/*
* Create socket for incoming connections and bind to it an address for
* clients to use. Returns -1 on failure.
*
* If '*family' is AF_UNIX and unix domain sockets are supported:
* 'port' is ignored and 'address' is the path to the socket file in the filesystem.
*
* Otherwise:
* address is a string representing the Inet/IPv6 address that the port
* is advertised for. To allow connections to all this host's internet
* addresses from clients use address == "INADDR_ANY", or for localhost
* access only use address == "INADDR_LOOPBACK".
* On input, 'family' is a pointer to the address family to use (AF_INET,
* AF_INET6) if the address specified is empty. If the spec is not
* empty then family is ignored and is set to the actual address family
* used. 'family' must be initialized to AF_UNSPEC, in this case.
*/
static int
OpenRequestSocket(int port, const char *address, int *family,
int backlog, __pmFdSet *fdset, int *maximum)
{
int fd = -1;
int one, sts;
__pmSockAddr *myAddr;
int isUnix = 0;
/*
* Using this flag will eliminate the need for more conditional
* compilation below, hopefully making the code easier to read and maintain.
*/
#if defined(HAVE_STRUCT_SOCKADDR_UN)
if (*family == AF_UNIX)
isUnix = 1;
#endif
if (isUnix) {
if ((myAddr = __pmSockAddrAlloc()) == NULL) {
__pmNoMem("OpenRequestSocket: can't allocate socket address",
sizeof(*myAddr), PM_FATAL_ERR);
}
/* Initialize the address. */
__pmSockAddrSetFamily(myAddr, *family);
__pmSockAddrSetPath(myAddr, address);
/* Create the socket. */
fd = __pmCreateUnixSocket();
}
else {
/*
* If the address is unspecified, then use the address family we
* have been given, otherwise the family will be determined by
* __pmStringToSockAddr.
*/
if (address == NULL || strcmp(address, "INADDR_ANY") == 0) {
if ((myAddr = __pmSockAddrAlloc()) == NULL) {
__pmNoMem("OpenRequestSocket: can't allocate socket address",
sizeof(*myAddr), PM_FATAL_ERR);
}
__pmSockAddrInit(myAddr, *family, INADDR_ANY, 0);
}
else if (strcmp(address, "INADDR_LOOPBACK") == 0) {
if ((myAddr = __pmSockAddrAlloc()) == NULL) {
__pmNoMem("OpenRequestSocket: can't allocate socket address",
sizeof(*myAddr), PM_FATAL_ERR);
}
__pmSockAddrInit(myAddr, *family, INADDR_LOOPBACK, 0);
}
else {
if ((myAddr = __pmStringToSockAddr(address)) == NULL) {
__pmNotifyErr(LOG_ERR, "OpenRequestSocket(%d, %s) invalid address\n",
port, address);
goto fail;
}
*family = __pmSockAddrGetFamily(myAddr);
}
__pmSockAddrSetPort(myAddr, port);
/* Create the socket. */
if (*family == AF_INET)
fd = __pmCreateSocket();
else if (*family == AF_INET6)
fd = __pmCreateIPv6Socket();
else {
__pmNotifyErr(LOG_ERR, "OpenRequestSocket(%d, %s) invalid address family: %d\n",
port, address, *family);
goto fail;
}
}
if (fd < 0) {
__pmNotifyErr(LOG_ERR, "OpenRequestSocket(%d, %s, %s) __pmCreateSocket: %s\n",
port, address, AddressFamily(*family), netstrerror());
goto fail;
}
/* Ignore dead client connections. */
one = 1;
#ifndef IS_MINGW
if (__pmSetSockOpt(fd, SOL_SOCKET, SO_REUSEADDR, (char *)&one,
(__pmSockLen)sizeof(one)) < 0) {
__pmNotifyErr(LOG_ERR,
"OpenRequestSocket(%d, %s, %s) __pmSetSockOpt(SO_REUSEADDR): %s\n",
port, address, AddressFamily(*family), netstrerror());
goto fail;
}
#else
if (__pmSetSockOpt(fd, SOL_SOCKET, SO_EXCLUSIVEADDRUSE, (char *)&one,
(__pmSockLen)sizeof(one)) < 0) {
__pmNotifyErr(LOG_ERR,
"OpenRequestSocket(%d, %s, %s) __pmSetSockOpt(EXCLUSIVEADDRUSE): %s\n",
port, address, AddressFamily(*family), netstrerror());
goto fail;
}
#endif
/* and keep alive please - bad networks eat fds */
if (__pmSetSockOpt(fd, SOL_SOCKET, SO_KEEPALIVE, (char *)&one,
(__pmSockLen)sizeof(one)) < 0) {
__pmNotifyErr(LOG_ERR,
"OpenRequestSocket(%d, %s, %s) __pmSetSockOpt(SO_KEEPALIVE): %s\n",
port, address, AddressFamily(*family), netstrerror());
goto fail;
}
sts = __pmBind(fd, (void *)myAddr, __pmSockAddrSize());
__pmSockAddrFree(myAddr);
myAddr = NULL;
if (sts < 0) {
sts = neterror();
__pmNotifyErr(LOG_ERR, "OpenRequestSocket(%d, %s, %s) __pmBind: %s\n",
port, address, AddressFamily(*family), netstrerror());
if (sts == EADDRINUSE)
__pmNotifyErr(LOG_ERR, "%s may already be running\n", pmProgname);
goto fail;
}
if (isUnix) {
/*
* For unix domain sockets, grant rw access to the socket for all,
* otherwise, on linux platforms, connection will not be possible.
* This must be done AFTER binding the address. See Unix(7) for details.
*/
sts = chmod(address, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
if (sts != 0) {
__pmNotifyErr(LOG_ERR,
"OpenRequestSocket(%d, %s, %s) chmod(%s): %s\n",
port, address, AddressFamily(*family), address, strerror(errno));
goto fail;
}
}
sts = __pmListen(fd, backlog); /* Max. pending connection requests */
if (sts < 0) {
__pmNotifyErr(LOG_ERR, "OpenRequestSocket(%d, %s, %s) __pmListen: %s\n",
port, address, AddressFamily(*family), netstrerror());
goto fail;
}
if (fd > *maximum)
*maximum = fd;
__pmFD_SET(fd, fdset);
return fd;
fail:
if (fd != -1) {
__pmCloseSocket(fd);
/* We must unlink the socket file. */
if (isUnix)
unlink(address);
}
if (myAddr)
__pmSockAddrFree(myAddr);
return -1;
}
int
reconnect(void)
{
int sts;
int ctx;
__pmContext *ctxp;
if ((ctx = pmWhichContext()) >= 0)
ctxp = __pmHandleToPtr(ctx);
if (ctx < 0 || ctxp == NULL) {
fprintf(stderr, "%s: reconnect botch: cannot get context: %s\n", pmProgname, pmErrStr(ctx));
exit(1);
}
sts = pmReconnectContext(ctx);
if (sts >= 0) {
time_t now;
time(&now);
fprintf(stderr, "%s: re-established connection to PMCD on \"%s\" at %s\n",
pmProgname, pmcd_host, ctime(&now));
pmcdfd = ctxp->c_pmcd->pc_fd;
__pmFD_SET(pmcdfd, &fds);
numfds = maxfd() + 1;
}
PM_UNLOCK(ctxp->c_lock);
return sts;
}
static void
CheckNewClient(__pmFdSet * fdset, int rfd, int family)
{
ClientInfo *cp;
if (__pmFD_ISSET(rfd, fdset)) {
if ((cp = AcceptNewClient(rfd)) == NULL)
/* failed to negotiate, already cleaned up */
return;
/* establish a new connection to pmcd */
if ((cp->pmcd_fd = __pmAuxConnectPMCDPort(cp->pmcd_hostname, cp->pmcd_port)) < 0) {
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_CONTEXT)
/* append to message started in AcceptNewClient() */
fprintf(stderr, " oops!\n"
"__pmAuxConnectPMCDPort(%s,%d) failed: %s\n",
cp->pmcd_hostname, cp->pmcd_port,
pmErrStr(-oserror()));
#endif
CleanupClient(cp, -oserror());
}
else {
if (cp->pmcd_fd > maxSockFd)
maxSockFd = cp->pmcd_fd;
__pmFD_SET(cp->pmcd_fd, &sockFds);
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_CONTEXT)
/* append to message started in AcceptNewClient() */
fprintf(stderr, " fd=%d\n", cp->pmcd_fd);
#endif
}
}
}
/* Establish a new socket connection to a client */
ClientInfo *
AcceptNewClient(int reqfd)
{
static unsigned int seq = 0;
int i, fd;
__pmSockLen addrlen;
struct timeval now;
i = NewClient();
addrlen = __pmSockAddrSize();
fd = __pmAccept(reqfd, client[i].addr, &addrlen);
if (fd == -1) {
if (neterror() == EPERM) {
__pmNotifyErr(LOG_NOTICE, "AcceptNewClient(%d): "
"Permission Denied\n", reqfd);
client[i].fd = -1;
DeleteClient(&client[i]);
return NULL;
}
else {
__pmNotifyErr(LOG_ERR, "AcceptNewClient(%d) __pmAccept: %s\n",
reqfd, netstrerror());
Shutdown();
exit(1);
}
}
if (fd > maxClientFd)
maxClientFd = fd;
pmcd_openfds_sethi(fd);
__pmFD_SET(fd, &clientFds);
__pmSetVersionIPC(fd, UNKNOWN_VERSION); /* before negotiation */
__pmSetSocketIPC(fd);
client[i].fd = fd;
client[i].status.connected = 1;
client[i].status.changes = 0;
memset(&client[i].attrs, 0, sizeof(__pmHashCtl));
/*
* Note seq needs to be unique, but we're using a free running counter
* and not bothering to check here ... unless we churn through
* 4,294,967,296 (2^32) clients while one client remains connected
* we won't have a problem
*/
client[i].seq = seq++;
__pmtimevalNow(&now);
client[i].start = now.tv_sec;
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL0)
fprintf(stderr, "AcceptNewClient(%d): client[%d] (fd %d)\n", reqfd, i, fd);
#endif
pmcd_trace(TR_ADD_CLIENT, i, 0, 0);
return &client[i];
}
int
reconnect(void)
{
int sts;
int ctx;
time_t now;
__pmContext *ctxp;
if ((ctx = pmWhichContext()) >= 0)
ctxp = __pmHandleToPtr(ctx);
if (ctx < 0 || ctxp == NULL) {
fprintf(stderr, "%s: reconnect botch: cannot get context: %s\n",
pmProgname, pmErrStr(ctx));
exit(1);
}
sts = pmReconnectContext(ctx);
if (sts >= 0) {
pmcdfd = ctxp->c_pmcd->pc_fd;
__pmFD_SET(pmcdfd, &fds);
numfds = maxfd() + 1;
}
PM_UNLOCK(ctxp->c_lock);
if (sts < 0)
return sts;
time(&now);
fprintf(stderr, "%s: re-established connection to PMCD on \"%s\" at %s",
pmProgname, pmcd_host, ctime(&now));
/*
* Metrics may have changed while PMCD was unreachable, so we
* need to recheck each metric to make sure that its PMID and
* semantics have not changed. We cannot recover if there is
* an incompatible change - must defer to controlling scripts
* or processes (a new-named archive will have to be created,
* from a new pmlogger process, and pmlogrewrite/pmlogextract
* will need to become involved if they need to be merged).
