next

Makefile

@@ -122,7 +122,7 @@ CFLAGS_MYSQL += -DMX_MYSQL_FAIL_WAIT_DEFAULT=5
 
 CFLAGS += -g
 CFLAGS += -O3
-CFLAGS += -Wall -Wextra -Wno-override-init
+CFLAGS += -Wall -Wextra -Wno-override-init -Werror
 CFLAGS += -DMXQ_VERSION=\"${MXQ_VERSION}\"
 CFLAGS += -DMXQ_VERSIONFULL=\"${MXQ_VERSIONFULL}\"
 CFLAGS += -DMXQ_VERSIONDATE=\"${MXQ_VERSIONDATE}\"
@@ -361,6 +361,14 @@ ppidcache.h += ppidcache.h
 
 ########################################################################
 
+## parser.tab.o
+
+# Disable "variable 'yynerrs' set but not used" diagnostic which
+# appears with bison-3.4.2 + llvm 15.0.4
+
+parser.tab.o: parser.tab.c parser.tab.h
+	$(call quiet-command,${CC} ${CFLAGS} -Wno-unused-but-set-variable -o $@ -c $<,"     CC $@")
+
 ### mx_getopt.o --------------------------------------------------------
 
 mx_getopt.o: $(mx_getopt.h)

helper/gpu-setup

@@ -235,6 +235,8 @@ job_init() {
     pid=$1
     uid=$2
 
+    echo "XXX $$ job_init $pid: called" >&2
+
     test -d /dev/shm/mxqd/gpu_devs || die "$0: Not initialized (no dir /dev/shm/mxqd/gpu_devs)"
 
     shopt -s nullglob
@@ -253,6 +255,7 @@ job_init() {
                 esac
             done
             cat $d/uuid
+            echo "XXX $$ job_init $pid: allocated gpu from $d" >&2
             exit
         fi
     done
@@ -263,12 +266,14 @@ job_release() {
     (( $# == 1 )) || usage
     pid=$1
 
+    echo "XXX $$ job_release $pid: called" >&2
+
     test -d /dev/shm/mxqd/gpu_devs || die "$0: Not initialized (no dir /dev/shm/mxqd/gpu_devs)"
     for d in /dev/shm/mxqd/gpu_devs/???; do
         if [ -e $d/pid ]; then
             test_pid="$(cat $d/pid 2>/dev/null)"
             if [ "$pid" = "$test_pid" ]; then
-                rm $d/pid
+                echo "XXX $$ job_release $pid: found my pid in $d, releasing" >&2
                 for f in $(cat $d/access-files); do
                     case $f in
                         /dev/nvidia-caps/nvidia-cap*)
@@ -281,11 +286,12 @@ job_release() {
                             ;;
                     esac
                 done
+                rm $d/pid
                 exit 0
             fi
         fi
     done
-    die "$0: job_release: job with $pid has no GPU locked"
+    die "$0: job_release: job with pid $pid has no GPU locked"
 }
 
 show() {

mx_getopt.c

@@ -127,8 +127,7 @@ static int find_short_option(struct mx_option *options, char **name, char **opta
 
    assert(short_opt);
 
-   if (optarg)
-       *optarg = NULL;
+   *optarg = NULL;
 
    for (i=0, idx=-1; options[i].long_opt || options[i].short_opt; i++) {
 

mx_mysql.c

@@ -630,24 +630,21 @@ int mx_mysql_option_set_default_file(struct mx_mysql *mysql, char *fname)
     return 0;
 }
 
-int mx_mysql_option_set_default_group(struct mx_mysql *mysql, char *group)
+void mx_mysql_option_set_default_group(struct mx_mysql *mysql, char *group)
 {
     assert(mysql);
 
     if (group && !(*group))
         group = NULL;
 
     mysql->default_group = group;
-
-    return 0;
 }
 
-int mx_mysql_option_set_reconnect(struct mx_mysql *mysql, int reconnect)
+void mx_mysql_option_set_reconnect(struct mx_mysql *mysql, int reconnect)
 {
     assert(mysql);
 
     mysql->reconnect = (bool)!!reconnect;
-    return 0;
 }
 
 static int mx_mysql_real_connect(struct mx_mysql *mysql, const char *host, const char *user, const char *passwd, const char *db, unsigned int port, const char *unix_socket, unsigned long client_flag)
@@ -1154,8 +1151,8 @@ struct mx_mysql_stmt *mx_mysql_statement_prepare(struct mx_mysql *mysql, char *s
 
