void
CpuAttributes::display( amask_t how_much )
{
if( IS_UPDATE(how_much) ) {
dprintf( D_KEYBOARD,
"Idle time: %s %-8d %s %d\n",
"Keyboard:", (int)c_idle,
"Console:", (int)c_console_idle );
dprintf( D_LOAD,
"%s %.2f %s %.2f %s %.2f\n",
"SystemLoad:", c_condor_load + c_owner_load,
"CondorLoad:", c_condor_load,
"OwnerLoad:", c_owner_load );
} else {
if( IsDebugLevel( D_LOAD ) ) {
dprintf( D_FULLDEBUG,
"%s %.2f %s %.2f %s %.2f\n",
"SystemLoad:", c_condor_load + c_owner_load,
"CondorLoad:", c_condor_load,
"OwnerLoad:", c_owner_load );
}
if( IsDebugLevel( D_KEYBOARD ) ) {
dprintf( D_FULLDEBUG,
"Idle time: %s %-8d %s %d\n",
"Keyboard:", (int)c_idle,
"Console:", (int)c_console_idle );
}
}
}
void
JobInfoCommunicator::checkForStarterDebugging( void )
{
if( ! job_ad ) {
EXCEPT( "checkForStarterDebugging() called with no job ad!" );
}
// For debugging, see if there's a special attribute in the
// job ad that sends us into an infinite loop, waiting for
// someone to attach with a debugger
int starter_should_wait = 0;
job_ad->LookupInteger( ATTR_STARTER_WAIT_FOR_DEBUG,
starter_should_wait );
if( starter_should_wait ) {
dprintf( D_ALWAYS, "Job requested starter should wait for "
"debugger with %s=%d, going into infinite loop\n",
ATTR_STARTER_WAIT_FOR_DEBUG, starter_should_wait );
while( 1 ) {
if ( !starter_should_wait ) {
break;
}
}
}
// Also, if the starter has D_JOB turned on, we want to dump
// out the job ad to the log file...
if( IsDebugLevel( D_JOB ) ) {
dprintf( D_JOB, "*** Job ClassAd ***\n" );
job_ad->dPrint( D_JOB );
dprintf( D_JOB, "--- End of ClassAd ---\n" );
}
}
void
Selector::delete_fd( int fd, IO_FUNC interest )
{
#if !defined(WIN32)
if ( fd < 0 || fd >= fd_select_size() ) {
EXCEPT( "Selector::delete_fd(): fd %d outside valid range 0-%d",
fd, _fd_select_size-1 );
}
#endif
m_single_shot = SINGLE_SHOT_SKIP;
if (IsDebugLevel(D_DAEMONCORE)) {
dprintf(D_DAEMONCORE | D_VERBOSE, "selector %p deleting fd %d\n", this, fd);
}
switch( interest ) {
case IO_READ:
FD_CLR( fd, save_read_fds );
break;
case IO_WRITE:
FD_CLR( fd, save_write_fds );
break;
case IO_EXCEPT:
FD_CLR( fd, save_except_fds );
break;
}
}
void
Selector::reset()
{
_select_retval = -2;
_select_errno = 0;
state = VIRGIN;
timeout_wanted = false;
timeout.tv_sec = timeout.tv_usec = 0;
max_fd = -1;
#if defined(WIN32)
FD_ZERO( save_read_fds );
FD_ZERO( save_write_fds );
FD_ZERO( save_except_fds );
#else
memset( save_read_fds, 0, fd_set_size * sizeof(fd_set) );
memset( save_write_fds, 0, fd_set_size * sizeof(fd_set) );
memset( save_except_fds, 0, fd_set_size * sizeof(fd_set) );
#endif
#ifdef SELECTOR_USE_POLL
m_single_shot = SINGLE_SHOT_VIRGIN;
#else
m_single_shot = SINGLE_SHOT_SKIP;
#endif
memset(&m_poll, '\0', sizeof(m_poll));
if (IsDebugLevel(D_DAEMONCORE)) {
dprintf(D_DAEMONCORE | D_VERBOSE, "selector %p resetting\n", this);
}
}
void
Selector::add_fd( int fd, IO_FUNC interest )
{
// update max_fd (the highest valid index in fd_set's array) and also
// make sure we're not overflowing our fd_set
// On Windows, we have to check the individual fd_set to see if it's
// full.
if( fd > max_fd ) {
max_fd = fd;
}
#if !defined(WIN32)
if ( fd < 0 || fd >= fd_select_size() ) {
EXCEPT( "Selector::add_fd(): fd %d outside valid range 0-%d",
fd, _fd_select_size-1 );
}
#endif
if(IsDebugLevel(D_DAEMONCORE)) {
char *fd_description = describe_fd(fd);
dprintf(D_FULLDEBUG, "selector %p adding fd %d (%s)\n",
this, fd, fd_description);
free(fd_description);
}
switch( interest ) {
case IO_READ:
#if defined(WIN32)
if ( save_read_fds->fd_count >= fd_select_size() ) {
EXCEPT( "Selector::add_fd(): read fd_set is full" );
}
#endif
FD_SET( fd, save_read_fds );
break;
case IO_WRITE:
#if defined(WIN32)
if ( save_write_fds->fd_count >= fd_select_size() ) {
EXCEPT( "Selector::add_fd(): write fd_set is full" );
}
#endif
FD_SET( fd, save_write_fds );
break;
case IO_EXCEPT:
#if defined(WIN32)
if ( save_except_fds->fd_count >= fd_select_size() ) {
EXCEPT( "Selector::add_fd(): except fd_set is full" );
}
#endif
FD_SET( fd, save_except_fds );
break;
}
}
bool
DCStartd::_suspendClaim( )
{
setCmdStr( "suspendClaim" );
if( ! checkClaimId() ) {
return false;
}
if( ! checkAddr() ) {
return false;
}
// if this claim is associated with a security session
ClaimIdParser cidp(claim_id);
char const *sec_session = cidp.secSessionId();
if (IsDebugLevel(D_COMMAND)) {
int cmd = SUSPEND_CLAIM;
dprintf (D_COMMAND, "DCStartd::_suspendClaim(%s,...) making connection to %s\n", getCommandStringSafe(cmd), _addr ? _addr : "NULL");
}
bool result;
ReliSock reli_sock;
reli_sock.timeout(20); // years of research... :)
if( ! reli_sock.connect(_addr) ) {
std::string err = "DCStartd::_suspendClaim: ";
err += "Failed to connect to startd (";
err += _addr ? _addr : "NULL";
err += ')';
newError( CA_CONNECT_FAILED, err.c_str() );
return false;
}
int cmd = SUSPEND_CLAIM;
result = startCommand( cmd, (Sock*)&reli_sock, 20, NULL, NULL, false, sec_session );
if( ! result ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::_suspendClaim: Failed to send command " );
return false;
}
// Now, send the ClaimId
if( ! reli_sock.put_secret(claim_id) ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::_suspendClaim: Failed to send ClaimId to the startd" );
return false;
}
if( ! reli_sock.end_of_message() ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::_suspendClaim: Failed to send EOM to the startd" );
return false;
}
return true;
}
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 (IsDebugLevel(D_COMMAND)) {
dprintf (D_COMMAND, "DCStarter::createJobOwnerSecSession(%s,...) making connection to %s\n",
getCommandStringSafe(CREATE_JOB_OWNER_SEC_SESSION), _addr ? _addr : "NULL");
}
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( !putClassAd(&sock, input) || !sock.end_of_message() ) {
error_msg = "Failed to compose CREATE_JOB_OWNER_SEC_SESSION to starter";
return false;
}
sock.decode();
ClassAd reply;
if( !getClassAd(&sock, reply) || !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
DCStartd::checkpointJob( const char* name_ckpt )
{
dprintf( D_FULLDEBUG, "Entering DCStartd::checkpointJob(%s)\n",
name_ckpt );
setCmdStr( "checkpointJob" );
if (IsDebugLevel(D_COMMAND)) {
int cmd = PCKPT_JOB;
dprintf (D_COMMAND, "DCStartd::checkpointJob(%s,...) making connection to %s\n", getCommandStringSafe(cmd), _addr ? _addr : "NULL");
}
bool result;
ReliSock reli_sock;
reli_sock.timeout(20); // years of research... :)
if( ! reli_sock.connect(_addr) ) {
std::string err = "DCStartd::checkpointJob: ";
err += "Failed to connect to startd (";
err += _addr ? _addr : "NULL";
err += ')';
newError( CA_CONNECT_FAILED, err.c_str() );
return false;
}
int cmd = PCKPT_JOB;
result = startCommand( cmd, (Sock*)&reli_sock );
if( ! result ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::checkpointJob: Failed to send command PCKPT_JOB to the startd" );
return false;
}
// Now, send the name
if( ! reli_sock.put(name_ckpt) ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::checkpointJob: Failed to send Name to the startd" );
return false;
}
if( ! reli_sock.end_of_message() ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::checkpointJob: Failed to send EOM to the startd" );
return false;
}
// we're done
dprintf( D_FULLDEBUG, "DCStartd::checkpointJob: "
"successfully sent command\n" );
return true;
}
bool
DCStartd::vacateClaim( const char* name_vacate )
{
setCmdStr( "vacateClaim" );
if (IsDebugLevel(D_COMMAND)) {
int cmd = VACATE_CLAIM;
dprintf (D_COMMAND, "DCStartd::vacateClaim(%s,...) making connection to %s\n", getCommandStringSafe(cmd), _addr ? _addr : "NULL");
}
bool result;
ReliSock reli_sock;
reli_sock.timeout(20); // years of research... :)
if( ! reli_sock.connect(_addr) ) {
std::string err = "DCStartd::vacateClaim: ";
err += "Failed to connect to startd (";
err += _addr ? _addr : "NULL";
err += ')';
newError( CA_CONNECT_FAILED, err.c_str() );
return false;
}
int cmd = VACATE_CLAIM;
result = startCommand( cmd, (Sock*)&reli_sock );
if( ! result ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::vacateClaim: Failed to send command PCKPT_JOB to the startd" );
return false;
}
if( ! reli_sock.put(name_vacate) ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::vacateClaim: Failed to send Name to the startd" );
return false;
}
if( ! reli_sock.end_of_message() ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::vacateClaim: Failed to send EOM to the startd" );
return false;
}
return true;
}
int
IpVerify::add_hash_entry(const struct in6_addr & sin6_addr, const char * user, perm_mask_t new_mask)
{
UserPerm_t * perm = NULL;
perm_mask_t old_mask = 0; // must init old_mask to zero!!!