*/
validate_metrics();
/*
* All metrics have been validated, however, this state change
* represents a potential gap in the stream of metrics. So we
* must store a <mark> record at this point.
*/
if ((sts = putmark()) < 0) {
fprintf(stderr, "putmark: %s\n", pmErrStr(sts));
exit(1);
}
return 0;
}
/*
* Opens a TCP socket and returns the descriptor
* Returns:
* socket descriptor, or
* -1 on error
*/
int makeSocket(const char *host)
{
int sock; /* Socket descriptor */
int ret;
int port;
char *p;
__pmFdSet wfds;
struct timeval tv;
struct timeval *ptv;
__pmSockAddr *myaddr;
__pmHostEnt *servInfo;
void *enumIx;
int flags = 0;
/* Check for port number specified in URL */
p = strchr(host, ':');
if(p)
{
port = atoi(p + 1);
*p = '\0';
}
else
port = PORT_NUMBER;
servInfo = __pmGetAddrInfo(host);
if(servInfo == NULL) { errorSource = H_ERRNO; return -1; }
sock = -1;
enumIx = NULL;
for (myaddr = __pmHostEntGetSockAddr(servInfo, &enumIx);
myaddr != NULL;
myaddr = __pmHostEntGetSockAddr(servInfo, &enumIx)) {
/* Create a socket */
if (__pmSockAddrIsInet(myaddr))
sock = __pmCreateSocket();
else if (__pmSockAddrIsIPv6(myaddr))
sock = __pmCreateIPv6Socket();
else
sock = -1;
if (sock < 0) {
__pmSockAddrFree(myaddr);
continue; /* Try the next address */
}
/* Attempt to connect */
flags = __pmConnectTo(sock, myaddr, port);
__pmSockAddrFree(myaddr);
if (flags < 0) {
/*
* Mark failure in case we fall out the end of the loop
* and try next address. sock has been closed in __pmConnectTo().
*/
sock = -1;
continue;
}
/* FNDELAY and we're in progress - wait on select */
tv.tv_sec = timeout;
tv.tv_usec = 0;
ptv = (tv.tv_sec || tv.tv_usec) ? &tv : NULL;
__pmFD_ZERO(&wfds);
__pmFD_SET(sock, &wfds);
ret = __pmSelectWrite(sock+1, &wfds, ptv);
/* Was the connection successful? */
if (ret < 0) {
if (oserror() == EINTR)
return _makeSocketErr(sock, FETCHER_ERROR, HF_CONNECTTIMEOUT);
return _makeSocketErr(sock, ERRNO, 0);
}
ret = __pmConnectCheckError(sock);
if (ret == 0)
break;
/* Unsuccessful connection. */
__pmCloseSocket(sock);
sock = -1;
} /* loop over addresses */
__pmHostEntFree(servInfo);
if(sock == -1) { errorSource = ERRNO; return -1; }
sock = __pmConnectRestoreFlags(sock, flags);
if(sock < 0) { errorSource = ERRNO; return -1; }
return sock;
}
static int
http_client_connectunix(const char *path, struct timeval *timeout)
{
#if defined(HAVE_STRUCT_SOCKADDR_UN)
__pmFdSet wfds;
__pmSockAddr *myAddr;
struct timeval stv, *ptv;
int fdFlags = 0;
int fd = -1;
int sts;
int rc;
/* Initialize the socket address. */
if ((myAddr = __pmSockAddrAlloc()) == NULL) {
if (pmDebug & DBG_TRACE_HTTP)
fprintf(stderr, "HTTP connect unix(%s): out of memory\n", path);
return -ENOMEM;
}
__pmSockAddrSetFamily(myAddr, AF_UNIX);
__pmSockAddrSetPath(myAddr, path);
if ((fd = __pmCreateUnixSocket()) < 0) {
if (pmDebug & DBG_TRACE_HTTP) {
char errmsg[PM_MAXERRMSGLEN];
fprintf(stderr, "HTTP connect unix(%s) unable to create socket: %s\n",
path, osstrerror_r(errmsg, sizeof(errmsg)));
}
__pmSockAddrFree(myAddr);
return fd;
}
/* Attempt to connect */
fdFlags = __pmConnectTo(fd, myAddr, -1);
__pmSockAddrFree(myAddr);
if (fdFlags < 0) {
__pmCloseSocket(fd);
return -ECONNREFUSED;
}
/* FNDELAY and we're in progress - wait on select */
stv = *timeout;
ptv = (stv.tv_sec || stv.tv_usec) ? &stv : NULL;
__pmFD_ZERO(&wfds);
__pmFD_SET(fd, &wfds);
sts = 0;
if ((rc = __pmSelectWrite(fd+1, &wfds, ptv)) == 1)
sts = __pmConnectCheckError(fd);
else if (rc == 0)
sts = ETIMEDOUT;
else
sts = (rc < 0) ? neterror() : EINVAL;
if (sts != 0) {
/* Unsuccessful connection. */
if (sts == ENOENT)
sts = ECONNREFUSED;
__pmCloseSocket(fd);
fd = -sts;
}
if (fd < 0)
return fd;
/*
* If we're here, it means we have a valid connection; restore the
* flags and make sure this file descriptor is closed if exec() is
* called
*/
return __pmConnectRestoreFlags(fd, fdFlags);
#else
if (pmDebug & DBG_TRACE_HTTP)
__pmNotifyErr(LOG_ERR, "HTTP connect unix(%s) not supported\n", path);
return -EOPNOTSUPP;
#endif
}
static int
http_client_connectto(const char *host, int port, struct timeval *timeout)
{
struct timeval stv, *ptv;
__pmHostEnt *servInfo;
__pmSockAddr *myAddr;
__pmFdSet readyFds, allFds;
void *enumIx;
int fdFlags[FD_SETSIZE];
int i, fd, sts, maxFd;
if ((servInfo = __pmGetAddrInfo(host)) == NULL) {
if (pmDebug & DBG_TRACE_HTTP)
fprintf(stderr, "HTTP connect(%s, %d): hosterror=%d, ``%s''\n",
host, port, hosterror(), hoststrerror());
return -EHOSTUNREACH;
}
/*
* We want to respect the connect timeout that has been configured, but we
* may have more than one address to try. Do this by creating a socket for
* each address and then using __pmSelectWrite() to wait for one of them to
* respond. That way, the timeout is applied to all of the addresses
* simultaneously. First, create the sockets, add them to the fd set and
* try to establish connectections.
