Re: [PATCH] linux-user: Tolerate CONFIG_LSM_MMAP_MIN_ADDR

[Richard Henderson]

On 10/21/24 05:17, Ilya Leoshkevich wrote:
> Running qemu-i386 on a system running with SELinux in enforcing mode
> fails with:
>
>      qemu-i386: tests/tcg/i386-linux-user/sigreturn-sigmask: Unable to find a 
> guest_base to satisfy all guest address mapping requirements
>        00000000-ffffffff
>
> The reason is that main() determines mmap_min_addr from
> /proc/sys/vm/mmap_min_addr, but SELinux additionally defines
> CONFIG_LSM_MMAP_MIN_ADDR, which is normally larger: 32K or 64K, but,
> in general, can be anything. There is no portable way to query its
> value: /boot/config, /proc/config and /proc/config.gz are distro- and
> environment-specific.
>
> For maximum compatibility, probing is required. Use pgb_find_fallback()
> for this purpose. The downside of this approach is that mmap_min_addr
> remains incorrect, but there don't seem to be any practical
> consequences from this. If a correct mmap_min_addr will be required in
> the future, probing will need to be moved to linux-user main().
>
> Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
> ---
>   linux-user/elfload.c | 4 ++++
>   1 file changed, 4 insertions(+)

This is

    https://gitlab.com/qemu-project/qemu/-/issues/2598

which we closed as a system configuration / kernel bug.

I'm open to working around the issue, because I can see it coming up again and 
again.

In pgb_find_fallback, we use a skip value of 4M or 4G, using that skip as the base at 
which to begin the search.  I think it might be better to use this as the initial start 
point in pgb_find_itree as well, rather than mmap_min_addr.

Since I never had a setup in which this triggered, would you be willing to test 
such a change?


r~