MyString user_key = user;
// assert(PermHashTable);
if ( PermHashTable->lookup(sin6_addr, perm) != -1 ) {
// found an existing entry.
if (has_user(perm, user, old_mask)) {
// remove it because we are going to edit the mask below
// and re-insert it.
perm->remove(user_key);
}
}
else {
perm = new UserPerm_t(42, compute_host_hash);
if (PermHashTable->insert(sin6_addr, perm) != 0) {
delete perm;
return FALSE;
}
}
perm->insert(user_key, old_mask | new_mask);
if( IsFulldebug(D_FULLDEBUG) || IsDebugLevel(D_SECURITY) ) {
MyString auth_str;
AuthEntryToString(sin6_addr,user,new_mask, auth_str);
dprintf(D_FULLDEBUG|D_SECURITY,
"Adding to resolved authorization table: %s\n",
auth_str.Value());
}
return TRUE;
}
void
Selector::reset()
{
_select_retval = -2;
_select_errno = 0;
state = VIRGIN;
timeout_wanted = FALSE;
timeout.tv_sec = timeout.tv_usec = 0;
max_fd = -1;
#if defined(WIN32)
FD_ZERO( save_read_fds );
FD_ZERO( save_write_fds );
FD_ZERO( save_except_fds );
#else
memset( save_read_fds, 0, fd_set_size * sizeof(fd_set) );
memset( save_write_fds, 0, fd_set_size * sizeof(fd_set) );
memset( save_except_fds, 0, fd_set_size * sizeof(fd_set) );
#endif
if (IsDebugLevel(D_DAEMONCORE)) {
dprintf(D_FULLDEBUG, "selector %p resetting\n", this);
}
}
// process ads from the collector, handing each to the callback
// callback will return 'false' if it took ownership of the ad.
QueryResult CondorQuery::
processAds (bool (*callback)(void*, ClassAd *), void* pv, const char * poolName, CondorError* errstack /*= NULL*/)
{
Sock* sock;
QueryResult result;
ClassAd queryAd(extraAttrs);
if ( !poolName ) {
return Q_NO_COLLECTOR_HOST;
}
// contact collector
Daemon my_collector( DT_COLLECTOR, poolName, NULL );
if( !my_collector.locate() ) {
// We were passed a bogus poolName, abort gracefully
return Q_NO_COLLECTOR_HOST;
}
// make the query ad
result = getQueryAd (queryAd);
if (result != Q_OK) return result;
if (IsDebugLevel(D_HOSTNAME)) {
dprintf( D_HOSTNAME, "Querying collector %s (%s) with classad:\n",
my_collector.addr(), my_collector.fullHostname() );
dPrintAd( D_HOSTNAME, queryAd );
dprintf( D_HOSTNAME, " --- End of Query ClassAd ---\n" );
}
int mytimeout = param_integer ("QUERY_TIMEOUT",60);
if (!(sock = my_collector.startCommand(command, Stream::reli_sock, mytimeout, errstack)) ||
!putClassAd (sock, queryAd) || !sock->end_of_message()) {
if (sock) {
delete sock;
}
return Q_COMMUNICATION_ERROR;
}
// get result
sock->decode ();
int more = 1;
while (more)
{
if (!sock->code (more)) {
sock->end_of_message();
delete sock;
return Q_COMMUNICATION_ERROR;
}
if (more) {
ClassAd * ad = new ClassAd;
if( !getClassAd(sock, *ad) ) {
sock->end_of_message();
delete ad;
delete sock;
return Q_COMMUNICATION_ERROR;
}
if (callback(pv, ad)) {
delete ad;
}
}
}
sock->end_of_message();
// finalize
sock->close();
delete sock;
return (Q_OK);
}
bool
DCStartd::deactivateClaim( bool graceful, bool *claim_is_closing )
{
dprintf( D_FULLDEBUG, "Entering DCStartd::deactivateClaim(%s)\n",
graceful ? "graceful" : "forceful" );
if( claim_is_closing ) {
*claim_is_closing = false;
}
setCmdStr( "deactivateClaim" );
if( ! checkClaimId() ) {
return false;
}
if( ! checkAddr() ) {
return false;
}
// if this claim is associated with a security session
ClaimIdParser cidp(claim_id);
char const *sec_session = cidp.secSessionId();
if (IsDebugLevel(D_COMMAND)) {
int cmd = graceful ? DEACTIVATE_CLAIM : DEACTIVATE_CLAIM_FORCIBLY;
dprintf (D_COMMAND, "DCStartd::deactivateClaim(%s,...) making connection to %s\n", getCommandStringSafe(cmd), _addr ? _addr : "NULL");
}
bool result;
ReliSock reli_sock;
reli_sock.timeout(20); // years of research... :)
if( ! reli_sock.connect(_addr) ) {
std::string err = "DCStartd::deactivateClaim: ";
err += "Failed to connect to startd (";
err += _addr ? _addr : "NULL";
err += ')';
newError( CA_CONNECT_FAILED, err.c_str() );
return false;
}
int cmd;
if( graceful ) {
cmd = DEACTIVATE_CLAIM;
} else {
cmd = DEACTIVATE_CLAIM_FORCIBLY;
}
result = startCommand( cmd, (Sock*)&reli_sock, 20, NULL, NULL, false, sec_session );
if( ! result ) {
std::string err = "DCStartd::deactivateClaim: ";
err += "Failed to send command ";
if( graceful ) {
err += "DEACTIVATE_CLAIM";
} else {
err += "DEACTIVATE_CLAIM_FORCIBLY";
}
err += " to the startd";
newError( CA_COMMUNICATION_ERROR, err.c_str() );
return false;
}
// Now, send the ClaimId
if( ! reli_sock.put_secret(claim_id) ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::deactivateClaim: Failed to send ClaimId to the startd" );
return false;
}
if( ! reli_sock.end_of_message() ) {
newError( CA_COMMUNICATION_ERROR,
"DCStartd::deactivateClaim: Failed to send EOM to the startd" );
return false;
}
reli_sock.decode();
ClassAd response_ad;
if( !getClassAd(&reli_sock, response_ad) || !reli_sock.end_of_message() ) {
dprintf( D_FULLDEBUG, "DCStartd::deactivateClaim: failed to read response ad.\n");
// The response ad is not critical and is expected to be missing
// if the startd is from before 7.0.5.