*/
__pmFD_ZERO(&allFds);
maxFd = -1;
enumIx = NULL;
for (myAddr = __pmHostEntGetSockAddr(servInfo, &enumIx);
myAddr != NULL;
myAddr = __pmHostEntGetSockAddr(servInfo, &enumIx)) {
/* Create a socket */
if (__pmSockAddrIsInet(myAddr))
fd = __pmCreateSocket();
else if (__pmSockAddrIsIPv6(myAddr))
fd = __pmCreateIPv6Socket();
else {
if (pmDebug & DBG_TRACE_HTTP)
fprintf(stderr, "HTTP connect(%s, %d): bad address family %d\n",
host, port, __pmSockAddrGetFamily(myAddr));
fd = -EINVAL;
}
if (fd < 0) {
__pmSockAddrFree(myAddr);
continue; /* Try the next address */
}
/* Attempt to connect */
fdFlags[fd] = __pmConnectTo(fd, myAddr, port);
__pmSockAddrFree(myAddr);
if (fdFlags[fd] < 0) {
/*
* Mark failure in case we fall out the end of the loop
* and try next address
*/
__pmCloseSocket(fd);
continue;
}
/* Add it to the fd set. */
__pmFD_SET(fd, &allFds);
if (fd > maxFd)
maxFd = fd;
}
__pmHostEntFree(servInfo);
/* If we were unable to open any sockets, then give up. */
if (maxFd == -1)
return -ECONNREFUSED;
/* FNDELAY and we're in progress - wait on select */
__pmFD_COPY(&readyFds, &allFds);
stv = *timeout;
ptv = (stv.tv_sec || stv.tv_usec) ? &stv : NULL;
sts = __pmSelectWrite(maxFd+1, &readyFds, ptv);
/* Figure out what happened. */
if (sts == 0)
fd = -ETIMEDOUT;
else if (sts < 0)
fd = -neterror();
else {
/* Scan fd set, find first successfully connected socket (if any). */
fd = -EINVAL;
for (i = 0; i <= maxFd; ++i) {
if (__pmFD_ISSET(i, &readyFds)) {
/* Successful connection? */
sts = __pmConnectCheckError(i);
if (sts == 0) {
fd = i;
break;
}
fd = -sts;
}
}
}
/* Clean up the unused fds. */
for (i = 0; i <= maxFd; ++i) {
if (i != fd && __pmFD_ISSET(i, &allFds))
__pmCloseSocket(i);
}
if (fd < 0)
return fd;
/*
* If we're here, it means we have a valid connection; restore the
* flags and make sure this file descriptor is closed if exec() is
* called
*/
return __pmConnectRestoreFlags(fd, fdFlags[fd]);
}
static int stomp_connect(const char *hostname, int port)
{
__pmSockAddr *myaddr;
__pmHostEnt *servinfo;
void *enumIx;
struct timeval tv;
struct timeval *ptv;
__pmFdSet wfds;
int ret;
int flags = 0;
if ((servinfo = __pmGetAddrInfo(hostname)) == NULL)
return -1;
fd = -1;
enumIx = NULL;
for (myaddr = __pmHostEntGetSockAddr(servinfo, &enumIx);
myaddr != NULL;
myaddr = __pmHostEntGetSockAddr(servinfo, &enumIx)) {
/* Create a socket */
if (__pmSockAddrIsInet(myaddr))
fd = __pmCreateSocket();
else if (__pmSockAddrIsIPv6(myaddr))
fd = __pmCreateIPv6Socket();
else
fd = -1;
if (fd < 0) {
__pmSockAddrFree(myaddr);
continue; /* Try the next address */
}
/* Attempt to connect */
flags = __pmConnectTo(fd, myaddr, port);
__pmSockAddrFree(myaddr);
if (flags < 0) {
/*
* Mark failure in case we fall out the end of the loop
* and try next address. fd has been closed in __pmConnectTo().
*/
fd = -1;
continue;
}
/* FNDELAY and we're in progress - wait on select */
tv.tv_sec = timeout;
tv.tv_usec = 0;
ptv = (tv.tv_sec || tv.tv_usec) ? &tv : NULL;
__pmFD_ZERO(&wfds);
__pmFD_SET(fd, &wfds);
ret = __pmSelectWrite(fd+1, &wfds, ptv);
/* Was the connection successful? */
if (ret <= 0) {
if (oserror() == EINTR)
return -2;
continue;
}
ret = __pmConnectCheckError(fd);
if (ret == 0)
break;
/* Unsuccessful connection. */
__pmCloseSocket(fd);
fd = -1;
} /* loop over addresses */
__pmHostEntFree(servinfo);
if(fd == -1)
return -4;
fd = __pmConnectRestoreFlags(fd, flags);
if(fd < 0)
return -5;
return fd;
}
int
DoFetch(ClientInfo *cip, __pmPDU* pb)
{
int i, j;
int sts;
int ctxnum;
__pmTimeval when;
int nPmids;
pmID *pmidList;
static pmResult *endResult = NULL;
static int maxnpmids = 0; /* sizes endResult */
DomPmidList *dList; /* NOTE: NOT indexed by agent index */
static int nDoms = 0;
static pmResult **results = NULL;
static int *resIndex = NULL;
__pmFdSet waitFds;
__pmFdSet readyFds;
int nWait;
int maxFd;
struct timeval timeout;
if (nAgents > nDoms) {
if (results != NULL)
free(results);
if (resIndex != NULL)
free(resIndex);
results = (pmResult **)malloc((nAgents + 1) * sizeof (pmResult *));
resIndex = (int *)malloc((nAgents + 1) * sizeof(int));
if (results == NULL || resIndex == NULL) {
__pmNoMem("DoFetch.results", (nAgents + 1) * sizeof (pmResult *) + (nAgents + 1) * sizeof(int), PM_FATAL_ERR);
}
nDoms = nAgents;
}
memset(results, 0, (nAgents + 1) * sizeof(results[0]));
sts = __pmDecodeFetch(pb, &ctxnum, &when, &nPmids, &pmidList);
if (sts < 0)
return sts;
/* Check that a profile has been received from the specified context */
if (ctxnum < 0 || ctxnum >= cip->szProfile ||
cip->profile[ctxnum] == NULL) {
__pmUnpinPDUBuf(pb);
if (ctxnum < 0 || ctxnum >= cip->szProfile)
__pmNotifyErr(LOG_ERR, "DoFetch: bad ctxnum=%d\n", ctxnum);
else
__pmNotifyErr(LOG_ERR, "DoFetch: no profile for ctxnum=%d\n", ctxnum);
return PM_ERR_NOPROFILE;
}
if (nPmids > maxnpmids) {
int need;
if (endResult != NULL)
free(endResult);
need = (int)sizeof(pmResult) + (nPmids - 1) * (int)sizeof(pmValueSet *);
if ((endResult = (pmResult *)malloc(need)) == NULL) {
__pmNoMem("DoFetch.endResult", need, PM_FATAL_ERR);
}
maxnpmids = nPmids;
}
dList = SplitPmidList(nPmids, pmidList);
/* For each domain in the split pmidList, dispatch the per-domain subset
* of pmIDs to the appropriate agent. For DSO agents, the pmResult will
* come back immediately. If a request cannot be sent to an agent, a
* suitable pmResult (containing metric not available values) will be
* returned.
