diff --git a/Makefile.am b/Makefile.am
index 2b2ca549ea6025f84d5081fb42a5a6bc25245be3..00dad97e1a6b81b8724cc31e27c598da5acc2b59 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -2,9 +2,15 @@ AUTOMAKE_OPTIONS = foreign
SUBDIRS = src testsuite doc
-EXTRA_DIST = auxdir etc
+EXTRA_DIST = \
+ auxdir \
+ etc \
+ doc \
+ autogen.sh
MAINTAINERCLEANFILES = Makefile.in aclocal.m4 config.guess \
config.h.in config.sub configure install-sh \
ltconfig ltmain.sh missing mkinstalldirs \
stamp-h.in
+
+include $(top_srcdir)/Make-rpm.mk
diff --git a/doc/man/man1/srun.1 b/doc/man/man1/srun.1
index 015230b3b614257d61e418a84323f740ec188ab0..3cf46711be34608497b0796428d70c2316cec44f 100644
--- a/doc/man/man1/srun.1
+++ b/doc/man/man1/srun.1
@@ -154,8 +154,8 @@ verbose operation. Multiple \fB-v\fR's will further increase the verbosity of
.TP
\fB\-d\fR, \fB\-\-slurmd-debug\fR=\fIlevel\fR
Specify a debug level for slurmd(8). \fIlevel\fR may be an integer value
-between 0 [quiet, only errors are displayed] and 6 [insanely verbose
-operation]. The slurmd debug information is copied to the stderr of
+between 0 [quiet, only errors are displayed] and 5 [insanely verbose
+operation]. The slurmd debug information is copied onto the stderr of
the job.
.TP
\fB\-W\fR, \fB\-\-wait\fR=\fIseconds\fR
diff --git a/src/common/daemonize.c b/src/common/daemonize.c
index 08b07ade2db59222a288ebf83c20c7ec233b1a47..5790b999e0b506b7bbfc91b4219620ef420d365a 100644
--- a/src/common/daemonize.c
+++ b/src/common/daemonize.c
@@ -83,8 +83,14 @@ daemon(int nochdir, int noclose)
}
+/*
+ * Read and return pid stored in pidfile.
+ * Returns 0 if file doesn't exist or pid cannot be read.
+ * If pidfd != NULL, the file will be kept open and the fd
+ * returned.
+ */
pid_t
-read_pidfile(const char *pidfile)
+read_pidfile(const char *pidfile, int *pidfd)
{
int fd;
FILE *fp = NULL;
@@ -113,6 +119,11 @@ read_pidfile(const char *pidfile)
fatal ("pidfile locked by %ld but contains pid=%ld",
(long) lpid, pid);
+ if (pidfd != NULL)
+ *pidfd = fd;
+ else
+ (void) close(fd); /* Ignore errors */
+
return (lpid);
}
diff --git a/src/common/daemonize.h b/src/common/daemonize.h
index a4f382ee7d2cf577723ba45c47734592c6afde68..903464826fac8a0bb3a19fc55bc4a022c8f56427 100644
--- a/src/common/daemonize.h
+++ b/src/common/daemonize.h
@@ -44,7 +44,9 @@ int create_pidfile(char *pidfilename);
/*
* Attempt to read an old pid from the configured pidfile
* Returns 0 if no pidfile exists (No running process)
+ * If pidfilefd is not NULL, returns open file descriptor for
+ * pidfile (when pid != 0).
