diff --git a/src/plugins/proctrack/aix/proctrack_aix.c b/src/plugins/proctrack/aix/proctrack_aix.c
index 2617757ff1641c0c31d76b02d5efb36da967a52b..a2bbc27c85faf5021681bb3d939ecd640a266495 100644
--- a/src/plugins/proctrack/aix/proctrack_aix.c
+++ b/src/plugins/proctrack/aix/proctrack_aix.c
@@ -43,10 +43,6 @@
#include <slurm/slurm_errno.h>
#include "src/common/log.h"
-#ifndef __USE_XOPEN_EXTENDED
-extern pid_t getsid(pid_t pid); /* missing from <unistd.h> */
-#endif
-
extern int proctrack_job_reg(int *jobid); /* register a job, include this proc */
extern int proctrack_job_unreg(int *jobid); /* unregister a job */
extern int proctrack_job_kill(int *jobid, int *signal); /* signal a job */
@@ -102,13 +98,12 @@ extern int fini ( void )
/*
* For this plugin, we ignore the job_id.
- * To generate a unique container ID, we use setsid.
+ * To generate a unique container ID, we use getpid.
*/
extern uint32_t slurm_create_container ( uint32_t job_id )
{
- pid_t pid = setsid();
+ pid_t pid = getpid();
int jobid = (int) pid;
- (void) setpgrp();
if (pid < 0) {
error("slurm_create_container: setsid: %m");
diff --git a/src/plugins/proctrack/sid/proctrack_sid.c b/src/plugins/proctrack/sid/proctrack_sid.c
index c6f38222b5e1281d511b0aa9a9da23db8f1fb7ae..ee2d81703c00796b9289bc9294cabf93a3f58823 100644
--- a/src/plugins/proctrack/sid/proctrack_sid.c
+++ b/src/plugins/proctrack/sid/proctrack_sid.c
@@ -1,5 +1,5 @@
/*****************************************************************************\
- * proctrack_sid.c - process tracking via session ID plugin.
+ * proctrack_sid.c - process tracking via process group ID plugin.
*****************************************************************************
* Copyright (C) 2005 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -43,11 +43,6 @@
#include <slurm/slurm_errno.h>
#include "src/common/log.h"
-#ifndef __USE_XOPEN_EXTENDED
-extern pid_t getsid(pid_t pid); /* missing from <unistd.h> */
-extern pid_t setsid(void); /* missing from <unistd.h> */
-#endif
-
/*
* These variables are required by the generic plugin interface. If they
* are not found in the plugin, the plugin loader will ignore it.
@@ -97,18 +92,16 @@ extern int fini ( void )
/*
* For this plugin, we ignore the job_id.
- * To generate a unique container ID, we use setsid.
+ *
+ * FIXME! - This is basically a no-op. We return -1 because
+ * only returning a 0 disturbs the caller. The caller throws
+ * away the return code anyway. This needs to be redone
+ * for slurm-0.6 when the task-creation code is rewritten to
+ * eliminate the user-owned session manager slurmd.
*/
extern uint32_t slurm_create_container ( uint32_t job_id )
{
- pid_t pid = setsid();
- (void) setpgrp();
-
- if (pid < 0) {
- error("slurm_create_container: setsid: %m");
- return (uint32_t) 0;
- }
- return (uint32_t) pid;
+ return (uint32_t) -1;
}
extern int slurm_add_container ( uint32_t id )
@@ -123,6 +116,11 @@ extern int slurm_signal_container ( uint32_t id, int signal )
if (!id) /* no container ID */
return ESRCH;
+ if (id == getpid() || id == getpgid(0)) {
+ error("slurm_signal_container would kill caller!");
+ return ESRCH;
+ }
+
return killpg(pid, signal);
}
@@ -134,7 +132,7 @@ extern int slurm_destroy_container ( uint32_t id )
extern uint32_t
slurm_find_container(pid_t pid)
{
- pid_t rc = getsid(pid);
+ pid_t rc = getpgid(pid);
if (rc == -1)
return (uint32_t) 0;
diff --git a/src/slurmd/mgr.c b/src/slurmd/mgr.c
index f75b45815aaf90622ece63106675d650ead7271e..8d5d081efa58a2429b2c37e0df67853f13186dae 100644
--- a/src/slurmd/mgr.c
+++ b/src/slurmd/mgr.c
@@ -655,6 +655,8 @@ _create_job_session(slurmd_job_t *job)
*/
if (shm_update_step_mpid(job->jobid, job->stepid, getpid()) < 0)
debug("shm_update_step_mpid: %m");
+ if (shm_update_step_spid(job->jobid, job->stepid, spid) < 0)
+ debug("shm_update_step_spid: %m");
job->smgr_pid = spid;
@@ -924,8 +926,8 @@ _kill_running_tasks(slurmd_job_t *job)
while ((s = list_next(i))) {
if ((s->jobid != job->jobid) || (s->stepid != job->stepid))
continue;
- if (s->task_list && s->task_list->pid)
- killpg(s->task_list->pid, SIGKILL);
+/* if (s->task_list && s->task_list->pid) */
+/* killpg(s->task_list->pid, SIGKILL); */
if (s->cont_id)
slurm_signal_container(s->cont_id, SIGKILL);
}
diff --git a/src/slurmd/req.c b/src/slurmd/req.c
