_timer(c); return len; } @@ -557,7 +574,7 @@ static void retransmit(struct utcp_connection *c) { struct { struct hdr hdr; - char data[]; + uint8_t data[]; } *pkt; pkt = malloc(sizeof pkt->hdr + c->utcp->mtu); @@ -575,8 +592,13 @@ static void retransmit(struct utcp_connection *c) { pkt->hdr.seq = c->snd.iss; pkt->hdr.ack = 0; pkt->hdr.ctl = SYN; - print_packet(c->utcp, "rtrx", pkt, sizeof pkt->hdr); - utcp->send(utcp, pkt, sizeof pkt->hdr); + 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->utcp, "rtrx", pkt, sizeof pkt->hdr + 4); + utcp->send(utcp, pkt, sizeof pkt->hdr + 4); break; case SYN_RECEIVED: @@ -646,15 +668,18 @@ cleanup: * that are shifted to before the start of the receive buffer. * * There are three cases: - * - the SACK entry is ahead of ^, in that case just change the offset. + * - 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); @@ -685,7 +710,7 @@ static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, cons debug("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 < len) + if(rxd < 0 || (size_t)rxd < len) abort(); // Make note of where we put it. @@ -702,6 +727,8 @@ static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, cons 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 @@ -726,7 +753,7 @@ static void handle_out_of_order(struct utcp_connection *c, uint32_t offset, cons 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; @@ -734,7 +761,7 @@ static void handle_in_order(struct utcp_connection *c, const void *data, size_t if(c->recv) { ssize_t rxd = c->recv(c, data, len); - if(rxd != len) { + if(rxd < 0 || (size_t)rxd != len) { // TODO: handle the application not accepting all data. abort(); } @@ -748,6 +775,12 @@ static void handle_in_order(struct utcp_connection *c, const void *data, size_t static void handle_incoming_data(struct utcp_connection *c, uint32_t seq, const void *data, size_t len) { + if(!is_reliable(c)) { + c->recv(c, data, len); + c->rcv.nxt = seq + len; + return; + } + uint32_t offset = seqdiff(seq, c->rcv.nxt); if(offset + len > c->rcvbuf.maxsize) abort(); @@ -796,6 +829,49 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) { return -1; } + // Check for auxiliary headers + + 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 = data; + break; + default: + errno = EBADMSG; + return -1; + } + + len -= auxlen; + data += auxlen; + + if(!(aux & 0x800)) + break; + + if(len < 2) { + errno = EBADMSG; + return -1; + } + + memcpy(&aux, data, 2); + len -= 2; + data += 2; + } + // Try to match the packet to an existing connection struct utcp_connection *c = find_connection(utcp, hdr.dst, hdr.src); @@ -824,19 +900,47 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) { 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; + } + // 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->rcv.wnd; + 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->utcp, "send", &pkt, sizeof hdr + 4); + utcp->send(utcp, &pkt, sizeof hdr + 4); + } else { + pkt.hdr.aux = 0; + print_packet(c->utcp, "send", &pkt, sizeof hdr); + utcp->send(utcp, &pkt, sizeof hdr); + } } else { // No, we don't want your packets, send a RST back len = 1; @@ -851,8 +955,10 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) { // 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. @@ -893,24 +999,24 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) { // cut already accepted front overlapping if(rcv_offset < 0) { - acceptable = rcv_offset + len >= 0; + acceptable = len > (size_t)-rcv_offset; if(acceptable) { data -= 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; } - - 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 + %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; - // Otherwise, send an ACK back in the hope things improve. - ack(c, true); - return 0; + // Otherwise, continue processing. + len = 0; } c->snd.wnd = hdr.wnd; // TODO: move below @@ -980,6 +1086,9 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) { } } + if(!(hdr.ctl & ACK)) + goto skip_ack; + // 3. Advance snd.una uint32_t advanced = seqdiff(hdr.ack, c->snd.una); @@ -1048,7 +1157,7 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) { break; } } else { - if(!len) { + if(!len && is_reliable(c)) { c->dupack++; if(c->dupack == 3) { debug("Triplicate ACK\n"); @@ -1068,10 +1177,11 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) { 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.last) stop_retransmit_timer(c); - else + else if(is_reliable(c)) start_retransmit_timer(c); } +skip_ack: // 5. Process SYN stuff if(hdr.ctl & SYN) { @@ -1094,7 +1204,7 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) { case LAST_ACK: case TIME_WAIT: // Ehm, no. We should never receive a second SYN. - goto reset; + return 0; default: #ifdef UTCP_DEBUG abort(); @@ -1211,6 +1321,7 @@ ssize_t utcp_recv(struct utcp *utcp, const void *data, size_t len) { reset: swap_ports(&hdr); hdr.wnd = 0; + hdr.aux = 0; if(hdr.ctl & ACK) { hdr.seq = hdr.ack; hdr.ctl = RST; @@ -1288,7 +1399,7 @@ int utcp_shutdown(struct utcp_connection *c, int dir) { } 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; @@ -1387,8 +1498,15 @@ struct timeval utcp_timeout(struct utcp *utcp) { 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)) { + uint32_t len = buffer_free(&c->sndbuf); + if(len) + c->poll(c, len); + } else if(c->state == CLOSED) { + c->poll(c, 0); + } + } if(timerisset(&c->conn_timeout) && timercmp(&c->conn_timeout, &next, <)) next = c->conn_timeout; @@ -1428,8 +1546,8 @@ struct utcp *utcp_init(utcp_accept_t accept, utcp_pre_accept_t pre_accept, utcp_ utcp->send = send; utcp->priv = priv; utcp->mtu = DEFAULT_MTU; - utcp->timeout = DEFAULT_USER_TIMEOUT; // s - utcp->rto = START_RTO; // us + utcp->timeout = DEFAULT_USER_TIMEOUT; // sec + utcp->rto = START_RTO; // usec return utcp; } @@ -1438,11 +1556,13 @@ void utcp_exit(struct utcp *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]->rcvbuf); - 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); + buffer_exit(&c->rcvbuf); + buffer_exit(&c->sndbuf); + free(c); } free(utcp->connections); free(utcp); @@ -1458,6 +1578,22 @@ void utcp_set_mtu(struct utcp *utcp, uint16_t mtu) { utcp->mtu = mtu; } +void utcp_reset_timers(struct utcp *utcp) { + if(!utcp) + return; + struct timeval now, then; + gettimeofday(&now, NULL); + then = now; + then.tv_sec += utcp->timeout; + for(int i = 0; i < utcp->nconnections; i++) { + utcp->connections[i]->rtrx_timeout = now; + utcp->connections[i]->conn_timeout = then; + utcp->connections[i]->rtt_start.tv_sec = 0; + } + if(utcp->rto > START_RTO) + utcp->rto = START_RTO; +} + int utcp_get_user_timeout(struct utcp *u) { return u ? u->timeout : 0; }