*/
-pid_t read_pidfile(char *pidfilename);
+pid_t read_pidfile(char *pidfilename, int *pidfilefd);
#endif /* !_HAVE_DAEMONIZE_H */
diff --git a/src/common/log.c b/src/common/log.c
index 1cebb6f6e6c798c46cd87ba48437cf2461f4a2ab..5659c778a69298f1291c55e430ff07b4c218e841 100644
--- a/src/common/log.c
+++ b/src/common/log.c
@@ -179,7 +179,7 @@ _log_init(char *prog, log_options_t opt, log_facility_t fac, char *logfile )
if (log->opt.syslog_level > LOG_LEVEL_QUIET)
log->facility = fac;
- if (logfile && log->opt.logfile_level > LOG_LEVEL_QUIET) {
+ if (logfile && (log->opt.logfile_level > LOG_LEVEL_QUIET)) {
FILE *fp;
fp = safeopen(logfile, "a", SAFEOPEN_LINK_OK);
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index dc91240987089c00812bf039826da9606d9aabbd..b6c5b2ad291a8d5e8eaaac523fb989a57673a404 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -193,7 +193,7 @@ int main(int argc, char *argv[])
if (chdir(slurmctld_conf.state_save_location))
fatal("chdir to %s error %m",
slurmctld_conf.state_save_location);
- error_code = daemon(0, 0);
+ error_code = daemon(1, 0);
log_alter(log_opts, LOG_DAEMON,
slurmctld_conf.slurmctld_logfile);
if (error_code)
diff --git a/src/slurmd/req.c b/src/slurmd/req.c
index 6ce669b8200600a9a52d106de96c468ffab6a41b..43f8cdc167b63530abed58e5ceaa4af19e868712 100644
--- a/src/slurmd/req.c
+++ b/src/slurmd/req.c
@@ -186,7 +186,8 @@ _launch_tasks(launch_tasks_request_msg_t *req, slurm_addr *cli)
static void
_rpc_launch_tasks(slurm_msg_t *msg, slurm_addr *cli)
{
- int rc;
+ int retval = 0;
+ int rc = 0;
uint16_t port;
char host[MAXHOSTNAMELEN];
uid_t req_uid;
@@ -199,9 +200,8 @@ _rpc_launch_tasks(slurm_msg_t *msg, slurm_addr *cli)
if ((super_user == false) && (req_uid != req->uid)) {
error("Security violation, launch task RCP from uid %u",
(unsigned int) req_uid);
- rc = ESLURM_USER_ID_MISSING; /* or invalid user */
- slurm_send_rc_msg(msg, rc);
- return;
+ retval = ESLURM_USER_ID_MISSING; /* or invalid user */
+ goto done;
}
slurm_get_addr(cli, &port, host, sizeof(host));
@@ -216,21 +216,23 @@ _rpc_launch_tasks(slurm_msg_t *msg, slurm_addr *cli)
conf->cred_state_list);
if ((rc != SLURM_SUCCESS) && (super_user == false)) {
+ retval = errno;
error("Invalid credential from %ld@%s", req_uid, host);
- slurm_send_rc_msg(msg, rc);
- return;
+ goto done;
}
/* Run job prolog if necessary */
if (run_prolog && (_run_prolog(req->job_id, req->uid) != 0)) {
error("[job %d] prolog failed", req->job_id);
- slurm_send_rc_msg(msg, ESLURMD_PROLOG_FAILED);
- return;
+ retval = ESLURMD_PROLOG_FAILED;
+ goto done;
}
- rc = _launch_tasks(req, cli);
+ if (_launch_tasks(req, cli) < 0)
+ retval = errno;
- slurm_send_rc_msg(msg, rc);
+ done:
+ slurm_send_rc_msg(msg, retval);
}
diff --git a/src/slurmd/semaphore.c b/src/slurmd/semaphore.c
index ae6c362041cb18743d2677f019def85f0a134e33..29794104b92e6456544398843ee60c28ee5970ab 100644
--- a/src/slurmd/semaphore.c
+++ b/src/slurmd/semaphore.c
@@ -48,6 +48,8 @@
#include <stdio.h>
#include "semaphore.h"
+#include "../common/log.h"
+
#define MAX_TRIES 3
@@ -157,6 +159,7 @@ sem_t * sem_open(const char *name, int oflag, ...)
}
}
+
/* Open (presumably) existing semaphore. Either O_CREAT was not specified,
* or O_CREAT was specified w/o O_EXCL and the semaphore already exists.
