void socket_free(gpointer data) {
Socket* socket = data;
MAGIC_ASSERT(socket);
MAGIC_ASSERT(socket->vtable);
if(socket->peerString) {
g_free(socket->peerString);
}
if(socket->boundString) {
g_free(socket->boundString);
}
if(socket->unixPath) {
g_free(socket->unixPath);
}
while(g_queue_get_length(socket->inputBuffer) > 0) {
packet_unref(g_queue_pop_head(socket->inputBuffer));
}
g_queue_free(socket->inputBuffer);
while(g_queue_get_length(socket->outputBuffer) > 0) {
packet_unref(g_queue_pop_head(socket->outputBuffer));
}
g_queue_free(socket->outputBuffer);
MAGIC_CLEAR(socket);
socket->vtable->free((Descriptor*)socket);
}
void tcp_free(TCP* tcp) {
MAGIC_ASSERT(tcp);
while(g_queue_get_length(tcp->throttledOutput) > 0) {
packet_unref(g_queue_pop_head(tcp->throttledOutput));
}
g_queue_free(tcp->throttledOutput);
while(g_queue_get_length(tcp->unorderedInput) > 0) {
packet_unref(g_queue_pop_head(tcp->unorderedInput));
}
g_queue_free(tcp->unorderedInput);
g_hash_table_destroy(tcp->retransmission);
if(tcp->child) {
MAGIC_ASSERT(tcp->child);
MAGIC_ASSERT(tcp->child->parent);
MAGIC_ASSERT(tcp->child->parent->server);
/* remove parents reference to child, if it exists */
g_hash_table_remove(tcp->child->parent->server->children, &(tcp->child->key));
_tcpchild_free(tcp->child);
}
if(tcp->server) {
_tcpserver_free(tcp->server);
}
MAGIC_CLEAR(tcp);
g_free(tcp);
}
void networkinterface_free(NetworkInterface* interface) {
MAGIC_ASSERT(interface);
/* unref all packets sitting in our input buffer */
while(interface->inBuffer && !g_queue_is_empty(interface->inBuffer)) {
Packet* packet = g_queue_pop_head(interface->inBuffer);
packet_unref(packet);
}
g_queue_free(interface->inBuffer);
/* unref all sockets wanting to send */
while(interface->rrQueue && !g_queue_is_empty(interface->rrQueue)) {
Socket* socket = g_queue_pop_head(interface->rrQueue);
descriptor_unref(socket);
}
g_queue_free(interface->rrQueue);
priorityqueue_free(interface->fifoQueue);
g_hash_table_destroy(interface->boundSockets);
dns_deregister(worker_getDNS(), interface->address);
address_unref(interface->address);
if(interface->pcapFile) {
fclose(interface->pcapFile);
}
MAGIC_CLEAR(interface);
g_free(interface);
}
static void _networkinterface_scheduleNextReceive(NetworkInterface* interface) {
/* the next packets need to be received and processed */
SimulationTime batchTime = worker_getConfig()->interfaceBatchTime;
/* receive packets in batches */
while(!g_queue_is_empty(interface->inBuffer) &&
interface->receiveNanosecondsConsumed <= batchTime) {
/* get the next packet */
Packet* packet = g_queue_pop_head(interface->inBuffer);
utility_assert(packet);
/* successfully received */
packet_addDeliveryStatus(packet, PDS_RCV_INTERFACE_RECEIVED);
_networkinterface_pcapWritePacket(interface, packet);
/* free up buffer space */
guint length = packet_getPayloadLength(packet) + packet_getHeaderSize(packet);
interface->inBufferLength -= length;
/* calculate how long it took to 'receive' this packet */
interface->receiveNanosecondsConsumed += (length * interface->timePerByteDown);
/* hand it off to the correct socket layer */
gint key = packet_getDestinationAssociationKey(packet);
Socket* socket = g_hash_table_lookup(interface->boundSockets, GINT_TO_POINTER(key));
/* if the socket closed, just drop the packet */
gint socketHandle = -1;
if(socket) {
socketHandle = *descriptor_getHandleReference((Descriptor*)socket);
socket_pushInPacket(socket, packet);
} else {
packet_addDeliveryStatus(packet, PDS_RCV_INTERFACE_DROPPED);
}
packet_unref(packet);
/* count our bandwidth usage by interface, and by socket handle if possible */
tracker_addInputBytes(host_getTracker(worker_getCurrentHost()),(guint64)length, socketHandle);
}
/*
* we need to call back and try to receive more, even if we didnt consume all
* of our batch time, because we might have more packets to receive then.
