Hello Roland, You knew we'd be bitten by the bug someday, and that's today. From exec/exec.c:
/* XXX all accesses of the mapped data need to use fault handling to abort the RPC when mapped file data generates bad page faults. I've marked some accesses with XXX/fault comments. --roland */ And it happens that we do get faults when trying to bunzip2 the binary, i.e. from the zipread() function. So what's the plan? If the offset is really bogus, I guess we could use a safe version of memcpy and longjmp(ziperr) when things go wrong? For now I have disabled gzip/bzip2 support from the exec used by the Debian hurd-i386 buildd. Samuel