index 9667b6f9d808e5398596ef7cb9950777ab106152..5e8ad327492f56272f670792b9d31d603fef8885 100644
--- a/src/slurmd/req.c
+++ b/src/slurmd/req.c
@@ -138,6 +138,7 @@ slurmd_req(slurm_msg_t *msg, slurm_addr *cli)
slurm_free_reattach_tasks_request_msg(msg->data);
break;
case REQUEST_KILL_JOB:
+ debug2("RPC: REQUEST_KILL_JOB");
_rpc_kill_job(msg, cli);
slurm_free_kill_job_msg(msg->data);
break;
@@ -757,14 +758,16 @@ _rpc_kill_tasks(slurm_msg_t *msg, slurm_addr *cli_addr)
kill_proc_tree((pid_t) step->cont_id, req->signal);
rc = SLURM_SUCCESS;
} else {
- if (slurm_signal_container(step->cont_id, req->signal) < 0)
- rc = errno;
-
- if (step->task_list
- && (step->task_list->pid > (pid_t) 0)
- && (killpg(step->task_list->pid, req->signal) < 0))
- rc = errno;
-
+ if ((req->signal == SIGKILL) || (req->signal == 0)) {
+ if (slurm_signal_container(step->cont_id,
+ req->signal) < 0)
+ rc = errno;
+ } else {
+ if (step->task_list
+ && (step->task_list->pid > (pid_t) 0)
+ && (killpg(step->task_list->pid, req->signal) < 0))
+ rc = errno;
+ }
if (rc == SLURM_SUCCESS)
verbose("Sent signal %d to %u.%u",
req->signal, req->job_id, req->job_step_id);
@@ -790,7 +793,8 @@ _kill_running_session_mgrs(uint32_t jobid, int signum, char *signame)
while ((s = list_next(i))) {
if ((s->jobid == jobid) && s->cont_id) {
- slurm_signal_container(s->cont_id, signum);
+ kill(s->spid, signum);
+ /* slurm_signal_container(s->cont_id, signum); */
cnt++;
}
}
@@ -1028,11 +1032,11 @@ _kill_all_active_steps(uint32_t jobid, int sig, bool batch)
if (slurm_signal_container(s->cont_id, sig) < 0)
error("kill jid %d cont_id %u: %m",
s->jobid, s->cont_id);
- if (s->task_list
- && (s->task_list->pid > (pid_t) 0)
- && (killpg(s->task_list->pid, sig) < 0))
- error("kill jid %d pgrp %d: %m",
- s->jobid, s->task_list->pid);
+/* if (s->task_list */
+/* && (s->task_list->pid > (pid_t) 0) */
+/* && (killpg(s->task_list->pid, sig) < 0)) */
+/* error("kill jid %d pgrp %d: %m", */
+/* s->jobid, s->task_list->pid); */
}
}
list_iterator_destroy(i);
diff --git a/src/slurmd/shm.c b/src/slurmd/shm.c
index 47256bb04b255c511558c67c36d35fba4365019c..514f7aa601d02d5ddf601b8b1f68af34002509d4 100644
--- a/src/slurmd/shm.c
+++ b/src/slurmd/shm.c
@@ -526,6 +526,21 @@ shm_update_step_mpid(uint32_t jobid, uint32_t stepid, int mpid)
return retval;
}
+int
+shm_update_step_spid(uint32_t jobid, uint32_t stepid, int spid)
+{
+ int i, retval = SLURM_SUCCESS;
+ _shm_lock();
+ if ((i = _shm_find_step(jobid, stepid)) >= 0)
+ slurmd_shm->step[i].spid = spid;
+ else {
+ slurm_seterrno(ESRCH);
+ retval = SLURM_FAILURE;
+ }
+ _shm_unlock();
+ return retval;
+}
+
int
shm_update_step_cont_id(uint32_t jobid, uint32_t stepid, uint32_t cont_id)
{
diff --git a/src/slurmd/shm.h b/src/slurmd/shm.h
index f8687caa3a342974fbd9014f614d0cbafc63033e..78e0d4e220609f8d6570c735465af09430430e7f 100644
--- a/src/slurmd/shm.h
+++ b/src/slurmd/shm.h
@@ -76,6 +76,7 @@ struct job_step {
uint32_t sw_id; /* Switch/Interconnect specific id */
int ntasks; /* number of tasks in this job */
pid_t mpid; /* Job manager pid */
+ pid_t spid; /* Session manager pid */
uint32_t cont_id; /* Job container id */
/* Executable's pathname */
@@ -198,6 +199,12 @@ int shm_update_step_cont_id(uint32_t jobid, uint32_t stepid, uint32_t cont_id);
int shm_update_step_mpid(uint32_t jobid, uint32_t stepid, int mpid);
+/*
+ * update job step "session manager" pid
+ */
+int shm_update_step_spid(uint32_t jobid, uint32_t stepid, int spid);
+
+
/*
* update job step state
*/
diff --git a/src/slurmd/smgr.c b/src/slurmd/smgr.c
index dc0ff900fed38cd66742362bddfa994ae8f2c08a..7770f45b89a0dd5fa6edf7ab9df928ed925d7367 100644
--- a/src/slurmd/smgr.c
+++ b/src/slurmd/smgr.c
@@ -324,14 +324,6 @@ _exec_all_tasks(slurmd_job_t *job)
(unsigned long) job->task[i]->gtid,
(unsigned long) pid);
- /*
- * Send pid to job manager
- */
- if (fd_write_n(fd, (char *)&pid, sizeof(pid_t)) < 0) {
- error("unable to update task pid!: %m");
- return SLURM_ERROR;
- }
-
job->task[i]->pid = pid;
/*
@@ -341,6 +333,14 @@ _exec_all_tasks(slurmd_job_t *job)
error ("Unable to put task %d (pid %ld) into pgrp %ld",
i, pid, job->task[0]->pid);
+ /*
+ * Send pid to job manager
+ */
+ if (fd_write_n(fd, (char *)&pid, sizeof(pid_t)) < 0) {
+ error("unable to update task pid!: %m");
+ return SLURM_ERROR;
+ }
+
/*
* Now it's ok to unblock this child, so it may call exec
*/