Hi Vijay and Mark,
On 4/21/2022 4:46 PM, Dhanraj, Vijay wrote:
Hi All,
I evaluated V4 patch changes with Gramine and ran into an issue when trying to set EPC page permission to PROT_NONE. It looks like with V3 patch series a change was introduced which requires kernel to have at least R permission when calling RESTRICT IOCTL. This change was done under the assumption that EPCM requires at least R permission for EMODPE/EACCEPT to succeed. But when testing with V2 version, EACCEPT worked fine with page permission set to PROT_NONE.
Thanks to @Shanahan, Mark for confirming that EPCM does not need to have R value to allow EACCEPT or EMODPE. Given this, can we please revert this change?
Thank you very much for pointing this out. I can revert the change to what was done in V2 where the only check is to ensure that W requires R. This is a requirement of EMODPR. Could you please check if this snippet results in things working for you again?
---8<--- diff --git a/arch/x86/kernel/cpu/sgx/ioctl.c b/arch/x86/kernel/cpu/sgx/ioctl.c index 83674d054c13..7c7c8a61196e 100644 --- a/arch/x86/kernel/cpu/sgx/ioctl.c +++ b/arch/x86/kernel/cpu/sgx/ioctl.c @@ -855,12 +855,8 @@ static long sgx_ioc_enclave_restrict_permissions(struct sgx_encl *encl, if (params.permissions & ~SGX_SECINFO_PERMISSION_MASK) return -EINVAL;
- /* - * Read access is required for the enclave to be able to use the page. - * SGX instructions like ENCLU[EMODPE] and ENCLU[EACCEPT] require - * read access. - */ - if (!(params.permissions & SGX_SECINFO_R)) + if ((params.permissions & SGX_SECINFO_W) && + !(params.permissions & SGX_SECINFO_R)) return -EINVAL;
if (params.result || params.count)