ClassAd *CollectorEngine::
collect (int command, Sock *sock, const condor_sockaddr& from, int &insert)
{
ClassAd *clientAd;
ClassAd *rval;
// Avoid lengthy blocking on communication with our peer.
// This command-handler should not get called until data
// is ready to read.
sock->timeout(1);
clientAd = new ClassAd;
if (!clientAd) return 0;
// get the ad
if( !clientAd->initFromStream(*sock) )
{
dprintf (D_ALWAYS,"Command %d on Sock not follwed by ClassAd (or timeout occured)\n",
command);
delete clientAd;
sock->end_of_message();
return 0;
}
// insert the authenticated user into the ad itself
const char* authn_user = sock->getFullyQualifiedUser();
if (authn_user) {
clientAd->Assign("AuthenticatedIdentity", authn_user);
} else {
// remove it from the ad if it's not authenticated.
clientAd->Delete("AuthenticatedIdentity");
}
rval = collect(command, clientAd, from, insert, sock);
// Don't leak the ad on error!
if ( ! rval ) {
delete clientAd;
}
// get the end_of_message()
if (!sock->end_of_message())
{
dprintf(D_FULLDEBUG,"Warning: Command %d; maybe shedding data on eom\n",
command);
}
return rval;
}
bool
DCStarter::createJobOwnerSecSession(int timeout,char const *job_claim_id,char const *starter_sec_session,char const *session_info,MyString &owner_claim_id,MyString &error_msg,MyString &starter_version,MyString &starter_addr)
{
ReliSock sock;
if( !connectSock(&sock, timeout, NULL) ) {
error_msg = "Failed to connect to starter";
return false;
}
if( !startCommand(CREATE_JOB_OWNER_SEC_SESSION, &sock,timeout,NULL,NULL,false,starter_sec_session) ) {
error_msg = "Failed to send CREATE_JOB_OWNER_SEC_SESSION to starter";
return false;
}
ClassAd input;
input.Assign(ATTR_CLAIM_ID,job_claim_id);
input.Assign(ATTR_SESSION_INFO,session_info);
sock.encode();
if( !input.put(sock) || !sock.end_of_message() ) {
error_msg = "Failed to compose CREATE_JOB_OWNER_SEC_SESSION to starter";
return false;
}
sock.decode();
ClassAd reply;
if( !reply.initFromStream(sock) || !sock.end_of_message() ) {
error_msg = "Failed to get response to CREATE_JOB_OWNER_SEC_SESSION from starter";
return false;
}
bool success = false;
reply.LookupBool(ATTR_RESULT,success);
if( !success ) {
reply.LookupString(ATTR_ERROR_STRING,error_msg);
return false;
}
reply.LookupString(ATTR_CLAIM_ID,owner_claim_id);
reply.LookupString(ATTR_VERSION,starter_version);
// get the full starter address from the starter in case it contains
// extra CCB info that we don't already know about
reply.LookupString(ATTR_STARTER_IP_ADDR,starter_addr);
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;
}
// This handler is called when a client wishes to write files from the
// transferd's storage.
int
TransferD::write_files_handler(int cmd, Stream *sock)
{
ReliSock *rsock = (ReliSock*)sock;
MyString capability;
int protocol = FTP_UNKNOWN;
TransferRequest *treq = NULL;
MyString fquser;
static int transfer_reaper_id = -1;
ThreadArg *thread_arg;
int tid;
ClassAd reqad;
ClassAd respad;
cmd = cmd; // quiet the compiler.
dprintf(D_ALWAYS, "Got TRANSFERD_WRITE_FILES!\n");
/////////////////////////////////////////////////////////////////////////
// make sure we are authenticated
/////////////////////////////////////////////////////////////////////////
if( ! rsock->triedAuthentication() ) {
CondorError errstack;
if( ! SecMan::authenticate_sock(rsock, WRITE, &errstack) ) {
// we failed to authenticate, we should bail out now
// since we don't know what user is trying to perform
// this action.
// TODO: it'd be nice to print out what failed, but we
// need better error propagation for that...
errstack.push( "TransferD::setup_transfer_request_handler()", 42,
"Failure to register transferd - Authentication failed" );
dprintf( D_ALWAYS, "setup_transfer_request_handler() "
"aborting: %s\n",
errstack.getFullText() );
refuse( rsock );
return CLOSE_STREAM;
}
}
fquser = rsock->getFullyQualifiedUser();
/////////////////////////////////////////////////////////////////////////
// Check to see if the capability the client tells us is something that
// we have knowledge of. We ONLY check the capability and not the
// identity of the person in question. This allows people of different
// identities to write files here as long as they had the right
// capability. While this might not sound secure, they STILL had to have
// authenticated as someone this daemon trusts.
// Similarly, check the protocol it wants to use as well as ensure that
// the direction the transfer request was supposed to be is being honored.
