diff --git a/src/slurmd/slurmd/req.c b/src/slurmd/slurmd/req.c
index 1c886b9120fb8d80e37acb77fb34c0f411d68b65..0192c43563322f342a93c2369733f796459d3225 100644
--- a/src/slurmd/slurmd/req.c
+++ b/src/slurmd/slurmd/req.c
@@ -318,7 +318,7 @@ _forkexec_slurmstepd(slurmd_step_type_t type, void *req,
if (pipe(to_stepd) < 0 || pipe(to_slurmd) < 0) {
error("_forkexec_slurmstepd pipe failed: %m");
- return -1;
+ return SLURM_FAILURE;
}
if ((pid = fork()) < 0) {
@@ -327,13 +327,12 @@ _forkexec_slurmstepd(slurmd_step_type_t type, void *req,
close(to_stepd[1]);
close(to_slurmd[0]);
close(to_slurmd[1]);
- return -1;
+ return SLURM_FAILURE;
} else if (pid > 0) {
- int ok;
int rc = 0;
/*
* Parent sends initialization data to the slurmstepd
- * over the to_stepd pipe, and waits for an "ok"
+ * over the to_stepd pipe, and waits for the return code
* reply on the to_slurmd pipe.
*/
if (close(to_stepd[0]) < 0)
@@ -344,14 +343,16 @@ _forkexec_slurmstepd(slurmd_step_type_t type, void *req,
if ((rc = _send_slurmstepd_init(to_stepd[1], type,
req, cli, self)) < 0) {
error("Unable to init slurmstepd");
- rc = -1;
+ rc = SLURM_FAILURE;
+ goto done;
}
- if (read(to_slurmd[0], &ok, sizeof(int)) != sizeof(int)) {
- error("Error reading \"ok\" message "
+ if (read(to_slurmd[0], &rc, sizeof(int)) != sizeof(int)) {
+ error("Error reading return code message "
" from slurmstepd: %m");
- rc = -2;
+ rc = SLURM_FAILURE;
}
+ done:
/* Reap child */
if (waitpid(pid, NULL, 0) < 0)
error("Unable to reap slurmd child process");
diff --git a/src/slurmd/slurmstepd/req.c b/src/slurmd/slurmstepd/req.c
index d15f02a5ca7a8159620903ee2e1436358328f152..74a2e9187bb73a9761323a4368ad1daf2228c13f 100644
--- a/src/slurmd/slurmstepd/req.c
+++ b/src/slurmd/slurmstepd/req.c
@@ -95,8 +95,6 @@ _create_socket(const char *name)
return -1;
fd_set_close_on_exec(fd);
- unlink(name); /* in case it already exists */
-
memset(&addr, 0, sizeof(addr));
addr.sun_family = AF_UNIX;
strcpy(addr.sun_path, name);
@@ -123,15 +121,29 @@ _domain_socket_create(const char *dir, const char *nodename,
/*
* Make sure that "dir" exists and is a directory.
*/
- if (stat(dir, &stat_buf) < 0)
- fatal("Domain socket directory %s: %m", dir);
- else if (!S_ISDIR(stat_buf.st_mode))
- fatal("%s is not a directory", dir);
+ if (stat(dir, &stat_buf) < 0) {
+ error("Domain socket directory %s: %m", dir);
+ return -1;
+ } else if (!S_ISDIR(stat_buf.st_mode)) {
+ error("%s is not a directory", dir);
+ return -1;
+ }
/*
* Now build the the name of socket, and create the socket.
*/
xstrfmtcat(name, "%s/%s_%u.%u", dir, nodename, jobid, stepid);
+
+ /*
+ * First check to see if the named socket already exists.