}
else {
bool start = true;
response_ad.LookupBool(ATTR_START,start);
if( claim_is_closing ) {
*claim_is_closing = !start;
}
}
// we're done
dprintf( D_FULLDEBUG, "DCStartd::deactivateClaim: "
"successfully sent command\n" );
return true;
}
// fetch all ads from the collector that satisfy the constraints
QueryResult CondorQuery::
fetchAds (ClassAdList &adList, const char *poolName, CondorError* errstack)
{
Sock* sock;
int more;
QueryResult result;
ClassAd queryAd(extraAttrs), *ad;
if ( !poolName ) {
return Q_NO_COLLECTOR_HOST;
}
// contact collector
Daemon my_collector( DT_COLLECTOR, poolName, NULL );
if( !my_collector.locate() ) {
// We were passed a bogus poolName, abort gracefully
return Q_NO_COLLECTOR_HOST;
}
// make the query ad
result = getQueryAd (queryAd);
if (result != Q_OK) return result;
if( IsDebugLevel( D_HOSTNAME ) ) {
dprintf( D_HOSTNAME, "Querying collector %s (%s) with classad:\n",
my_collector.addr(), my_collector.fullHostname() );
queryAd.dPrint( D_HOSTNAME );
dprintf( D_HOSTNAME, " --- End of Query ClassAd ---\n" );
}
int mytimeout = param_integer ("QUERY_TIMEOUT",60);
if (!(sock = my_collector.startCommand(command, Stream::reli_sock, mytimeout, errstack)) ||
!queryAd.put (*sock) || !sock->end_of_message()) {
if (sock) {
delete sock;
}
return Q_COMMUNICATION_ERROR;
}
// get result
sock->decode ();
more = 1;
while (more)
{
if (!sock->code (more)) {
sock->end_of_message();
delete sock;
return Q_COMMUNICATION_ERROR;
}
if (more) {
ad = new ClassAd;
if( !ad->initFromStream(*sock) ) {
sock->end_of_message();
delete ad;
delete sock;
return Q_COMMUNICATION_ERROR;
}
adList.Insert (ad);
}
}
sock->end_of_message();
// finalize
sock->close();
delete sock;
return (Q_OK);
}
bool
DCTransferQueue::RequestTransferQueueSlot(bool downloading,filesize_t sandbox_size,char const *fname,char const *jobid,char const *queue_user,int timeout,MyString &error_desc)
{
ASSERT(fname);
ASSERT(jobid);
if( GoAheadAlways( downloading ) ) {
m_xfer_downloading = downloading;
m_xfer_fname = fname;
m_xfer_jobid = jobid;
return true;
}
CheckTransferQueueSlot();
if( m_xfer_queue_sock ) {
// A request has already been made.
// Currently, this is a no-op, because any upload/download slot
// is as good as any other. In the future, there may be
// different queues for different paths.
ASSERT( m_xfer_downloading == downloading );
m_xfer_fname = fname;
m_xfer_jobid = jobid;
return true;
}
time_t started = time(NULL);
CondorError errstack;
// Our caller has to finish this operation in the specified
// amount of time or risk not responding to the file transfer
// peer in time, so ignore the timeout multiplier and set the
// timeout exactly as specified.
m_xfer_queue_sock = reliSock( timeout, 0, &errstack, false, true );
if( !m_xfer_queue_sock ) {
formatstr(m_xfer_rejected_reason,
"Failed to connect to transfer queue manager for job %s (%s): %s.",
jobid, fname, errstack.getFullText().c_str() );
error_desc = m_xfer_rejected_reason;
dprintf(D_ALWAYS,"%s\n",m_xfer_rejected_reason.c_str());
return false;
}
if( timeout ) {
timeout -= time(NULL)-started;
if( timeout <= 0 ) {
timeout = 1;
}
}
if (IsDebugLevel(D_COMMAND)) {
int cmd = TRANSFER_QUEUE_REQUEST;
dprintf (D_COMMAND, "DCTransferQueue::RequestTransferQueueSlot(%s,...) making connection to %s\n", getCommandStringSafe(cmd), _addr ? _addr : "NULL");
}
bool connected = startCommand(
TRANSFER_QUEUE_REQUEST, m_xfer_queue_sock, timeout, &errstack );
if( !connected )
{
delete m_xfer_queue_sock;
m_xfer_queue_sock = NULL;
formatstr(m_xfer_rejected_reason,
"Failed to initiate transfer queue request for job %s (%s): %s.",
jobid, fname, errstack.getFullText().c_str() );
error_desc = m_xfer_rejected_reason;
dprintf(D_ALWAYS,"%s\n",m_xfer_rejected_reason.c_str());
return false;
}
m_xfer_downloading = downloading;
m_xfer_fname = fname;
m_xfer_jobid = jobid;
ClassAd msg;
msg.Assign(ATTR_DOWNLOADING,downloading);
msg.Assign(ATTR_FILE_NAME,fname);
msg.Assign(ATTR_JOB_ID,jobid);
msg.Assign(ATTR_USER,queue_user);
msg.Assign(ATTR_SANDBOX_SIZE,sandbox_size);
m_xfer_queue_sock->encode();
if( !putClassAd(m_xfer_queue_sock, msg) || !m_xfer_queue_sock->end_of_message() )
{
formatstr(m_xfer_rejected_reason,
"Failed to write transfer request to %s for job %s "
"(initial file %s).",
m_xfer_queue_sock->peer_description(),
m_xfer_jobid.c_str(), m_xfer_fname.c_str());
error_desc = m_xfer_rejected_reason;
dprintf(D_ALWAYS,"%s\n",m_xfer_rejected_reason.c_str());
return false;
}
m_xfer_queue_sock->decode();
// Request has been initiated. Now sender should call
// PollForTransferQueueSlot() to get response.
m_xfer_queue_pending = true;
return true;
}
void
Selector::add_fd( int fd, IO_FUNC interest )
{
// update max_fd (the highest valid index in fd_set's array) and also
// make sure we're not overflowing our fd_set
// On Windows, we have to check the individual fd_set to see if it's
// full.
if( fd > max_fd ) {
max_fd = fd;
}
#if !defined(WIN32)
if ( fd < 0 || fd >= fd_select_size() ) {
EXCEPT( "Selector::add_fd(): fd %d outside valid range 0-%d",
fd, _fd_select_size-1 );
}
#endif
if(IsDebugLevel(D_DAEMONCORE)) {
char *fd_description = describe_fd(fd);
dprintf(D_DAEMONCORE | D_VERBOSE, "selector %p adding fd %d (%s)\n",
this, fd, fd_description);
free(fd_description);
}
bool new_fd = false;
if ((m_single_shot == SINGLE_SHOT_OK) && (m_poll.fd != fd)) {
new_fd = true;
}
m_poll.fd = fd;
switch( interest ) {
case IO_READ:
#if defined(WIN32)
if ( save_read_fds->fd_count >= fd_select_size() ) {
EXCEPT( "Selector::add_fd(): read fd_set is full" );
}
#endif
m_poll.events |= POLLIN;
FD_SET( fd, save_read_fds );
break;
case IO_WRITE:
#if defined(WIN32)
if ( save_write_fds->fd_count >= fd_select_size() ) {
EXCEPT( "Selector::add_fd(): write fd_set is full" );
}
#endif
m_poll.events |= POLLOUT;
FD_SET( fd, save_write_fds );
break;
case IO_EXCEPT:
#if defined(WIN32)
if ( save_except_fds->fd_count >= fd_select_size() ) {
EXCEPT( "Selector::add_fd(): except fd_set is full" );
}
#endif
m_poll.events |= POLLERR;
FD_SET( fd, save_except_fds );
break;
}
if ((m_single_shot == SINGLE_SHOT_VIRGIN) || ((m_single_shot == SINGLE_SHOT_OK) && (new_fd == false)))
{
m_single_shot = SINGLE_SHOT_OK;
}
else
{
m_single_shot = SINGLE_SHOT_SKIP;
}
}
void DCMessenger::startCommand( classy_counted_ptr<DCMsg> msg )
{
MyString error;
msg->setMessenger( this );
if( msg->deliveryStatus() == DCMsg::DELIVERY_CANCELED ) {
msg->callMessageSendFailed( this );
return;
}
time_t deadline = msg->getDeadline();
if( deadline && deadline < time(NULL) ) {
msg->addError(CEDAR_ERR_DEADLINE_EXPIRED,
"deadline for delivery of this message expired");
msg->callMessageSendFailed( this );
return;
}
// For a UDP message, we may need to register two sockets, one for
// the SafeSock and another for a ReliSock to establish the
// security session.
