linux-kernel - Re: [PATCH 02/10] x86/mm/cpa: Reject incorrect encryption change requests

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

Date:   Wed, 18 Oct 2023 10:44:51 +0200
From:   Ingo Molnar <mingo@...nel.org>
To:     Rick Edgecombe <rick.p.edgecombe@...el.com>
Cc:     x86@...nel.org, tglx@...utronix.de, mingo@...hat.com, bp@...en8.de,
        dave.hansen@...ux.intel.com, hpa@...or.com, luto@...nel.org,
        peterz@...radead.org, kirill.shutemov@...ux.intel.com,
        elena.reshetova@...el.com, isaku.yamahata@...el.com,
        seanjc@...gle.com, Michael Kelley <mikelley@...rosoft.com>,
        thomas.lendacky@....com, decui@...rosoft.com,
        sathyanarayanan.kuppuswamy@...ux.intel.com, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org, linux-s390@...r.kernel.org
Subject: Re: [PATCH 02/10] x86/mm/cpa: Reject incorrect encryption change
 requests


* Rick Edgecombe <rick.p.edgecombe@...el.com> wrote:

> Kernel memory is "encrypted" by default.  Some callers may "decrypt" it
> in order to share it with things outside the kernel like a device or an
> untrusted VMM.
> 
> There is nothing to stop set_memory_encrypted() from being passed memory
> that is already "encrypted" (aka. "private" on TDX).  In fact, some
> callers do this because ... $REASONS.  Unfortunately, part of the TDX
> decrypted=>encrypted transition is truly one way*.  It can't handle
> being asked to encrypt an already encrypted page
> 
> Allow __set_memory_enc_pgtable() to detect already-encrypted memory
> before it hits the TDX code.
> 
> * The one way part is "page acceptance"
> 
> [commit log written by Dave Hansen]
> Signed-off-by: Rick Edgecombe <rick.p.edgecombe@...el.com>
> ---
>  arch/x86/mm/pat/set_memory.c | 41 +++++++++++++++++++++++++++++++++++-
>  1 file changed, 40 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/mm/pat/set_memory.c b/arch/x86/mm/pat/set_memory.c
> index bda9f129835e..1238b0db3e33 100644
> --- a/arch/x86/mm/pat/set_memory.c
> +++ b/arch/x86/mm/pat/set_memory.c
> @@ -2122,6 +2122,21 @@ int set_memory_global(unsigned long addr, int numpages)
>  				    __pgprot(_PAGE_GLOBAL), 0);
>  }
>  
> +static bool kernel_vaddr_encryped(unsigned long addr, bool enc)
> +{
> +	unsigned int level;
> +	pte_t *pte;
> +
> +	pte = lookup_address(addr, &level);
> +	if (!pte)
> +		return false;
> +
> +	if (enc)
> +		return pte_val(*pte) == cc_mkenc(pte_val(*pte));
> +
> +	return pte_val(*pte) == cc_mkdec(pte_val(*pte));
> +}
> +
>  /*
>   * __set_memory_enc_pgtable() is used for the hypervisors that get
>   * informed about "encryption" status via page tables.
> @@ -2130,7 +2145,7 @@ static int __set_memory_enc_pgtable(unsigned long addr, int numpages, bool enc)
>  {
>  	pgprot_t empty = __pgprot(0);
>  	struct cpa_data cpa;
> -	int ret;
> +	int ret, numpages_in_state = 0;
>  
>  	/* Should not be working on unaligned addresses */
>  	if (WARN_ONCE(addr & ~PAGE_MASK, "misaligned address: %#lx\n", addr))
> @@ -2143,6 +2158,30 @@ static int __set_memory_enc_pgtable(unsigned long addr, int numpages, bool enc)
>  	cpa.mask_clr = enc ? pgprot_decrypted(empty) : pgprot_encrypted(empty);
>  	cpa.pgd = init_mm.pgd;
>  
> +	/*
> +	 * If any page is already in the right state, bail with an error
> +	 * because the code doesn't handled it. This is likely because

Grammar mistake here.

> +	 * something has gone wrong and isn't worth optimizing for.
> +	 *
> +	 * If all the memory pages are already in the desired state return
> +	 * success.

Missing comma.

> +	 *
> +	 * kernel_vaddr_encryped() does not synchronize against huge page
> +	 * splits so take pgd_lock. A caller doing strange things could

Missing comma.

Thanks,

	Ingo