diff --git a/configure.ac b/configure.ac
index be7b4dc2a3958836d74bc5f247e690f14e5f2c20..50baf2814fedf6c63dd5639e670292623b488940 100644
--- a/configure.ac
+++ b/configure.ac
@@ -103,8 +103,12 @@ CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
LIBS="$PTHREAD_LIBS $LIBS"
case "$host" in
- *-*-aix*) LDFLAGS="$LDFLAGS -Wl,-brtl -Wl,-bexpall " ;;
+ *-*-aix*) LDFLAGS="$LDFLAGS -Wl,-brtl -Wl,-bexpall"
+ CMD_LDFLAGS="-Wl,-bgcbypass:1000" # keep all common functions
+ AC_DEFINE(USE_ALIAS, 0, [Define slurm_ prefix function aliases for plusins]) ;;
+ *) AC_DEFINE(USE_ALIAS, 1, [Define slurm_ prefix function aliases for plugins]) ;;
esac
+AC_SUBST(CMD_LDFLAGS)
AC_SLURM_SEMAPHORE
diff --git a/slurm/bnr.h b/slurm/bnr.h
index 25b4b727125f5b9ff84b64076bfc87eb6248206b..ffadeb1e0eb2e9f3f40dd1450ecf64a645eaac27 100644
--- a/slurm/bnr.h
+++ b/slurm/bnr.h
@@ -13,11 +13,11 @@ typedef int BNR_gid;
#define BNR_SUCCESS 0
#define BNR_ERROR 1
-int BNR_Init(BNR_gid *mygid);
-int BNR_Put(BNR_gid gid, char *attr, char *val);
-int BNR_Fence(BNR_gid gid);
-int BNR_Get(BNR_gid gid, char *attr, char *val);
-int BNR_Finalize();
-int BNR_Rank(BNR_gid group, int *myrank);
-int BNR_Nprocs(BNR_gid group, int *nprocs);
+extern int BNR_Init(BNR_gid *mygid);
+extern int BNR_Put(BNR_gid gid, char *attr, char *val);
+extern int BNR_Fence(BNR_gid gid);
+extern int BNR_Get(BNR_gid gid, char *attr, char *val);
+extern int BNR_Finalize();
+extern int BNR_Rank(BNR_gid group, int *myrank);
+extern int BNR_Nprocs(BNR_gid group, int *nprocs);
diff --git a/src/common/cbuf.c b/src/common/cbuf.c
index 9aca5a9bc4dbcc84a7a6bbaa6e875473b73b63c8..bfcb23cf6217670f171f1f386736c8201a6b1fc1 100644
--- a/src/common/cbuf.c
+++ b/src/common/cbuf.c
@@ -1275,7 +1275,7 @@ cbuf_get_fd (void *dstbuf, int *psrcfd, int len)
do {
n = read(*psrcfd, dstbuf, len);
- } while ((n < 0) && (errno == EINTR));
+ } while ((n < 0) && ((errno == EINTR) || (errno == EAGAIN)));
return(n);
}
diff --git a/src/common/log.c b/src/common/log.c
index 8f5a18a44b2f3991991f8b9438f98202985367d5..3237f662db67fa7a4770bc66b4837a3aca9aac47 100644
--- a/src/common/log.c
+++ b/src/common/log.c
@@ -414,6 +414,28 @@ static char *vxstrfmt(const char *fmt, va_list ap)
} else
xstrcat(buf, "%d");
break;
+ case 'u':
+ if (unprocessed == 0) {
+ snprintf(tmp, sizeof(tmp), "%u",
+ va_arg(ap, int));
+ xstrcat(buf, tmp);
+ } else
+ xstrcat(buf, "%u");
+ break;
+ case 'l':
+ if ((unprocessed == 0) && (*(p+1) == 'u')) {
+ snprintf(tmp, sizeof(tmp), "%lu",
+ va_arg(ap, long unsigned));
+ xstrcat(buf, tmp);
+ p++;
+ } else if ((unprocessed==0) && (*(p+1)=='d')) {
+ snprintf(tmp, sizeof(tmp), "%ld",
+ va_arg(ap, long int));
+ xstrcat(buf, tmp);
+ p++;
+ } else
+ xstrcat(buf, "%l");
+ break;
default: /* try to handle the rest */
xstrcatchar(buf, '%');
xstrcatchar(buf, *p);
diff --git a/src/common/macros.h b/src/common/macros.h
index 45c18c9c50400cafc500ae12b3b2af46c2a3e9b2..d654fa37d0d14e7be486862edfca711547488e28 100644
--- a/src/common/macros.h
+++ b/src/common/macros.h
@@ -192,17 +192,47 @@ typedef enum {false, true} bool;
} \
} _STMT_END
+# ifdef PTHREAD_SCOPE_SYSTEM
+# define slurm_attr_init(attr) \
+ _STMT_START { \
+ if (pthread_attr_init(attr)) \
+ fatal("pthread_attr_init: %m"); \
+ /* we want 1:1 threads if there is a choice */ \
+ if (pthread_attr_setscope(attr, PTHREAD_SCOPE_SYSTEM)) \
+ error("pthread_attr_setscope: %m"); \
+ if (pthread_attr_setstacksize(attr, 1024*1024)) \