Stream::stream_type st = msg->getStreamType();
if( daemonCore->TooManyRegisteredSockets(-1,&error,st==Stream::safe_sock?2:1) ) {
// Try again in a sec
// Eventually, it would be better to queue this centrally
// (i.e. in DaemonCore) rather than having an independent
// timer for each case. Then it would be possible to control
// priority of different messages etc.
dprintf(D_FULLDEBUG, "Delaying delivery of %s to %s, because %s\n",
msg->name(),peerDescription(),error.Value());
startCommandAfterDelay( 1, msg );
return;
}
// Currently, there may be only one pending operation per messenger.
ASSERT(!m_callback_msg.get());
ASSERT(!m_callback_sock);
ASSERT(m_pending_operation == NOTHING_PENDING);
m_pending_operation = START_COMMAND_PENDING;
m_callback_msg = msg;
m_callback_sock = m_sock.get();
if( !m_callback_sock ) {
if (IsDebugLevel(D_COMMAND)) {
const char * addr = m_daemon->addr();
const int cmd = msg->m_cmd;
dprintf (D_COMMAND, "DCMessenger::startCommand(%s,...) making non-blocking connection to %s\n", getCommandStringSafe(cmd), addr ? addr : "NULL");
}
const bool nonblocking = true;
m_callback_sock = m_daemon->makeConnectedSocket(st,msg->getTimeout(),msg->getDeadline(),&msg->m_errstack,nonblocking);
if( !m_callback_sock ) {
msg->callMessageSendFailed( this );
return;
}
}
incRefCount();
m_daemon->startCommand_nonblocking (
msg->m_cmd,
m_callback_sock,
msg->getTimeout(),
&msg->m_errstack,
&DCMessenger::connectCallback,
this,
msg->name(),
msg->getRawProtocol(),
msg->getSecSessionId());
}
bool
DCStarter::peek(bool transfer_stdout, ssize_t &stdout_offset, bool transfer_stderr, ssize_t &stderr_offset, const std::vector<std::string> &filenames, std::vector<ssize_t> &offsets, size_t max_bytes, bool &retry_sensible, PeekGetFD &next, std::string &error_msg, unsigned timeout, const std::string &sec_session_id, DCTransferQueue *xfer_q)
{
compat_classad::ClassAd ad;
ad.InsertAttr(ATTR_JOB_OUTPUT, transfer_stdout);
ad.InsertAttr("OutOffset", stdout_offset);
ad.InsertAttr(ATTR_JOB_ERROR, transfer_stderr);
ad.InsertAttr("ErrOffset", stderr_offset);
ad.InsertAttr(ATTR_VERSION, CondorVersion());
size_t total_files = 0;
total_files += transfer_stdout ? 1 : 0;
total_files += transfer_stderr ? 1 : 0;
if (filenames.size())
{
total_files += filenames.size();
std::vector<classad::ExprTree *> filelist; filelist.reserve(filenames.size());
std::vector<classad::ExprTree *> offsetlist; offsetlist.reserve(filenames.size());
std::vector<ssize_t>::const_iterator it2 = offsets.begin();
for (std::vector<std::string>::const_iterator it = filenames.begin();
it != filenames.end() && it2 != offsets.end();
it++, it2++)
{
classad::Value value;
value.SetStringValue(*it);
filelist.push_back(classad::Literal::MakeLiteral(value));
value.SetIntegerValue(*it2);
offsetlist.push_back(classad::Literal::MakeLiteral(value));
}
classad::ExprTree *list(classad::ExprList::MakeExprList(filelist));
ad.Insert("TransferFiles", list);
list = classad::ExprList::MakeExprList(offsetlist);
ad.Insert("TransferOffsets", list);
}
ad.InsertAttr(ATTR_MAX_TRANSFER_BYTES, static_cast<long long>(max_bytes));
ReliSock sock;
if (IsDebugLevel(D_COMMAND)) {
dprintf (D_COMMAND, "DCStarter::peek(%s,...) making connection to %s\n",
getCommandStringSafe(STARTER_PEEK), _addr ? _addr : "NULL");
}
if( !connectSock(&sock, timeout, NULL) ) {
error_msg = "Failed to connect to starter";
return false;
}
if( !startCommand(STARTER_PEEK, &sock, timeout, NULL, NULL, false, sec_session_id.c_str()) ) {
error_msg = "Failed to send START_PEEK to starter";
return false;
}
sock.encode();
if (!putClassAd(&sock, ad) || !sock.end_of_message()) {
error_msg = "Failed to send request to starter";
return false;
}
compat_classad::ClassAd response;
sock.decode();
if (!getClassAd(&sock, response) || !sock.end_of_message())
{
error_msg = "Failed to read response for peeking at logs.";
return false;
}
dPrintAd(D_FULLDEBUG, response);
bool success = false;
if (!response.EvaluateAttrBool(ATTR_RESULT, success) || !success)
{
response.EvaluateAttrBool(ATTR_RETRY, retry_sensible);
error_msg = "Remote operation failed.";
response.EvaluateAttrString(ATTR_ERROR_STRING, error_msg);
return false;
}
classad::Value valueX;
classad_shared_ptr<classad::ExprList> list;
if (!response.EvaluateAttr("TransferFiles", valueX) || !valueX.IsSListValue(list))
{
error_msg = "Unable to evaluate starter response";
return false;
}
classad_shared_ptr<classad::ExprList> offlist;
if (!response.EvaluateAttr("TransferOffsets", valueX) || !valueX.IsSListValue(offlist))
{
error_msg = "Unable to evaluate starter response (missing offsets)";
return false;
}
size_t remaining = max_bytes;
size_t file_count = 0;
classad::ExprList::const_iterator it2 = offlist->begin();
for (classad::ExprList::const_iterator it = list->begin();
it != list->end() && it2 != offlist->end();
it++, it2++)
{
classad::Value value;
(*it2)->Evaluate(value);
off_t off = -1;
value.IsIntegerValue(off);
(*it)->Evaluate(value);
std::string filename;
int64_t xfer_fd = -1;
if (!value.IsStringValue(filename) && value.IsIntegerValue(xfer_fd))
{
if (xfer_fd == 0) filename = "_condor_stdout";
if (xfer_fd == 1) filename = "_condor_stderr";
}
int fd = next.getNextFD(filename);
filesize_t size = -1;
int retval;
if ((retval = sock.get_file(&size, fd, false, false, remaining, xfer_q)) && (retval != GET_FILE_MAX_BYTES_EXCEEDED))
{
error_msg = "Internal error when transferring file " + filename;
}
else if (size >= 0)
{
remaining -= max_bytes;
file_count++;
off += size;
}
else
{
error_msg = "Failed to transfer file " + filename;
}
if (xfer_fd == 0)
{
stdout_offset = off;
//dprintf(D_FULLDEBUG, "New stdout offset: %ld\n", stdout_offset);
}
else if (xfer_fd == 1)
{
stderr_offset = off;
}
else
{
std::vector<ssize_t>::iterator it4 = offsets.begin();
for (std::vector<std::string>::const_iterator it3 = filenames.begin();
it3 != filenames.end() && it4 != offsets.end();
it3++, it4++)
{
if (*it3 == filename) *it4 = off;
}
}
}
size_t remote_file_count;
if (!sock.get(remote_file_count) || !sock.end_of_message())
{
error_msg = "Unable to get remote file count.";
return false;
}
if (file_count != remote_file_count)
{
formatstr(error_msg, "Received %ld files, but remote side thought it sent %ld files\n", file_count, remote_file_count);
return false;
}
if ((total_files != file_count) && !error_msg.size())
{
error_msg = "At least one file transfer failed.";
return false;
}
return true;
}
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 (IsDebugLevel(D_COMMAND)) {
dprintf (D_COMMAND, "DCStarter::startSSHD(%s,...) making connection to %s\n",
getCommandStringSafe(START_SSHD), _addr ? _addr : "NULL");
}
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( !putClassAd(&sock, input) || !sock.end_of_message() ) {
error_msg = "Failed to send START_SSHD request to starter";
return false;
}
ClassAd result;
sock.decode();
if( !getClassAd(&sock, result) || !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.formatstr("%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.formatstr("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.formatstr("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.formatstr("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.formatstr("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.formatstr("Failed to write to %s: %s",
known_hosts_file,strerror(errno));
fclose( fp );
free( decode_buf );
return false;
}
if( fclose(fp)!=0 ) {
error_msg.formatstr("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
}
int
do_Q_request(ReliSock *syscall_sock,bool &may_fork)
{
int request_num = -1;
int rval;
syscall_sock->decode();
assert( syscall_sock->code(request_num) );
dprintf(D_SYSCALLS, "Got request #%d\n", request_num);
switch( request_num ) {
case CONDOR_InitializeConnection:
{
// dprintf( D_ALWAYS, "InitializeConnection()\n" );
bool authenticated = true;
// Authenticate socket, if not already done by daemonCore
if( !syscall_sock->triedAuthentication() ) {
if( IsDebugLevel(D_SECURITY) ) {
MyString methods;
SecMan::getAuthenticationMethods( WRITE, &methods );
dprintf(D_SECURITY,"Calling authenticate(%s) in qmgmt_receivers\n", methods.Value());
}
CondorError errstack;
if( ! SecMan::authenticate_sock(syscall_sock, WRITE, &errstack) ) {
// Failed to authenticate
dprintf( D_ALWAYS, "SCHEDD: authentication failed: %s\n",
errstack.getFullText().c_str() );
authenticated = false;
}
}
if ( authenticated ) {
InitializeConnection( syscall_sock->getOwner(),
syscall_sock->getDomain() );
} else {
InitializeConnection( NULL, NULL );
}
return 0;
}
case CONDOR_InitializeReadOnlyConnection:
{
// dprintf( D_ALWAYS, "InitializeReadOnlyConnection()\n" );
// Since InitializeConnection() does nothing, and we need
// to record the fact that this is a read-only connection,
// but we have to do it in the socket (since we don't have
// any other persistent data structure, and it's probably
// the right place anyway), set the FQU.
