NetPBM contains large numbers of maths overflow errors, some of which are deeply theoretical as they involve passing 2Gb file names, others of which are straight forward x * y * depth type overflows, of the kind which have shown up in numerous other imaging libraries. Finally there are a couple of signed value overflows which appear safe but since signed maths overflow is not defined for C may not be. An initial set of patches were sent to vendors and to the NetPBM maintainers. Today Martin Schulze found two minor errors in the changes released so far. These don't appear to leave open holes, just cause correctness problems. While netpbm is not setuid it is used by some applications for print formatting and also for converting untrusted images received from third parties. Although the patches appear to address the main problems as patch author I believe the right path is probably to recognize that netpbm is very old code, written in times with a different threat model and use something else instead. Al Viro found the original bug. Alan Cox did the initial fixes. Martin Schulze and Sebastian Kramer provided additional fixes to the patches. www.MachinaeSupremacy.com provided the music to keep me sane through the initial tedious process. The patches are over 100K so please ask your vendor or check the main netpbm site. Alan -- 'On the other hand, you sometimes wish the world, like nethack, had some sort of "Genocide All Stupid People" key sequence.' - Alec Muffett
This archive was generated by hypermail 2b30 : Fri Feb 28 2003 - 07:44:25 PST