https://bugs.exim.org/show_bug.cgi?id=2250
--- Comment #13 from David Carter <[email protected]> --- (In reply to David Carter from comment #12) > I am currently running ae06ddc47aa43bb5e2dcbe2643e41649d1947a9d (latest > available commit from origin/exim-4_90+fixes), and that still has the > problem. > > I am now trying the same version with: > > DISABLE_DKIM=yes > > to see if that helps. We have been running this way for three hours now, and the problem has gone away. So it looks like the problem is somewhere in the DKIM code. I guess that I shouldn't be surprised: I have reported half a dozen separate crash problems and memory management issues against Exim 4.88, 4.89 and 4.90 over the last two years, and they have all been in the DKIM code. At face value it looks like something in that code is running: *((char *)s) = '\0'; for some bad value of s. However I would expect that to run after the DATA phase of the SMTP conversation. Unfortunately that suggests a rather more unpleasant memory management issue. Those are always fun to debug... Suggestions? It wouldn't be hard to generate a core dump of one of these broken processes, but I'm not convinced that the back trace will tell us anything useful. If we need to bisect with DKIM enabled then I am going to have to work around the following bug which causes email loss: > This is an important prerequisite to bisection because of: > > commit e5fc5d4ba779be4c57bd08ad2da70b6e1a85a549 > Author: Jeremy Harris <[email protected]> > Date: Wed Feb 7 23:09:55 2018 +0000 > > DKIM: fix buffer overflow in verify > > Caused crash in free() by corrupting malloc metadata. > > Reported-by: University of Cambridge > Broken-by: 80a47a2c96 That patch doesn't apply cleanly to "80a47a2c96". It looks like quite a long has changed since then, so I'm going to need some help applying the same fix to intermediate versions. -- You are receiving this mail because: You are on the CC list for the bug. -- ## List details at https://lists.exim.org/mailman/listinfo/exim-dev Exim details at http://www.exim.org/ ##