//
// We need to record if this is a read-only connection so that
// we can avoid expanding $$ in GetJobAd; simply checking if the
// connection is authenticated isn't sufficient, because the
// security session cache means that read-only connection could
// be authenticated by a previous authenticated connection from
// the same address (when using host-based security) less than
// the expiration period ago.
syscall_sock->setFullyQualifiedUser( "read-only" );
// same as InitializeConnection but no authenticate()
InitializeConnection( NULL, NULL );
may_fork = true;
return 0;
}
case CONDOR_SetEffectiveOwner:
{
MyString owner;
int terrno;
assert( syscall_sock->get(owner) );
assert( syscall_sock->end_of_message() );
rval = QmgmtSetEffectiveOwner( owner.Value() );
terrno = errno;
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );
char const *fqu = syscall_sock->getFullyQualifiedUser();
dprintf(D_SYSCALLS, "\tSetEffectiveOwner\n");
dprintf(D_SYSCALLS, "\tauthenticated user = '******'\n", fqu ? fqu : "");
dprintf(D_SYSCALLS, "\trequested owner = '%s'\n", owner.Value());
dprintf(D_SYSCALLS, "\trval %d, errno %d\n", rval, terrno);
return 0;
}
case CONDOR_NewCluster:
{
int terrno;
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = NewCluster( );
terrno = errno;
dprintf(D_SYSCALLS,
"\tNewCluster: rval = %d, errno = %d\n",rval,terrno );
if ( rval > 0 ) {
dprintf( D_AUDIT, *syscall_sock,
"Submitting new job %d.0\n", rval );
}
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
dprintf(D_FULLDEBUG,"schedd: NewCluster rval %d errno %d\n",rval,terrno);
return 0;
}
case CONDOR_NewProc:
{
int cluster_id = -1;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = NewProc( cluster_id );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
if ( rval > 0 ) {
dprintf( D_AUDIT, *syscall_sock,
"Submitting new job %d.%d\n", cluster_id, rval );
}
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
dprintf(D_FULLDEBUG,"schedd: NewProc rval %d errno %d\n",rval,terrno);
return 0;
}
case CONDOR_DestroyProc:
{
int cluster_id = -1;
int proc_id = -1;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = DestroyProc( cluster_id, proc_id );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
dprintf(D_FULLDEBUG,"schedd: DestroyProc cluster %d proc %d rval %d errno %d\n",cluster_id,proc_id,rval,terrno);
return 0;
}
case CONDOR_DestroyCluster:
{
int cluster_id = -1;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = DestroyCluster( cluster_id );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
return 0;
}
#if 0
case CONDOR_DestroyClusterByConstraint:
{
char *constraint=NULL;
int terrno;
assert( syscall_sock->code(constraint) );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = DestroyClusterByConstraint( constraint );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
free( (char *)constraint );
assert( syscall_sock->end_of_message() );;
return 0;
}
#endif
case CONDOR_SetAttributeByConstraint:
case CONDOR_SetAttributeByConstraint2:
{
char *attr_name=NULL;
char *attr_value=NULL;
char *constraint=NULL;
int terrno;
SetAttributeFlags_t flags = 0;
assert( syscall_sock->code(constraint) );
dprintf( D_SYSCALLS, " constraint = %s\n",constraint);
assert( syscall_sock->code(attr_value) );
assert( syscall_sock->code(attr_name) );
if( request_num == CONDOR_SetAttributeByConstraint2 ) {
assert( syscall_sock->code( flags ) );
}
assert( syscall_sock->end_of_message() );;
if (strcmp (attr_name, ATTR_MYPROXY_PASSWORD) == 0) {
errno = 0;
dprintf( D_SYSCALLS, "SetAttributeByConstraint (MyProxyPassword) not supported...\n");
rval = 0;
terrno = errno;
} else {
errno = 0;
rval = SetAttributeByConstraint( constraint, attr_name, attr_value, flags );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
if ( rval == 0 ) {
dprintf( D_AUDIT, *syscall_sock,
"Set Attribute By Constraint %s, "
"%s = %s\n",
constraint, attr_name, attr_value);
}
}
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
free( (char *)constraint );
free( (char *)attr_value );
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_SetAttribute:
case CONDOR_SetAttribute2:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
char *attr_value=NULL;
int terrno;
SetAttributeFlags_t flags = 0;
const char *users_username;
const char *condor_username;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_value) );
assert( syscall_sock->code(attr_name) );
if( request_num == CONDOR_SetAttribute2 ) {
assert( syscall_sock->code( flags ) );
}
users_username = syscall_sock->getOwner();
condor_username = get_condor_username();
if (attr_name) dprintf(D_SYSCALLS,"\tattr_name = %s\n",attr_name);
if (attr_value) dprintf(D_SYSCALLS,"\tattr_value = %s\n",attr_value);
assert( syscall_sock->end_of_message() );;
// ckireyev:
// We do NOT want to include MyProxy password in the ClassAd (since it's a secret)
// I'm not sure if this is the best place to do this, but....
if (attr_name && attr_value && strcmp (attr_name, ATTR_MYPROXY_PASSWORD) == 0) {
errno = 0;
dprintf( D_SYSCALLS, "Got MyProxyPassword, stashing...\n");
rval = SetMyProxyPassword (cluster_id, proc_id, attr_value);
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
}
else {
errno = 0;
rval = SetAttribute( cluster_id, proc_id, attr_name, attr_value, flags );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
// If we're modifying a previously-submitted job AND either
// the client's username is not HTCondor's (i.e. not a
// daemon) OR the client says we should log...
