bool
CCBListener::HandleCCBRequest( ClassAd &msg )
{
MyString address;
MyString connect_id;
MyString request_id;
MyString name;
if( !msg.LookupString( ATTR_MY_ADDRESS, address) ||
!msg.LookupString( ATTR_CLAIM_ID, connect_id) ||
!msg.LookupString( ATTR_REQUEST_ID, request_id) )
{
MyString msg_str;
msg.sPrint(msg_str);
EXCEPT("CCBListener: invalid CCB request from %s: %s\n",
m_ccb_address.Value(),
msg_str.Value() );
}
msg.LookupString( ATTR_NAME, name );
if( name.find(address.Value())<0 ) {
name.formatstr_cat(" with reverse connect address %s",address.Value());
}
dprintf(D_FULLDEBUG|D_NETWORK,
"CCBListener: received request to connect to %s, request id %s.\n",
name.Value(), request_id.Value());
return DoReversedCCBConnect( address.Value(), connect_id.Value(), request_id.Value(), name.Value() );
}
bool
CCBListener::HandleCCBRegistrationReply( ClassAd &msg )
{
if( !msg.LookupString(ATTR_CCBID,m_ccbid) ) {
MyString msg_str;
msg.sPrint(msg_str);
EXCEPT("CCBListener: no ccbid in registration reply: %s\n",
msg_str.Value() );
}
msg.LookupString(ATTR_CLAIM_ID,m_reconnect_cookie);
dprintf(D_ALWAYS,
"CCBListener: registered with CCB server %s as ccbid %s\n",
m_ccb_address.Value(),
m_ccbid.Value() );
m_waiting_for_registration = false;
m_registered = true;
daemonCore->daemonContactInfoChanged();
return true;
}
bool
CCBListener::ReadMsgFromCCB()
{
if( !m_sock ) {
return false;
}
m_sock->timeout(CCB_TIMEOUT);
ClassAd msg;
if( !msg.initFromStream( *m_sock ) || !m_sock->end_of_message() ) {
dprintf(D_ALWAYS,
"CCBListener: failed to receive message from CCB server %s\n",
m_ccb_address.Value());
Disconnected();
return false;
}
m_last_contact_from_peer = time(NULL);
RescheduleHeartbeat();
int cmd = -1;
msg.LookupInteger( ATTR_COMMAND, cmd );
switch( cmd ) {
case CCB_REGISTER:
return HandleCCBRegistrationReply( msg );
case CCB_REQUEST:
return HandleCCBRequest( msg );
case ALIVE:
dprintf(D_FULLDEBUG,"CCBListener: received heartbeat from server.\n");
return true;
}
MyString msg_str;
msg.sPrint(msg_str);
dprintf( D_ALWAYS,
"CCBListener: Unexpected message received from CCB "
"server: %s\n",
msg_str.Value() );
return false;
}
void
CCBServer::HandleRequestResultsMsg( CCBTarget *target )
{
// Reply from target daemon about whether it succeeded in
// connecting to the requested client.
Sock *sock = target->getSock();
ClassAd msg;
sock->decode();
if( !msg.initFromStream( *sock ) || !sock->end_of_message() ) {
// disconnect
dprintf(D_FULLDEBUG,
"CCB: received disconnect from target daemon %s "
"with ccbid %lu.\n",
sock->peer_description(), target->getCCBID() );
RemoveTarget( target );
return;
}
int command = 0;
if( msg.LookupInteger( ATTR_COMMAND, command ) && command == ALIVE ) {
SendHeartbeatResponse( target );
return;
}
target->decPendingRequestResults();
bool success = false;
MyString error_msg;
MyString reqid_str;
CCBID reqid;
MyString connect_id;
msg.LookupBool( ATTR_RESULT, success );
msg.LookupString( ATTR_ERROR_STRING, error_msg );
msg.LookupString( ATTR_REQUEST_ID, reqid_str );
msg.LookupString( ATTR_CLAIM_ID, connect_id );
if( !CCBIDFromString( reqid, reqid_str.Value() ) ) {
MyString msg_str;
msg.sPrint(msg_str);
dprintf(D_ALWAYS,
"CCB: received reply from target daemon %s with ccbid %lu "
"without a valid request id: %s\n",
sock->peer_description(),
target->getCCBID(),
msg_str.Value());
RemoveTarget( target );
return;
}
CCBServerRequest *request = GetRequest( reqid );
if( request && request->getSock()->readReady() ) {
// Request socket must have just closed. To avoid noise in
// logs when we fail to write to it, delete the request now.