*/
__pmFD_ZERO(&waitFds);
nWait = 0;
maxFd = -1;
for (i = 0; dList[i].domain != -1; i++) {
j = mapdom[dList[i].domain];
results[j] = SendFetch(&dList[i], &agent[j], cip, ctxnum);
if (results[j] == NULL) { /* Wait for agent's response */
int fd = agent[j].outFd;
agent[j].status.busy = 1;
__pmFD_SET(fd, &waitFds);
if (fd > maxFd)
maxFd = fd;
nWait++;
}
}
/* Construct pmResult for bad-pmID list */
if (dList[i].listSize != 0)
results[nAgents] = MakeBadResult(dList[i].listSize, dList[i].list, PM_ERR_NOAGENT);
/* Wait for results to roll in from agents */
while (nWait > 0) {
__pmFD_COPY(&readyFds, &waitFds);
if (nWait > 1) {
timeout.tv_sec = _pmcd_timeout;
timeout.tv_usec = 0;
retry:
setoserror(0);
sts = __pmSelectRead(maxFd+1, &readyFds, &timeout);
if (sts == 0) {
__pmNotifyErr(LOG_INFO, "DoFetch: select timeout");
/* Timeout, terminate agents with undelivered results */
for (i = 0; i < nAgents; i++) {
if (agent[i].status.busy) {
/* Find entry in dList for this agent */
for (j = 0; dList[j].domain != -1; j++)
if (dList[j].domain == agent[i].pmDomainId)
break;
results[i] = MakeBadResult(dList[j].listSize,
dList[j].list,
PM_ERR_NOAGENT);
pmcd_trace(TR_RECV_TIMEOUT, agent[i].outFd, PDU_RESULT, 0);
CleanupAgent(&agent[i], AT_COMM, agent[i].inFd);
}
}
break;
}
else if (sts < 0) {
if (neterror() == EINTR)
goto retry;
/* this is not expected to happen! */
__pmNotifyErr(LOG_ERR, "DoFetch: fatal select failure: %s\n",
netstrerror());
Shutdown();
exit(1);
}
}
/* Read results from agents that have them ready */
for (i = 0; i < nAgents; i++) {
AgentInfo *ap = &agent[i];
int pinpdu;
if (!ap->status.busy || !__pmFD_ISSET(ap->outFd, &readyFds))
continue;
ap->status.busy = 0;
__pmFD_CLR(ap->outFd, &waitFds);
nWait--;
pinpdu = sts = __pmGetPDU(ap->outFd, ANY_SIZE, _pmcd_timeout, &pb);
if (sts > 0)
pmcd_trace(TR_RECV_PDU, ap->outFd, sts, (int)((__psint_t)pb & 0xffffffff));
if (sts == PDU_RESULT) {
if ((sts = __pmDecodeResult(pb, &results[i])) >= 0)
if (results[i]->numpmid != aFreq[i]) {
pmFreeResult(results[i]);
sts = PM_ERR_IPC;
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL0)
__pmNotifyErr(LOG_ERR, "DoFetch: \"%s\" agent given %d pmIDs, returned %d\n",
ap->pmDomainLabel, aFreq[i], results[i]->numpmid);
#endif
}
}
else {
if (sts == PDU_ERROR) {
int s;
if ((s = __pmDecodeError(pb, &sts)) < 0)
sts = s;
else if (sts >= 0)
sts = PM_ERR_GENERIC;
pmcd_trace(TR_RECV_ERR, ap->outFd, PDU_RESULT, sts);
}
else if (sts >= 0) {
pmcd_trace(TR_WRONG_PDU, ap->outFd, PDU_RESULT, sts);
sts = PM_ERR_IPC;
}
}
if (pinpdu > 0)
__pmUnpinPDUBuf(pb);
if (sts < 0) {
/* Find entry in dList for this agent */
for (j = 0; dList[j].domain != -1; j++)
if (dList[j].domain == agent[i].pmDomainId)
break;
results[i] = MakeBadResult(dList[j].listSize,
dList[j].list, sts);
if (sts == PM_ERR_PMDANOTREADY) {
/* the agent is indicating it can't handle PDUs for now */
int k;
extern int CheckError(AgentInfo *ap, int sts);
for (k = 0; k < dList[j].listSize; k++)
results[i]->vset[k]->numval = PM_ERR_AGAIN;
sts = CheckError(&agent[i], sts);
}
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL0) {
fprintf(stderr, "RESULT error from \"%s\" agent : %s\n",
ap->pmDomainLabel, pmErrStr(sts));
}
#endif
if (sts == PM_ERR_IPC || sts == PM_ERR_TIMEOUT)
CleanupAgent(ap, AT_COMM, ap->outFd);
}
}
}
endResult->numpmid = nPmids;
__pmtimevalNow(&endResult->timestamp);
/* The order of the pmIDs in the per-domain results is the same as in the
* original request, but on a per-domain basis. resIndex is an array of
* indices (one per agent) of the next metric to be retrieved from each
* per-domain result's vset.
*/
memset(resIndex, 0, (nAgents + 1) * sizeof(resIndex[0]));
for (i = 0; i < nPmids; i++) {
j = mapdom[((__pmID_int *)&pmidList[i])->domain];
endResult->vset[i] = results[j]->vset[resIndex[j]++];
}
pmcd_trace(TR_XMIT_PDU, cip->fd, PDU_RESULT, endResult->numpmid);
sts = 0;
if (cip->status.changes) {
/* notify client of PMCD state change */
sts = __pmSendError(cip->fd, FROM_ANON, (int)cip->status.changes);
if (sts > 0)
sts = 0;
cip->status.changes = 0;
}
if (sts == 0)
sts = __pmSendResult(cip->fd, FROM_ANON, endResult);
if (sts < 0) {
pmcd_trace(TR_XMIT_ERR, cip->fd, PDU_RESULT, sts);
CleanupClient(cip, sts);
}
/*
* pmFreeResult() all the accumulated results.
*/
for (i = 0; dList[i].domain != -1; i++) {
j = mapdom[dList[i].domain];
if (agent[j].ipcType == AGENT_DSO && agent[j].status.connected &&
!agent[j].status.madeDsoResult)
/* Living DSO's manage their own pmResult skeleton unless
* MakeBadResult was called to create the result. The value sets
* within the skeleton need to be freed though!
*/
__pmFreeResultValues(results[j]);
else
/* For others it is dynamically allocated in __pmDecodeResult or
* MakeBadResult
*/
pmFreeResult(results[j]);
}
if (results[nAgents] != NULL)
pmFreeResult(results[nAgents]);
__pmUnpinPDUBuf(pmidList);
return 0;
}
int
main(int argc, char **argv)
{
int c;
int sts;
int sep = __pmPathSeparator();
int use_localtime = 0;
int isdaemon = 0;
char *pmnsfile = PM_NS_DEFAULT;
char *username;
char *logfile = "pmlogger.log";
/* default log (not archive) file name */
char *endnum;
int i;
task_t *tp;
optcost_t ocp;
__pmFdSet readyfds;
char *p;
char *runtime = NULL;
int ctx; /* handle corresponding to ctxp below */
__pmContext *ctxp; /* pmlogger has just this one context */
int niter;
pid_t target_pid = 0;
__pmGetUsername(&username);
/*
* Warning:
* If any of the pmlogger options change, make sure the
* corresponding changes are made to pmnewlog when pmlogger
* options are passed through from the control file
*/
while ((c = pmgetopt_r(argc, argv, &opts)) != EOF) {
switch (c) {
case 'c': /* config file */
if (access(opts.optarg, F_OK) == 0)
configfile = opts.optarg;
else {
/* does not exist as given, try the standard place */
char *sysconf = pmGetConfig("PCP_VAR_DIR");
int sz = strlen(sysconf)+strlen("/config/pmlogger/")+strlen(opts.optarg)+1;
if ((configfile = (char *)malloc(sz)) == NULL)
__pmNoMem("config file name", sz, PM_FATAL_ERR);
snprintf(configfile, sz,
"%s%c" "config%c" "pmlogger%c" "%s",
sysconf, sep, sep, sep, opts.optarg);
if (access(configfile, F_OK) != 0) {
/* still no good, error handling happens below */
free(configfile);
configfile = opts.optarg;
}
}
break;
case 'D': /* debug flag */
sts = __pmParseDebug(opts.optarg);
if (sts < 0) {
pmprintf("%s: unrecognized debug flag specification (%s)\n",
pmProgname, opts.optarg);
opts.errors++;
}
else
pmDebug |= sts;
break;
case 'h': /* hostname for PMCD to contact */
pmcd_host_conn = opts.optarg;
break;
