void
DeleteClient(ClientInfo *cp)
{
int i;
for (i = 0; i < nClients; i++)
if (cp == &client[i])
break;
if (i == nClients) {
fprintf(stderr, "DeleteClient: Botch: tried to delete non-existent client @" PRINTF_P_PFX "%p\n", cp);
return;
}
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_CONTEXT)
fprintf(stderr, "DeleteClient [%d]\n", i);
#endif
if (cp->fd >= 0) {
__pmFD_CLR(cp->fd, &sockFds);
__pmCloseSocket(cp->fd);
}
if (cp->pmcd_fd >= 0) {
__pmFD_CLR(cp->pmcd_fd, &sockFds);
__pmCloseSocket(cp->pmcd_fd);
}
if (i == nClients-1) {
i--;
while (i >= 0 && !client[i].status.connected)
i--;
nClients = (i >= 0) ? i + 1 : 0;
}
if (cp->fd == maxSockFd || cp->pmcd_fd == maxSockFd) {
maxSockFd = maxReqPortFd;
for (i = 0; i < nClients; i++) {
if (cp->status.connected == 0)
continue;
if (client[i].fd > maxSockFd)
maxSockFd = client[i].fd;
if (client[i].pmcd_fd > maxSockFd)
maxSockFd = client[i].pmcd_fd;
}
}
__pmSockAddrFree(cp->addr);
cp->addr = NULL;
cp->status.connected = 0;
cp->fd = -1;
cp->pmcd_fd = -1;
if (cp->pmcd_hostname != NULL) {
free(cp->pmcd_hostname);
cp->pmcd_hostname = NULL;
}
}
void
disconnect(int sts)
{
time_t now;
#if CAN_RECONNECT
int ctx;
__pmContext *ctxp;
#endif
time(&now);
if (sts != 0)
fprintf(stderr, "%s: Error: %s\n", pmProgname, pmErrStr(sts));
fprintf(stderr, "%s: Lost connection to PMCD on \"%s\" at %s",
pmProgname, pmcd_host, ctime(&now));
#if CAN_RECONNECT
if (primary) {
fprintf(stderr, "This is fatal for the primary logger.");
exit(1);
}
close(pmcdfd);
__pmFD_CLR(pmcdfd, &fds);
pmcdfd = -1;
numfds = maxfd() + 1;
if ((ctx = pmWhichContext()) >= 0)
ctxp = __pmHandleToPtr(ctx);
if (ctx < 0 || ctxp == NULL) {
fprintf(stderr, "%s: disconnect botch: cannot get context: %s\n", pmProgname, pmErrStr(ctx));
exit(1);
}
ctxp->c_pmcd->pc_fd = -1;
PM_UNLOCK(ctxp->c_lock);
#else
exit(1);
#endif
}
void
disconnect(int sts)
{
time_t now;
int ctx;
__pmContext *ctxp;
time(&now);
if (sts != 0)
fprintf(stderr, "%s: Error: %s\n", pmProgname, pmErrStr(sts));
fprintf(stderr, "%s: Lost connection to PMCD on \"%s\" at %s",
pmProgname, pmcd_host, ctime(&now));
if (pmcdfd != -1) {
close(pmcdfd);
__pmFD_CLR(pmcdfd, &fds);
pmcdfd = -1;
}
numfds = maxfd() + 1;
if ((ctx = pmWhichContext()) >= 0)
ctxp = __pmHandleToPtr(ctx);
if (ctx < 0 || ctxp == NULL) {
fprintf(stderr, "%s: disconnect botch: cannot get context: %s\n", pmProgname, pmErrStr(ctx));
exit(1);
}
ctxp->c_pmcd->pc_fd = -1;
PM_UNLOCK(ctxp->c_lock);
}
void
DeleteClient(ClientInfo *cp)
{
int i;
for (i = 0; i < nClients; i++)
if (cp == &client[i])
break;
if (i == nClients) {
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_APPL0) {
__pmNotifyErr(LOG_ERR, "DeleteClient: tried to delete non-existent client\n");
Shutdown();
exit(1);
}
#endif
return;
}
if (cp->fd != -1) {