*/
SimulationTime receiveTime = (SimulationTime) floor(interface->receiveNanosecondsConsumed);
if(receiveTime >= SIMTIME_ONE_NANOSECOND) {
/* we are 'receiving' the packets */
interface->flags |= NIF_RECEIVING;
/* call back when the packets are 'received' */
InterfaceReceivedEvent* event = interfacereceived_new(interface);
/* event destination is our node */
worker_scheduleEvent((Event*)event, receiveTime, 0);
}
}
gssize udp_receiveUserData(UDP* udp, gpointer buffer, gsize nBytes, in_addr_t* ip, in_port_t* port) {
MAGIC_ASSERT(udp);
Packet* packet = socket_removeFromInputBuffer((Socket*)udp);
if(!packet) {
return -1;
}
/* copy lesser of requested and available amount to application buffer */
guint packetLength = packet_getPayloadLength(packet);
gsize copyLength = MIN(nBytes, packetLength);
guint bytesCopied = packet_copyPayload(packet, 0, buffer, copyLength);
utility_assert(bytesCopied == copyLength);
packet_addDeliveryStatus(packet, PDS_RCV_SOCKET_DELIVERED);
/* fill in address info */
if(ip) {
*ip = packet_getSourceIP(packet);
}
if(port) {
*port = packet_getSourcePort(packet);
}
/* destroy packet, throwing away any bytes not claimed by the app */
packet_unref(packet);
/* update the tracker output buffer stats */
Tracker* tracker = host_getTracker(worker_getCurrentHost());
Socket* socket = (Socket* )udp;
Descriptor* descriptor = (Descriptor *)socket;
gsize outLength = socket_getOutputBufferLength(socket);
gsize outSize = socket_getOutputBufferSize(socket);
tracker_updateSocketOutputBuffer(tracker, descriptor->handle, outLength, outSize);
debug("user read %u inbound UDP bytes", bytesCopied);
return (gssize)bytesCopied;
}
/* return TRUE if the packet should be retransmitted */
gboolean tcp_processPacket(TCP* tcp, Packet* packet) {
MAGIC_ASSERT(tcp);
/* fetch the TCP info from the packet */
PacketTCPHeader header;
packet_getTCPHeader(packet, &header);
guint packetLength = packet_getPayloadLength(packet);
/* if we run a server, the packet could be for an existing child */
tcp = _tcp_getSourceTCP(tcp, header.sourceIP, header.sourcePort);
/* now we have the true TCP for the packet */
MAGIC_ASSERT(tcp);
/* print packet info for debugging */
debug("%s <-> %s: processing packet# %u length %u",
tcp->super.boundString, tcp->super.peerString, header.sequence, packetLength);
/* if packet is reset, don't process */
if(header.flags & PTCP_RST) {
/* @todo: not sure if this is handled correctly */
debug("received RESET packet");
if(!(tcp->state & TCPS_LISTEN) && !(tcp->error & TCPE_CONNECTION_RESET)) {
tcp->error |= TCPE_CONNECTION_RESET;
tcp->flags |= TCPF_REMOTE_CLOSED;
_tcp_setState(tcp, TCPS_TIMEWAIT);
/* it will send no more user data after what we have now */
tcp->receive.end = tcp->receive.next;
}
packet_unref(packet);
return FALSE;
}
/* if we are a server, we have to remember who we got this from so we can
* respond back to them. this is because we could be bound to several
* interfaces and otherwise cant decide which to send on.