 int mx_mysql_statement_close(struct mx_mysql_stmt **stmt)
 {
-    assert(stmt);
-    assert(*stmt);
+    if (*stmt == NULL)
+        return 0;
 
     mx__mysql_stmt_free_result(*stmt);
     mx__mysql_stmt_close(*stmt);
@@ -1169,8 +1166,8 @@ int mx_mysql_statement_close(struct mx_mysql_stmt **stmt)
 
 int mx_mysql_statement_close_no_bind_cleanup(struct mx_mysql_stmt **stmt)
 {
-    assert(stmt);
-    assert(*stmt);
+    if (*stmt == NULL)
+        return 0;
 
     mx__mysql_stmt_free_result(*stmt);
     mx__mysql_stmt_close(*stmt);

mx_mysql.h

@@ -77,9 +77,9 @@ struct mx_mysql_stmt {
 
 int mx_mysql_initialize(struct mx_mysql **mysql);
 
-int mx_mysql_option_set_default_file(struct mx_mysql *mysql, char *fname);
-int mx_mysql_option_set_default_group(struct mx_mysql *mysql, char *group);
-int mx_mysql_option_set_reconnect(struct mx_mysql *mysql, int reconnect);
+int mx_mysql_option_set_default_file(struct mx_mysql *mysql, char *fname)  __attribute__ ((warn_unused_result));
+void mx_mysql_option_set_default_group(struct mx_mysql *mysql, char *group);
+void mx_mysql_option_set_reconnect(struct mx_mysql *mysql, int reconnect);
 
 int mx_mysql_connect(struct mx_mysql **mysql);
 int mx_mysql_connect_forever_sec(struct mx_mysql **mysql, unsigned int seconds);

mx_util.c

@@ -931,7 +931,7 @@ char **mx_strvec_new(void)
 {
     char **strvec;
 
-    strvec = calloc(sizeof(*strvec), 1);
+    strvec = calloc(1, sizeof(*strvec));
     if (!strvec)
         return NULL;
 
@@ -1414,3 +1414,11 @@ char *mx_pipe_external(char *helper, char **argv) {
     errno = err;
     return NULL;
 }
+
+void mx_die(const char *restrict fmt, ...) {
+    va_list ap;
+    va_start(ap, fmt);
+    vfprintf(stderr, fmt, ap);
+    va_end(ap);
+    _exit(1);
+}

mx_util.h

@@ -183,6 +183,8 @@ int mx_fs_get_sizes(const char *path, unsigned long *free, unsigned long *total)
 time_t mx_clock_boottime(void);
 int mx_call_external(char *helper, char **argv);
 char *mx_pipe_external(char *args, char **argv);
+void mx_die(const char *restrict fmt, ...) __attribute__ ((noreturn,format (printf, 1, 2)));
+
 
 #if __GLIBC__ <2 || __GLIBC__ == 2 && __GLIBC_MINOR__ < 34
 #include <sys/syscall.h>

mxq_reaper.c

@@ -8,11 +8,17 @@
 #include <sys/wait.h>
 #include <errno.h>
 #include <string.h>
+#include <stdarg.h>
 
 static const char REAPER_PNAME[] = "mxqd reaper";
 
-__attribute__((noreturn)) static void die(char *msg)  {
-    perror(msg);
+__attribute__ ((noreturn))
+__attribute__ ((format (printf, 1, 2)))
+static void die(const char *restrict fmt, ...) {
+    va_list ap;
+    va_start(ap, fmt);
+    vfprintf(stderr, fmt, ap);
+    va_end(ap);
     _exit(1);
 }
 
@@ -36,18 +42,18 @@ int main(int argc, char **argv) {
     struct timeval endtime;
 
     if (prctl(PR_SET_NAME, REAPER_PNAME, NULL, NULL, NULL) == -1)
-        die("PR_SET_NAME");
+        die("PR_SET_NAME: %m\n");
     user_pid = fork();
     if (user_pid == 0) {
         if (setreuid(uid, uid) == -1)
-            die("setreuid");
+            die("setreuid: %m\n");
         execvp(user_argv[0], user_argv);
-        die(user_argv[0]);
+        die("%s: %m\n", user_argv[0]);
     }
     if (user_pid == -1)
-        die("fork");
+        die("fork: %m\n");
     if (gettimeofday(&starttime, NULL) == -1)
-        die("gettimeofday");
+        die("gettimeofday: %m\n");
     while (1) {
         int status;
         pid_t pid = wait(&status);
@@ -57,48 +63,64 @@ int main(int argc, char **argv) {
             user_status = status;
     }
     if (gettimeofday(&endtime, NULL) == -1)
-        die("gettimeofday");
+        die("gettimeofday: %m\n");
     timersub(&endtime, &starttime, &user_time);
     if (getrusage(RUSAGE_CHILDREN, &user_rusage) == -1)
-        die("getrusage");
+        die("getrusage: %m\n");
 
    if (user_time.tv_sec<30) {
         int wait=30-user_time.tv_sec;
         sleep(wait);
     }
 