RemoveRequest( request );
request = NULL;
}
char const *request_desc = "(client which has gone away)";
if( request ) {
request_desc = request->getSock()->peer_description();
}
if( success ) {
dprintf(D_FULLDEBUG,"CCB: received 'success' from target daemon %s "
"with ccbid %lu for "
"request %s from %s.\n",
sock->peer_description(),
target->getCCBID(),
reqid_str.Value(),
request_desc);
}
else {
dprintf(D_FULLDEBUG,"CCB: received error from target daemon %s "
"with ccbid %lu for "
"request %s from %s: %s\n",
sock->peer_description(),
target->getCCBID(),
reqid_str.Value(),
request_desc,
error_msg.Value());
}
if( !request ) {
if( success ) {
// expected: the client has gone away; it got what it wanted
return;
}
dprintf( D_FULLDEBUG,
"CCB: client for request %s to target daemon %s with ccbid "
"%lu disappeared before receiving error details.\n",
reqid_str.Value(),
sock->peer_description(),
target->getCCBID());
return;
}
if( connect_id != request->getConnectID() ) {
MyString msg_str;
msg.sPrint(msg_str);
dprintf( D_FULLDEBUG,
"CCB: received wrong connect id (%s) from target daemon %s "
"with ccbid %lu for "
"request %s\n",
connect_id.Value(),
sock->peer_description(),
target->getCCBID(),
reqid_str.Value());
RemoveTarget( target );
return;
}
RequestFinished( request, success, error_msg.Value() );
}
int
CCBServer::HandleRequest(int cmd,Stream *stream)
{
ReliSock *sock = (ReliSock *)stream;
ASSERT( cmd == CCB_REQUEST );
// Avoid lengthy blocking on communication with our peer.
// This command-handler should not get called until data
// is ready to read.
sock->timeout(1);
ClassAd msg;
sock->decode();
if( !msg.initFromStream( *sock ) || !sock->end_of_message() ) {
dprintf(D_ALWAYS,
"CCB: failed to receive request "
"from %s.\n", sock->peer_description() );
return FALSE;
}
MyString name;
if( msg.LookupString(ATTR_NAME,name) ) {
// client name is purely for debugging purposes
name.formatstr_cat(" on %s",sock->peer_description());
sock->set_peer_description(name.Value());
}
MyString target_ccbid_str;
MyString return_addr;
MyString connect_id; // id target daemon should present to requester
CCBID target_ccbid;
// NOTE: using ATTR_CLAIM_ID for connect id so that it is
// automatically treated as a secret over the network.
// It must be presented by the target daemon when connecting
// to the requesting client, so the client can confirm that
// the connection is in response to its request.
if( !msg.LookupString(ATTR_CCBID,target_ccbid_str) ||
!msg.LookupString(ATTR_MY_ADDRESS,return_addr) ||
!msg.LookupString(ATTR_CLAIM_ID,connect_id) )
{
MyString ad_str;
msg.sPrint(ad_str);
dprintf(D_ALWAYS,
"CCB: invalid request from %s: %s\n",
sock->peer_description(), ad_str.Value() );
return FALSE;
}
if( !CCBIDFromString(target_ccbid,target_ccbid_str.Value()) ) {
dprintf(D_ALWAYS,
"CCB: request from %s contains invalid CCBID %s\n",
sock->peer_description(), target_ccbid_str.Value() );
return FALSE;
}
CCBTarget *target = GetTarget( target_ccbid );
if( !target ) {
dprintf(D_ALWAYS,
"CCB: rejecting request from %s for ccbid %s because no daemon is "
"currently registered with that id "
"(perhaps it recently disconnected).\n",
sock->peer_description(), target_ccbid_str.Value());
MyString error_msg;
error_msg.formatstr(
"CCB server rejecting request for ccbid %s because no daemon is "
"currently registered with that id "
"(perhaps it recently disconnected).", target_ccbid_str.Value());
RequestReply( sock, false, error_msg.Value(), 0, target_ccbid );
return FALSE;
}
SetSmallBuffers(sock);
CCBServerRequest *request =
new CCBServerRequest(
sock,
target_ccbid,
return_addr.Value(),
connect_id.Value() );
AddRequest( request, target );
dprintf(D_FULLDEBUG,
"CCB: received request id %lu from %s for target ccbid %s "
"(registered as %s)\n",
request->getRequestID(),
request->getSock()->peer_description(),
target_ccbid_str.Value(),
target->getSock()->peer_description());
ForwardRequestToTarget( request, target );
return KEEP_STREAM;
}
bool DCSchedd::getJobConnectInfo(
PROC_ID jobid,
int subproc,
char const *session_info,
int timeout,
CondorError *errstack,
MyString &starter_addr,
MyString &starter_claim_id,