+ error("pthread_attr_setstacksize: %m"); \
+ } _STMT_END
+# else
+# define slurm_attr_init(attr) \
+ _STMT_START { \
+ if (pthread_attr_init(attr)) \
+ fatal("pthread_attr_init: %m"); \
+ if (pthread_attr_setstacksize(attr, 1024*1024)) \
+ error("pthread_attr_setstacksize: %m"); \
+ } _STMT_END
+# endif
+
#else /* !WITH_PTHREADS */
# define slurm_mutex_init(mutex)
# define slurm_mutex_destroy(mutex)
# define slurm_mutex_lock(mutex)
# define slurm_mutex_unlock(mutex)
+# define slurm_attr_init(attr)
#endif /* WITH_PTHREADS */
#ifndef strong_alias
-# define strong_alias(name, aliasname) \
- extern __typeof (name) aliasname __attribute ((alias (#name)))
+# if USE_ALIAS
+# define strong_alias(name, aliasname) \
+ extern __typeof (name) aliasname __attribute ((alias (#name)))
+# else
+ /* dummy function definition,
+ * confirm "aliasname" is free and waste "name" */
+# define strong_alias(name, aliasname) \
+ extern void aliasname(int name)
+# endif
#endif
+
#endif /* !_MACROS_H */
diff --git a/src/common/slurm_cred.c b/src/common/slurm_cred.c
index 3a6ddc4ecf2fbbf4b546da4527733fc03774bc3d..a9901160ba7096646d3637c1518165d422205016 100644
--- a/src/common/slurm_cred.c
+++ b/src/common/slurm_cred.c
@@ -31,8 +31,10 @@
#include <slurm/slurm_errno.h>
-#include <stdarg.h>
#include <fcntl.h>
+#include <stdarg.h>
+#include <stdlib.h>
+#include <sys/time.h>
/*
* OpenSSL includes
@@ -395,14 +397,21 @@ slurm_cred_faker(slurm_cred_arg_t *arg)
cred->ctime = time(NULL);
cred->siglen = SLURM_IO_KEY_SIZE;
- if ((fd = open("/dev/urandom", O_RDONLY)) < 0)
- error ("unable to open /dev/random: %m");
-
cred->signature = xmalloc(cred->siglen * sizeof(char));
- read(fd, cred->signature, cred->siglen);
- if (close(fd) < 0)
- error ("close(/dev/random): %m");
+ if ((fd = open("/dev/urandom", O_RDONLY)) >= 0) {
+ read(fd, cred->signature, cred->siglen);
+ if (close(fd) < 0)
+ error ("close(/dev/urandom): %m");
+ } else { /* Note: some systems lack this file */
+ unsigned int i;
+ struct timeval tv;
+ gettimeofday(&tv, NULL);
+ i = (unsigned int) (tv.tv_sec + tv.tv_usec);
+ srand((unsigned int) i);
+ for (i=0; i<cred->siglen; i++)
+ cred->signature[i] = (rand() & 0xff);
+ }
slurm_mutex_unlock(&cred->mutex);
return cred;
@@ -966,6 +975,7 @@ _slurm_cred_sign(slurm_cred_ctx_t ctx, slurm_cred_t cred)
rc = SLURM_ERROR;
}
+ EVP_MD_CTX_cleanup(&ectx);
free_buf(buffer);
return rc;
diff --git a/src/common/slurm_protocol_interface.h b/src/common/slurm_protocol_interface.h
index c8f36b3e00246a381b9b94037d3f72d0b0556c8d..b302fa0276532bbc67b6f9c63187fe1b4fd1cd8e 100644
--- a/src/common/slurm_protocol_interface.h
+++ b/src/common/slurm_protocol_interface.h
@@ -65,7 +65,7 @@
#include "src/common/pack.h"
#include "src/common/slurm_protocol_common.h"
-#define SLURM_MESSAGE_TIMEOUT_MSEC_STATIC 3000
+#define SLURM_MESSAGE_TIMEOUT_MSEC_STATIC 5000
/****************\
** Data Types **
diff --git a/src/common/slurm_protocol_socket_implementation.c b/src/common/slurm_protocol_socket_implementation.c
index 47ca7ddd17d10f381c5d87f73ad0b0dcac0b819d..e9f15b84eda528e3026823f95fa7049358c815c7 100644
--- a/src/common/slurm_protocol_socket_implementation.c
+++ b/src/common/slurm_protocol_socket_implementation.c
@@ -252,7 +252,7 @@ int _slurm_send_timeout(slurm_fd fd, char *buf, size_t size,
goto done;
}
if ((rc = poll(&ufds, 1, timeout)) <= 0) {
- if ((rc == 0) || (errno == EINTR))
+ if ((rc == 0) || (errno == EINTR) || (errno == EAGAIN))
continue;
else {
debug("_slurm_send_timeout at %d of %d, "
