Re: PATCH: Allow setting $0 when POSIX_ARGZERO is not set

[Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx>]

On Jun 16,  1:44am, Mikael Magnusson wrote:
}
} I don't think this should hurt anything.

When sourcing a script the global argzero (which has been initialized
from runscript) is not malloc'd memory.

==25941== Invalid free() / delete / delete[]
==25941==    at 0x4004EFA: free (vg_replace_malloc.c:235)
==25941==    by 0x80950D7: zsfree (mem.c:1828)
==25941==    by 0x80A4DE9: argzerosetfn (params.c:4055)
==25941==    by 0x80A0CDF: assignstrvalue (params.c:2349)
==25941==    by 0x80A22D2: assignsparam (params.c:2806)
==25941==    by 0x806642F: addvars (exec.c:2317)
==25941==    by 0x80635F0: execsimple (exec.c:1117)
==25941==    by 0x80639AD: execlist (exec.c:1247)
==25941==    by 0x80634D1: execode (exec.c:1074)
==25941==    by 0x807F4EC: loop (init.c:207)
==25941==    by 0x808290B: zsh_main (init.c:1674)
==25941==    by 0x804C2E9: main (main.c:93)
==25941==  Address 0xBEEAFA62 is on thread 1's stack

I'm not sure if this comes up anywhere else.  I wasn't trivially able
to make it fail.

I think it would cause leaks to have parseargs() call ztrdup() for its
runscript argument, so it looks like setupshin() should do so when
assigning to argzero.

There are probably other race conditions if a signal were to arrive
while the shell is still initializing state.  Maybe we should get a
queue_signals() in there somewhere early.


diff --git a/Src/init.c b/Src/init.c
index 102276a..0fe4d75 100644
--- a/Src/init.c
+++ b/Src/init.c
@@ -1117,8 +1117,9 @@ setupshin(char *runscript)
 	    exit(127);
 	}
 	scriptfilename = sfname;
-	zsfree(argzero); /* ztrdup'd in parseargs */
-	argzero = runscript;
+	sfname = argzero; /* copy to avoid race condition */
+	argzero = ztrdup(runscript);
+	zsfree(sfname); /* argzero ztrdup'd in parseargs */
     }
     /*
      * We only initialise line numbering once there is a script to