full-disclosure - S-plus /tmp usage

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <200301052139.h05LdLr263128@milan.maths.usyd.edu.au>
From: psz at maths.usyd.edu.au (Paul Szabo)
Subject: S-plus /tmp usage

INTRODUCTION

S-PLUS is a Statistical analysis, graphics and programming tool
http://www.statsci.com/
http://www.insightful.com/

PROBLEM

As installed on UNIX machines, Splus uses files in /tmp in an unsafe way.

DETAILS

The main Sqpe binary, and various shell script modules, use files in /tmp:

splus/6.0/cmd/Sqpe
	Clobbers /tmp/__F$$:
	open("/tmp/__F8499", O_RDWR|O_CREAT|O_TRUNC, 0666) = 3

splus/6.0/cmd/PRINT
	Clobbers /tmp/PRINT.$$.out

splus/6.0/cmd/mustfix.hlinks
	Clobbers /tmp/SUBST$PID.TXT /tmp/ed.cmds$PID

splus/6.0/cmd/sas_get
	May clobber and use /tmp/file.1 /tmp/file.2

splus/6.0/cmd/sas_vars
	May clobber and use /tmp/file.1

splus/6.0/cmd/sgml2html
	Clobbers /tmp/sgml2html$$tmp /tmp/sgml2html$$tmp1 /tmp/sgml2html$$tmp2

Suppose an attacker creates a symlink from any of the "clobbered" files to
one owned by the victim: guesses the PID that will be used, does

  ln -s ~victim/.profile /tmp/__F123

and waits for the victim to use Splus, then the victim's .profile will be
trashed. Some or all of these attacks may then be escalated to arbitrary
command execution; if root ever uses Splus then the damage is much greater.

It might be argued that it is hard to guess what PID will be used next.
It is easy enough to create a few thousand symlinks with likely PIDs; in
fact the attacker could create a symlink for every possible PID (as these
normally range from 0 to 32k or 64k).

VENDOR COMMUNICATION

 5 Dec 2002  StatSci/Insightful notified about shell scripts
 9 Dec 2002  Insightful notified about Sqpe
11 Dec 2002  We are currently investigating
17 Dec 2002  continuing to look into your queries
18 Dec 2002  anticipate tmpfile() ... in the next release
26 Dec 2002  Another list might be http://www.biostat.wustl.edu/s-news/

WORKAROUND/PATCH

The scripts could be patched trivially using one of the textbook methods,
e.g. using a safe directory:
  mkdir -m 700 /tmp/mydir$$ || exit 1
  ... do things to /tmp/mydir$$/myfile ...
  rm -rf /tmp/mydir$$

Fixing Sqpe is harder. Could (safely) pre-create /tmp/__F$$ e.g.:

*** splus/6.0/cmd/NEW.old	Tue Oct 10 16:06:37 2000
--- splus/6.0/cmd/NEW	Tue Dec 24 09:15:59 2002
***************
*** 9,13 ****
--- 9,19 ----
  then
  	echo $target not found; exit 1
  fi
+ set -e
+ umask 077
+ mkdir /tmp/F$$
+ touch /tmp/F$$/__F$$
+ mv -i /tmp/F$$/__F$$ /tmp </dev/null
+ rmdir /tmp/F$$
  exec $target

but Sqpe would still be open to races as it repeatedly open()s and
unlink()s that file. A proper fix will have to come from the vendor.

SIGNATURE

Paul Szabo - psz@...hs.usyd.edu.au  http://www.maths.usyd.edu.au:8000/u/psz/
School of Mathematics and Statistics  University of Sydney   2006  Australia