@@ -329,7 +329,7 @@ int _slurm_recv_timeout(slurm_fd fd, char *buffer, size_t size,
}
if ((rc = poll(&ufds, 1, timeout)) <= 0) {
- if ((errno == EINTR) || (rc == 0))
+ if ((errno == EINTR) || (errno == EAGAIN) || (rc == 0))
continue;
else {
debug("_slurm_recv_timeout at %d of %d, "
diff --git a/src/common/slurm_protocol_util.c b/src/common/slurm_protocol_util.c
index 638c2a3da2e4d23b619c73316ec3af636efdea26..cf0da53c6f0e299415ec75d4ae5045abef30f1e1 100644
--- a/src/common/slurm_protocol_util.c
+++ b/src/common/slurm_protocol_util.c
@@ -62,6 +62,7 @@ void init_header(header_t * header, slurm_msg_type_t msg_type,
header->version = SLURM_PROTOCOL_VERSION;
header->flags = flags;
header->msg_type = msg_type;
+ header->body_length = 0; /* over-written later */
}
/*
diff --git a/src/common/slurm_xlator.h b/src/common/slurm_xlator.h
index 2774f0b18680002f015d0a49450e94825782fc4d..52b00fb34dff2025213d3b3f37fec15e06c13f65 100644
--- a/src/common/slurm_xlator.h
+++ b/src/common/slurm_xlator.h
@@ -10,7 +10,7 @@
*
* All SLURM functions referenced from the switch and auth plugins should
* have aliases established. Functions not referenced from the plugins
- * need not be aliased.
+ * need not be aliased.
*
* To use this header file:
* 1. In the module containing the exported function code, establish an
@@ -24,6 +24,8 @@
* and remove other slurm header files (they should all be in this header).
* This logic will have the plugin link only to the function names with
* the "slurm_" prefix.
+ *
+ * NOTE: Not all operating systems support this function aliasing (e.g. AIX).
*****************************************************************************
* Copyright (C) 2004 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
@@ -51,6 +53,12 @@
#ifndef __SLURM_XLATOR_H__
#define __SLURM_XLATOR_H__
+#if HAVE_CONFIG_H
+# include "config.h"
+#endif
+
+#if USE_ALIAS
+
/* arg_desc.[ch] functions*/
#define arg_count slurm_arg_count
#define arg_idx_by_name slurm_arg_idx_by_name
@@ -233,6 +241,8 @@
#define xstrdup slurm_xstrdup
#define xbasename slurm_xbasename
+#endif /* USE_ALIAS */
+
/* Include the function definitions after redefining their names. */
#include "src/common/arg_desc.h"
#include "src/common/bitstring.h"
diff --git a/src/common/xmalloc.c b/src/common/xmalloc.c
index 7678fc6d6c0a9779d7e64f0a7150a2555503167b..42eb90a6c3013c3ffbe478c856ee1e559f5cf804 100644
--- a/src/common/xmalloc.c
+++ b/src/common/xmalloc.c
@@ -264,7 +264,7 @@ void slurm_xfree(void **item, const char *file, int line, const char *func)
static void malloc_assert_failed(char *expr, const char *file,
int line, const char *caller, const char *func)
{
- fatal("%s() Error: from %s:%d: %s(): Assertion (%s) failed",
+ error("%s() Error: from %s:%d: %s(): Assertion (%s) failed",
func, file, line, caller, expr);
abort();
}
diff --git a/src/scancel/Makefile.am b/src/scancel/Makefile.am
index ab37c7449bae4b7571b13e85a4cb5bc3df85e4a1..2681e24baeb98a54057c0f4e47428b4e2fd3a48b 100644
--- a/src/scancel/Makefile.am
+++ b/src/scancel/Makefile.am
@@ -14,6 +14,8 @@ scancel_LDADD = \
noinst_HEADERS = scancel.h
scancel_SOURCES = scancel.c opt.c
+scancel_LDFLAGS = -export-dynamic $(CMD_LDFLAGS)
+
force:
$(scancel_LDADD) : force
@cd `dirname $@` && $(MAKE) `basename $@`
diff --git a/src/scontrol/Makefile.am b/src/scontrol/Makefile.am
index 858c07f1635e6a4169f3f1a363d480e22559639f..781a0b559f2f6ce17177c2c81281cb6e8e5943b8 100644
--- a/src/scontrol/Makefile.am