/////////////////////////////////////////////////////////////////////////
rsock->decode();
// soak the request ad from the client about what it wants to transfer
reqad.initFromStream(*rsock);
rsock->end_of_message();
reqad.LookupString(ATTR_TREQ_CAPABILITY, capability);
rsock->encode();
// do I know of such a capability?
if (m_treqs.lookup(capability, treq) != 0) {
// didn't find it. Log it and tell them to leave and close up shop
respad.Assign(ATTR_TREQ_INVALID_REQUEST, TRUE);
respad.Assign(ATTR_TREQ_INVALID_REASON, "Invalid capability!");
respad.put(*rsock);
rsock->end_of_message();
dprintf(D_ALWAYS, "Client identity '%s' tried to write some files "
"using capability '%s', but there was no such capability. "
"Access denied.\n", fquser.Value(), capability.Value());
return CLOSE_STREAM;
}
reqad.LookupInteger(ATTR_TREQ_FTP, protocol);
// am I willing to use this protocol?
switch(protocol) {
case FTP_CFTP: // FileTrans protocol, I'm happy.
break;
default:
respad.Assign(ATTR_TREQ_INVALID_REQUEST, TRUE);
respad.Assign(ATTR_TREQ_INVALID_REASON,
"Invalid file transfer protocol!");
respad.put(*rsock);
rsock->end_of_message();
dprintf(D_ALWAYS, "Client identity '%s' tried to write some files "
"using protocol '%d', but I don't support that protocol. "
"Access denied.\n", fquser.Value(), protocol);
return CLOSE_STREAM;
}
// nsure that this transfer request was of the uploading variety
if (treq->get_direction() != FTPD_UPLOAD) {
respad.Assign(ATTR_TREQ_INVALID_REQUEST, TRUE);
respad.Assign(ATTR_TREQ_INVALID_REASON,
"Transfer Request was not an uploading request!");
respad.put(*rsock);
rsock->end_of_message();
dprintf(D_ALWAYS, "Client identity '%s' tried to write some files "
"to a transfer request that wasn't expecting to be written. "
"Access denied.\n", fquser.Value());
}
/////////////////////////////////////////////////////////////////////////
// Tell the client everything was ok.
/////////////////////////////////////////////////////////////////////////
respad.Assign(ATTR_TREQ_INVALID_REQUEST, FALSE);
respad.put(*rsock);
rsock->end_of_message();
/////////////////////////////////////////////////////////////////////////
// Set up a thread (a process under unix) to read ALL of the job files
// for all of the ads in the TransferRequest.
/////////////////////////////////////////////////////////////////////////
// now create a thread, passing in the sock, which uses the file transfer
// object to accept the files.
if (transfer_reaper_id == -1) {
// only set this up ONCE so each and every thread gets one.
transfer_reaper_id = daemonCore->Register_Reaper(
"write_files_reaper",
(ReaperHandlercpp) &TransferD::write_files_reaper,
"write_files_reaper",
this
);
}
thread_arg = new ThreadArg(protocol, treq);
// Start a new thread (process on Unix) to do the work
tid = daemonCore->Create_Thread(
(ThreadStartFunc)&TransferD::write_files_thread,
(void *)thread_arg,
rsock,
transfer_reaper_id
);
if (tid == FALSE) {
// XXX How do I handle this failure?
}
// associate the tid with the request so I can deal with it propery in
// the reaper
m_client_to_transferd_threads.insert(tid, treq);
// The stream is inherited to the thread, who does the transfer and
// finishes the protocol, but in the parent, I'm closing it.
return CLOSE_STREAM;
}
// download the files associated with the jobads to the sandbox at td_sinful
// with the supplied capability.
// The work_ad should contain:
// ATTR_TREQ_CAPABILITY
// ATTR_TREQ_FTP
// ATTR_TREQ_JOBID_ALLOW_LIST
bool
DCTransferD::download_job_files(ClassAd *work_ad, CondorError * errstack)
{
ReliSock *rsock = NULL;
int timeout = 60 * 60 * 8; // transfers take a long time...
int i;
ClassAd reqad, respad;
std::string cap;
int ftp;
int invalid;
int protocol;
std::string reason;
int num_transfers;
ClassAd jad;
const char *lhstr = NULL;
ExprTree *tree = NULL;
//////////////////////////////////////////////////////////////////////////
// Connect to the transferd and authenticate
//////////////////////////////////////////////////////////////////////////
// This call with automatically connect to _addr, which was set in the
// constructor of this object to be the transferd in question.
rsock = (ReliSock*)startCommand(TRANSFERD_READ_FILES, Stream::reli_sock,
timeout, errstack);
if( ! rsock ) {
dprintf( D_ALWAYS, "DCTransferD::download_job_files: "
"Failed to send command (TRANSFERD_READ_FILES) "
"to the schedd\n" );
errstack->push("DC_TRANSFERD", 1,
"Failed to start a TRANSFERD_READ_FILES command.");
return false;
}
// First, if we're not already authenticated, force that now.
if (!forceAuthentication( rsock, errstack )) {
dprintf( D_ALWAYS, "DCTransferD::download_job_files() authentication "
"failure: %s\n", errstack->getFullText().c_str() );
errstack->push("DC_TRANSFERD", 1,
"Failed to authenticate properly.");
return false;
}
rsock->encode();
//////////////////////////////////////////////////////////////////////////
// Query the transferd about the capability/protocol and see if I can
// download my files. It will respond with a classad saying good or bad.
