linux-kernel - Re: [PATCH v2 2/2] x86: sgx_vepc: implement SGX_IOC_VEPC

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: <00419ea5-9b51-0175-0500-0882fd0b4290@redhat.com>
Date:   Sat, 16 Oct 2021 09:14:46 +0200
From:   Paolo Bonzini <pbonzini@...hat.com>
To:     Sean Christopherson <seanjc@...gle.com>
Cc:     linux-kernel@...r.kernel.org, kvm@...r.kernel.org,
        dave.hansen@...ux.intel.com, x86@...nel.org, yang.zhong@...el.com,
        jarkko@...nel.org
Subject: Re: [PATCH v2 2/2] x86: sgx_vepc: implement SGX_IOC_VEPC_REMOVE ioctl

On 16/10/21 00:29, Sean Christopherson wrote:
> On Tue, Oct 12, 2021, Paolo Bonzini wrote:
>> diff --git a/arch/x86/kernel/cpu/sgx/virt.c b/arch/x86/kernel/cpu/sgx/virt.c
>> index 59cdf3f742ac..81a0a0f22007 100644
>> --- a/arch/x86/kernel/cpu/sgx/virt.c
>> +++ b/arch/x86/kernel/cpu/sgx/virt.c
>> @@ -150,6 +150,46 @@ static int sgx_vepc_free_page(struct sgx_epc_page *epc_page)
>>   	return 0;
>>   }
>>   
>> +static long sgx_vepc_remove_all(struct sgx_vepc *vepc)
>> +{
>> +	struct sgx_epc_page *entry;
>> +	unsigned long index;
>> +	long failures = 0;
>> +
>> +	xa_for_each(&vepc->page_array, index, entry) {
> 
> Might be worth a comment that xa_for_each() is safe to use concurrently with
> xa_load/xa_store, i.e. this doesn't need to take vepc->lock.

I considered that to be part of the xarray contract (xa_store uses 
rcu_assign_pointer so it has release semantics, and vepc->page_array is 
essentially "store once").

> It does raise the
> question of whether or not the kernel is responsible for providing deterministic
> results if userspace/guest is accessing previously-unallocated pages.

Garbage in, garbage out -- but you're right below that garbage in, WARN 
out is not acceptable.  I'm sending a v3 with documentation changes too.

Paolo

>> +		int ret = sgx_vepc_remove_page(entry);
> 
> I don't see anything that prevents userspace from doing SGX_IOC_VEPC_REMOVE_ALL
> on multiple threads with the same vEPC.  That means userspace can induce a #GP
> due to concurrent access.  Taking vepc->lock would solve that particular problem,
> but I think that's a moot point because the EREMOVE locking rules are relative to
> the SECS, not the individual page (because of refcounting).  SGX_IOC_VEPC_REMOVE_ALL
> on any two arbitrary vEPCs could induce a fault if they have children belonging to
> the same enclave, i.e. share an SECS.
> 
> Sadly, I think this needs to be:
> 
> 		if (ret == SGX_CHILD_PRESENT)
> 			failures++;
> 		else if (ret)
> 			return -EBUSY;
> 
>> +		switch (ret) {
>> +		case 0:
>> +			break;
>> +
>> +		case SGX_CHILD_PRESENT:
>> +			failures++;
>> +			break;
>> +
>> +		case SGX_ENCLAVE_ACT:
>> +			/*
>> +			 * Unlike in sgx_vepc_free_page, userspace could be calling
>> +			 * the ioctl while logical processors are running in the
>> +			 * enclave; do not warn.
>> +			 */
>> +			return -EBUSY;
>> +
>> +		default:
>> +			WARN_ONCE(1, EREMOVE_ERROR_MESSAGE, ret, ret);
>> +			failures++;
>> +			break;
>> +		}
>> +		cond_resched();
>> +	}
>> +
>> +	/*
>> +	 * Return the number of pages that failed to be removed, so
>> +	 * userspace knows that there are still SECS pages lying
>> +	 * around.
>> +	 */
>> +	return failures;
>> +}
>