if( (cluster_id != active_cluster_num) && (rval == 0) &&
( strcmp(users_username, condor_username) || (flags & SHOULDLOG) ) ) {
dprintf( D_AUDIT, *syscall_sock,
"Set Attribute for job %d.%d, "
"%s = %s\n",
cluster_id, proc_id, attr_name, attr_value);
}
}
free( (char *)attr_value );
free( (char *)attr_name );
if( flags & SetAttribute_NoAck ) {
if( rval < 0 ) {
return -1;
}
}
else {
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );
}
return 0;
}
case CONDOR_SetTimerAttribute:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
int duration = 0;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
if (attr_name) dprintf(D_SYSCALLS,"\tattr_name = %s\n",attr_name);
assert( syscall_sock->code(duration) );
dprintf(D_SYSCALLS,"\tduration = %d\n",duration);
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = SetTimerAttribute( cluster_id, proc_id, attr_name, duration );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
dprintf( D_AUDIT, *syscall_sock,
"Set Timer Attribute for job %d.%d, "
"attr_name = %s, duration = %d\n",
cluster_id, proc_id, attr_name, duration);
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_BeginTransaction:
{
int terrno;
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = 0; // BeginTransaction returns void (sigh), so always success
BeginTransaction( );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_AbortTransaction:
{
int terrno;
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = 0; // AbortTransaction returns void (sigh), so always success
AbortTransaction( );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_CommitTransactionNoFlags:
case CONDOR_CommitTransaction:
{
int terrno;
int flags;
if( request_num == CONDOR_CommitTransaction ) {
assert( syscall_sock->code(flags) );
}
else {
flags = 0;
}
assert( syscall_sock->end_of_message() );;
errno = 0;
CondorError errstack;
rval = CheckTransaction( flags, & errstack );
terrno = errno;
dprintf( D_SYSCALLS, "\tflags = %d, rval = %d, errno = %d\n", flags, rval, terrno );
if( rval >= 0 ) {
errno = 0;
CommitTransaction( flags );
// CommitTransaction() never returns on failure
rval = 0;
terrno = errno;
dprintf( D_SYSCALLS, "\tflags = %d, rval = %d, errno = %d\n", flags, rval, terrno );
}
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
const CondorVersionInfo *vers = syscall_sock->get_peer_version();
if (vers && vers->built_since_version(8, 3, 4))
{
// Send a classad, for less backwards-incompatibility.
int code = 1;
const char * reason = "QMGMT rejected job submission.";
if( errstack.subsys() ) {
code = 2;
reason = errstack.message();
}
ClassAd reply;
reply.Assign( "ErrorCode", code );
reply.Assign( "ErrorReason", reason );
assert( putClassAd( syscall_sock, reply ) );
}
}
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetAttributeFloat:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
float value = 0.0;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
assert( syscall_sock->end_of_message() );;
errno = 0;
if( QmgmtMayAccessAttribute( attr_name ) ) {
rval = GetAttributeFloat( cluster_id, proc_id, attr_name, &value );
}
else {
rval = -1;
}
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( syscall_sock->code(value) );
}
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetAttributeInt:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
int value = 0;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
dprintf( D_SYSCALLS, " attr_name = %s\n", attr_name );
assert( syscall_sock->end_of_message() );;
errno = 0;
if( QmgmtMayAccessAttribute( attr_name ) ) {
rval = GetAttributeInt( cluster_id, proc_id, attr_name, &value );
}
else {
rval = -1;
}
terrno = errno;
if (rval < 0) {
dprintf( D_SYSCALLS, "GetAttributeInt(%d, %d, %s) not found.\n",
cluster_id, proc_id, attr_name);
} else {
dprintf( D_SYSCALLS, " value: %d\n", value );
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
}
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( syscall_sock->code(value) );
}
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetAttributeString:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
char *value = NULL;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
assert( syscall_sock->end_of_message() );;
errno = 0;
if( QmgmtMayAccessAttribute( attr_name ) ) {
rval = GetAttributeStringNew( cluster_id, proc_id, attr_name, &value );
}
else {
rval = -1;
}
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( syscall_sock->code(value) );
}
free( (char *)value );
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetAttributeExpr:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
assert( syscall_sock->end_of_message() );;
char *value = NULL;
errno = 0;
if( QmgmtMayAccessAttribute( attr_name ) ) {
rval = GetAttributeExprNew( cluster_id, proc_id, attr_name, &value );
}
else {
rval = -1;
}
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
if ( !syscall_sock->code(rval) ) {
free(value);
return -1;
}
if( rval < 0 ) {
if ( !syscall_sock->code(terrno) ) {
free(value);
return -1;
}
}
if( rval >= 0 ) {
if ( !syscall_sock->code(value) ) {
free(value);
return -1;
}
}
free( (char *)value );
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetDirtyAttributes:
{
int cluster_id = -1;
int proc_id = -1;
ClassAd updates;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = GetDirtyAttributes( cluster_id, proc_id, &updates );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
if ( !syscall_sock->code(rval) ) {
return -1;
}
if( rval < 0 ) {
if ( !syscall_sock->code(terrno) ) {
return -1;
}
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, updates) );
}
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_DeleteAttribute:
{
int cluster_id = -1;
int proc_id = -1;
char *attr_name=NULL;
int terrno;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->code(attr_name) );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = DeleteAttribute( cluster_id, proc_id, attr_name );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
free( (char *)attr_name );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetJobAd:
{
int cluster_id = -1;
int proc_id = -1;
ClassAd *ad = NULL;
int terrno;
bool delete_ad = false;
assert( syscall_sock->code(cluster_id) );
dprintf( D_SYSCALLS, " cluster_id = %d\n", cluster_id );
assert( syscall_sock->code(proc_id) );
dprintf( D_SYSCALLS, " proc_id = %d\n", proc_id );
assert( syscall_sock->end_of_message() );;
// dprintf( D_ALWAYS, "(%d.%d) isAuthenticated() = %d\n", cluster_id, proc_id, syscall_sock->isAuthenticated() );
// dprintf( D_ALWAYS, "(%d.%d) getOwner() = %s\n", cluster_id, proc_id, syscall_sock->getOwner() );
errno = 0;
// Only fetch the jobad for legal values of cluster/proc
if( cluster_id >= 1 ) {
if( proc_id >= 0 ) {
const char * fqu = syscall_sock->getFullyQualifiedUser();
if( fqu != NULL && strcmp( fqu, "read-only" ) != 0 ) {
// expand $$() macros in the jobad as required by GridManager.
// The GridManager depends on the fact that the following call
// expands $$ and saves the expansions to disk in case of
// restart.
ad = GetJobAd_as_ClassAd( cluster_id, proc_id, true, true );
delete_ad = true;
// note : since we expanded the ad, ad is now a deep
// copy of the ad in memory, so we must delete it below.
} else {
ad = GetJobAd_as_ClassAd( cluster_id, proc_id, false, false );
}
} else if( proc_id == -1 ) {
// allow cluster ad to be queried as required by preen, but
// do NOT ask to expand $$() macros in a cluster ad!
ad = GetJobAd_as_ClassAd( cluster_id, proc_id, false, false );
}
}
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE) );
}
// If we called GetJobAd() with the third bool argument set
// to True (expandedAd), it does a deep copy of the ad in the
// queue in order to expand the $$() attributes. So we must
// delete it.
