On 02/05/2025 00:49, Kees Cook wrote: > On Thu, May 01, 2025 at 12:03:32PM +0100, Ryan Roberts wrote: >> I agree, as long as COMPAT_BRK is not set (which is the common case IFAICT). >> When COMPAT_BRK is enabled, I think you are breaking the purpose of that >> Kconfig? Perhaps it's not a real-world problem though... > > When you turned off ASLR, what mechanism did you use? Personality or > randomize_va_space=0? randomize_va_space=0 > >>> It's possible it could break running the loader directly against some >>> libc5-based binaries. If this turns out to be a real-world issue, we can >>> find a better solution (perhaps pre-allocating a large brk). >> >> But how large is large enough... > > Right -- Chrome has a 500MB brk on my laptop. :P Or with randomization > off, it could allocate to the top of the mmap space just to keep > "future" mmap allocations from landing in any holes... > >> Perhaps it is safer to only move the brk if !IS_ENABLED(CONFIG_COMPAT_BRK) ? >> Then wait to see if there are any real-world COMPAT_BRK users that hit the issue? > > Yeah, that might be the best middle-ground. >