//////////////////////////////////////////////////////////////////////////
work_ad->LookupString(ATTR_TREQ_CAPABILITY, cap);
work_ad->LookupInteger(ATTR_TREQ_FTP, ftp);
reqad.Assign(ATTR_TREQ_CAPABILITY, cap);
reqad.Assign(ATTR_TREQ_FTP, ftp);
// This request ad to the transferd should contain:
// ATTR_TREQ_CAPABILITY
// ATTR_TREQ_FTP
reqad.put(*rsock);
rsock->end_of_message();
rsock->decode();
// This response ad from the transferd should contain:
// ATTR_TREQ_INVALID_REQUEST (set to true)
// ATTR_TREQ_INVALID_REASON
//
// OR
//
// ATTR_TREQ_INVALID_REQUEST (set to false)
// ATTR_TREQ_NUM_TRANSFERS
//
respad.initFromStream(*rsock);
rsock->end_of_message();
respad.LookupInteger(ATTR_TREQ_INVALID_REQUEST, invalid);
if (invalid == TRUE) {
// The transferd rejected my attempt to upload the fileset
delete rsock;
respad.LookupString(ATTR_TREQ_INVALID_REASON, reason);
errstack->push("DC_TRANSFERD", 1, reason.c_str());
return false;
}
respad.LookupInteger(ATTR_TREQ_NUM_TRANSFERS, num_transfers);
//////////////////////////////////////////////////////////////////////////
// Based upon the protocol I've chosen, use that method to download the
// files. When using the FileTrans protocol, a child process on the
// transferd side will be sending me individual job ads and then
// instantiating a filetransfer object for that ad.
//////////////////////////////////////////////////////////////////////////
dprintf(D_ALWAYS, "Receiving fileset");
work_ad->LookupInteger(ATTR_TREQ_FTP, protocol);
switch(protocol) {
case FTP_CFTP: // download the files using the FileTransfer Object
for (i = 0; i < num_transfers; i++) {
// Grab a job ad the server is sending us so we know what
// to receive.
jad.initFromStream(*rsock);
rsock->end_of_message();
// translate the job ad by replacing the
// saved SUBMIT_ attributes so the download goes into the
// correct place.
jad.ResetExpr();
while( jad.NextExpr(lhstr, tree) ) {
if ( lhstr && strncasecmp("SUBMIT_",lhstr,7)==0 ) {
// this attr name starts with SUBMIT_
// compute new lhs (strip off the SUBMIT_)
const char *new_attr_name = strchr(lhstr,'_');
ExprTree * pTree;
ASSERT(new_attr_name);
new_attr_name++;
// insert attribute
pTree = tree->Copy();
jad.Insert(new_attr_name, pTree, false);
}
} // while next expr
// instantiate a filetransfer object and have it accept the
// files.
FileTransfer ftrans;
if ( !ftrans.SimpleInit(&jad, false, false, rsock) )
{
delete rsock;
errstack->push("DC_TRANSFERD", 1,
"Failed to initate uploading of files.");
return false;
}
// We want files to be copied to their final places, so apply
// any filename remaps when downloading.
if ( !ftrans.InitDownloadFilenameRemaps(&jad) ) {
return false;
}
ftrans.setPeerVersion( version() );
if ( !ftrans.DownloadFiles() ) {
delete rsock;
errstack->push("DC_TRANSFERD", 1,
"Failed to download files.");
return false;
}
dprintf(D_ALWAYS | D_NOHEADER, ".");
}
rsock->end_of_message();
dprintf(D_ALWAYS | D_NOHEADER, "\n");
break;
default:
// Bail due to user error. This client doesn't support the unknown
// protocol.
delete rsock;
errstack->push("DC_TRANSFERD", 1,
"Unknown file transfer protocol selected.");
return false;
break;
}
//////////////////////////////////////////////////////////////////////////
// Get the response from the transferd once it sees a completed
// movement of files to the child process.
//////////////////////////////////////////////////////////////////////////
rsock->decode();
respad.initFromStream(*rsock);
rsock->end_of_message();
// close up shop
delete rsock;
respad.LookupInteger(ATTR_TREQ_INVALID_REQUEST, invalid);
if ( invalid == TRUE ) {
respad.LookupString(ATTR_TREQ_INVALID_REASON, reason);
errstack->push("DC_TRANSFERD", 1, reason.c_str());
return false;
}
return true;
}
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;
}
int
CCBServer::HandleRegistration(int cmd,Stream *stream)
{
ReliSock *sock = (ReliSock *)stream;
ASSERT( cmd == CCB_REGISTER );
// 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 registration "
"from %s.\n", sock->peer_description() );
return FALSE;
}
SetSmallBuffers(sock);
MyString name;
if( msg.LookupString(ATTR_NAME,name) ) {
// target daemon name is purely for debugging purposes
name.formatstr_cat(" on %s",sock->peer_description());
sock->set_peer_description(name.Value());
}
CCBTarget *target = new CCBTarget(sock);
MyString reconnect_cookie_str,reconnect_ccbid_str;
CCBID reconnect_cookie,reconnect_ccbid;
bool reconnected = false;
if( msg.LookupString(ATTR_CLAIM_ID,reconnect_cookie_str) &&
CCBIDFromString(reconnect_cookie,reconnect_cookie_str.Value()) &&
msg.LookupString( ATTR_CCBID,reconnect_ccbid_str) &&
CCBIDFromContactString(reconnect_ccbid,reconnect_ccbid_str.Value()) )
{
target->setCCBID( reconnect_ccbid );
reconnected = ReconnectTarget( target, reconnect_cookie );
}
if( !reconnected ) {
AddTarget( target );
}
CCBReconnectInfo *reconnect_info = GetReconnectInfo( target->getCCBID() );
ASSERT( reconnect_info );
sock->encode();
ClassAd reply_msg;
MyString ccb_contact;
CCBIDToString( reconnect_info->getReconnectCookie(),reconnect_cookie_str );
// We send our address as part of the CCB contact string, rather
// than letting the target daemon fill it in. This is to give us
// potential flexibility on the CCB server side to do things like
// assign different targets to different CCB server sub-processes,
// each with their own command port.
