libglusterfs: handle failure properly in gf_system()

- Child must be terminated if execve(2) fails, otherwise it will deadlock
  parent
- Status, as of given by waitpid(2), is aggregated data, should not be
  returned as is. In fact, there is not much point in threading up the
  exitvalue: callers usually return gf_system result as is, and according
  to glusterfs conventions, they should return -1 on failure. Therefore
  now we check only for success/fail, and return 0/-1.

Signed-off-by: Csaba Henk <csaba@gluster.com>
Signed-off-by: Vijay Bellur <vijay@dev.gluster.com>

BUG: 1665 (glusterd hangs if spawning a child is failed)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=1665

1 files changed, 6 insertions, 1 deletions

diff --git a/libglusterfs/src/common-utils.c b/libglusterfs/src/common-utils.c
index 022fdf35537..75476a0291e 100644
--- a/libglusterfs/src/common-utils.c
+++ b/libglusterfs/src/common-utils.c
@@ -1510,11 +1510,16 @@ gf_system (const char *command)
 
                 /* Code will not come here at all */
                 gf_log ("", GF_LOG_ERROR, "execv of (%s) failed", command);
+
+                kill (getpid(), SIGKILL);
         }
         if (pid > 0) {
                 /* Current, ie, parent process */
                 pid = waitpid (pid, &status, 0);
-                ret = status;
+                if (WIFEXITED(status) && WEXITSTATUS(status) == EXIT_SUCCESS)
+                        ret = 0;
+                else
+                        ret = -1;
         }
 out:
         if (dupcmd)