MyString &starter_version,
MyString &slot_name,
MyString &error_msg,
bool &retry_is_sensible)
{
ClassAd input;
ClassAd output;
input.Assign(ATTR_CLUSTER_ID,jobid.cluster);
input.Assign(ATTR_PROC_ID,jobid.proc);
if( subproc != -1 ) {
input.Assign(ATTR_SUB_PROC_ID,subproc);
}
input.Assign(ATTR_SESSION_INFO,session_info);
ReliSock sock;
if( !connectSock(&sock,timeout,errstack) ) {
error_msg = "Failed to connect to schedd";
dprintf( D_ALWAYS, "%s\n",error_msg.Value());
return false;
}
if( !startCommand(GET_JOB_CONNECT_INFO, &sock, timeout, errstack) ) {
error_msg = "Failed to send GET_JOB_CONNECT_INFO to schedd";
dprintf( D_ALWAYS, "%s\n",error_msg.Value());
return false;
}
if( !forceAuthentication(&sock, errstack) ) {
error_msg = "Failed to authenticate";
dprintf( D_ALWAYS, "%s\n",error_msg.Value());
return false;
}
sock.encode();
if( !input.put(sock) || !sock.end_of_message() ) {
error_msg = "Failed to send GET_JOB_CONNECT_INFO to schedd";
dprintf( D_ALWAYS, "%s\n",error_msg.Value());
return false;
}
sock.decode();
if( !output.initFromStream(sock) || !sock.end_of_message() ) {
error_msg = "Failed to get response from schedd";
dprintf( D_ALWAYS, "%s\n",error_msg.Value());
return false;
}
if( IsFulldebug(D_FULLDEBUG) ) {
std::string adstr;
output.SetPrivateAttributesInvisible(true);
output.sPrint(adstr);
output.SetPrivateAttributesInvisible(false);
dprintf(D_FULLDEBUG,"Response for GET_JOB_CONNECT_INFO:\n%s\n",
adstr.c_str());
}
bool result=false;
output.LookupBool(ATTR_RESULT,result);
if( !result ) {
output.LookupString(ATTR_ERROR_STRING,error_msg);
retry_is_sensible = false;
output.LookupBool(ATTR_RETRY,retry_is_sensible);
}
else {
output.LookupString(ATTR_STARTER_IP_ADDR,starter_addr);
output.LookupString(ATTR_CLAIM_ID,starter_claim_id);
output.LookupString(ATTR_VERSION,starter_version);
output.LookupString(ATTR_REMOTE_HOST,slot_name);
}
return result;
}
bool
DCTransferQueue::PollForTransferQueueSlot(int timeout,bool &pending,MyString &error_desc)
{
if( GoAheadAlways( m_xfer_downloading ) ) {
return true;
}
CheckTransferQueueSlot();
if( !m_xfer_queue_pending ) {
// status of request is known
pending = false;
if( !m_xfer_queue_go_ahead ) {
error_desc = m_xfer_rejected_reason;
}
return m_xfer_queue_go_ahead;
}
Selector selector;
selector.add_fd( m_xfer_queue_sock->get_file_desc(), Selector::IO_READ );
time_t start = time(NULL);
do {
int t = timeout - (time(NULL) - start);
selector.set_timeout( t >= 0 ? t : 0 );
selector.execute();
} while( selector.signalled() );
if( selector.timed_out() ) {
// It is expected that we may time out while waiting for a
// response. The caller should keep calling this function
// periodically until we get a result.
pending = true;
return false;
}
m_xfer_queue_sock->decode();
ClassAd msg;
if( !msg.initFromStream(*m_xfer_queue_sock) ||
!m_xfer_queue_sock->end_of_message() )
{
formatstr(m_xfer_rejected_reason,
"Failed to receive transfer queue response from %s for job %s "
"(initial file %s).",
m_xfer_queue_sock->peer_description(),
m_xfer_jobid.c_str(),
m_xfer_fname.c_str());
goto request_failed;
}
int result; // this should be one of the values in XFER_QUEUE_ENUM
if( !msg.LookupInteger(ATTR_RESULT,result) ) {
std::string msg_str;
msg.sPrint(msg_str);
formatstr(m_xfer_rejected_reason,
"Invalid transfer queue response from %s for job %s (%s): %s",
m_xfer_queue_sock->peer_description(),
m_xfer_jobid.c_str(),
m_xfer_fname.c_str(),
msg_str.c_str());
goto request_failed;
}
if( result == XFER_QUEUE_GO_AHEAD ) {
m_xfer_queue_go_ahead = true;
}
else {
m_xfer_queue_go_ahead = false;
std::string reason;
msg.LookupString(ATTR_ERROR_STRING,reason);
formatstr(m_xfer_rejected_reason,
"Request to transfer files for %s (%s) was rejected by %s: %s",
m_xfer_jobid.c_str(), m_xfer_fname.c_str(),
m_xfer_queue_sock->peer_description(),
reason.c_str());
goto request_failed;
}
m_xfer_queue_pending = false;
pending = m_xfer_queue_pending;
return true;
request_failed:
error_desc = m_xfer_rejected_reason;
dprintf(D_ALWAYS, "%s\n", m_xfer_rejected_reason.c_str());
m_xfer_queue_pending = false;
m_xfer_queue_go_ahead = false;
pending = m_xfer_queue_pending;
return false;
}