CCBIDToContactString( m_address.Value(), target->getCCBID(), ccb_contact );
reply_msg.Assign(ATTR_CCBID,ccb_contact.Value());
reply_msg.Assign(ATTR_COMMAND,CCB_REGISTER);
reply_msg.Assign(ATTR_CLAIM_ID,reconnect_cookie_str.Value());
if( !reply_msg.put( *sock ) || !sock->end_of_message() ) {
dprintf(D_ALWAYS,
"CCB: failed to send registration response "
"to %s.\n", sock->peer_description() );
RemoveTarget( target );
return KEEP_STREAM; // we have already closed this socket
}
return KEEP_STREAM;
}
bool DCStarter::startSSHD(char const *known_hosts_file,char const *private_client_key_file,char const *preferred_shells,char const *slot_name,char const *ssh_keygen_args,ReliSock &sock,int timeout,char const *sec_session_id,MyString &remote_user,MyString &error_msg,bool &retry_is_sensible)
{
retry_is_sensible = false;
#ifndef HAVE_SSH_TO_JOB
error_msg = "This version of Condor does not support ssh key exchange.";
return false;
#else
if( !connectSock(&sock, timeout, NULL) ) {
error_msg = "Failed to connect to starter";
return false;
}
if( !startCommand(START_SSHD, &sock,timeout,NULL,NULL,false,sec_session_id) ) {
error_msg = "Failed to send START_SSHD to starter";
return false;
}
ClassAd input;
if( preferred_shells && *preferred_shells ) {
input.Assign(ATTR_SHELL,preferred_shells);
}
if( slot_name && *slot_name ) {
// This is a little silly.
// We are telling the remote side the name of the slot so
// that it can put it in the welcome message.
input.Assign(ATTR_NAME,slot_name);
}
if( ssh_keygen_args && *ssh_keygen_args ) {
input.Assign(ATTR_SSH_KEYGEN_ARGS,ssh_keygen_args);
}
sock.encode();
if( !input.put(sock) || !sock.end_of_message() ) {
error_msg = "Failed to send START_SSHD request to starter";
return false;
}
ClassAd result;
sock.decode();
if( !result.initFromStream(sock) || !sock.end_of_message() ) {
error_msg = "Failed to read response to START_SSHD from starter";
return false;
}
bool success = false;
result.LookupBool(ATTR_RESULT,success);
if( !success ) {
std::string remote_error_msg;
result.LookupString(ATTR_ERROR_STRING,remote_error_msg);
error_msg.sprintf("%s: %s",slot_name,remote_error_msg.c_str());
retry_is_sensible = false;
result.LookupBool(ATTR_RETRY,retry_is_sensible);
return false;
}
result.LookupString(ATTR_REMOTE_USER,remote_user);
std::string public_server_key;
if( !result.LookupString(ATTR_SSH_PUBLIC_SERVER_KEY,public_server_key) ) {
error_msg = "No public ssh server key received in reply to START_SSHD";
return false;
}
std::string private_client_key;
if( !result.LookupString(ATTR_SSH_PRIVATE_CLIENT_KEY,private_client_key) ) {
error_msg = "No ssh client key received in reply to START_SSHD";
return false;
}
// store the private client key
unsigned char *decode_buf = NULL;
int length = -1;
condor_base64_decode(private_client_key.c_str(),&decode_buf,&length);
if( !decode_buf ) {
error_msg = "Error decoding ssh client key.";
return false;
}
FILE *fp = safe_fcreate_fail_if_exists(private_client_key_file,"a",0400);
if( !fp ) {
error_msg.sprintf("Failed to create %s: %s",
private_client_key_file,strerror(errno));
free( decode_buf );
return false;
}
if( fwrite(decode_buf,length,1,fp)!=1 ) {
error_msg.sprintf("Failed to write to %s: %s",
private_client_key_file,strerror(errno));
fclose( fp );
free( decode_buf );
return false;
}
if( fclose(fp)!=0 ) {
error_msg.sprintf("Failed to close %s: %s",
private_client_key_file,strerror(errno));
free( decode_buf );
return false;
}
fp = NULL;
free( decode_buf );
decode_buf = NULL;
// store the public server key in the known_hosts file
length = -1;
condor_base64_decode(public_server_key.c_str(),&decode_buf,&length);
if( !decode_buf ) {
error_msg = "Error decoding ssh server key.";
return false;
}
fp = safe_fcreate_fail_if_exists(known_hosts_file,"a",0600);
if( !fp ) {
error_msg.sprintf("Failed to create %s: %s",
known_hosts_file,strerror(errno));
free( decode_buf );
return false;
}
// prepend a host name pattern (*) to the public key to make a valid
// record in the known_hosts file
fprintf(fp,"* ");
if( fwrite(decode_buf,length,1,fp)!=1 ) {
error_msg.sprintf("Failed to write to %s: %s",
known_hosts_file,strerror(errno));
fclose( fp );
free( decode_buf );
return false;
}
if( fclose(fp)!=0 ) {
error_msg.sprintf("Failed to close %s: %s",
known_hosts_file,strerror(errno));
free( decode_buf );
return false;
}
fp = NULL;
free( decode_buf );
decode_buf = NULL;
return true;
#endif
}
// when a transferd registers itself, it identifies who it is. The connection
// is then held open and the schedd periodically might send more transfer
// requests to the transferd. Also, if the transferd dies, the schedd is
// informed quickly and reliably due to the closed connection.
bool
DCSchedd::register_transferd(MyString sinful, MyString id, int timeout,
ReliSock **regsock_ptr, CondorError *errstack)
{
ReliSock *rsock;
int invalid_request = 0;
ClassAd regad;
ClassAd respad;
std::string errstr;
std::string reason;
if (regsock_ptr != NULL) {
// Our caller wants a pointer to the socket we used to succesfully
// register the claim. The NULL pointer will represent failure and
// this will only be set to something real if everything was ok.
