chmod: bugfix and new tests.

* sysif.c (chmod_wrap): When processing set (=), only punch a
hole in the target permission area once per clause, so as not
to clobber previously set modes. We do this by checking for
the chm_perm state. Whenever '=' is processed, the state machine
enters into that state; when any permission letter is then
processed, it transitions out of that state. This gets the
"u=rwsx" test to pass.

* tests/018/chmod.tl: New tests.

1 files changed, 7 insertions, 5 deletions

diff --git a/sysif.c b/sysif.c
index 286dcca6..ee219ff3 100644
--- a/sysif.c
+++ b/sysif.c
@@ -671,11 +671,13 @@ static val chmod_wrap(val target, val mode)
           case chm_add: cmode |= bits; break;
           case chm_sub: cmode &= ~bits; break;
           case chm_set:
-            cmode &= ~mask;
-            if (implicit_all || (who & CHM_O) != 0)
-              cmode &= ~S_ISVTX; /* GNU Coreutils 8.28 chmod behavior */
-            if (!S_ISDIR(cmode))
-              cmode &= ~(S_ISUID | S_ISGID);
+            if (cs == chm_perm) {
+              cmode &= ~mask;
+              if (implicit_all || (who & CHM_O) != 0)
+                cmode &= ~S_ISVTX; /* GNU Coreutils 8.28 chmod behavior */
+              if (!S_ISDIR(cmode))
+                cmode &= ~(S_ISUID | S_ISGID);
+            }
             cmode |= bits;
             break;
           }