linux-kernel - Re: [PATCH v2 12/16] KVM: arm64: Mark host bss and rodata section as shared

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <YQFKPoUUjJyceVjR@google.com>
Date:   Wed, 28 Jul 2021 13:14:54 +0100
From:   Quentin Perret <qperret@...gle.com>
To:     maz@...nel.org, james.morse@....com, alexandru.elisei@....com,
        suzuki.poulose@....com, catalin.marinas@....com, will@...nel.org
Cc:     linux-arm-kernel@...ts.infradead.org, kvmarm@...ts.cs.columbia.edu,
        linux-kernel@...r.kernel.org, ardb@...nel.org, qwandor@...gle.com,
        tabba@...gle.com, dbrazdil@...gle.com, kernel-team@...roid.com
Subject: Re: [PATCH v2 12/16] KVM: arm64: Mark host bss and rodata section as
 shared

On Monday 26 Jul 2021 at 10:29:01 (+0100), Quentin Perret wrote:
> +static int finalize_mappings(void)
> +{
> +	enum kvm_pgtable_prot prot;
> +	int ret;
> +
> +	/*
> +	 * The host's .bss and .rodata sections are now conceptually owned by
> +	 * the hypervisor, so mark them as 'borrowed' in the host stage-2. We
> +	 * can safely use host_stage2_idmap_locked() at this point since the
> +	 * host stage-2 has not been enabled yet.
> +	 */
> +	prot = pkvm_mkstate(KVM_PGTABLE_PROT_RWX, PKVM_PAGE_SHARED_BORROWED);
> +	ret = host_stage2_idmap_locked(__hyp_pa(__start_rodata),
> +				       __hyp_pa(__end_rodata), prot);
> +	if (ret)
> +		return ret;
> +
> +	return host_stage2_idmap_locked(__hyp_pa(__hyp_bss_end),
> +					__hyp_pa(__bss_stop), prot);
> +}
> +
>  void __noreturn __pkvm_init_finalise(void)
>  {
>  	struct kvm_host_data *host_data = this_cpu_ptr(&kvm_host_data);
> @@ -167,6 +199,10 @@ void __noreturn __pkvm_init_finalise(void)
>  	if (ret)
>  		goto out;
>  
> +	ret = finalize_mappings();
> +	if (ret)
> +		goto out;

While working on v3 of this series it occurred to me that we can
actually do vastly better than this. Specifically, the annotation of
shared pages currently happens in two places (recreate_hyp_mappings()
and finalize_mappings()) with nothing to guarantee they are in sync. At
the same time, the annotation of pages owned by the hypervisor is left
to the host itself using the __pkvm_mark_hyp hypercall. But clearly, by
the point we arrive to finalize_mappings() above, all the information I
need is already stored in the hyp pgtable. That is, it should be fairly
easy to walk the hyp stage-1, and for each valid mapping create a
matching annotation in the host stage-2 to mark the page shared or owned
by the hypervisor.

I'll have a go at implementing this in v3, which would guarantee
consistency across page-tables once the hypervisor is initialized, and
also allow to get rid of __pkvm_mark_hyp entirely. But if anybody thinks
this is a bad idea in the meantime, please shout!

Thanks,
Quentin