Re: [PATCH V3 15/30] x86/sgx: Support adding of pages to an initialized enclave

[Jarkko Sakkinen <jarkko@kernel.org>]

On Tue, 2022-04-05 at 13:03 +0300, Jarkko Sakkinen wrote:
> On Tue, 2022-04-05 at 08:05 +0300, Jarkko Sakkinen wrote:
> > On Mon, 2022-04-04 at 09:49 -0700, Reinette Chatre wrote:
> > > With SGX1 an enclave needs to be created with its maximum memory demands
> > > allocated. Pages cannot be added to an enclave after it is initialized.
> > > SGX2 introduces a new function, ENCLS[EAUG], that can be used to add
> > > pages to an initialized enclave. With SGX2 the enclave still needs to
> > > set aside address space for its maximum memory demands during enclave
> > > creation, but all pages need not be added before enclave initialization.
> > > Pages can be added during enclave runtime.
> > > 
> > > Add support for dynamically adding pages to an initialized enclave,
> > > architecturally limited to RW permission at creation but allowed to
> > > obtain RWX permissions after enclave runs EMODPE. Add pages via the
> > > page fault handler at the time an enclave address without a backing
> > > enclave page is accessed, potentially directly reclaiming pages if
> > > no free pages are available.
> > > 
> > > The enclave is still required to run ENCLU[EACCEPT] on the page before
> > > it can be used. A useful flow is for the enclave to run ENCLU[EACCEPT]
> > > on an uninitialized address. This will trigger the page fault handler
> > > that will add the enclave page and return execution to the enclave to
> > > repeat the ENCLU[EACCEPT] instruction, this time successful.
> > > 
> > > If the enclave accesses an uninitialized address in another way, for
> > > example by expanding the enclave stack to a page that has not yet been
> > > added, then the page fault handler would add the page on the first
> > > write but upon returning to the enclave the instruction that triggered
> > > the page fault would be repeated and since ENCLU[EACCEPT] was not run
> > > yet it would trigger a second page fault, this time with the SGX flag
> > > set in the page fault error code. This can only be recovered by entering
> > > the enclave again and directly running the ENCLU[EACCEPT] instruction on
> > > the now initialized address.
> > > 
> > > Accessing an uninitialized address from outside the enclave also
> > > triggers this flow but the page will remain inaccessible (access will
> > > result in #PF) until accepted from within the enclave via
> > > ENCLU[EACCEPT].
> > > 
> > > Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
> > > ---
> > > Changes since V2:
> > > - Remove runtime tracking of EPCM permissions
> > >   (sgx_encl_page->vm_run_prot_bits) (Jarkko).
> > > - Move export of sgx_encl_{grow,shrink}() to separate patch. (Jarkko)
> > > - Use sgx_encl_page_alloc(). (Jarkko)
> > > - Set max allowed permissions to be RWX (Jarkko). Update changelog
> > >   to indicate the change and use comment in code as
> > >   created by Jarkko in:
> > > https://lore.kernel.org/linux-sgx/20220306053211.135762-4-jarkko@kernel.org
> > > - Do not set protection bits but let it be inherited by VMA (Jarkko)
> > > 
> > > Changes since V1:
> > > - Fix subject line "to initialized" -> "to an initialized" (Jarkko).
> > > - Move text about hardware's PENDING state to the patch that introduces
> > >   the ENCLS[EAUG] wrapper (Jarkko).
> > > - Ensure kernel-doc uses brackets when referring to function.
