fs_mkdirs: use O_NOFOLLOW and O_CLOEXEC

Don't follow symlinks. Suggestion from Jann Horn.

Also, add O_CLOEXEC. This prevents file descriptor leakage should
this code ever run in a multithreaded environment.

I'm not sure if either of these changes actually address any
security concerns, but it's harmless, so go ahead and add it.

Bug: 15675141
Change-Id: I7ba4e9d10439b7150f59759b54e3ad8ccba411e3

libcutils/fs.c

@@ -212,7 +212,7 @@ int fs_mkdirs(const char* path, mode_t mode) {
 
             /* Yay, segment is ready for us to step into */
             int next_fd;
-            if ((next_fd = openat(fd, segment, 0)) == -1) {
+            if ((next_fd = openat(fd, segment, O_NOFOLLOW | O_CLOEXEC)) == -1) {
                 ALOGE("Failed to openat(%s): %s", buf, strerror(errno));
                 res = -errno;
                 goto done_close;