case 'l': /* log file name */
logfile = opts.optarg;
break;
case 'L': /* linger if not primary logger */
linger = 1;
break;
case 'm': /* note for port map file */
note = opts.optarg;
isdaemon = ((strcmp(note, "pmlogger_check") == 0) ||
(strcmp(note, "pmlogger_daily") == 0));
break;
case 'n': /* alternative name space file */
pmnsfile = opts.optarg;
break;
case 'p':
target_pid = (int)strtol(opts.optarg, &endnum, 10);
if (*endnum != '\0') {
pmprintf("%s: invalid process identifier (%s)\n",
pmProgname, opts.optarg);
opts.errors++;
} else if (!__pmProcessExists(target_pid)) {
pmprintf("%s: PID error - no such process (%d)\n",
pmProgname, target_pid);
opts.errors++;
}
break;
case 'P': /* this is the primary pmlogger */
primary = 1;
isdaemon = 1;
break;
case 'r': /* report sizes of pmResult records */
rflag = 1;
break;
case 's': /* exit size */
sts = ParseSize(opts.optarg, &exit_samples, &exit_bytes, &exit_time);
if (sts < 0) {
pmprintf("%s: illegal size argument '%s' for exit size\n",
pmProgname, opts.optarg);
opts.errors++;
}
else if (exit_time.tv_sec > 0) {
__pmAFregister(&exit_time, NULL, run_done_callback);
}
break;
case 'T': /* end time */
runtime = opts.optarg;
break;
case 't': /* change default logging interval */
if (pmParseInterval(opts.optarg, &delta, &p) < 0) {
pmprintf("%s: illegal -t argument\n%s", pmProgname, p);
free(p);
opts.errors++;
}
break;
case 'U': /* run as named user */
username = opts.optarg;
isdaemon = 1;
break;
case 'u': /* flush output buffers after each fetch */
/*
* all archive write I/O is unbuffered now, so maintain -u
* for backwards compatibility only
*/
break;
case 'v': /* volume switch after given size */
sts = ParseSize(opts.optarg, &vol_switch_samples, &vol_switch_bytes,
&vol_switch_time);
if (sts < 0) {
pmprintf("%s: illegal size argument '%s' for volume size\n",
pmProgname, opts.optarg);
opts.errors++;
}
else if (vol_switch_time.tv_sec > 0) {
vol_switch_afid = __pmAFregister(&vol_switch_time, NULL,
vol_switch_callback);
}
break;
case 'V':
archive_version = (int)strtol(opts.optarg, &endnum, 10);
if (*endnum != '\0' || archive_version != PM_LOG_VERS02) {
pmprintf("%s: -V requires a version number of %d\n",
pmProgname, PM_LOG_VERS02);
opts.errors++;
}
break;
case 'x': /* recording session control fd */
rsc_fd = (int)strtol(opts.optarg, &endnum, 10);
if (*endnum != '\0' || rsc_fd < 0) {
pmprintf("%s: -x requires a non-negative numeric argument\n", pmProgname);
opts.errors++;
}
else {
time(&rsc_start);
}
break;
case 'y':
use_localtime = 1;
break;
case '?':
default:
opts.errors++;
break;
}
}
if (primary && pmcd_host != NULL) {
pmprintf(
"%s: -P and -h are mutually exclusive; use -P only when running\n"
"%s on the same (local) host as the PMCD to which it connects.\n",
pmProgname, pmProgname);
opts.errors++;
}
if (!opts.errors && opts.optind != argc - 1) {
pmprintf("%s: insufficient arguments\n", pmProgname);
opts.errors++;
}
if (opts.errors) {
pmUsageMessage(&opts);
exit(1);
}
if (rsc_fd != -1 && note == NULL) {
/* add default note to indicate running with -x */
static char xnote[10];
snprintf(xnote, sizeof(xnote), "-x %d", rsc_fd);
note = xnote;
}
/* if we are running as a daemon, change user early */
if (isdaemon)
__pmSetProcessIdentity(username);
__pmOpenLog("pmlogger", logfile, stderr, &sts);
if (sts != 1) {
fprintf(stderr, "%s: Warning: log file (%s) creation failed\n", pmProgname, logfile);
/* continue on ... writing to stderr */
}
/* base name for archive is here ... */
archBase = argv[opts.optind];
if (pmcd_host_conn == NULL)
pmcd_host_conn = "local:";
/* initialise access control */
if (__pmAccAddOp(PM_OP_LOG_ADV) < 0 ||
__pmAccAddOp(PM_OP_LOG_MAND) < 0 ||
__pmAccAddOp(PM_OP_LOG_ENQ) < 0) {
fprintf(stderr, "%s: access control initialisation failed\n", pmProgname);
exit(1);
}
if (pmnsfile != PM_NS_DEFAULT) {
if ((sts = pmLoadASCIINameSpace(pmnsfile, 1)) < 0) {
fprintf(stderr, "%s: Cannot load namespace from \"%s\": %s\n", pmProgname, pmnsfile, pmErrStr(sts));
exit(1);
}
}
if ((ctx = pmNewContext(PM_CONTEXT_HOST, pmcd_host_conn)) < 0) {
fprintf(stderr, "%s: Cannot connect to PMCD on host \"%s\": %s\n", pmProgname, pmcd_host_conn, pmErrStr(ctx));
exit(1);
}
pmcd_host = (char *)pmGetContextHostName(ctx);
if (strlen(pmcd_host) == 0) {
fprintf(stderr, "%s: pmGetContextHostName(%d) failed\n",
pmProgname, ctx);
exit(1);
}
if (rsc_fd == -1) {
/* no -x, so register client id with pmcd */
__pmSetClientIdArgv(argc, argv);
}
/*
* discover fd for comms channel to PMCD ...
*/
if ((ctxp = __pmHandleToPtr(ctx)) == NULL) {
fprintf(stderr, "%s: botch: __pmHandleToPtr(%d) returns NULL!\n", pmProgname, ctx);
exit(1);
}
pmcdfd = ctxp->c_pmcd->pc_fd;
PM_UNLOCK(ctxp->c_lock);
if (configfile != NULL) {
if ((yyin = fopen(configfile, "r")) == NULL) {
fprintf(stderr, "%s: Cannot open config file \"%s\": %s\n",
pmProgname, configfile, osstrerror());
exit(1);
}
}
else {
/* **ANY** Lex would read from stdin automagically */
configfile = "<stdin>";
}
__pmOptFetchGetParams(&ocp);
ocp.c_scope = 1;
__pmOptFetchPutParams(&ocp);
/* prevent early timer events ... */
__pmAFblock();
if (yyparse() != 0)
exit(1);
if (configfile != NULL)
fclose(yyin);
yyend();
#ifdef PCP_DEBUG
fprintf(stderr, "Config parsed\n");
#endif
fprintf(stderr, "Starting %slogger for host \"%s\" via \"%s\"\n",
primary ? "primary " : "", pmcd_host, pmcd_host_conn);
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_LOG) {
fprintf(stderr, "optFetch Cost Parameters: pmid=%d indom=%d fetch=%d scope=%d\n",
ocp.c_pmid, ocp.c_indom, ocp.c_fetch, ocp.c_scope);
fprintf(stderr, "\nAfter loading config ...\n");
for (tp = tasklist; tp != NULL; tp = tp->t_next) {
if (tp->t_numvalid == 0)
continue;
fprintf(stderr, " state: %sin log, %savail, %s, %s",
PMLC_GET_INLOG(tp->t_state) ? "" : "not ",
PMLC_GET_AVAIL(tp->t_state) ? "" : "un",
PMLC_GET_MAND(tp->t_state) ? "mand" : "adv",
PMLC_GET_ON(tp->t_state) ? "on" : "off");
fprintf(stderr, " delta: %ld usec",
(long)1000 * tp->t_delta.tv_sec + tp->t_delta.tv_usec);
fprintf(stderr, " numpmid: %d\n", tp->t_numpmid);
for (i = 0; i < tp->t_numpmid; i++) {
fprintf(stderr, " %s (%s):\n", pmIDStr(tp->t_pmidlist[i]), tp->t_namelist[i]);
}
__pmOptFetchDump(stderr, tp->t_fetch);
}
}
#endif
if (!primary && tasklist == NULL && !linger) {
fprintf(stderr, "Nothing to log, and not the primary logger instance ... good-bye\n");
exit(1);
}
if ((sts = __pmLogCreate(pmcd_host, archBase, archive_version, &logctl)) < 0) {
fprintf(stderr, "__pmLogCreate: %s\n", pmErrStr(sts));
exit(1);
}
else {
/*
* try and establish $TZ from the remote PMCD ...