*/
@@ -239,13 +242,15 @@ int sem_unlink(const char *name)
if ((key = ftok(name, 1)) == -1) {
return(-1);
}
- if (unlink(name) == -1) {
- return(-1);
- }
if ((semid = semget(key, 0, 0)) == -1) {
- return(-1);
+ goto done;
}
if (semctl(semid, 0, IPC_RMID) == -1) {
+ goto done;
+ }
+
+ done:
+ if (unlink(name) == -1) {
return(-1);
}
return(0);
@@ -262,7 +267,7 @@ int sem_wait(sem_t *sem)
}
op.sem_num = 0;
op.sem_op = -1;
- op.sem_flg = 0;
+ op.sem_flg = SEM_UNDO;
if (semop(sem->id, &op, 1) == -1) {
return(-1);
}
diff --git a/src/slurmd/shm.c b/src/slurmd/shm.c
index 4a0f9380bdc981bd4e2c0ef633798ed5c582cb5a..5034ede844a69050a4c487a940aea483c8225648 100644
--- a/src/slurmd/shm.c
+++ b/src/slurmd/shm.c
@@ -156,11 +156,9 @@ shm_fini(void)
debug3("%ld calling shm_fini() (attached by %ld)",
getpid(), attach_pid);
- /* xassert(attach_pid == getpid()); */
- /* if ((attach_pid == getpid()) && (--slurmd_shm->users == 0))
- * destroy = 1;
- */
+ debug("[%ld] shm_fini: shm_users = %d", getpid(), slurmd_shm->users);
+
if (--slurmd_shm->users == 0)
destroy = 1;
@@ -191,7 +189,7 @@ shm_cleanup(void)
char *s;
if ((s = _create_ipc_name(SHM_LOCKNAME))) {
- verbose("request to destroy shm lock `%s'", s);
+ info("request to destroy shm lock `%s'", s);
if (sem_unlink(s) < 0)
error("sem_unlink: %m");
xfree(s);
diff --git a/src/slurmd/slurmd.c b/src/slurmd/slurmd.c
index bd05ff97d2ee7ebb52f8b0dccec4b2194bd40a08..b5efe2ea34df801f236681edc6363c7a8a37d323 100644
--- a/src/slurmd/slurmd.c
+++ b/src/slurmd/slurmd.c
@@ -66,7 +66,7 @@
# define MAXHOSTNAMELEN 64
#endif
-#define DEFAULT_SPOOLDIR "/tmp"
+#define DEFAULT_SPOOLDIR "/tmp/slurmd"
#define DEFAULT_PIDFILE "/var/run/slurmd.pid"
typedef struct connection {
@@ -136,12 +136,13 @@ main (int argc, char *argv[])
_print_conf();
_kill_old_slurmd();
+ _create_msg_socket();
+
+ conf->pid = getpid();
create_pidfile(conf->pidfile);
info("%s started on %T", xbasename(argv[0]));
- _create_msg_socket();
- conf->pid = getpid();
if (_slurmd_init() < 0)
exit(1);
@@ -217,11 +218,12 @@ _wait_for_all_threads()
ListIterator i;
pthread_t *ptid;
- debug("Cancelling %d running threads", list_count(conf->threads));
+ debug("Waiting for %d running threads", list_count(conf->threads));
i = list_iterator_create(conf->threads);
while ((ptid = list_next(i))) {
- pthread_cancel(*ptid);
+ pthread_join(*ptid, NULL);
+ debug2("thread %d finished", *ptid);
}
list_iterator_destroy(i);
}
@@ -452,6 +454,7 @@ _init_conf()
conf->prolog = NULL;
conf->port = 0;
conf->daemonize = 1;
+ conf->shm_cleanup = 0;
conf->lfd = -1;
conf->log_opts = lopts;
conf->pidfile = xstrdup(DEFAULT_PIDFILE);
@@ -486,7 +489,7 @@ _process_cmdline(int ac, char **av)
conf->logfile = xstrdup(optarg);
break;
case 'c':
- shm_cleanup();
+ conf->shm_cleanup = 1;
break;
default:
_usage(c);
@@ -529,6 +532,8 @@ _slurmd_init()
slurm_init_verifier(&conf->vctx, conf->pubkey);
_restore_cred_state(&conf->cred_state_list);
conf->threads = list_create((ListDelF) _tid_free);
+ if (conf->shm_cleanup)
+ shm_cleanup();
if (shm_init() < 0)
return SLURM_FAILURE;
return SLURM_SUCCESS;
@@ -680,11 +685,18 @@ _set_slurmd_spooldir(void)
static void
_kill_old_slurmd(void)
{
- pid_t oldpid = read_pidfile(conf->pidfile);
+ int fd;
+ pid_t oldpid = read_pidfile(conf->pidfile, &fd);
if (oldpid != (pid_t) 0) {
info ("killing old slurmd[%ld]", (long) oldpid);
kill(oldpid, SIGTERM);
- sleep(2);
+
+ /*
+ * Wait for previous daemon to terminate
+ */
+ if (fd_get_readw_lock(fd) < 0)
+ fatal ("unable to wait for readw lock: %m");
+ (void) close(fd); /* Ignore errors */
}
}
diff --git a/src/slurmd/slurmd.h b/src/slurmd/slurmd.h
index 2d2d62ba1c7e47005e64797ea92c0e292de65f37..144d72b8fd43657d5fe3d79e7826ef13b29d98ee 100644
--- a/src/slurmd/slurmd.h
+++ b/src/slurmd/slurmd.h
@@ -75,6 +75,7 @@ typedef struct slurmd_config {
log_options_t log_opts; /* current logging options */
int debug_level; /* logging detail level */
int daemonize:1; /* daemonize flag */
+ int shm_cleanup:1;
List cred_state_list; /* credential stat list */
List threads; /* list of active threads */