*regsock_ptr = NULL;
}
// This call with automatically connect to _addr, which was set in the
// constructor of this object to be the schedd in question.
rsock = (ReliSock*)startCommand(TRANSFERD_REGISTER, Stream::reli_sock,
timeout, errstack);
if( ! rsock ) {
dprintf( D_ALWAYS, "DCSchedd::register_transferd: "
"Failed to send command (TRANSFERD_REGISTER) "
"to the schedd\n" );
errstack->push("DC_SCHEDD", 1,
"Failed to start a TRANSFERD_REGISTER command.");
return false;
}
// First, if we're not already authenticated, force that now.
if (!forceAuthentication( rsock, errstack )) {
dprintf( D_ALWAYS, "DCSchedd::register_transferd authentication "
"failure: %s\n", errstack->getFullText().c_str() );
errstack->push("DC_SCHEDD", 1,
"Failed to authenticate properly.");
return false;
}
rsock->encode();
// set up my registration request.
regad.Assign(ATTR_TREQ_TD_SINFUL, sinful);
regad.Assign(ATTR_TREQ_TD_ID, id);
// This is the initial registration identification ad to the schedd
// It contains:
// ATTR_TREQ_TD_SINFUL
// ATTR_TREQ_TD_ID
regad.put(*rsock);
rsock->end_of_message();
// Get the response from the schedd.
rsock->decode();
// This is the response ad from the schedd:
// It contains:
// ATTR_TREQ_INVALID_REQUEST
//
// OR
//
// ATTR_TREQ_INVALID_REQUEST
// ATTR_TREQ_INVALID_REASON
respad.initFromStream(*rsock);
rsock->end_of_message();
respad.LookupInteger(ATTR_TREQ_INVALID_REQUEST, invalid_request);
if (invalid_request == FALSE) {
// not an invalid request
if (regsock_ptr)
*regsock_ptr = rsock;
return true;
}
respad.LookupString(ATTR_TREQ_INVALID_REASON, reason);
errstack->pushf("DC_SCHEDD", 1, "Schedd refused registration: %s", reason.c_str());
return false;
}
bool
DCSchedd::receiveJobSandbox(const char* constraint, CondorError * errstack, int * numdone /*=0*/)
{
if(numdone) { *numdone = 0; }
ExprTree *tree = NULL;
const char *lhstr;
int reply;
int i;
ReliSock rsock;
int JobAdsArrayLen;
bool use_new_command = true;
if ( version() ) {
CondorVersionInfo vi( version() );
if ( vi.built_since_version(6,7,7) ) {
use_new_command = true;
} else {
use_new_command = false;
}
}
// // // // // // // //
// On the wire protocol
// // // // // // // //
rsock.timeout(20); // years of research... :)
if( ! rsock.connect(_addr) ) {
dprintf( D_ALWAYS, "DCSchedd::receiveJobSandbox: "
"Failed to connect to schedd (%s)\n", _addr );
return false;
}
if ( use_new_command ) {
if( ! startCommand(TRANSFER_DATA_WITH_PERMS, (Sock*)&rsock, 0,
errstack) ) {
dprintf( D_ALWAYS, "DCSchedd::receiveJobSandbox: "
"Failed to send command (TRANSFER_DATA_WITH_PERMS) "
"to the schedd\n" );
return false;
}
} else {
if( ! startCommand(TRANSFER_DATA, (Sock*)&rsock, 0, errstack) ) {
dprintf( D_ALWAYS, "DCSchedd::receiveJobSandbox: "
"Failed to send command (TRANSFER_DATA) "
"to the schedd\n" );
return false;
}
}
// First, if we're not already authenticated, force that now.
if (!forceAuthentication( &rsock, errstack )) {
dprintf( D_ALWAYS,
"DCSchedd::receiveJobSandbox: authentication failure: %s\n",
errstack ? errstack->getFullText().c_str() : "" );
return false;
}
rsock.encode();
// Send our version if using the new command
if ( use_new_command ) {
// Need to use a named variable, else the wrong version of
// code() is called.
char *my_version = strdup( CondorVersion() );
if ( !rsock.code(my_version) ) {
dprintf(D_ALWAYS,"DCSchedd:receiveJobSandbox: "
"Can't send version string to the schedd\n");
free( my_version );
return false;
}
free( my_version );
}
// Send the constraint
char * nc_constraint = strdup( constraint ); // de-const
if ( !rsock.code(nc_constraint) ) {
free( nc_constraint );
dprintf(D_ALWAYS,"DCSchedd:receiveJobSandbox: "
"Can't send JobAdsArrayLen to the schedd\n");
return false;
}
free( nc_constraint );
if ( !rsock.end_of_message() ) {
std::string errmsg;
formatstr(errmsg,
"Can't send initial message (version + constraint) to schedd (%s)",
_addr);
dprintf(D_ALWAYS,"DCSchedd::receiveJobSandbox: %s\n", errmsg.c_str());
if( errstack ) {
errstack->push(
"DCSchedd::receiveJobSandbox",
CEDAR_ERR_EOM_FAILED,
errmsg.c_str());
}
return false;
}
// Now, read how many jobs matched the constraint.