* Note the label record has been set up, but not written yet
*/
char *name = "pmcd.timezone";
pmID pmid;
pmResult *resp;
__pmtimevalNow(&epoch);
sts = pmUseContext(ctx);
if (sts >= 0)
sts = pmLookupName(1, &name, &pmid);
if (sts >= 0)
sts = pmFetch(1, &pmid, &resp);
if (sts >= 0) {
if (resp->vset[0]->numval > 0) { /* pmcd.timezone present */
strcpy(logctl.l_label.ill_tz, resp->vset[0]->vlist[0].value.pval->vbuf);
/* prefer to use remote time to avoid clock drift problems */
epoch = resp->timestamp; /* struct assignment */
if (! use_localtime)
pmNewZone(logctl.l_label.ill_tz);
}
#ifdef PCP_DEBUG
else if (pmDebug & DBG_TRACE_LOG) {
fprintf(stderr,
"main: Could not get timezone from host %s\n",
pmcd_host);
}
#endif
pmFreeResult(resp);
}
}
/* do ParseTimeWindow stuff for -T */
if (runtime) {
struct timeval res_end; /* time window end */
struct timeval start;
struct timeval end;
struct timeval last_delta;
char *err_msg; /* parsing error message */
time_t now;
struct timeval now_tv;
time(&now);
now_tv.tv_sec = now;
now_tv.tv_usec = 0;
start = now_tv;
end.tv_sec = INT_MAX;
end.tv_usec = INT_MAX;
sts = __pmParseTime(runtime, &start, &end, &res_end, &err_msg);
if (sts < 0) {
fprintf(stderr, "%s: illegal -T argument\n%s", pmProgname, err_msg);
exit(1);
}
last_delta = res_end;
tsub(&last_delta, &now_tv);
__pmAFregister(&last_delta, NULL, run_done_callback);
last_stamp = res_end;
}
fprintf(stderr, "Archive basename: %s\n", archBase);
#ifndef IS_MINGW
/* detach yourself from the launching process */
if (isdaemon)
setpgid(getpid(), 0);
#endif
/* set up control port */
init_ports();
__pmFD_ZERO(&fds);
for (i = 0; i < CFD_NUM; ++i) {
if (ctlfds[i] >= 0)
__pmFD_SET(ctlfds[i], &fds);
}
#ifndef IS_MINGW
__pmFD_SET(pmcdfd, &fds);
#endif
if (rsc_fd != -1)
__pmFD_SET(rsc_fd, &fds);
numfds = maxfd() + 1;
if ((sts = do_preamble()) < 0)
fprintf(stderr, "Warning: problem writing archive preamble: %s\n",
pmErrStr(sts));
sts = 0; /* default exit status */
parse_done = 1; /* enable callback processing */
__pmAFunblock();
for ( ; ; ) {
int nready;
#ifdef PCP_DEBUG
if ((pmDebug & DBG_TRACE_APPL2) && (pmDebug & DBG_TRACE_DESPERATE)) {
fprintf(stderr, "before __pmSelectRead(%d,...): run_done_alarm=%d vol_switch_alarm=%d log_alarm=%d\n", numfds, run_done_alarm, vol_switch_alarm, log_alarm);
}
#endif
niter = 0;
while (log_alarm && niter++ < 10) {
__pmAFblock();
log_alarm = 0;
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL2)
fprintf(stderr, "delayed callback: log_alarm\n");
#endif
for (tp = tasklist; tp != NULL; tp = tp->t_next) {
if (tp->t_alarm) {
tp->t_alarm = 0;
do_work(tp);
}
}
__pmAFunblock();
}
if (vol_switch_alarm) {
__pmAFblock();
vol_switch_alarm = 0;
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL2)
fprintf(stderr, "delayed callback: vol_switch_alarm\n");
#endif
newvolume(VOL_SW_TIME);
__pmAFunblock();
}
if (run_done_alarm) {
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL2)
fprintf(stderr, "delayed callback: run_done_alarm\n");
#endif
run_done(0, NULL);
/*NOTREACHED*/
}
__pmFD_COPY(&readyfds, &fds);
nready = __pmSelectRead(numfds, &readyfds, NULL);
#ifdef PCP_DEBUG
if ((pmDebug & DBG_TRACE_APPL2) && (pmDebug & DBG_TRACE_DESPERATE)) {
fprintf(stderr, "__pmSelectRead(%d,...) done: nready=%d run_done_alarm=%d vol_switch_alarm=%d log_alarm=%d\n", numfds, nready, run_done_alarm, vol_switch_alarm, log_alarm);
}
#endif
__pmAFblock();
if (nready > 0) {
/* handle request on control port */
for (i = 0; i < CFD_NUM; ++i) {
if (ctlfds[i] >= 0 && __pmFD_ISSET(ctlfds[i], &readyfds)) {
if (control_req(ctlfds[i])) {
/* new client has connected */
__pmFD_SET(clientfd, &fds);
if (clientfd >= numfds)
numfds = clientfd + 1;
}
}
}
if (clientfd >= 0 && __pmFD_ISSET(clientfd, &readyfds)) {
/* process request from client, save clientfd in case client
* closes connection, resetting clientfd to -1
*/
int fd = clientfd;
if (client_req()) {
/* client closed connection */
__pmFD_CLR(fd, &fds);
__pmCloseSocket(clientfd);
clientfd = -1;
numfds = maxfd() + 1;
qa_case = 0;
}
}
#ifndef IS_MINGW
if (pmcdfd >= 0 && __pmFD_ISSET(pmcdfd, &readyfds)) {
/*
* do not expect this, given synchronous commumication with the
* pmcd ... either pmcd has terminated, or bogus PDU ... or its
* Win32 and we are operating under the different conditions of
* our AF.c implementation there, which has to deal with a lack
* of signal support on Windows - race condition exists between
* this check and the async event timer callback.
*/
__pmPDU *pb;
__pmPDUHdr *php;
sts = __pmGetPDU(pmcdfd, ANY_SIZE, TIMEOUT_NEVER, &pb);
if (sts <= 0) {
if (sts < 0)
fprintf(stderr, "Error: __pmGetPDU: %s\n", pmErrStr(sts));
disconnect(sts);
}
else {
php = (__pmPDUHdr *)pb;
fprintf(stderr, "Error: Unsolicited %s PDU from PMCD\n",
__pmPDUTypeStr(php->type));
disconnect(PM_ERR_IPC);
}
if (sts > 0)
__pmUnpinPDUBuf(pb);
}
#endif
if (rsc_fd >= 0 && __pmFD_ISSET(rsc_fd, &readyfds)) {
/*
* some action on the recording session control fd
* end-of-file means launcher has quit, otherwise we
* expect one of these commands
* V<number>\n - version
* F<folio>\n - folio name
* P<name>\n - launcher's name
* R\n - launcher can replay
* D\n - detach from launcher
* Q\n - quit pmlogger
*/
char rsc_buf[MAXPATHLEN];
char *rp = rsc_buf;
char myc;
int fake_x = 0;
for (rp = rsc_buf; ; rp++) {
if (read(rsc_fd, &myc, 1) <= 0) {
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL2)
fprintf(stderr, "recording session control: eof\n");
#endif
if (rp != rsc_buf) {
*rp = '\0';
fprintf(stderr, "Error: incomplete recording session control message: \"%s\"\n", rsc_buf);
}
fake_x = 1;
break;
}
if (rp >= &rsc_buf[MAXPATHLEN]) {
fprintf(stderr, "Error: absurd recording session control message: \"%100.100s ...\"\n", rsc_buf);
fake_x = 1;
break;
}
if (myc == '\n') {
*rp = '\0';
break;
}
*rp = myc;
}
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL2) {
if (fake_x == 0)
fprintf(stderr, "recording session control: \"%s\"\n", rsc_buf);
}
#endif
if (fake_x)
do_dialog('X');
else if (strcmp(rsc_buf, "Q") == 0 ||
strcmp(rsc_buf, "D") == 0 ||
strcmp(rsc_buf, "?") == 0)
do_dialog(rsc_buf[0]);
else if (rsc_buf[0] == 'F')
folio_name = strdup(&rsc_buf[1]);
else if (rsc_buf[0] == 'P')
rsc_prog = strdup(&rsc_buf[1]);
else if (strcmp(rsc_buf, "R") == 0)
rsc_replay = 1;
else if (rsc_buf[0] == 'V' && rsc_buf[1] == '0') {
/*
* version 0 of the recording session control ...
* this is all we grok at the moment
*/
;
}
else {
fprintf(stderr, "Error: illegal recording session control message: \"%s\"\n", rsc_buf);
do_dialog('X');
}
}
}
else if (vol_switch_flag) {
newvolume(VOL_SW_SIGHUP);
vol_switch_flag = 0;
}
else if (nready < 0 && neterror() != EINTR)
fprintf(stderr, "Error: select: %s\n", netstrerror());
__pmAFunblock();
if (target_pid && !__pmProcessExists(target_pid))
exit(EXIT_SUCCESS);
if (exit_code)
break;
}
exit(exit_code);
}
static int
pduread(int fd, char *buf, int len, int mode, int timeout)
{
/*
* handle short reads that may split a PDU ...
*/
int status = 0;
int have = 0;
__pmFdSet onefd;
static int done_default = 0;
static struct timeval def_wait = { 10, 0 };
if (timeout == TRACE_TIMEOUT_DEFAULT) {
if (!done_default) {
double def_timeout;
char *timeout_str;
char *end_ptr;
if ((timeout_str = getenv(TRACE_ENV_REQTIMEOUT)) != NULL) {
def_timeout = strtod(timeout_str, &end_ptr);
if (*end_ptr != '\0' || def_timeout < 0.0) {
status = PMTRACE_ERR_ENVFORMAT;
return status;
}
else {
pmtimevalFromReal(def_timeout, &def_wait);
}
}
done_default = 1;
}
}
while (len) {
struct timeval wait;
/*
* either never timeout (i.e. block forever), or timeout
*/
if (timeout != TRACE_TIMEOUT_NEVER) {
if (timeout > 0) {
wait.tv_sec = timeout;
wait.tv_usec = 0;
}
else
wait = def_wait;
__pmFD_ZERO(&onefd);
__pmFD_SET(fd, &onefd);
status = __pmSelectRead(fd+1, &onefd, &wait);
if (status == 0)
return PMTRACE_ERR_TIMEOUT;
else if (status < 0) {
setoserror(neterror());
return status;
}
}
status = (int)__pmRead(fd, buf, len);
if (status <= 0) { /* EOF or error */
setoserror(neterror());
return status;
}
if (mode == -1)
/* special case, see __pmtracegetPDU */
return status;
have += status;
buf += status;
len -= status;
}
return have;
}
int
conn_cisco(cisco_t * cp)
{
__pmFdSet wfds;
__pmSockAddr *myaddr;
void *enumIx;
int flags = 0;
int fd;
int ret;
fd = -1;
enumIx = NULL;
for (myaddr = __pmHostEntGetSockAddr(cp->hostinfo, &enumIx);
myaddr != NULL;
myaddr = __pmHostEntGetSockAddr(cp->hostinfo, &enumIx)) {
/* Create a socket */
if (__pmSockAddrIsInet(myaddr))
fd = __pmCreateSocket();
else if (__pmSockAddrIsIPv6(myaddr))
fd = __pmCreateIPv6Socket();
else
fd = -1;
if (fd < 0) {
__pmSockAddrFree(myaddr);
continue; /* Try the next address */
}
/* Attempt to connect */
flags = __pmConnectTo(fd, myaddr, cp->port);
__pmSockAddrFree(myaddr);
if (flags < 0) {
/*
* Mark failure in case we fall out the end of the loop
* and try next address. fd has been closed in __pmConnectTo().