+ */
+ if (stat(name, &stat_buf) == 0) {
+ error("Socket %s already exists", name);
+ xfree(name);
+ errno = ESLURMD_STEP_EXISTS;
+ return -1;
+ }
+
fd = _create_socket(name);
if (fd < 0)
fatal("Could not create domain socket: %m");
@@ -161,15 +173,19 @@ _msg_thr_internal(void *job_arg)
debug("Message thread exited");
}
-void
+int
msg_thr_create(slurmd_job_t *job)
{
int fd;
eio_obj_t *eio_obj;
pthread_attr_t attr;
+ errno = 0;
fd = _domain_socket_create(conf->spooldir, "nodename",
job->jobid, job->stepid);
+ if (fd == -1)
+ return SLURM_ERROR;
+
fd_set_nonblocking(fd);
eio_obj = eio_obj_create(fd, &msg_socket_ops, (void *)job);
@@ -178,8 +194,12 @@ msg_thr_create(slurmd_job_t *job)
slurm_attr_init(&attr);
if (pthread_create(&job->msgid, &attr,
- &_msg_thr_internal, (void *)job) != 0)
- fatal("pthread_create: %m");
+ &_msg_thr_internal, (void *)job) != 0) {
+ error("pthread_create: %m");
+ return SLURM_ERROR;
+ }
+
+ return SLURM_SUCCESS;
}
static bool
diff --git a/src/slurmd/slurmstepd/req.h b/src/slurmd/slurmstepd/req.h
index cf72b9d273034d0d1ca90fa17854ca3af539306d..5dc51bc580b684f2c7a4af39a5d61226741d25eb 100644
--- a/src/slurmd/slurmstepd/req.h
+++ b/src/slurmd/slurmstepd/req.h
@@ -30,6 +30,6 @@
#include "src/slurmd/slurmstepd/slurmstepd_job.h"
-void msg_thr_create(slurmd_job_t *job);
+int msg_thr_create(slurmd_job_t *job);
#endif /* _STEP_REQ_H */
diff --git a/src/slurmd/slurmstepd/slurmstepd.c b/src/slurmd/slurmstepd/slurmstepd.c
index 0b13e968c89c9b74962fcb311d7c5f909bd6f8a3..5e96d6cdb92ab1bf69860ce8a88f25b80c6aa873 100644
--- a/src/slurmd/slurmstepd/slurmstepd.c
+++ b/src/slurmd/slurmstepd/slurmstepd.c
@@ -43,6 +43,7 @@
static int _init_from_slurmd(int sock, char **argv, slurm_addr **_cli,
slurm_addr **_self, slurm_msg_t **_msg);
static void _send_ok_to_slurmd(int sock);
+static void _send_fail_to_slurmd(int sock);
static slurmd_job_t *_step_setup(slurm_addr *cli, slurm_addr *self,
slurm_msg_t *msg);
static void _step_cleanup(slurmd_job_t *job, slurm_msg_t *msg, int rc);
@@ -71,7 +72,11 @@ main (int argc, char *argv[])
job = _step_setup(cli, self, msg);
- msg_thr_create(job); /* sets job->msg_handle and job->msgid */
+ /* sets job->msg_handle and job->msgid */
+ if (msg_thr_create(job) == SLURM_ERROR) {
+ _send_fail_to_slurmd(STDOUT_FILENO);
+ return -1;
+ }
_send_ok_to_slurmd(STDOUT_FILENO);
close(STDOUT_FILENO);
@@ -99,13 +104,25 @@ main (int argc, char *argv[])
static void
_send_ok_to_slurmd(int sock)
{
- int ok = 42;
+ int ok = SLURM_SUCCESS;
safe_write(sock, &ok, sizeof(int));
return;
rwfail:
error("Unable to send \"ok\" to slurmd");
}
+static void
+_send_fail_to_slurmd(int sock)
+{
+ int fail = SLURM_FAILURE;
+
+ if (errno)
+ fail = errno;
+ safe_write(sock, &fail, sizeof(int));
+ return;
+rwfail:
+ error("Unable to send \"fail\" to slurmd");
+}
static int
_init_from_slurmd(int sock, char **argv,