rsock.decode();
if ( !rsock.code(JobAdsArrayLen) ) {
std::string errmsg;
formatstr(errmsg,
"Can't receive JobAdsArrayLen from the schedd (%s)",
_addr);
dprintf(D_ALWAYS,"DCSchedd::receiveJobSandbox: %s\n", errmsg.c_str());
if( errstack ) {
errstack->push(
"DCSchedd::receiveJobSandbox",
CEDAR_ERR_GET_FAILED,
errmsg.c_str());
}
return false;
}
rsock.end_of_message();
dprintf(D_FULLDEBUG,"DCSchedd:receiveJobSandbox: "
"%d jobs matched my constraint (%s)\n",
JobAdsArrayLen, constraint);
// Now read all the files via the file transfer object
for (i=0; i<JobAdsArrayLen; i++) {
FileTransfer ftrans;
ClassAd job;
// grab job ClassAd
if ( !job.initFromStream(rsock) ) {
std::string errmsg;
formatstr(errmsg, "Can't receive job ad %d from the schedd", i);
dprintf(D_ALWAYS, "DCSchedd::receiveJobSandbox: %s\n", errmsg.c_str());
if( errstack ) {
errstack->push(
"DCSchedd::receiveJobSandbox",
CEDAR_ERR_GET_FAILED,
errmsg.c_str());
}
return false;
}
rsock.end_of_message();
// translate the job ad by replacing the
// saved SUBMIT_ attributes
job.ResetExpr();
while( job.NextExpr(lhstr, tree) ) {
if ( lhstr && strncasecmp("SUBMIT_",lhstr,7)==0 ) {
// this attr name starts with SUBMIT_
// compute new lhs (strip off the SUBMIT_)
const char *new_attr_name = strchr(lhstr,'_');
ExprTree * pTree;
ASSERT(new_attr_name);
new_attr_name++;
// insert attribute
pTree = tree->Copy();
job.Insert(new_attr_name, pTree, false);
}
} // while next expr
if ( !ftrans.SimpleInit(&job,false,false,&rsock) ) {
if( errstack ) {
int cluster = -1, proc = -1;
job.LookupInteger(ATTR_CLUSTER_ID,cluster);
job.LookupInteger(ATTR_PROC_ID,proc);
errstack->pushf(
"DCSchedd::receiveJobSandbox",
FILETRANSFER_INIT_FAILED,
"File transfer initialization failed for target job %d.%d",
cluster, proc );
}
return false;
}
// We want files to be copied to their final places, so apply
// any filename remaps when downloading.
if ( !ftrans.InitDownloadFilenameRemaps(&job) ) {
return false;
}
if ( use_new_command ) {
ftrans.setPeerVersion( version() );
}
if ( !ftrans.DownloadFiles() ) {
if( errstack ) {
FileTransfer::FileTransferInfo ft_info = ftrans.GetInfo();
int cluster = -1, proc = -1;
job.LookupInteger(ATTR_CLUSTER_ID,cluster);
job.LookupInteger(ATTR_PROC_ID,proc);
errstack->pushf(
"DCSchedd::receiveJobSandbox",
FILETRANSFER_DOWNLOAD_FAILED,
"File transfer failed for target job %d.%d: %s",
cluster, proc, ft_info.error_desc.Value() );
}
return false;
}
}
rsock.end_of_message();
rsock.encode();
reply = OK;
rsock.code(reply);
rsock.end_of_message();
if(numdone) { *numdone = JobAdsArrayLen; }
return true;
}
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;
}
ClassAd *CollectorEngine::
collect (int command,ClassAd *clientAd,const condor_sockaddr& from,int &insert,Sock *sock)
{
ClassAd *retVal;
ClassAd *pvtAd;
int insPvt;
AdNameHashKey hk;
HashString hashString;
static int repeatStartdAds = -1; // for debugging
ClassAd *clientAdToRepeat = NULL;
if (repeatStartdAds == -1) {
repeatStartdAds = param_integer("COLLECTOR_REPEAT_STARTD_ADS",0);
}
if( !ValidateClassAd(command,clientAd,sock) ) {
return NULL;
}
// mux on command
switch (command)
{
case UPDATE_STARTD_AD:
case UPDATE_STARTD_AD_WITH_ACK:
#if !defined(WANT_OLD_CLASSADS)
clientAd->AddTargetRefs( TargetJobAttrs );
#endif
if ( repeatStartdAds > 0 ) {
clientAdToRepeat = new ClassAd(*clientAd);
}
if (!makeStartdAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=updateClassAd (StartdAds, "StartdAd ", "Start",
clientAd, hk, hashString, insert, from );
// if we want to store private ads
if (!sock)
{
dprintf (D_ALWAYS, "Want private ads, but no socket given!\n");
break;
}
else
{
if (!(pvtAd = new ClassAd))
{
EXCEPT ("Memory error!");
}
if( !pvtAd->initFromStream(*sock) )
{
dprintf(D_FULLDEBUG,"\t(Could not get startd's private ad)\n");
delete pvtAd;
break;
}
// Fix up some stuff in the private ad that we depend on.