+++ b/src/scontrol/Makefile.am
@@ -12,5 +12,7 @@ LDADD = \
$(top_builddir)/src/api/libslurm.la \
$(READLINE_LIBS)
-
scontrol_SOURCES = scontrol.c
+
+scontrol_LDFLAGS = -export-dynamic $(CMD_LDFLAGS)
+
diff --git a/src/sinfo/Makefile.am b/src/sinfo/Makefile.am
index 963529ba6af11c437cb2f45399754386ac6ba0c9..78b6bfe91cd677217b38d2f23f78c49ff9448ef6 100644
--- a/src/sinfo/Makefile.am
+++ b/src/sinfo/Makefile.am
@@ -18,3 +18,5 @@ force:
$(sinfo_LDADD) : force
@cd `dirname $@` && $(MAKE) `basename $@`
+sinfo_LDFLAGS = -export-dynamic $(CMD_LDFLAGS)
+
diff --git a/src/slurmctld/Makefile.am b/src/slurmctld/Makefile.am
index 47d748ad90b5e18b4ed9f2bd88c2b431362b5f2b..23f4356b11b0f86d3d614e26594d3edc904e23be 100644
--- a/src/slurmctld/Makefile.am
+++ b/src/slurmctld/Makefile.am
@@ -42,8 +42,7 @@ slurmctld_SOURCES = \
state_save.h \
step_mgr.c
-
-slurmctld_LDFLAGS = -export-dynamic
+slurmctld_LDFLAGS = -export-dynamic $(CMD_LDFLAGS)
force:
$(slurmctld_LDADD) : force
diff --git a/src/slurmctld/agent.c b/src/slurmctld/agent.c
index 9c3a8ca90cd3c1de94cd8167d39e1253f3ba9790..afe7efbcff1ca526fa1411174b28a5a6ad01de2c 100644
--- a/src/slurmctld/agent.c
+++ b/src/slurmctld/agent.c
@@ -188,15 +188,10 @@ void *agent(void *args)
thread_ptr = agent_info_ptr->thread_struct;
/* start the watchdog thread */
- if (pthread_attr_init(&attr_wdog))
- fatal("pthread_attr_init error %m");
+ slurm_attr_init(&attr_wdog);
if (pthread_attr_setdetachstate
(&attr_wdog, PTHREAD_CREATE_JOINABLE))
error("pthread_attr_setdetachstate error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- if (pthread_attr_setscope(&attr_wdog, PTHREAD_SCOPE_SYSTEM))
- error("pthread_attr_setscope error %m");
-#endif
while (pthread_create(&thread_wdog, &attr_wdog, _wdog,
(void *) agent_info_ptr)) {
error("pthread_create error %m");
@@ -222,16 +217,10 @@ void *agent(void *args)
* _thread_per_node_rpc() */
task_specific_ptr = _make_task_data(agent_info_ptr, i);
- if (pthread_attr_init(&thread_ptr[i].attr))
- fatal("pthread_attr_init error %m");
+ slurm_attr_init(&thread_ptr[i].attr);
if (pthread_attr_setdetachstate(&thread_ptr[i].attr,
PTHREAD_CREATE_DETACHED))
error("pthread_attr_setdetachstate error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- if (pthread_attr_setscope(&thread_ptr[i].attr,
- PTHREAD_SCOPE_SYSTEM))
- error("pthread_attr_setscope error %m");
-#endif
while ((rc = pthread_create(&thread_ptr[i].thread,
&thread_ptr[i].attr,
_thread_per_node_rpc,
@@ -856,15 +845,10 @@ void agent_queue_request(agent_arg_t *agent_arg_ptr)
if (agent_cnt < MAX_AGENT_CNT) { /* execute now */
pthread_attr_t attr_agent;
pthread_t thread_agent;
- if (pthread_attr_init(&attr_agent))
- fatal("pthread_attr_init error %m");
+ slurm_attr_init(&attr_agent);
if (pthread_attr_setdetachstate
(&attr_agent, PTHREAD_CREATE_DETACHED))
error("pthread_attr_setdetachstate error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- if (pthread_attr_setscope(&attr_agent, PTHREAD_SCOPE_SYSTEM))
- error("pthread_attr_setscope error %m");
-#endif
if (pthread_create(&thread_agent, &attr_agent,
agent, (void *) agent_arg_ptr) == 0)
return;
@@ -896,15 +880,10 @@ static void _spawn_retry_agent(agent_arg_t * agent_arg_ptr)
debug2("Spawning RPC agent for msg_type %u",
agent_arg_ptr->msg_type);
- if (pthread_attr_init(&attr_agent))
- fatal("pthread_attr_init error %m");
+ slurm_attr_init(&attr_agent);
if (pthread_attr_setdetachstate(&attr_agent,
PTHREAD_CREATE_DETACHED))