__pmFD_CLR(cp->fd, &clientFds);
__pmCloseSocket(cp->fd);
}
if (i == nClients-1) {
i--;
while (i >= 0 && !client[i].status.connected)
i--;
nClients = (i >= 0) ? i + 1 : 0;
}
if (cp->fd == maxClientFd) {
maxClientFd = -1;
for (i = 0; i < nClients; i++) {
if (client[i].fd > maxClientFd)
maxClientFd = client[i].fd;
}
}
for (i = 0; i < cp->szProfile; i++) {
if (cp->profile[i] != NULL) {
__pmFreeProfile(cp->profile[i]);
cp->profile[i] = NULL;
}
}
__pmFreeAttrsSpec(&cp->attrs);
__pmHashClear(&cp->attrs);
__pmSockAddrFree(cp->addr);
cp->addr = NULL;
cp->status.connected = 0;
cp->fd = -1;
NotifyEndContext(cp-client);
}
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);
}
/*
* generate dialog/message when launching application wishes to break
* its association with pmlogger
*
* cmd is one of the following:
* D detach pmlogger and let it run forever
* Q terminate pmlogger
* ? display status
* X fatal error or application exit ... user must decide
* to detach or quit
*/
void
do_dialog(char cmd)
{
FILE *msgf = NULL;
time_t now;
static char lbuf[100+MAXPATHLEN];
double archsize;
char *q;
char *p = NULL;
int nchar;
char *msg;
#if HAVE_MKSTEMP
char tmp[MAXPATHLEN];
#endif
time(&now);
now -= rsc_start;
if (now == 0)
/* hack is close enough! */
now = 1;
archsize = vol_bytes + ftell(logctl.l_mfp);
nchar = add_msg(&p, 0, "");
p[0] = '\0';
snprintf(lbuf, sizeof(lbuf), "PCP recording for the archive folio \"%s\" and the host", folio_name);
nchar = add_msg(&p, nchar, lbuf);
snprintf(lbuf, sizeof(lbuf), " \"%s\" has been in progress for %ld %s",
pmcd_host,
now < 240 ? now : now/60, now < 240 ? "seconds" : "minutes");
nchar = add_msg(&p, nchar, lbuf);
nchar = add_msg(&p, nchar, " and in that time the pmlogger process has created an");
nchar = add_msg(&p, nchar, " archive of ");
q = do_size(archsize);
nchar = add_msg(&p, nchar, q);
nchar = add_msg(&p, nchar, ".");
if (rsc_replay) {
nchar = add_msg(&p, nchar, "\n\nThis archive may be replayed with the following command:\n");
snprintf(lbuf, sizeof(lbuf), " $ pmafm %s replay", folio_name);
nchar = add_msg(&p, nchar, lbuf);
}
if (cmd == 'D') {
nchar = add_msg(&p, nchar, "\n\nThe application that launched pmlogger has asked pmlogger");
nchar = add_msg(&p, nchar, " to continue independently and the PCP archive will grow at");
nchar = add_msg(&p, nchar, " the rate of ");
q = do_size((archsize * 3600) / now);
nchar = add_msg(&p, nchar, q);
nchar = add_msg(&p, nchar, " per hour or ");
q = do_size((archsize * 3600 * 24) / now);
nchar = add_msg(&p, nchar, q);
nchar = add_msg(&p, nchar, " per day.");
}
if (cmd == 'X') {
nchar = add_msg(&p, nchar, "\n\nThe application that launched pmlogger has exited and you");
nchar = add_msg(&p, nchar, " must decide if the PCP recording session should be terminated");