-    char *tmpfilename;
-    if (asprintf(&tmpfilename, "%s.tmp", spoolfilename) == -1)
-        die("");
-
-    FILE *out = fopen(tmpfilename,"w");
-    if (out == NULL)
-        die(tmpfilename);
-    fprintf(out,"1 %d %d %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld\n",
-        getpid(),
-        user_status,
-        user_time.tv_sec, user_time.tv_usec,
-        user_rusage.ru_utime.tv_sec, user_rusage.ru_utime.tv_usec,
-        user_rusage.ru_stime.tv_sec, user_rusage.ru_stime.tv_usec,
-        user_rusage.ru_maxrss,
-        user_rusage.ru_ixrss,
-        user_rusage.ru_idrss,
-        user_rusage.ru_isrss,
-        user_rusage.ru_minflt,
-        user_rusage.ru_majflt,
-        user_rusage.ru_nswap,
-        user_rusage.ru_inblock,
-        user_rusage.ru_oublock,
-        user_rusage.ru_msgsnd,
-        user_rusage.ru_msgrcv,
-        user_rusage.ru_nsignals,
-        user_rusage.ru_nvcsw,
-        user_rusage.ru_nivcsw
-    );
-    fflush(out);
-    fsync(fileno(out));
-    fclose(out);
-    if (rename(tmpfilename, spoolfilename) == -1)
-        die(spoolfilename);
+    while (1) {
+        // if anything fails, do not write to stderr, as this is the users
+        // stderr, just wait and retry.
+
+        char *tmpfilename;
+        if (asprintf(&tmpfilename, "%s.tmp", spoolfilename) == -1)
+            goto retry_1;
+
+        FILE *out = fopen(tmpfilename, "w");
+        if (out == NULL)
+            goto retry_2;
+        if (fprintf(out,"1 %d %d %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld\n",
+                getpid(),
+                user_status,
+                user_time.tv_sec, user_time.tv_usec,
+                user_rusage.ru_utime.tv_sec, user_rusage.ru_utime.tv_usec,
+                user_rusage.ru_stime.tv_sec, user_rusage.ru_stime.tv_usec,
+                user_rusage.ru_maxrss,
+                user_rusage.ru_ixrss,
+                user_rusage.ru_idrss,
+                user_rusage.ru_isrss,
+                user_rusage.ru_minflt,
+                user_rusage.ru_majflt,
+                user_rusage.ru_nswap,
+                user_rusage.ru_inblock,
+                user_rusage.ru_oublock,
+                user_rusage.ru_msgsnd,
+                user_rusage.ru_msgrcv,
+                user_rusage.ru_nsignals,
+                user_rusage.ru_nvcsw,
+                user_rusage.ru_nivcsw) < 0)
+            goto retry_3;
+        if (fflush(out) == EOF)
+            goto retry_3;
+        if (fsync(fileno(out)) == -1)
+            goto retry_3;
+        if (fclose(out) == EOF)
+            goto retry_2;
+        if (rename(tmpfilename, spoolfilename) == -1)
+            goto retry_2;
+        break;
+
+retry_3:
+        fclose(out);
+retry_2:
+        free(tmpfilename);
+retry_1:
+        sleep(10);
+    }
     return 0;
 }

mxqadmin.c

@@ -369,7 +369,8 @@ int main(int argc, char *argv[])
     res = mx_mysql_initialize(&mysql);
     assert(res == 0);
 
-    mx_mysql_option_set_default_file(mysql, arg_mysql_default_file);
+    if (mx_mysql_option_set_default_file(mysql, arg_mysql_default_file) < 0)
+        mx_die("%s: %s\n", arg_mysql_default_file, mx_mysql_error());
     mx_mysql_option_set_default_group(mysql, arg_mysql_default_group);
 
     res = mx_mysql_connect_forever(&mysql);