error("pthread_attr_setdetachstate error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- if (pthread_attr_setscope(&attr_agent, PTHREAD_SCOPE_SYSTEM))
- error("pthread_attr_setscope error %m");
-#endif
while (pthread_create(&thread_agent, &attr_agent,
agent, (void *) agent_arg_ptr)) {
error("pthread_create error %m");
diff --git a/src/slurmctld/backup.c b/src/slurmctld/backup.c
index 047fc01f53d33f5826c65ec52c540a639ed4f5b9..6c7d7329ae51f706d11b4efbe5bd177840bb399b 100644
--- a/src/slurmctld/backup.c
+++ b/src/slurmctld/backup.c
@@ -90,13 +90,7 @@ void run_backup(void)
/*
* create attached thread to process RPCs
*/
- if (pthread_attr_init(&thread_attr_rpc))
- fatal("pthread_attr_init error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- /* we want 1:1 threads if there is a choice */
- if (pthread_attr_setscope(&thread_attr_rpc, PTHREAD_SCOPE_SYSTEM))
- error("pthread_attr_setscope error %m");
-#endif
+ slurm_attr_init(&thread_attr_rpc);
if (pthread_create(&slurmctld_config.thread_id_rpc,
&thread_attr_rpc, _background_rpc_mgr, NULL))
fatal("pthread_create error %m");
@@ -104,13 +98,7 @@ void run_backup(void)
/*
* create attached thread for signal handling
*/
- if (pthread_attr_init(&thread_attr_sig))
- fatal("pthread_attr_init error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- /* we want 1:1 threads if there is a choice */
- if (pthread_attr_setscope(&thread_attr_sig, PTHREAD_SCOPE_SYSTEM))
- error("pthread_attr_setscope error %m");
-#endif
+ slurm_attr_init(&thread_attr_sig);
if (pthread_create(&slurmctld_config.thread_id_sig,
&thread_attr_sig, _background_signal_hand, NULL))
fatal("pthread_create %m");
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index 2a4085d935d6984a605ab25bcf0df08723dad3f2..3aa75a0b373a3630e24a0632768fcb7d48ed9171 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -253,14 +253,7 @@ int main(int argc, char *argv[])
slurm_mutex_lock(&slurmctld_config.thread_count_lock);
slurmctld_config.server_thread_count++;
slurm_mutex_unlock(&slurmctld_config.thread_count_lock);
- if (pthread_attr_init(&thread_attr_rpc))
- fatal("pthread_attr_init error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- /* we want 1:1 threads if there is a choice */
- if (pthread_attr_setscope
- (&thread_attr_rpc, PTHREAD_SCOPE_SYSTEM))
- error("pthread_attr_setscope error %m");
-#endif
+ slurm_attr_init(&thread_attr_rpc);
if (pthread_create(&slurmctld_config.thread_id_rpc,
&thread_attr_rpc,_slurmctld_rpc_mgr, NULL))
fatal("pthread_create error %m");
@@ -268,14 +261,7 @@ int main(int argc, char *argv[])
/*
* create attached thread for signal handling
*/
- if (pthread_attr_init(&thread_attr_sig))
- fatal("pthread_attr_init error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- /* we want 1:1 threads if there is a choice */
- if (pthread_attr_setscope(&thread_attr_sig,
- PTHREAD_SCOPE_SYSTEM))
- error("pthread_attr_setscope error %m");
-#endif
+ slurm_attr_init(&thread_attr_sig);
if (pthread_create(&slurmctld_config.thread_id_sig,
&thread_attr_sig, _slurmctld_signal_hand,
NULL))
@@ -284,14 +270,7 @@ int main(int argc, char *argv[])
/*
* create attached thread for state save
*/
- if (pthread_attr_init(&thread_attr_save))
- fatal("pthread_attr_init error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- /* we want 1:1 threads if there is a choice */
- if (pthread_attr_setscope(&thread_attr_save,
- PTHREAD_SCOPE_SYSTEM))
- error("pthread_attr_setscope error %m");
-#endif