nchar = add_msg(&p, nchar, " or continued. If recording is continued the PCP archive will");
nchar = add_msg(&p, nchar, " grow at the rate of ");
q = do_size((archsize * 3600) / now);
nchar = add_msg(&p, nchar, q);
nchar = add_msg(&p, nchar, " per hour or ");
q = do_size((archsize * 3600 * 24) / now);
nchar = add_msg(&p, nchar, q);
nchar = add_msg(&p, nchar, " per day.");
}
if (cmd == 'Q') {
nchar = add_msg(&p, nchar, "\n\nThe application that launched pmlogger has terminated");
nchar = add_msg(&p, nchar, " this PCP recording session.\n");
}
if (cmd != 'Q') {
nchar = add_msg(&p, nchar, "\n\nAt any time this pmlogger process may be terminated with the");
nchar = add_msg(&p, nchar, " following command:\n");
snprintf(lbuf, sizeof(lbuf), " $ pmsignal -s TERM %" FMT_PID "\n", getpid());
nchar = add_msg(&p, nchar, lbuf);
}
if (cmd == 'X')
nchar = add_msg(&p, nchar, "\n\nTerminate this PCP recording session now?");
if (nchar > 0) {
char * xconfirm = __pmNativePath(pmGetConfig("PCP_XCONFIRM_PROG"));
int fd = -1;
#if HAVE_MKSTEMP
snprintf(tmp, sizeof(tmp), "%s%cmsgXXXXXX", pmGetConfig("PCP_TMPFILE_DIR"), __pmPathSeparator());
msg = tmp;
fd = mkstemp(tmp);
#else
if ((msg = tmpnam(NULL)) != NULL)
fd = open(msg, O_WRONLY|O_CREAT|O_EXCL, 0600);
#endif
if (fd >= 0)
msgf = fdopen(fd, "w");
if (msgf == NULL) {
fprintf(stderr, "\nError: failed create temporary message file for recording session dialog\n");
fprintf(stderr, "Reason? %s\n", osstrerror());
if (fd != -1)
close(fd);
goto failed;
}
fputs(p, msgf);
fclose(msgf);
msgf = NULL;
if (cmd == 'X')
snprintf(lbuf, sizeof(lbuf), "%s -c -header \"%s - %s\" -file %s -icon question "
"-B Yes -b No 2>/dev/null",
xconfirm, dialog_title, rsc_prog, msg);
else
snprintf(lbuf, sizeof(lbuf), "%s -c -header \"%s - %s\" -file %s -icon info "
"-b Close 2>/dev/null",
xconfirm, dialog_title, rsc_prog, msg);
if ((msgf = popen(lbuf, "r")) == NULL) {
fprintf(stderr, "\nError: failed to start command for recording session dialog\n");
fprintf(stderr, "Command: \"%s\"\n", lbuf);
goto failed;
}
if (fgets(lbuf, sizeof(lbuf), msgf) == NULL) {
fprintf(stderr, "\n%s: pmconfirm(1) failed for recording session dialog\n",
cmd == '?' ? "Warning" : "Error");
failed:
fprintf(stderr, "Dialog:\n");
fputs(p, stderr);
strcpy(lbuf, "Yes");
}
else {
/* strip at first newline */
for (q = lbuf; *q && *q != '\n'; q++)
;
*q = '\0';
}
if (msgf != NULL)
pclose(msgf);
unlink(msg);
}
else {
fprintf(stderr, "Error: failed to create recording session dialog message!\n");
fprintf(stderr, "Reason? %s\n", osstrerror());
strcpy(lbuf, "Yes");
}
free(p);
if (cmd == 'Q' || (cmd == 'X' && strcmp(lbuf, "Yes") == 0)) {
run_done(0, "Recording session terminated");
/*NOTREACHED*/
}
if (cmd != '?') {
/* detach, silently go off to the races ... */
close(rsc_fd);
__pmFD_CLR(rsc_fd, &fds);
rsc_fd = -1;
}
}