// We started doing this in 7.2.0, so once we no longer
// care about compatibility with stuff from before then,
// the startd could stop bothering to send these attributes.
// Queries of private ads depend on the following:
pvtAd->SetMyTypeName( STARTD_ADTYPE );
// Negotiator matches up private ad with public ad by
// using the following.
if( retVal ) {
pvtAd->CopyAttribute( ATTR_MY_ADDRESS, retVal );
pvtAd->CopyAttribute( ATTR_NAME, retVal );
}
// insert the private ad into its hashtable --- use the same
// hash key as the public ad
(void) updateClassAd (StartdPrivateAds, "StartdPvtAd ",
"StartdPvt", pvtAd, hk, hashString, insPvt,
from );
}
// create fake duplicates of this ad, each with a different name, if
// we are told to do so. this feature exists for developer
// scalability testing.
if ( repeatStartdAds > 0 && clientAdToRepeat ) {
ClassAd *fakeAd;
int n;
char newname[150],oldname[130];
oldname[0] = '\0';
clientAdToRepeat->LookupString("Name",oldname,sizeof(oldname));
for (n=0;n<repeatStartdAds;n++) {
fakeAd = new ClassAd(*clientAdToRepeat);
snprintf(newname,sizeof(newname),
"Name=\"fake%d-%s\"",n,oldname);
fakeAd->InsertOrUpdate(newname);
makeStartdAdHashKey (hk, fakeAd);
hashString.Build( hk );
if (! updateClassAd (StartdAds, "StartdAd ", "Start",
fakeAd, hk, hashString, insert, from ) )
{
// don't leak memory if there is some failure
delete fakeAd;
}
}
delete clientAdToRepeat;
clientAdToRepeat = NULL;
}
break;
case MERGE_STARTD_AD:
#if !defined(WANT_OLD_CLASSADS)
clientAd->AddTargetRefs( TargetJobAttrs );
#endif
if (!makeStartdAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=mergeClassAd (StartdAds, "StartdAd ", "Start",
clientAd, hk, hashString, insert, from );
break;
#ifdef HAVE_EXT_POSTGRESQL
case UPDATE_QUILL_AD:
if (!makeQuillAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=updateClassAd (QuillAds, "QuillAd ", "Quill",
clientAd, hk, hashString, insert, from );
break;
#endif /* HAVE_EXT_POSTGRESQL */
case UPDATE_SCHEDD_AD:
if (!makeScheddAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=updateClassAd (ScheddAds, "ScheddAd ", "Schedd",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_SUBMITTOR_AD:
// use the same hashkey function as a schedd ad
if (!makeScheddAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
// CRUFT: Before 7.3.2, submitter ads had a MyType of
// "Scheduler". The only way to tell the difference
// was that submitter ads didn't have ATTR_NUM_USERS.
// Coerce MyStype to "Submitter" for ads coming from
// these older schedds.
// Before 7.7.3, submitter ads for parallel universe
// jobs had a MyType of "Scheduler".
clientAd->SetMyTypeName( SUBMITTER_ADTYPE );
// since submittor ads always follow a schedd ad, and a master check is
// performed for schedd ads, we don't need a master check in here
hashString.Build( hk );
retVal=updateClassAd (SubmittorAds, "SubmittorAd ", "Submittor",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_LICENSE_AD:
// use the same hashkey function as a schedd ad
if (!makeLicenseAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
// since submittor ads always follow a schedd ad, and a master check is
// performed for schedd ads, we don't need a master check in here
hashString.Build( hk );
retVal=updateClassAd (LicenseAds, "LicenseAd ", "License",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_MASTER_AD:
if (!makeMasterAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=updateClassAd (MasterAds, "MasterAd ", "Master",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_CKPT_SRVR_AD:
if (!makeCkptSrvrAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=updateClassAd (CkptServerAds, "CkptSrvrAd ", "CkptSrvr",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_COLLECTOR_AD:
if (!makeCollectorAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=updateClassAd (CollectorAds, "CollectorAd ", "Collector",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_STORAGE_AD:
if (!makeStorageAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=updateClassAd (StorageAds, "StorageAd ", "Storage",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_NEGOTIATOR_AD:
if (!makeNegotiatorAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
// first, purge all the existing negotiator ads, since we
// want to enforce that *ONLY* 1 negotiator is in the
// collector any given time.
purgeHashTable( NegotiatorAds );
retVal=updateClassAd (NegotiatorAds, "NegotiatorAd ", "Negotiator",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_HAD_AD:
if (!makeHadAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=updateClassAd (HadAds, "HadAd ", "HAD",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_GRID_AD:
if (!makeGridAdHashKey(hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=updateClassAd (GridAds, "GridAd ", "Grid",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_AD_GENERIC:
{
const char *type_str = clientAd->GetMyTypeName();
if (type_str == NULL) {
dprintf(D_ALWAYS, "collect: UPDATE_AD_GENERIC: ad has no type\n");
insert = -3;
retVal = 0;
break;
}
MyString type(type_str);
CollectorHashTable *cht = findOrCreateTable(type);
if (cht == NULL) {
dprintf(D_ALWAYS, "collect: findOrCreateTable failed\n");
insert = -3;
retVal = 0;
break;
}
if (!makeGenericAdHashKey (hk, clientAd))
{
dprintf(D_ALWAYS, "Could not make haskey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build(hk);
retVal = updateClassAd(*cht, type_str, type_str, clientAd,
hk, hashString, insert, from);
break;
}
case UPDATE_XFER_SERVICE_AD:
if (!makeXferServiceAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
retVal=updateClassAd (XferServiceAds, "XferServiceAd ",
"XferService",
clientAd, hk, hashString, insert, from );
break;
case UPDATE_LEASE_MANAGER_AD:
if (!makeLeaseManagerAdHashKey (hk, clientAd))
{
dprintf (D_ALWAYS, "Could not make hashkey --- ignoring ad\n");
insert = -3;
retVal = 0;
break;
}
hashString.Build( hk );
// first, purge all the existing LeaseManager ads, since we
// want to enforce that *ONLY* 1 manager is in the
// collector any given time.