+ slurm_attr_init(&thread_attr_save);
if (pthread_create(&slurmctld_config.thread_id_save,
&thread_attr_save, slurmctld_state_save,
NULL))
@@ -492,17 +471,10 @@ static void *_slurmctld_rpc_mgr(void *no_data)
debug3("_slurmctld_rpc_mgr pid = %u", getpid());
/* threads to process individual RPC's are detached */
- if (pthread_attr_init(&thread_attr_rpc_req))
- fatal("pthread_attr_init %m");
+ slurm_attr_init(&thread_attr_rpc_req);
if (pthread_attr_setdetachstate
(&thread_attr_rpc_req, PTHREAD_CREATE_DETACHED))
fatal("pthread_attr_setdetachstate %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- /* we want 1:1 threads if there is a choice */
- if (pthread_attr_setscope
- (&thread_attr_rpc_req, PTHREAD_SCOPE_SYSTEM))
- error("pthread_attr_setscope error %m");
-#endif
/* initialize port for RPCs */
lock_slurmctld(config_read_lock);
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 5382775de8af46231ea2d726ad65c81a38275a7d..dbefae2d8ef3c13ee8c79de14cea3409913569d7 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -809,17 +809,22 @@ void _add_job_hash(struct job_record *job_ptr)
struct job_record *find_job_record(uint32_t job_id)
{
int i;
+ struct job_record *job_ptr;
/* First try to find via hash table */
- if (job_hash[JOB_HASH_INX(job_id)] &&
- job_hash[JOB_HASH_INX(job_id)]->job_id == job_id)
- return job_hash[JOB_HASH_INX(job_id)];
+ job_ptr = job_hash[JOB_HASH_INX(job_id)];
+ if (job_ptr && (job_ptr->job_id == job_id)) {
+ xassert (job_ptr->magic == JOB_MAGIC);
+ return job_ptr;
+ }
/* linear search of overflow hash table overflow */
for (i = 0; i < max_hash_over; i++) {
- if (job_hash_over[i] != NULL &&
- job_hash_over[i]->job_id == job_id)
- return job_hash_over[i];
+ job_ptr = job_hash_over[i];
+ if (job_ptr && (job_ptr->job_id == job_id)) {
+ xassert (job_ptr->magic == JOB_MAGIC);
+ return job_ptr;
+ }
}
return NULL;
@@ -1044,8 +1049,12 @@ void dump_job_desc(job_desc_msg_t * job_specs)
(long) job_specs->kill_on_node_fail : -1L;
task_dist = (job_specs->task_dist != (uint16_t) NO_VAL) ?
(long) job_specs->task_dist : -1L;
- debug3(" kill_on_node_fail=%ld task_dist=%ld script=%.40s...",
- kill_on_node_fail, task_dist, job_specs->script);
+ if (job_specs->script) /* log has problem with string len & null */
+ debug3(" kill_on_node_fail=%ld task_dist=%ld script=%.40s...",
+ kill_on_node_fail, task_dist, job_specs->script);
+ else
+ debug3(" kill_on_node_fail=%ld task_dist=%ld script=%s",
+ kill_on_node_fail, task_dist, job_specs->script);
if (job_specs->argc == 1)
debug3(" argv=\"%s\"",
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index a95a6e28489b8c3c9011895a473dc04bc68bdcd5..63a5b3f854f91114b9a4588765c376c035967541 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -42,6 +42,7 @@
#include <fcntl.h>
#include "src/common/hostlist.h"
+#include "src/common/macros.h"
#include "src/common/pack.h"
#include "src/common/xassert.h"
#include "src/common/xstring.h"
@@ -111,7 +112,6 @@ char * bitmap2node_name (bitstr_t *bitmap)
hostlist_uniq(hl);
hostlist_ranged_string(hl, 8192, buf);
hostlist_destroy(hl);
-
return xstrdup(buf);
}
@@ -1362,16 +1362,10 @@ void msg_to_slurmd (slurm_msg_type_t msg_type)
xfree (kill_agent_args);
else {
debug ("Spawning agent msg_type=%d", msg_type);
- if (pthread_attr_init (&kill_attr_agent))
- fatal ("pthread_attr_init error %m");
+ slurm_attr_init (&kill_attr_agent);
if (pthread_attr_setdetachstate (&kill_attr_agent,
PTHREAD_CREATE_DETACHED))