*/
if(tcp->server) {
tcp->server->lastPeerIP = header.sourceIP;
tcp->server->lastPeerPort = header.sourcePort;
tcp->server->lastIP = header.destinationIP;
}
/* go through the state machine, tracking processing and response */
gboolean wasProcessed = FALSE;
enum ProtocolTCPFlags responseFlags = PTCP_NONE;
switch(tcp->state) {
case TCPS_LISTEN: {
/* receive SYN, send SYNACK, move to SYNRECEIVED */
if(header.flags & PTCP_SYN) {
MAGIC_ASSERT(tcp->server);
wasProcessed = TRUE;
/* we need to multiplex a new child */
Node* node = worker_getPrivate()->cached_node;
gint multiplexedHandle = node_createDescriptor(node, DT_TCPSOCKET);
TCP* multiplexed = (TCP*) node_lookupDescriptor(node, multiplexedHandle);
multiplexed->child = _tcpchild_new(multiplexed, tcp, header.sourceIP, header.sourcePort);
g_assert(g_hash_table_lookup(tcp->server->children, &(multiplexed->child->key)) == NULL);
g_hash_table_replace(tcp->server->children, &(multiplexed->child->key), multiplexed->child);
multiplexed->receive.start = header.sequence;
multiplexed->receive.next = multiplexed->receive.start + 1;
debug("%s <-> %s: server multiplexed child socket %s <-> %s",
tcp->super.boundString, tcp->super.peerString,
multiplexed->super.boundString, multiplexed->super.peerString);
_tcp_setState(multiplexed, TCPS_SYNRECEIVED);
/* parent will send response */
responseFlags = PTCP_SYN|PTCP_ACK;
}
break;
}
case TCPS_SYNSENT: {
/* receive SYNACK, send ACK, move to ESTABLISHED */
if((header.flags & PTCP_SYN) && (header.flags & PTCP_ACK)) {
wasProcessed = TRUE;
tcp->receive.start = header.sequence;
tcp->receive.next = tcp->receive.start + 1;
responseFlags |= PTCP_ACK;
_tcp_setState(tcp, TCPS_ESTABLISHED);
}
/* receive SYN, send ACK, move to SYNRECEIVED (simultaneous open) */
else if(header.flags & PTCP_SYN) {
wasProcessed = TRUE;
tcp->receive.start = header.sequence;
tcp->receive.next = tcp->receive.start + 1;
responseFlags |= PTCP_ACK;
_tcp_setState(tcp, TCPS_SYNRECEIVED);
}
break;
}
case TCPS_SYNRECEIVED: {
/* receive ACK, move to ESTABLISHED */
if(header.flags & PTCP_ACK) {
wasProcessed = TRUE;
_tcp_setState(tcp, TCPS_ESTABLISHED);
/* if this is a child, mark it accordingly */
if(tcp->child) {
tcp->child->state = TCPCS_PENDING;
g_queue_push_tail(tcp->child->parent->server->pending, tcp->child);
/* user should accept new child from parent */
descriptor_adjustStatus(&(tcp->child->parent->super.super.super), DS_READABLE, TRUE);
}
}
break;
}
case TCPS_ESTABLISHED: {
/* receive FIN, send FINACK, move to CLOSEWAIT */
if(header.flags & PTCP_FIN) {
wasProcessed = TRUE;
/* other side of connections closed */
tcp->flags |= TCPF_REMOTE_CLOSED;
responseFlags |= (PTCP_FIN|PTCP_ACK);
_tcp_setState(tcp, TCPS_CLOSEWAIT);
/* remote will send us no more user data after this sequence */
tcp->receive.end = header.sequence;
}
break;
}
case TCPS_FINWAIT1: {
/* receive FINACK, move to FINWAIT2 */
if((header.flags & PTCP_FIN) && (header.flags & PTCP_ACK)) {