purgeHashTable( LeaseManagerAds );
retVal=updateClassAd (LeaseManagerAds, "LeaseManagerAd ",
"LeaseManager",
clientAd, hk, hashString, insert, from );
break;
case QUERY_STARTD_ADS:
case QUERY_SCHEDD_ADS:
case QUERY_MASTER_ADS:
case QUERY_GATEWAY_ADS:
case QUERY_SUBMITTOR_ADS:
case QUERY_CKPT_SRVR_ADS:
case QUERY_STARTD_PVT_ADS:
case QUERY_COLLECTOR_ADS:
case QUERY_NEGOTIATOR_ADS:
case QUERY_HAD_ADS:
case QUERY_XFER_SERVICE_ADS:
case QUERY_LEASE_MANAGER_ADS:
case QUERY_GENERIC_ADS:
case INVALIDATE_STARTD_ADS:
case INVALIDATE_SCHEDD_ADS:
case INVALIDATE_MASTER_ADS:
case INVALIDATE_GATEWAY_ADS:
case INVALIDATE_CKPT_SRVR_ADS:
case INVALIDATE_SUBMITTOR_ADS:
case INVALIDATE_COLLECTOR_ADS:
case INVALIDATE_NEGOTIATOR_ADS:
case INVALIDATE_HAD_ADS:
case INVALIDATE_XFER_SERVICE_ADS:
case INVALIDATE_LEASE_MANAGER_ADS:
case INVALIDATE_ADS_GENERIC:
// these are not implemented in the engine, but we allow another
// daemon to detect that these commands have been given
insert = -2;
retVal = 0;
break;
default:
dprintf (D_ALWAYS, "Received illegal command: %d\n", command);
insert = -1;
retVal = 0;
}
// return the updated ad
return retVal;
}
void
VMRegister::requestHostClassAds(void)
{
// find host startd daemon
if( !m_vm_host_daemon )
m_vm_host_daemon = vmapi_findDaemon( m_vm_host_name, DT_STARTD);
if( !m_vm_host_daemon ) {
dprintf( D_FULLDEBUG, "Can't find host(%s) Startd daemon\n", m_vm_host_name );
return;
}
ClassAd query_ad;
query_ad.SetMyTypeName(QUERY_ADTYPE);
query_ad.SetTargetTypeName(STARTD_ADTYPE);
query_ad.Assign(ATTR_REQUIREMENTS, true);
char *addr = m_vm_host_daemon->addr();
Daemon hstartd(DT_STARTD, addr);
ReliSock ssock;
ssock.timeout( VM_SOCKET_TIMEOUT );
ssock.encode();
if( !ssock.connect(addr) ) {
dprintf( D_FULLDEBUG, "Failed to connect to host startd(%s)\n to get host classAd", addr);
return;
}
if(!hstartd.startCommand( QUERY_STARTD_ADS, &ssock )) {
dprintf( D_FULLDEBUG, "Failed to send QUERY_STARTD_ADS command to host startd(%s)\n", addr);
return;
}
if( !query_ad.put(ssock) ) {
dprintf(D_FULLDEBUG, "Failed to send query Ad to host startd(%s)\n", addr);
}
if( !ssock.end_of_message() ) {
dprintf(D_FULLDEBUG, "Failed to send query EOM to host startd(%s)\n", addr);
}
// Read host classAds
ssock.timeout( VM_SOCKET_TIMEOUT );
ssock.decode();
int more = 1, num_ads = 0;
ClassAdList adList;
ClassAd *ad;
while (more) {
if( !ssock.code(more) ) {
ssock.end_of_message();
return;
}
if(more) {
ad = new ClassAd;
if( !ad->initFromStream(ssock) ) {
ssock.end_of_message();
delete ad;
return;
}
adList.Insert(ad);
num_ads++;
}
}
ssock.end_of_message();
dprintf(D_FULLDEBUG, "Got %d classAds from host\n", num_ads);
// Although we can get more than one classAd from host machine,
// we use only the first one classAd
adList.Rewind();
ad = adList.Next();
#if !defined(WANT_OLD_CLASSADS)
ad->AddTargetRefs( TargetJobAttrs );
#endif
// Get each Attribute from the classAd
// added "HOST_" in front of each Attribute name
const char *name;
ExprTree *expr;
ad->ResetExpr();
while( ad->NextExpr(name, expr) ) {
MyString attr;
attr += "HOST_";
attr += name;
// Insert or Update an attribute to host_classAd in a VMRegister object
ExprTree * pTree = expr->Copy();
host_classad->Insert(attr.Value(), pTree, true);
}
}