error ("pthread_attr_setdetachstate error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- if (pthread_attr_setscope (&kill_attr_agent,
- PTHREAD_SCOPE_SYSTEM))
- error ("pthread_attr_setscope error %m");
-#endif
while (pthread_create (&kill_thread_agent, &kill_attr_agent,
agent, (void *)kill_agent_args)) {
error ("pthread_create error %m");
diff --git a/src/slurmctld/ping_nodes.c b/src/slurmctld/ping_nodes.c
index 729d7f291c9f0764c408637206c0dbe44e000e73..1a3a06c764618b90e1940862a6caab905f9b4bb5 100644
--- a/src/slurmctld/ping_nodes.c
+++ b/src/slurmctld/ping_nodes.c
@@ -230,16 +230,10 @@ void ping_nodes (void)
sizeof(host_str), host_str);
debug2 ("Spawning ping agent for %s", host_str);
ping_begin();
- if (pthread_attr_init (&ping_attr_agent))
- fatal ("pthread_attr_init error %m");
+ slurm_attr_init (&ping_attr_agent);
if (pthread_attr_setdetachstate (&ping_attr_agent,
PTHREAD_CREATE_DETACHED))
error ("pthread_attr_setdetachstate error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- if (pthread_attr_setscope (&ping_attr_agent,
- PTHREAD_SCOPE_SYSTEM))
- error ("pthread_attr_setscope error %m");
-#endif
while (pthread_create (&ping_thread_agent, &ping_attr_agent,
agent, (void *)ping_agent_args)) {
error ("pthread_create error %m");
@@ -257,16 +251,10 @@ void ping_nodes (void)
sizeof(host_str), host_str);
debug2 ("Spawning registration agent for %s", host_str);
ping_begin();
- if (pthread_attr_init (®_attr_agent))
- fatal ("pthread_attr_init error %m");
+ slurm_attr_init (®_attr_agent);
if (pthread_attr_setdetachstate (®_attr_agent,
PTHREAD_CREATE_DETACHED))
error ("pthread_attr_setdetachstate error %m");
-#ifdef PTHREAD_SCOPE_SYSTEM
- if (pthread_attr_setscope (®_attr_agent,
- PTHREAD_SCOPE_SYSTEM))
- error ("pthread_attr_setscope error %m");
-#endif
while (pthread_create (®_thread_agent, ®_attr_agent,
agent, (void *)reg_agent_args)) {
error ("pthread_create error %m");
diff --git a/src/slurmd/Makefile.am b/src/slurmd/Makefile.am
index a2f0169f16341cc2da48a5a45b50428a71042987..f10755ce8100eed3a53dd5325caaebadc13babfe 100644
--- a/src/slurmd/Makefile.am
+++ b/src/slurmd/Makefile.am
@@ -31,7 +31,7 @@ slurmd_SOURCES = \
ulimits.c ulimits.h \
setproctitle.c setproctitle.h
-slurmd_LDFLAGS = -export-dynamic
+slurmd_LDFLAGS = -export-dynamic $(CMD_LDFLAGS)
force:
$(slurmd_LDADD) : force
diff --git a/src/slurmd/io.c b/src/slurmd/io.c
index a28be9c901337a9457c06e4e765f6aab6a433af0..9c1e7865c7730e668438abe942444173907cc4c3 100644
--- a/src/slurmd/io.c
+++ b/src/slurmd/io.c
@@ -52,6 +52,7 @@
#include "src/common/io_hdr.h"
#include "src/common/cbuf.h"
#include "src/common/log.h"
+#include "src/common/macros.h"
#include "src/common/fd.h"
#include "src/common/list.h"
#include "src/common/xmalloc.h"
@@ -254,13 +255,7 @@ io_spawn_handler(slurmd_job_t *job)
if (_io_prepare_tasks(job) < 0)
return SLURM_FAILURE;
- if ((errno = pthread_attr_init(&attr)) != 0)
- error("pthread_attr_init: %m");
-
-#ifdef PTHREAD_SCOPE_SYSTEM
- if ((errno = pthread_attr_setscope(&attr, PTHREAD_SCOPE_SYSTEM)) != 0)
- error("pthread_attr_setscope: %m");
-#endif
+ slurm_attr_init(&attr);
xassert(_validate_io_list(job->objs));
pthread_create(&job->ioid, &attr, &_io_thr, (void *)job);
@@ -1412,7 +1407,8 @@ _task_error(io_obj_t *obj, List objs)
if (getsockopt(obj->fd, SOL_SOCKET, SO_ERROR, &err, &size) < 0)
error ("getsockopt: %m");
- _update_error_state(t, E_POLL, err);
+ else
+ _update_error_state(t, E_POLL, err);
_obj_close(obj, objs);
return -1;
