Re: [RFC PATCH 4/6] KVM: guest_memfd: Implemnet bmap inode operation

From: Sean Christopherson <seanjc@google.com>
To: isaku.yamahata@intel.com
Cc: kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
	 isaku.yamahata@gmail.com, Michael Roth <michael.roth@amd.com>,
	 Paolo Bonzini <pbonzini@redhat.com>,
	erdemaktas@google.com, Sagi Shahar <sagis@google.com>,
	 David Matlack <dmatlack@google.com>,
	Kai Huang <kai.huang@intel.com>,
	 Zhi Wang <zhi.wang.linux@gmail.com>,
	chen.bo@intel.com, linux-coco@lists.linux.dev,
	 Chao Peng <chao.p.peng@linux.intel.com>,
	Ackerley Tng <ackerleytng@google.com>,
	 Vishal Annapurve <vannapurve@google.com>,
	Yuan Yao <yuan.yao@linux.intel.com>,
	 Jarkko Sakkinen <jarkko@kernel.org>,
	Xu Yilun <yilun.xu@intel.com>,
	 Quentin Perret <qperret@google.com>,
	wei.w.wang@intel.com, Fuad Tabba <tabba@google.com>
Subject: Re: [RFC PATCH 4/6] KVM: guest_memfd: Implemnet bmap inode operation
Date: Wed, 13 Sep 2023 10:46:11 -0700	[thread overview]
Message-ID: <ZQH1YzB5YaeCwHii@google.com> (raw)
In-Reply-To: <852b6fa117bf3767a99353d908bc566a5dd9c61a.1694599703.git.isaku.yamahata@intel.com>

On Wed, Sep 13, 2023, isaku.yamahata@intel.com wrote:
> From: Isaku Yamahata <isaku.yamahata@intel.com>
> 
> To inject memory failure, physical address of the page is needed.
> Implement bmap() method to convert the file offset into physical address.
> 
> Signed-off-by: Isaku Yamahata <isaku.yamahata@intel.com>
> ---
>  virt/kvm/Kconfig     |  4 ++++
>  virt/kvm/guest_mem.c | 28 ++++++++++++++++++++++++++++
>  2 files changed, 32 insertions(+)
> 
> diff --git a/virt/kvm/Kconfig b/virt/kvm/Kconfig
> index 624df45baff0..eb008f0e7cc3 100644
> --- a/virt/kvm/Kconfig
> +++ b/virt/kvm/Kconfig
> @@ -115,3 +115,7 @@ config KVM_GENERIC_PRIVATE_MEM
>  
>  config HAVE_GENERIC_PRIVATE_MEM_HANDLE_ERROR
>  	bool
> +
> +config KVM_GENERIC_PRIVATE_MEM_BMAP
> +	depends on KVM_GENERIC_PRIVATE_MEM
> +	bool
> diff --git a/virt/kvm/guest_mem.c b/virt/kvm/guest_mem.c
> index 3678287d7c9d..90dfdfab1f8c 100644
> --- a/virt/kvm/guest_mem.c
> +++ b/virt/kvm/guest_mem.c
> @@ -355,12 +355,40 @@ static int kvm_gmem_error_page(struct address_space *mapping, struct page *page)
>  	return MF_DELAYED;
>  }
>  
> +#ifdef CONFIG_KVM_GENERIC_PRIVATE_MEM_BMAP
> +static sector_t kvm_gmem_bmap(struct address_space *mapping, sector_t block)
> +{
> +	struct folio *folio;
> +	sector_t pfn = 0;
> +
> +	filemap_invalidate_lock_shared(mapping);
> +
> +	if (block << PAGE_SHIFT > i_size_read(mapping->host))
> +		goto out;
> +
> +	folio = filemap_get_folio(mapping, block);
> +	if (IS_ERR_OR_NULL(folio))
> +		goto out;
> +
> +	pfn = folio_pfn(folio) + (block - folio->index);
> +	folio_put(folio);
> +
> +out:
> +	filemap_invalidate_unlock_shared(mapping);
> +	return pfn;

IIUC, hijacking bmap() is a gigantic hack to propagate a host pfn to userspace
without adding a new ioctl() or syscall.  If we want to support target injection,
I would much, much rather add a KVM ioctl(), e.g. to let userspace inject errors
for a gfn.  Returning a pfn for something that AFAICT has nothing to do with pfns
is gross, e.g. the whole "0 is the error code" thing is technically wrong because
'0' is a perfectly valid pfn.

My vote is to drop this and not extend the injection information for the initial
merge, i.e. rely on point testing to verify kvm_gmem_error_page(), and defer adding
uAPI to let selftests inject errors.

> +
> +}
> +#endif
> +
>  static const struct address_space_operations kvm_gmem_aops = {
>  	.dirty_folio = noop_dirty_folio,
>  #ifdef CONFIG_MIGRATION
>  	.migrate_folio	= kvm_gmem_migrate_folio,
>  #endif
>  	.error_remove_page = kvm_gmem_error_page,
> +#ifdef CONFIG_KVM_GENERIC_PRIVATE_MEM_BMAP
> +	.bmap = kvm_gmem_bmap,
> +#endif
>  };
>  
>  static int  kvm_gmem_getattr(struct mnt_idmap *idmap,
> -- 
> 2.25.1
> 