wasProcessed = TRUE;
_tcp_setState(tcp, TCPS_FINWAIT2);
}
/* receive FIN, send FINACK, move to CLOSING (simultaneous close) */
else if(header.flags & PTCP_FIN) {
wasProcessed = TRUE;
responseFlags |= (PTCP_FIN|PTCP_ACK);
tcp->flags |= TCPF_REMOTE_CLOSED;
_tcp_setState(tcp, TCPS_CLOSING);
/* it will send no more user data after this sequence */
tcp->receive.end = header.sequence;
}
break;
}
case TCPS_FINWAIT2: {
/* receive FIN, send FINACK, move to TIMEWAIT */
if(header.flags & PTCP_FIN) {
wasProcessed = TRUE;
responseFlags |= (PTCP_FIN|PTCP_ACK);
tcp->flags |= TCPF_REMOTE_CLOSED;
_tcp_setState(tcp, TCPS_TIMEWAIT);
/* it will send no more user data after this sequence */
tcp->receive.end = header.sequence;
}
break;
}
case TCPS_CLOSING: {
/* receive FINACK, move to TIMEWAIT */
if((header.flags & PTCP_FIN) && (header.flags & PTCP_ACK)) {
wasProcessed = TRUE;
_tcp_setState(tcp, TCPS_TIMEWAIT);
}
break;
}
case TCPS_TIMEWAIT: {
break;
}
case TCPS_CLOSEWAIT: {
break;
}
case TCPS_LASTACK: {
/* receive FINACK, move to CLOSED */
if((header.flags & PTCP_FIN) && (header.flags & PTCP_ACK)) {
wasProcessed = TRUE;
_tcp_setState(tcp, TCPS_CLOSED);
/* we closed, cant use tcp anymore, no retransmit */
packet_unref(packet);
return FALSE;
}
break;
}
case TCPS_CLOSED: {
/* stray packet, drop without retransmit */
packet_unref(packet);
return FALSE;
break;
}
default: {
break;
}
}
gint nPacketsAcked = 0;
/* check if we can update some TCP control info */
if(header.flags & PTCP_ACK) {
wasProcessed = TRUE;
if((header.acknowledgement > tcp->send.unacked) && (header.acknowledgement <= tcp->send.next)) {
/* some data we sent got acknowledged */
nPacketsAcked = header.acknowledgement - tcp->send.unacked;
/* the packets just acked are 'released' from retransmit queue */
for(guint i = tcp->send.unacked; i < header.acknowledgement; i++) {
_tcp_removeRetransmit(tcp, i);
}
tcp->send.unacked = header.acknowledgement;
/* update congestion window and keep track of when it was updated */
tcp->congestion.lastWindow = header.window;
tcp->congestion.lastSequence = header.sequence;
tcp->congestion.lastAcknowledgement = header.acknowledgement;
}
}
gboolean doRetransmitData = FALSE;
/* check if the packet carries user data for us */
if(packetLength > 0) {
/* it has data, check if its in the correct range */
if(header.sequence >= (tcp->receive.next + tcp->receive.window)) {
/* its too far ahead to accept now, but they should re-send it */
wasProcessed = TRUE;
doRetransmitData = TRUE;
} else if(header.sequence >= tcp->receive.next) {
/* its in our window, so we can accept the data */
wasProcessed = TRUE;
/*
* if this is THE next packet, we MUST accept it to avoid
* deadlocks (unless we are blocked b/c user should read)
*/
gboolean isNextPacket = (header.sequence == tcp->receive.next) ? TRUE : FALSE;
gboolean waitingUserRead = (socket_getInputBufferSpace(&(tcp->super)) > 0) ? TRUE : FALSE;
