openssh unix dev - Mar 2001 - Tru64 UNIX SIA in 2.5.2p1 is hosed (still)

The recent patch posted by Steve VanDevender <stevev at
darkwing.uoregon.edu>
for fixing the session code on Tru64 isn't quite right -- it still fails 
in the case of NO tty being allocated.

The problem is that s->tty is a char[TTYSZ] rather than a char *, and 
hence can't hold a NULL.  Calling sia_ses_init() with the tty being an 
empty string doesn't signify no tty, and hence will cause a failure.  The
"no tty" case should have tty passed as NULL.

One possible fix for this is to change the call to sia_ses_init from:

        if (sia_ses_init(&ent, saved_argc, saved_argv, host, user, tty, 0,
            NULL) != SIASUCCESS)
to:
        if (sia_ses_init(&ent, saved_argc, saved_argv, host, user,
            tty[0] ? tty : NULL, 0, NULL) != SIASUCCESS)

However, I'm not convinced that tty won't be some random value here if 
session structure has been used before, since s->tty isn't zeroed in
session_new().  Thus you may possibly also need to add:

	s->tty[0] = '\0';

into the session initialisation in session_new(), or maybe set it before
the call to do_child() in do_exec_no_pty().

On further thought, perhaps the call to sia_ses_init should be left alone 
and the call to session_setup_sia() in session.c changed from:

        session_setup_sia(pw->pw_name, s->tty);
to:
        session_setup_sia(pw->pw_name, s->ttyfd != -1 ? s->tty : NULL);

Can someone who knows the code better than I do shed some light on the
correct solution here.

 - Mike

-- 
Mike Battersby <mib at unimelb.edu.au>        The University of Melbourne
Fetch my pgp key from: http://ariel.ucs.unimelb.edu.au/~mib/pgpkey.txt


-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 222 bytes
Desc: not available
Url :
http://lists.mindrot.org/pipermail/openssh-unix-dev/attachments/20010321/149aff6c/attachment.bin

On Wed, 21 Mar 2001, Mike Battersby wrote:
> The recent patch posted by Steve VanDevender <stevev at
darkwing.uoregon.edu>
> for fixing the session code on Tru64 isn't quite right -- it still
fails
> in the case of NO tty being allocated.
>
> The problem is that s->tty is a char[TTYSZ] rather than a char *, and
> hence can't hold a NULL.  Calling sia_ses_init() with the tty being an
> empty string doesn't signify no tty, and hence will cause a failure. 
The
> "no tty" case should have tty passed as NULL.
Thanks for spotting this. The following takes care of the SIA case.

I am pretty sure it fixes AIX as well, but the manpage that Gert sent to
the list was a little ambiguous - only saying that TTY should be
"null"
in the cases where no tty is present. Perhaps an AIX guru could enlighten
us here?

Index: session.c
==================================================================RCS file:
/var/cvs/openssh/session.c,v
retrieving revision 1.93
diff -u -r1.93 session.c
--- session.c	2001/03/21 00:11:57	1.93
+++ session.c	2001/03/21 05:10:07
@@ -1053,7 +1053,7 @@
 	   switch, so we let login(1) to this for us. */
 	if (!options.use_login) {
 #ifdef HAVE_OSF_SIA
-		session_setup_sia(pw->pw_name, s->tty);
+		session_setup_sia(pw->pw_name, s->ttyfd == -1 ? NULL : s->tty);
 #else /* HAVE_OSF_SIA */
 #ifdef HAVE_CYGWIN
 		if (is_winnt) {
@@ -1137,7 +1137,8 @@
 			cp = xmalloc(22 + strlen(s->tty) +
 			    2 * strlen(pw->pw_name));
 			i = sprintf(cp, "LOGNAME=%s%cNAME=%s%cTTY=%s%c%c",
-			    pw->pw_name, 0, pw->pw_name, 0, s->tty, 0,0);
+			    pw->pw_name, 0, pw->pw_name, 0,
+			    s->ttyfd == -1 ? "" : s->tty, 0,0);
 			if (usrinfo(SETUINFO, cp, i) == -1)
 				fatal("Couldn't set usrinfo: %s",
 				    strerror(errno));

-- 
| Damien Miller <djm at mindrot.org> \ ``E-mail attachments are the poor
man's
| http://www.mindrot.org          /   distributed filesystem'' - Dan
Geer