if (delete_ad) delete ad;
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetJobByConstraint:
{
char *constraint=NULL;
ClassAd *ad;
int terrno;
assert( syscall_sock->code(constraint) );
assert( syscall_sock->end_of_message() );;
errno = 0;
ad = GetJobByConstraint_as_ClassAd( constraint );
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE) );
}
FreeJobAd(ad);
free( (char *)constraint );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetNextJob:
{
ClassAd *ad;
int initScan = 0;
int terrno;
assert( syscall_sock->code(initScan) );
dprintf( D_SYSCALLS, " initScan = %d\n", initScan );
assert( syscall_sock->end_of_message() );;
errno = 0;
ad = GetNextJob( initScan );
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE) );
}
FreeJobAd(ad);
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetNextJobByConstraint:
{
char *constraint=NULL;
ClassAd *ad;
int initScan = 0;
int terrno;
assert( syscall_sock->code(initScan) );
dprintf( D_SYSCALLS, " initScan = %d\n", initScan );
if ( !(syscall_sock->code(constraint)) ) {
if (constraint != NULL) {
free(constraint);
constraint = NULL;
}
return -1;
}
assert( syscall_sock->end_of_message() );;
errno = 0;
ad = GetNextJobByConstraint( constraint, initScan );
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE) );
}
FreeJobAd(ad);
free( (char *)constraint );
assert( syscall_sock->end_of_message() );;
return 0;
}
case CONDOR_GetNextDirtyJobByConstraint:
{
char *constraint=NULL;
ClassAd *ad;
int initScan = 0;
int terrno;
assert( syscall_sock->code(initScan) );
dprintf( D_SYSCALLS, " initScan = %d\n", initScan );
if ( !(syscall_sock->code(constraint)) ) {
if (constraint != NULL) {
free(constraint);
constraint = NULL;
}
return -1;
}
assert( syscall_sock->end_of_message() );
errno = 0;
ad = GetNextDirtyJobByConstraint( constraint, initScan );
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE) );
}
FreeJobAd(ad);
free( (char *)constraint );
assert( syscall_sock->end_of_message() );
return 0;
}
case CONDOR_SendSpoolFile:
{
char *filename=NULL;
int terrno;
assert( syscall_sock->code(filename) );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = SendSpoolFile( filename );
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
#if 0
syscall_sock->encode();
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
assert( syscall_sock->end_of_message() );;
#endif
free( (char *)filename );
return 0;
}
case CONDOR_SendSpoolFileIfNeeded:
{
int terrno;
ClassAd ad;
assert( getClassAd(syscall_sock, ad) );
assert( syscall_sock->end_of_message() );;
errno = 0;
rval = SendSpoolFileIfNeeded(ad);
terrno = errno;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
return 0;
}
case CONDOR_GetAllJobsByConstraint:
{
char *constraint=NULL;
char *projection=NULL;
ClassAd *ad;
int terrno;
int initScan = 1;
classad::References proj;
if ( !(syscall_sock->code(constraint)) ) {
if (constraint != NULL) {
free(constraint);
constraint = NULL;
}
return -1;
}
if ( !(syscall_sock->code(projection)) ) {
if (projection != NULL) {
free(constraint);
free(projection);
projection = NULL;
}
return -1;
}
dprintf( D_SYSCALLS, " constraint = %s\n", constraint );
dprintf( D_SYSCALLS, " projection = %s\n", projection ? projection : "");
assert( syscall_sock->end_of_message() );;
// if there is a projection, convert it into a set of attribute names
if (projection) {
StringTokenIterator list(projection);
const std::string * attr;
while ((attr = list.next_string())) { proj.insert(*attr); }
}
syscall_sock->encode();
do {
errno = 0;
ad = GetNextJobByConstraint( constraint, initScan );
initScan=0; // one first time through, otherwise 0
terrno = errno;
rval = ad ? 0 : -1;
dprintf( D_SYSCALLS, "\trval = %d, errno = %d\n", rval, terrno );
assert( syscall_sock->code(rval) );
if( rval < 0 ) {
assert( syscall_sock->code(terrno) );
}
if( rval >= 0 ) {
assert( putClassAd(syscall_sock, *ad, PUT_CLASSAD_NO_PRIVATE, proj.empty() ? NULL : &proj) );
FreeJobAd(ad);
}
} while (rval >= 0);
assert( syscall_sock->end_of_message() );;
free( (char *)constraint );
free( (char *)projection );
return 0;
}
case CONDOR_CloseSocket:
{
assert( syscall_sock->end_of_message() );;
return -1;
}
} /* End of switch */
return -1;
} /* End of function */
/* Generic read/write wrappers for condor. These function emulate-ish the
* read/write system calls under unix except that they are portable, use
* a timeout, and make sure that all data is read or written.
*
* A few notes on the behavior differing from POSIX:
* - These will never fail due to EINTR.
* - If in non_blocking mode, there may be a short read or write returned.
* - The corresponding POSIX functon returns 0 bytes read when the peer closed
* the socket; these return -2.
* - If zero bytes were read/written in non-blocking mode, this will return 0. This differs
* from POSIX.
* - Providing a zero-sized argument to this function will cause the program to abort().
*
* Returns < 0 on failure. -1 = general error, -2 = peer closed socket
*/
int
condor_read( char const *peer_description, SOCKET fd, char *buf, int sz, int timeout, int flags, bool non_blocking )
{
Selector selector;
int nr = 0, nro;
unsigned int start_time=0, cur_time=0;
char sinbuf[SINFUL_STRING_BUF_SIZE];
if( IsDebugLevel(D_NETWORK) ) {
dprintf(D_NETWORK,
"condor_read(fd=%d %s,,size=%d,timeout=%d,flags=%d,non_blocking=%d)\n",
fd,
not_null_peer_description(peer_description,fd,sinbuf),
sz,
timeout,
flags,
non_blocking);
}
/* PRE Conditions. */
ASSERT(fd >= 0); /* Need valid file descriptor */
ASSERT(buf != NULL); /* Need real memory to put data into */
ASSERT(sz > 0); /* Need legit size on buffer */
if (non_blocking) {
#ifdef WIN32
unsigned long mode = 1; // nonblocking mode
if (ioctlsocket(fd, FIONBIO, &mode) < 0)
return -1;
#else
int fcntl_flags;
if ( (fcntl_flags=fcntl(fd, F_GETFL)) < 0 )
return -1;
// set nonblocking mode
if ( ((fcntl_flags & O_NONBLOCK) == 0) && fcntl(fd, F_SETFL, fcntl_flags | O_NONBLOCK) == -1 )
return -1;
#endif
nr = -2;
while (nr == -2 || (nr == -1 && errno == EINTR)) {
nr = recv(fd, buf, sz, flags);
}
if ( nr <= 0 ) {
int the_error;
char const *the_errorstr;
#ifdef WIN32
the_error = WSAGetLastError();
the_errorstr = "";
#else
the_error = errno;
the_errorstr = strerror(the_error);
#endif
if ( nr == 0 && !(flags & MSG_PEEK)) {
nr = -2;
dprintf( D_FULLDEBUG, "condor_read(): "
"Socket closed when trying to read %d bytes from %s in non-blocking mode\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf) );
} else if ( !errno_is_temporary(the_error) ) {
dprintf( D_ALWAYS, "condor_read() failed: recv() %d bytes from %s "
"returned %d, "
"timeout=%d, errno=%d %s.\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf),
nr, timeout, the_error, the_errorstr );
}
else
{
nr = 0;
}
}
#ifdef WIN32
mode = 0; // reset blocking mode
if (ioctlsocket(fd, FIONBIO, &mode) < 0)
return -1;
#else
// reset flags to prior value
if ( ((fcntl_flags & O_NONBLOCK) == 0) && (fcntl(fd, F_SETFL, fcntl_flags) == -1) )
return -1;
#endif
return nr;
}
selector.add_fd( fd, Selector::IO_READ );
if ( timeout > 0 ) {
start_time = time(NULL);
cur_time = start_time;
}
while( nr < sz ) {
if( timeout > 0 ) {
if( cur_time == 0 ) {
cur_time = time(NULL);
}
// If it hasn't yet been longer then we said we would wait...
if( start_time + timeout > cur_time ) {
selector.set_timeout( (start_time + timeout) - cur_time );
} else {
dprintf( D_ALWAYS,
"condor_read(): timeout reading %d bytes from %s.\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf) );
return -1;
}
cur_time = 0;
if( IsDebugVerbose( D_NETWORK ) ) {
dprintf(D_NETWORK, "condor_read(): fd=%d\n", fd);
}
selector.execute();
if( IsDebugVerbose( D_NETWORK ) ) {
dprintf(D_NETWORK, "condor_read(): select returned %d\n",
selector.select_retval());
}
if ( selector.timed_out() ) {
dprintf( D_ALWAYS,
"condor_read(): timeout reading %d bytes from %s.\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf) );
return -1;
} else if ( selector.signalled() ) {
continue;
} else if ( !selector.has_ready() ) {
int the_error;
char const *the_errorstr;
#ifdef WIN32
the_error = WSAGetLastError();
the_errorstr = "";
#else
the_error = errno;
the_errorstr = strerror(the_error);
#endif
dprintf( D_ALWAYS, "condor_read() failed: select() "
"returns %d, reading %d bytes from %s (errno=%d %s).\n",
selector.select_retval(),
sz,
not_null_peer_description(peer_description,fd,sinbuf),
the_error, the_errorstr );
return -1;
}
}
start_thread_safe("recv");
nro = recv(fd, &buf[nr], sz - nr, flags);
// Save the error value before stop_thread_safe(), as that may
// overwrite it.
int the_error;
#ifdef WIN32
the_error = WSAGetLastError();
#else
the_error = errno;
#endif
stop_thread_safe("recv");
if( nro <= 0 ) {
// If timeout > 0, and we made it here, then
// we know we were woken by select(). Now, if
// select() wakes up on a read fd, and then recv()
// subsequently returns 0, that means that the
// socket has been closed by our peer.
// If timeout == 0, then recv() should have
// blocked until 1 or more bytes arrived.