}
@@ -1525,8 +1521,7 @@ _client_error(io_obj_t *obj, List objs)
if (getsockopt(obj->fd, SOL_SOCKET, SO_ERROR, &err, &size) < 0)
error ("getsockopt: %m");
-
- if (err != ECONNRESET) /* Do not log connection resets */
+ else if (err != ECONNRESET) /* Do not log connection resets */
_update_error_state(io, E_POLL, err);
return 0;
@@ -1578,7 +1573,13 @@ _update_error_state(struct io_info *io, enum error_type type, int err)
{
xassert(io != NULL);
xassert(io->magic == IO_MAGIC);
- xassert(err > 0);
+
+ /* getsockopt(,,SO_ERROR,&err,) returns err value of -1
+ * under AIX under some circumstances */
+ if (err <= 0) {
+ error("Unspecified I/O error <task %d>", io->id);
+ return 0;
+ }
if ( (io->err.e_type == type)
&& (io->err.e_last == err ) ) {
diff --git a/src/slurmd/slurmd.c b/src/slurmd/slurmd.c
index 695e80852a0a8d5a46a9deaa55af1f34632ed0bc..5cb0a25cca2a0b8e0657a217c7ccc1162893472c 100644
--- a/src/slurmd/slurmd.c
+++ b/src/slurmd/slurmd.c
@@ -287,12 +287,7 @@ _handle_connection(slurm_fd fd, slurm_addr *cli)
arg->fd = fd;
arg->cli_addr = cli;
- if ((rc = pthread_attr_init(&attr)) != 0) {
- error("pthread_attr_init: %s", slurm_strerror(rc));
- xfree(arg);
- return;
- }
-
+ slurm_attr_init(&attr);
rc = pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
if (rc != 0) {
errno = rc;
diff --git a/src/squeue/Makefile.am b/src/squeue/Makefile.am
index e65d0d20cc06e47976ff7ccddd2fe8fb77492abc..bf5f857ed27e5a2d255fbfc49ec7a6f836e2f68f 100644
--- a/src/squeue/Makefile.am
+++ b/src/squeue/Makefile.am
@@ -14,3 +14,5 @@ LDADD = $(top_builddir)/src/common/libcommon.la \
noinst_HEADERS = squeue.h print.h
squeue_SOURCES = squeue.c print.c opts.c sort.c
+squeue_LDFLAGS = -export-dynamic $(CMD_LDFLAGS)
+
diff --git a/src/srun/Makefile.am b/src/srun/Makefile.am
index 0ee22cd47682f426c9a2658848585cdfe85e7fa6..35a19e276a652ac4fc95b02a76ff7424b36100a7 100644
--- a/src/srun/Makefile.am
+++ b/src/srun/Makefile.am
@@ -44,6 +44,8 @@ srun_LDADD = \
$(SRUN_WRAPPER_OBJ) \
$(convenience_libs)
+srun_LDFLAGS = -export-dynamic $(CMD_LDFLAGS)
+
EXTRA_srun_SOURCES = srun.wrapper.c
srun_DEPENDENCIES = $(SRUN_WRAPPER_OBJ)
diff --git a/src/srun/io.c b/src/srun/io.c
index eedee8d41f03eb752762ad3d17bdcdcd092596d8..4535e4d8d586d0129e58f3f7246fa1dc1ee655f4 100644
--- a/src/srun/io.c
+++ b/src/srun/io.c
@@ -122,7 +122,7 @@ _handle_pollerr(fd_info_t *info)
if (getsockopt(fd, SOL_SOCKET, SO_ERROR, (void *)&err, &size) < 0)
error("_handle_error_poll: getsockopt: %m");
- if (err)
+ if (err > 0)
debug3("%d: poll error on fd %d: %s",
info->taskid, fd, slurm_strerror(err));
else
@@ -557,7 +557,7 @@ _read_io_header(int fd, job_t *job, char *host)
io_hdr_t hdr;
if (cbuf_write_from_fd(cb, fd, size, NULL) < 0) {
- error ("Bad stream header read: %m");
+ error ("Bad stream header write: %m");
goto fail;
}
diff --git a/src/srun/msg.c b/src/srun/msg.c
index c2146899193852661f0541f8023735d2a0121a86..41efc6075ef89c2759a36998674b0bf9a3e73844 100644
--- a/src/srun/msg.c
+++ b/src/srun/msg.c
@@ -610,12 +610,19 @@ static int
_do_poll(job_t *job, struct pollfd *fds, int timeout)
{
nfds_t nfds = (job->njfds + 1);
- int rc;
+ int rc, to;
- while ((rc = poll(fds, nfds, timeout * 1000)) < 0) {
+ if (timeout > 0)
+ to = timeout * 1000;
+ else
+ to = timeout;
+
+ while ((rc = poll(fds, nfds, to)) < 0) {
switch (errno) {
+ case EAGAIN:
case EINTR: continue;
case ENOMEM:
+ case EINVAL:
case EFAULT: fatal("poll: %m");
default: error("poll: %m. Continuing...");
continue;