gboolean packetFits = (packetLength <= _tcp_getBufferSpaceIn(tcp)) ? TRUE : FALSE;
if((isNextPacket && !waitingUserRead) || (packetFits)) {
/* make sure its in order */
_tcp_bufferPacketIn(tcp, packet);
} else {
debug("no space for packet even though its in our window");
doRetransmitData = TRUE;
}
}
}
/* if it is a spurious packet, send a reset */
if(!wasProcessed) {
g_assert(responseFlags == PTCP_NONE);
responseFlags = PTCP_RST;
}
/* try to update congestion window based on potentially new info */
_tcp_updateCongestionWindow(tcp, nPacketsAcked);
/* now flush as many packets as we can to socket */
_tcp_flush(tcp);
/* send ack if they need updates but we didn't send any yet (selective acks) */
if((tcp->receive.next > tcp->send.lastAcknowledgement) ||
(tcp->receive.window != tcp->send.lastWindow))
{
responseFlags |= PTCP_ACK;
}
/* send control packet if we have one */
if(responseFlags != PTCP_NONE) {
debug("%s <-> %s: sending response control packet",
tcp->super.boundString, tcp->super.peerString);
Packet* response = _tcp_createPacket(tcp, responseFlags, NULL, 0);
_tcp_bufferPacketOut(tcp, response);
_tcp_flush(tcp);
}
/* we should free packets that are done but were not buffered */
if(!doRetransmitData && packetLength <= 0) {
packet_unref(packet);
}
return doRetransmitData;
}
gssize tcp_receiveUserData(TCP* tcp, gpointer buffer, gsize nBytes, in_addr_t* ip, in_port_t* port) {
MAGIC_ASSERT(tcp);
/*
* TODO
* We call descriptor_adjustStatus too many times here, to handle the readable
* state of the socket at times when we have a partially read packet. Consider
* adding a required hook for socket subclasses so the socket layer can
* query TCP for readability status.
*/
/* make sure we pull in all readable user data */
_tcp_flush(tcp);
gsize remaining = nBytes;
gsize bytesCopied = 0;
gsize totalCopied = 0;
gsize offset = 0;
gsize copyLength = 0;
/* check if we have a partial packet waiting to get finished */
if(remaining > 0 && tcp->partialUserDataPacket) {
guint partialLength = packet_getPayloadLength(tcp->partialUserDataPacket);
guint partialBytes = partialLength - tcp->partialOffset;
g_assert(partialBytes > 0);
copyLength = MIN(partialBytes, remaining);
bytesCopied = packet_copyPayload(tcp->partialUserDataPacket, tcp->partialOffset, buffer, copyLength);
totalCopied += bytesCopied;
remaining -= bytesCopied;
offset += bytesCopied;
if(bytesCopied >= partialBytes) {
/* we finished off the partial packet */
packet_unref(tcp->partialUserDataPacket);
tcp->partialUserDataPacket = NULL;
tcp->partialOffset = 0;
} else {
/* still more partial bytes left */
tcp->partialOffset += bytesCopied;
g_assert(remaining == 0);
}
}
while(remaining > 0) {
/* if we get here, we should have read the partial packet above, or
* broken out below */
g_assert(tcp->partialUserDataPacket == NULL);
g_assert(tcp->partialOffset == 0);
/* get the next buffered packet - we'll always need it.