// Thus no matter what, if nro==0, then the
// socket must be closed.
if ( nro == 0 ) {
dprintf( D_FULLDEBUG, "condor_read(): "
"Socket closed when trying to read %d bytes from %s\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf) );
return -2;
}
char const *the_errorstr;
#ifdef WIN32
the_errorstr = "";
#else
the_errorstr = strerror(the_error);
#endif
if ( errno_is_temporary(the_error) ) {
dprintf( D_FULLDEBUG, "condor_read(): "
"recv() returned temporary error %d %s,"
"still trying to read from %s\n",
the_error,the_errorstr,
not_null_peer_description(peer_description,fd,sinbuf) );
continue;
}
dprintf( D_ALWAYS, "condor_read() failed: recv(fd=%d) returned %d, "
"errno = %d %s, reading %d bytes from %s.\n",
fd, nro, the_error, the_errorstr, sz,
not_null_peer_description(peer_description,fd,sinbuf) );
if( the_error == ETIMEDOUT ) {
if( timeout <= 0 ) {
dprintf( D_ALWAYS,
"condor_read(): read timeout during blocking read from %s\n",
not_null_peer_description(peer_description,fd,sinbuf));
}
else {
int lapse = (int)(time(NULL)-start_time);
dprintf( D_ALWAYS,
"condor_read(): UNEXPECTED read timeout after %ds during non-blocking read from %s (desired timeout=%ds)\n",
lapse,
not_null_peer_description(peer_description,fd,sinbuf),
timeout);
}
}
return -1;
}
nr += nro;
}
/* Post Conditions */
ASSERT( nr == sz ); // we should have read *ALL* the data
return nr;
}
int
condor_write( char const *peer_description, SOCKET fd, const char *buf, int sz, int timeout, int flags, bool non_blocking )
{
int nw = 0, nwo = 0;
unsigned int start_time = 0, cur_time = 0;
char tmpbuf[1];
int nro;
bool select_for_read = true;
bool needs_select = true;
char sinbuf[SINFUL_STRING_BUF_SIZE];
if( IsDebugLevel( D_NETWORK ) ) {
dprintf(D_NETWORK,
"condor_write(fd=%d %s,,size=%d,timeout=%d,flags=%d,non_blocking=%d)\n",
fd,
not_null_peer_description(peer_description,fd,sinbuf),
sz,
timeout,
flags,
non_blocking);
}
/* Pre-conditions. */
ASSERT(sz > 0); /* Can't write buffers that are have no data */
ASSERT(fd >= 0); /* Need valid file descriptor */
ASSERT(buf != NULL); /* Need valid buffer to write */
if (non_blocking) {
#ifdef WIN32
unsigned long mode = 1; // nonblocking mode
if (ioctlsocket(fd, FIONBIO, &mode) < 0)
return -1;
#else
int fcntl_flags;
if ( (fcntl_flags=fcntl(fd, F_GETFL)) < 0 )
return -1;
// set nonblocking mode
if ( ((fcntl_flags & O_NONBLOCK) == 0) && fcntl(fd, F_SETFL, fcntl_flags | O_NONBLOCK) == -1 )
return -1;
#endif
nw = -2;
while (nw == -2 || (nw == -1 && errno == EINTR)) {
nw = send(fd, buf, sz, flags);
}
if ( nw <= 0 ) {
int the_error;
char const *the_errorstr;
#ifdef WIN32
the_error = WSAGetLastError();
the_errorstr = "";
#else
the_error = errno;
the_errorstr = strerror(the_error);
#endif
if ( !errno_is_temporary(the_error) ) {
dprintf( D_ALWAYS, "condor_write() failed: send() %d bytes to %s "
"returned %d, "
"timeout=%d, errno=%d %s.\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf),
nw, timeout, the_error, the_errorstr );
}
else
{
nw = 0;
}
}
if (nw < 0) {
dprintf(D_NETWORK, "condor_write (non-blocking) wrote %d bytes.\n", nw);
}
#ifdef WIN32
mode = 0; // reset blocking mode
if (ioctlsocket(fd, FIONBIO, &mode) < 0)
return -1;
#else
// reset flags to prior value
if ( ((fcntl_flags & O_NONBLOCK) == 0) && fcntl(fd, F_SETFL, fcntl_flags) == -1 )
return -1;
#endif
return nw;
}
Selector selector;
selector.add_fd( fd, Selector::IO_READ );
selector.add_fd( fd, Selector::IO_WRITE );
selector.add_fd( fd, Selector::IO_EXCEPT );
if(timeout > 0) {
start_time = time(NULL);
cur_time = start_time;
}
while( nw < sz ) {
needs_select = true;
if( timeout > 0 ) {
while( needs_select ) {
if( cur_time == 0 ) {
cur_time = time(NULL);
}
if( start_time + timeout > cur_time ) {
selector.set_timeout( (start_time + timeout) - cur_time );
} else {
dprintf( D_ALWAYS, "condor_write(): "
"timed out writing %d bytes to %s\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf) );
return -1;
}
cur_time = 0;
// The write and except sets are added at the top of
// this function, since we always want to select on
// them.
if( select_for_read ) {
// Also, put it in the read fds, so we'll wake
// up if the socket is closed
selector.add_fd( fd, Selector::IO_READ );
} else {
selector.delete_fd( fd, Selector::IO_READ );
}
selector.execute();
// unless we decide we need to select() again, we
// want to break out of our while() loop now that
// we've actually performed a select()
needs_select = false;
if ( selector.timed_out() ) {
dprintf( D_ALWAYS, "condor_write(): "
"timed out writing %d bytes to %s\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf) );
return -1;
} else if ( selector.signalled() ) {
needs_select = true;
continue;
} else if ( selector.has_ready() ) {
if ( selector.fd_ready( fd, Selector::IO_READ ) ) {
dprintf(D_NETWORK, "condor_write(): socket %d is readable\n", fd);
// see if the socket was closed
nro = recv(fd, tmpbuf, 1, MSG_PEEK);
if( nro == -1 ) {
int the_error;
char const *the_errorstr;
#ifdef WIN32
the_error = WSAGetLastError();
the_errorstr = "";
#else
the_error = errno;
the_errorstr = strerror(the_error);
#endif
if(errno_is_temporary( the_error )) {
continue;
}
dprintf( D_ALWAYS, "condor_write(): "
"Socket closed when trying "
"to write %d bytes to %s, fd is %d, "
"errno=%d %s\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf),
fd, the_error, the_errorstr );
return -1;
}
if( ! nro ) {
dprintf( D_ALWAYS, "condor_write(): "
"Socket closed when trying "
"to write %d bytes to %s, fd is %d\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf),
fd );
return -1;
}
/*
otherwise, there's real data to consume
on the read side, and we don't want to
put our fd in the readfds anymore or
select() will never block. also, we
need to re-do the select()
*/
needs_select = true;
select_for_read = false;
}
} else {
dprintf( D_ALWAYS, "condor_write() failed: select() "
"returns %d, "
"writing %d bytes to %s.\n",
selector.select_retval(),
sz,
not_null_peer_description(peer_description,fd,sinbuf) );
return -1;
}
}
}
start_thread_safe("send");
nwo = send(fd, &buf[nw], sz - nw, flags);
// Save the error value before stop_thread_safe(), as that may
// overwrite it.
int the_error;
#ifdef WIN32
the_error = WSAGetLastError();
#else
the_error = errno;
#endif
stop_thread_safe("send");
if( nwo <= 0 ) {
char const *the_errorstr;
#ifdef WIN32
the_errorstr = "";
#else
the_errorstr = strerror(the_error);
#endif
if ( errno_is_temporary(the_error) ) {
dprintf( D_FULLDEBUG, "condor_write(): "
"send() returned temporary error %d %s,"
"still trying to write %d bytes to %s\n", the_error,
the_errorstr, sz,
not_null_peer_description(peer_description,fd,sinbuf) );
continue;
}
dprintf( D_ALWAYS, "condor_write() failed: send() %d bytes to %s "
"returned %d, "
"timeout=%d, errno=%d %s.\n",
sz,
not_null_peer_description(peer_description,fd,sinbuf),
nwo, timeout, the_error, the_errorstr );
return -1;
}
nw += nwo;
}
/* POST conditions. */
ASSERT( nw == sz ); /* Make sure that we wrote everything */
return nw;
}