*/
setoserror(ECONNREFUSED);
fd = -1;
continue;
}
/* FNDELAY and we're in progress - wait on select */
__pmFD_ZERO(&wfds);
__pmFD_SET(fd, &wfds);
ret = __pmSelectWrite(fd+1, &wfds, NULL);
/* Was the connection successful? */
if (ret == 0)
setoserror(ETIMEDOUT);
else if (ret > 0) {
ret = __pmConnectCheckError(fd);
if (ret == 0)
break;
setoserror(ret);
}
/* Unsuccessful connection. */
__pmCloseSocket(fd);
fd = -1;
} /* loop over addresses */
if (fd == -1) {
fprintf(stderr, "conn_cisco(%s): connect: %s\n",
cp->host, netstrerror());
return -1;
}
fd = __pmConnectRestoreFlags(fd, flags);
if (fd < 0) {
fprintf(stderr, "conn_cisco(%s): setsockopt: %s\n",
cp->host, netstrerror());
return -1;
}
return fd;
}
static int
_pmauxtraceconnect(void)
{
int port = TRACE_PORT;
char hostname[MAXHOSTNAMELEN];
struct timeval timeout = { 3, 0 }; /* default 3 secs */
__pmSockAddr *myaddr;
__pmHostEnt *servinfo;
void *enumIx;
#ifndef IS_MINGW
struct itimerval _pmolditimer;
void (*old_handler)(int foo);
#endif
int rc, sts;
int flags = 0;
char *sptr, *endptr, *endnum;
struct timeval canwait = { 5, 000000 };
struct timeval stv;
struct timeval *pstv;
__pmFdSet wfds;
#ifdef PMTRACE_DEBUG
if (__pmstate & PMTRACE_STATE_NOAGENT) {
fprintf(stderr, "_pmtraceconnect: connecting to PMDA (skipped)\n");
return 0;
}
else if (__pmstate & PMTRACE_STATE_COMMS)
fprintf(stderr, "_pmtraceconnect: connecting to PMDA ...\n");
#endif
/*
* get optional stuff from environment ...
* PCP_TRACE_HOST, PCP_TRACE_PORT, PCP_TRACE_TIMEOUT, and
* PCP_TRACE_NOAGENT
*/
if ((sptr = getenv(TRACE_ENV_HOST)) != NULL)
strcpy(hostname, sptr);
else {
(void)gethostname(hostname, MAXHOSTNAMELEN);
hostname[MAXHOSTNAMELEN-1] = '\0';
}
if ((sptr = getenv(TRACE_ENV_PORT)) != NULL) {
port = (int)strtol(sptr, &endnum, 0);
if (*endnum != '\0' || port < 0) {
fprintf(stderr, "trace warning: bad PCP_TRACE_PORT ignored.");
port = TRACE_PORT;
}
}
if ((sptr = getenv(TRACE_ENV_TIMEOUT)) != NULL) {
double timesec = strtod(sptr, &endptr);
if (*endptr != '\0' || timesec < 0.0)
fprintf(stderr, "trace warning: bogus PCP_TRACE_TIMEOUT.");
else {
timeout.tv_sec = (time_t)timesec;
timeout.tv_usec = (int)((timesec - (double)timeout.tv_sec)*1000000);
}
}
if (getenv(TRACE_ENV_NOAGENT) != NULL)
__pmstate |= PMTRACE_STATE_NOAGENT;
if ((servinfo = __pmGetAddrInfo(hostname)) == NULL) {
#ifdef PMTRACE_DEBUG
if (__pmstate & PMTRACE_STATE_COMMS)
fprintf(stderr, "_pmtraceconnect(__pmGetAddrInfo(hostname=%s): "
"hosterror=%d, ``%s''\n", hostname, hosterror(),
hoststrerror());
#endif
return -EHOSTUNREACH;
}
/* Try each address in turn until one connects. */
sts = EHOSTUNREACH;
__pmfd = -1;
enumIx = NULL;
for (myaddr = __pmHostEntGetSockAddr(servinfo, &enumIx);
myaddr != NULL;
myaddr = __pmHostEntGetSockAddr(servinfo, &enumIx)) {
/* Create a socket */
if (__pmSockAddrIsInet(myaddr))
__pmfd = __pmCreateSocket();
else if (__pmSockAddrIsIPv6(myaddr))
__pmfd = __pmCreateIPv6Socket();
else {
fprintf(stderr, "_pmtraceconnect(invalid address family): %d\n",
__pmSockAddrGetFamily(myaddr));
}
if (__pmfd < 0) {
sts = neterror();
__pmSockAddrFree(myaddr);
continue; /* Try the next address */
}
/* Set the port. */
__pmSockAddrSetPort(myaddr, port);
#ifndef IS_MINGW
/* arm interval timer */
_pmmyitimer.it_value.tv_sec = timeout.tv_sec;
_pmmyitimer.it_value.tv_usec = timeout.tv_usec;
_pmmyitimer.it_interval.tv_sec = 0;
_pmmyitimer.it_interval.tv_usec = 0;
old_handler = signal(SIGALRM, _pmtracealarm);
setitimer(ITIMER_REAL, &_pmmyitimer, &_pmolditimer);
#endif
#ifdef PMTRACE_DEBUG
if (__pmstate & PMTRACE_STATE_COMMS) {
char *name = __pmHostEntGetName (servinfo);
fprintf(stderr, "_pmtraceconnect: PMDA host=%s port=%d timeout=%d"
"secs\n", name == NULL ? "unknown" : name, port, (int)timeout.tv_sec);
if (name != NULL)
free(name);
}
#endif
/* Attempt to connect */
flags = __pmConnectTo(__pmfd, myaddr, port);
__pmSockAddrFree(myaddr);
if (flags < 0) {
/*
* Mark failure in case we fall out the end of the loop
* and try next address. __pmfd has been closed in __pmConnectTo().