* this could mark the socket as unreadable if this is its last packet.*/
Packet* packet = socket_removeFromInputBuffer((Socket*)tcp);
if(!packet) {
/* no more packets or partial packets */
break;
}
guint packetLength = packet_getPayloadLength(packet);
copyLength = MIN(packetLength, remaining);
bytesCopied = packet_copyPayload(packet, 0, buffer + offset, copyLength);
totalCopied += bytesCopied;
remaining -= bytesCopied;
offset += bytesCopied;
if(bytesCopied < packetLength) {
/* we were only able to read part of this packet */
tcp->partialUserDataPacket = packet;
tcp->partialOffset = bytesCopied;
break;
}
/* we read the entire packet, and are now finished with it */
packet_unref(packet);
}
/* now we update readability of the socket */
if((tcp->super.inputBufferLength > 0) || (tcp->partialUserDataPacket != NULL)) {
/* we still have readable data */
descriptor_adjustStatus(&(tcp->super.super.super), DS_READABLE, TRUE);
} else {
/* all of our ordered user data has been read */
if((tcp->unorderedInputLength == 0) && (tcp->error & TCPE_RECEIVE_EOF)) {
/* there is no more unordered data either, and we need to signal EOF */
if(totalCopied > 0) {
/* we just received bytes, so we can't EOF until the next call.
* make sure we stay readable so we DO actually EOF the socket */
descriptor_adjustStatus(&(tcp->super.super.super), DS_READABLE, TRUE);
} else {
/* OK, no more data and nothing just received. */
if(tcp->flags & TCPF_EOF_SIGNALED) {
/* we already signaled close, now its an error */
return -2;
} else {
/* we have not signaled close, do that now and close out the socket */
_tcp_endOfFileSignalled(tcp);
return 0;
}
}
} else {
/* our socket still has unordered data or is still open, but empty for now */
descriptor_adjustStatus(&(tcp->super.super.super), DS_READABLE, FALSE);
}
}
debug("%s <-> %s: receiving %lu user bytes", tcp->super.boundString, tcp->super.peerString, totalCopied);
return (gssize) (totalCopied == 0 ? -1 : totalCopied);
}
static void _networkinterface_scheduleNextSend(NetworkInterface* interface) {
/* the next packet needs to be sent according to bandwidth limitations.
* we need to spend time sending it before sending the next. */
SimulationTime batchTime = worker_getConfig()->interfaceBatchTime;
/* loop until we find a socket that has something to send */
while(interface->sendNanosecondsConsumed <= batchTime) {
gint socketHandle = -1;
/* choose which packet to send next based on our queuing discipline */
Packet* packet;
switch(interface->qdisc) {
case NIQ_RR: {
packet = _networkinterface_selectRoundRobin(interface, &socketHandle);
break;
}
case NIQ_FIFO:
default: {
packet = _networkinterface_selectFirstInFirstOut(interface, &socketHandle);
break;
}
}
if(!packet) {
break;
}
packet_addDeliveryStatus(packet, PDS_SND_INTERFACE_SENT);
/* now actually send the packet somewhere */
if(networkinterface_getIPAddress(interface) == packet_getDestinationIP(packet)) {
/* packet will arrive on our own interface */
PacketArrivedEvent* event = packetarrived_new(packet);
/* event destination is our node */
worker_scheduleEvent((Event*)event, 1, 0);
} else {
/* let the worker schedule with appropriate delays */
worker_schedulePacket(packet);
}
/* successfully sent, calculate how long it took to 'send' this packet */
guint length = packet_getPayloadLength(packet) + packet_getHeaderSize(packet);
interface->sendNanosecondsConsumed += (length * interface->timePerByteUp);
tracker_addOutputBytes(host_getTracker(worker_getCurrentHost()),(guint64)length, socketHandle);
_networkinterface_pcapWritePacket(interface, packet);
/* sending side is done with its ref */
packet_unref(packet);
}
/*
* we need to call back and try to send more, even if we didnt consume all
* of our batch time, because we might have more packets to send then.
*/
SimulationTime sendTime = (SimulationTime) floor(interface->sendNanosecondsConsumed);
if(sendTime >= SIMTIME_ONE_NANOSECOND) {
/* we are 'sending' the packets */
interface->flags |= NIF_SENDING;
/* call back when the packets are 'sent' */
InterfaceSentEvent* event = interfacesent_new(interface);
/* event destination is our node */
worker_scheduleEvent((Event*)event, sendTime, 0);
}
}
