YARN-2701. Addendum patch. Potential race condition in startLocalizer when using LinuxContainerExecutor. Contributed by Xuan Gong

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/impl/container-executor.c

@@ -451,36 +451,85 @@ char *get_tmp_directory(const char *work_dir) {
  * with the desired permissions.
  */
 int mkdirs(const char* path, mode_t perm) {
-  char *buffer = strdup(path);
+  struct stat sb;
-  char *token;
+  char * npath;
-  int cwd = open("/", O_RDONLY);
+  char * p;
-  if (cwd == -1) {
+  if (stat(path, &sb) == 0) {
-    fprintf(LOGFILE, "Can't open / in %s - %s\n", path, strerror(errno));
+    return check_dir(path, sb.st_mode, perm, 1);
-    free(buffer);
+  }
+  npath = strdup(path);
+  if (npath == NULL) {
+    fprintf(LOGFILE, "Not enough memory to copy path string");
     return -1;
   }
-  for(token = strtok(buffer, "/"); token != NULL; token = strtok(NULL, "/")) {
+  /* Skip leading slashes. */
-    if (mkdirat(cwd, token, perm) != 0) {
+  p = npath;
-      if (errno != EEXIST) {
+  while (*p == '/') {
-	    fprintf(LOGFILE, "Can't create directory %s in %s - %s\n",
+    p++;
-			    token, path, strerror(errno));
+  }
-	    close(cwd);
+
-	    free(buffer);
+  while (NULL != (p = strchr(p, '/'))) {
+    *p = '\0';
+    if (create_validate_dir(npath, perm, path, 0) == -1) {
+      free(npath);
       return -1;
     }
+    *p++ = '/'; /* restore slash */
+    while (*p == '/')
+      p++;
   }
-    int new_dir = openat(cwd, token, O_RDONLY);
+
-    close(cwd);
+  /* Create the final directory component. */
-    cwd = new_dir;
+  if (create_validate_dir(npath, perm, path, 1) == -1) {
-    if (cwd == -1) {
+    free(npath);
-      fprintf(LOGFILE, "Can't open %s in %s - %s\n", token, path,
+    return -1;
+  }
+  free(npath);
+  return 0;
+}
+
+/*
+* Create the parent directory if they do not exist. Or check the permission if
+* the race condition happens.
+* Give 0 or 1 to represent whether this is the final component. If it is, we
+* need to check the permission.
+*/
+int create_validate_dir(char* npath, mode_t perm, char* path, int finalComponent) {
+  struct stat sb;
+  if (stat(npath, &sb) != 0) {
+    if (mkdir(npath, perm) != 0) {
+      if (errno != EEXIST || stat(npath, &sb) != 0) {
+        fprintf(LOGFILE, "Can't create directory %s - %s\n", npath,
                 strerror(errno));
-      free(buffer);
+        return -1;
+      }
+      // The directory npath should exist.
+      if (check_dir(npath, sb.st_mode, perm, finalComponent) == -1) {
+        return -1;
+      }
+    }
+  } else {
+    if (check_dir(npath, sb.st_mode, perm, finalComponent) == -1) {
+      return -1;
+    }
+  }
+  return 0;
+}
+
+// check whether the given path is a directory
+// also check the access permissions whether it is the same as desired permissions
+int check_dir(char* npath, mode_t st_mode, mode_t desired, int finalComponent) {
+  if (!S_ISDIR(st_mode)) {
+    fprintf(LOGFILE, "Path %s is file not dir\n", npath);
+    return -1;
+  } else if (finalComponent == 1) {
+    int filePermInt = st_mode & (S_IRWXU | S_IRWXG | S_IRWXO);
+    int desiredInt = desired & (S_IRWXU | S_IRWXG | S_IRWXO);
+    if (filePermInt != desiredInt) {
+      fprintf(LOGFILE, "Path %s does not have desired permission.\n", npath);
       return -1;
     }
   }
-  free(buffer);
-  close(cwd);
   return 0;
 }
 

hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/impl/container-executor.h

@@ -203,3 +203,9 @@ int create_directory_for_user(const char* path);
 int change_user(uid_t user, gid_t group);
 
 int mount_cgroup(const char *pair, const char *hierarchy);
+
+int check_dir(char* npath, mode_t st_mode, mode_t desired,
+   int finalComponent);
+
+int create_validate_dir(char* npath, mode_t perm, char* path,
+   int finalComponent);