*/
sts = -flags;
__pmfd = -1;
continue;
}
/* FNDELAY and we're in progress - wait on select */
stv = canwait;
pstv = (stv.tv_sec || stv.tv_usec) ? &stv : NULL;
__pmFD_ZERO(&wfds);
__pmFD_SET(__pmfd, &wfds);
if ((rc = __pmSelectWrite(__pmfd+1, &wfds, pstv)) == 1) {
sts = __pmConnectCheckError(__pmfd);
}
else if (rc == 0) {
sts = ETIMEDOUT;
}
else {
sts = (rc < 0) ? neterror() : EINVAL;
}
#ifndef IS_MINGW
/* re-arm interval timer */
setitimer(ITIMER_REAL, &off_itimer, &_pmmyitimer);
signal(SIGALRM, old_handler);
if (_pmolditimer.it_value.tv_sec != 0 && _pmolditimer.it_value.tv_usec != 0) {
_pmolditimer.it_value.tv_usec -= timeout.tv_usec - _pmmyitimer.it_value.tv_usec;
while (_pmolditimer.it_value.tv_usec < 0) {
_pmolditimer.it_value.tv_usec += 1000000;
_pmolditimer.it_value.tv_sec--;
}
while (_pmolditimer.it_value.tv_usec > 1000000) {
_pmolditimer.it_value.tv_usec -= 1000000;
_pmolditimer.it_value.tv_sec++;
}
_pmolditimer.it_value.tv_sec -= timeout.tv_sec - _pmmyitimer.it_value.tv_sec;
if (_pmolditimer.it_value.tv_sec < 0) {
/* missed the user's itimer, pretend there is 1 msec to go! */
_pmolditimer.it_value.tv_sec = 0;
_pmolditimer.it_value.tv_usec = 1000;
}
setitimer(ITIMER_REAL, &_pmolditimer, &_pmmyitimer);
}
#endif
/* Was the connection successful? */
if (sts == 0)
break;
/* Unsuccessful connection. */
__pmCloseSocket(__pmfd);
__pmfd = -1;
} /* loop over addresses */
__pmHostEntFree(servinfo);
/* Was the connection successful? */
if (__pmfd < 0) {
#ifdef PMTRACE_DEBUG
if (__pmstate & PMTRACE_STATE_COMMS)
fprintf(stderr, "_pmtraceconnect(socket failed): %s\n",
netstrerror());
#endif
return -sts;
}
_pmtimedout = 0;
/* Restore the original file status flags. */
if (__pmSetFileStatusFlags(__pmfd, flags) < 0) {
#ifdef PMTRACE_DEBUG
if (__pmstate & PMTRACE_STATE_COMMS)
fprintf(stderr, ":_pmtraceconnect: cannot restore file status flags\n");
#endif
return -oserror();
}
/* make sure this file descriptor is closed if exec() is called */
if ((flags = __pmGetFileDescriptorFlags(__pmfd)) != -1)
sts = __pmSetFileDescriptorFlags(__pmfd, flags | FD_CLOEXEC);
else
sts = -1;
if (sts == -1)
return -oserror();
if (__pmtraceprotocol(TRACE_PROTOCOL_QUERY) == TRACE_PROTOCOL_ASYNC) {
/* in the asynchronoous protocol - ensure no delay after close */
if ((flags = __pmGetFileStatusFlags(__pmfd)) != -1)
sts = __pmSetFileStatusFlags(__pmfd, flags | FNDELAY);
else
sts = -1;
if (sts == -1)
return -oserror();
#ifdef PMTRACE_DEBUG
if (__pmstate & PMTRACE_STATE_COMMS)
fprintf(stderr, "_pmtraceconnect: async protocol setup complete\n");
#endif
}
else
#ifdef PMTRACE_DEBUG
if (__pmstate & PMTRACE_STATE_COMMS)
fprintf(stderr, "_pmtraceconnect: sync protocol setup complete\n");
#endif
/* trace PMDA sends an ACK on successful connect */
sts = _pmtracegetack(sts, 0);
return sts;
}
static void
ClientLoop(void)
{
int i, fd, sts;
int maxFd;
int checkAgents;
int reload_ns = 0;
__pmFdSet readableFds;
for (;;) {
/* Figure out which file descriptors to wait for input on. Keep
* track of the highest numbered descriptor for the select call.
*/
readableFds = clientFds;
maxFd = maxClientFd + 1;
/* If an agent was not ready, it may send an ERROR PDU to indicate it
* is now ready. Add such agents to the list of file descriptors.
*/
checkAgents = 0;
for (i = 0; i < nAgents; i++) {
AgentInfo *ap = &agent[i];
if (ap->status.notReady) {
fd = ap->outFd;
__pmFD_SET(fd, &readableFds);
if (fd > maxFd)
maxFd = fd + 1;
checkAgents = 1;
if (pmDebug & DBG_TRACE_APPL0)
__pmNotifyErr(LOG_INFO,
"not ready: check %s agent on fd %d (max = %d)\n",
ap->pmDomainLabel, fd, maxFd);
}
}
sts = __pmSelectRead(maxFd, &readableFds, NULL);
if (sts > 0) {
if (pmDebug & DBG_TRACE_APPL0)
for (i = 0; i <= maxClientFd; i++)
if (__pmFD_ISSET(i, &readableFds))
fprintf(stderr, "DATA: from %s (fd %d)\n",
FdToString(i), i);
__pmServerAddNewClients(&readableFds, CheckNewClient);
if (checkAgents)
reload_ns = HandleReadyAgents(&readableFds);
HandleClientInput(&readableFds);
}
else if (sts == -1 && neterror() != EINTR) {
__pmNotifyErr(LOG_ERR, "ClientLoop select: %s\n", netstrerror());
break;
}
if (restart) {
restart = 0;
reload_ns = 1;
SignalRestart();
}
if (reload_ns) {
reload_ns = 0;
SignalReloadPMNS();
}
if (timeToDie) {
SignalShutdown();
break;
}
if (AgentDied) {
AgentDied = 0;
for (i = 0; i < nAgents; i++) {
if (!agent[i].status.connected)
mapdom[agent[i].pmDomainId] = nAgents;
}
}
}
}
/* Establish a new socket connection to a client */
ClientInfo *
AcceptNewClient(int reqfd)
{
int i;
int fd;
__pmSockLen addrlen;
int ok = 0;
char buf[MY_BUFLEN];
char *bp;
char *endp;
char *abufp;
i = NewClient();
addrlen = __pmSockAddrSize();
fd = __pmAccept(reqfd, client[i].addr, &addrlen);
if (fd == -1) {
__pmNotifyErr(LOG_ERR, "AcceptNewClient(%d) __pmAccept failed: %s",
reqfd, netstrerror());
Shutdown();
exit(1);
}
__pmSetSocketIPC(fd);
if (fd > maxSockFd)
maxSockFd = fd;
__pmFD_SET(fd, &sockFds);
client[i].fd = fd;
client[i].pmcd_fd = -1;
client[i].status.connected = 1;
client[i].status.allowed = 0;
client[i].pmcd_hostname = NULL;
/*
* version negotiation (converse to negotiate_proxy() logic in
* libpcp
*
* __pmRecv client version message
* __pmSend my server version message
* __pmRecv pmcd hostname and pmcd port
*/
for (bp = buf; bp < &buf[MY_BUFLEN]; bp++) {
if (__pmRecv(fd, bp, 1, 0) != 1) {
*bp = '\0'; /* null terminate what we have */
bp = &buf[MY_BUFLEN]; /* flag error */
break;
}
/* end of line means no more ... */
if (*bp == '\n' || *bp == '\r') {
*bp = '\0';
break;
}
}
if (bp < &buf[MY_BUFLEN]) {
/* looks OK so far ... is this a version we can support? */
if (strcmp(buf, "pmproxy-client 1") == 0) {
client[i].version = 1;
ok = 1;
}
}
if (!ok) {
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_CONTEXT) {
abufp = __pmSockAddrToString(client[i].addr);
__pmNotifyErr(LOG_INFO, "Bad version string from client at %s",
abufp);
free(abufp);
fprintf(stderr, "AcceptNewClient: bad version string was \"");
for (bp = buf; *bp && bp < &buf[MY_BUFLEN]; bp++)
fputc(*bp & 0xff, stderr);
fprintf(stderr, "\"\n");
}
#endif
DeleteClient(&client[i]);
return NULL;
}
if (__pmSend(fd, MY_VERSION, strlen(MY_VERSION), 0) != strlen(MY_VERSION)) {
abufp = __pmSockAddrToString(client[i].addr);
__pmNotifyErr(LOG_WARNING, "AcceptNewClient: failed to send version "
"string (%s) to client at %s\n", MY_VERSION, abufp);
free(abufp);
DeleteClient(&client[i]);
return NULL;
}
for (bp = buf; bp < &buf[MY_BUFLEN]; bp++) {
if (__pmRecv(fd, bp, 1, 0) != 1) {
*bp = '\0'; /* null terminate what we have */
bp = &buf[MY_BUFLEN]; /* flag error */
break;
}
/* end of line means no more ... */
if (*bp == '\n' || *bp == '\r') {
*bp = '\0';
break;
}
}
if (bp < &buf[MY_BUFLEN]) {
/* looks OK so far ... get hostname and port */
for (bp = buf; *bp && *bp != ' '; bp++)
;
if (bp != buf) {
*bp = '\0';
client[i].pmcd_hostname = strdup(buf);
if (client[i].pmcd_hostname == NULL)
__pmNoMem("PMCD.hostname", strlen(buf), PM_FATAL_ERR);
bp++;
client[i].pmcd_port = (int)strtoul(bp, &endp, 10);
if (*endp != '\0') {
abufp = __pmSockAddrToString(client[i].addr);
__pmNotifyErr(LOG_WARNING, "AcceptNewClient: bad pmcd port "
"\"%s\" from client at %s", bp, abufp);
free(abufp);
DeleteClient(&client[i]);
return NULL;
}
}
/* error, fall through */
}
if (client[i].pmcd_hostname == NULL) {
abufp = __pmSockAddrToString(client[i].addr);
__pmNotifyErr(LOG_WARNING, "AcceptNewClient: failed to get PMCD "
"hostname (%s) from client at %s", buf, abufp);
free(abufp);
DeleteClient(&client[i]);
return NULL;
}
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_CONTEXT) {
/*
* note error message gets appended to once pmcd connection is
* made in ClientLoop()
*/
abufp = __pmSockAddrToString(client[i].addr);
fprintf(stderr, "AcceptNewClient [%d] fd=%d from %s to %s (port %s)",
i, fd, abufp, client[i].pmcd_hostname, bp);
free(abufp);
}
#endif
return &client[i];
}