> > > 
> > >  arch/x86/kernel/cpu/sgx/encl.c | 124 +++++++++++++++++++++++++++++++++
> > >  1 file changed, 124 insertions(+)
> > > 
> > > diff --git a/arch/x86/kernel/cpu/sgx/encl.c b/arch/x86/kernel/cpu/sgx/encl.c
> > > index 546423753e4c..fa4f947f8496 100644
> > > --- a/arch/x86/kernel/cpu/sgx/encl.c
> > > +++ b/arch/x86/kernel/cpu/sgx/encl.c
> > > @@ -194,6 +194,119 @@ struct sgx_encl_page *sgx_encl_load_page(struct sgx_encl *encl,
> > >         return __sgx_encl_load_page(encl, entry);
> > >  }
> > >  
> > > +/**
> > > + * sgx_encl_eaug_page() - Dynamically add page to initialized enclave
> > > + * @vma:       VMA obtained from fault info from where page is accessed
> > > + * @encl:      enclave accessing the page
> > > + * @addr:      address that triggered the page fault
> > > + *
> > > + * When an initialized enclave accesses a page with no backing EPC page
> > > + * on a SGX2 system then the EPC can be added dynamically via the SGX2
> > > + * ENCLS[EAUG] instruction.
> > > + *
> > > + * Returns: Appropriate vm_fault_t: VM_FAULT_NOPAGE when PTE was installed
> > > + * successfully, VM_FAULT_SIGBUS or VM_FAULT_OOM as error otherwise.
> > > + */
> > > +static vm_fault_t sgx_encl_eaug_page(struct vm_area_struct *vma,
> > > +                                    struct sgx_encl *encl, unsigned long addr)
> > > +{
> > > +       struct sgx_pageinfo pginfo = {0};
> > > +       struct sgx_encl_page *encl_page;
> > > +       struct sgx_epc_page *epc_page;
> > > +       struct sgx_va_page *va_page;
> > > +       unsigned long phys_addr;
> > > +       u64 secinfo_flags;
> > > +       vm_fault_t vmret;
> > > +       int ret;
> > > +
> > > +       if (!test_bit(SGX_ENCL_INITIALIZED, &encl->flags))
> > > +               return VM_FAULT_SIGBUS;
> > > +
> > > +       /*
> > > +        * Ignore internal permission checking for dynamically added pages.
> > > +        * They matter only for data added during the pre-initialization
> > > +        * phase. The enclave decides the permissions by the means of
> > > +        * EACCEPT, EACCEPTCOPY and EMODPE.
> > > +        */
> > > +       secinfo_flags = SGX_SECINFO_R | SGX_SECINFO_W | SGX_SECINFO_X;
> > > +       encl_page = sgx_encl_page_alloc(encl, addr - encl->base, secinfo_flags);
> > > +       if (IS_ERR(encl_page))
> > > +               return VM_FAULT_OOM;
> > > +
> > > +       epc_page = sgx_alloc_epc_page(encl_page, true);
> > > +       if (IS_ERR(epc_page)) {
> > > +               kfree(encl_page);
> > > +               return VM_FAULT_SIGBUS;
> > > +       }
> > > +
> > > +       va_page = sgx_encl_grow(encl);
> > > +       if (IS_ERR(va_page)) {
> > > +               ret = PTR_ERR(va_page);
> > > +               goto err_out_free;
> > > +       }
> > > +
> > > +       mutex_lock(&encl->lock);
> > > +
> > > +       /*
> > > +        * Copy comment from sgx_encl_add_page() to maintain guidance in
> > > +        * this similar flow:
> > > +        * Adding to encl->va_pages must be done under encl->lock.  Ditto for
> > > +        * deleting (via sgx_encl_shrink()) in the error path.
> > > +        */
> > > +       if (va_page)
> > > +               list_add(&va_page->list, &encl->va_pages);
> > > +
> > > +       ret = xa_insert(&encl->page_array, PFN_DOWN(encl_page->desc),
> > > +                       encl_page, GFP_KERNEL);
> > > +       /*
> > > +        * If ret == -EBUSY then page was created in another flow while
> > > +        * running without encl->lock
> > > +        */
> > > +       if (ret)
> > > +               goto err_out_unlock;
> > > +
> > > +       pginfo.secs = (unsigned long)sgx_get_epc_virt_addr(encl->secs.epc_page);
> > > +       pginfo.addr = encl_page->desc & PAGE_MASK;
> > > +       pginfo.metadata = 0;
> > > +
> > > +       ret = __eaug(&pginfo, sgx_get_epc_virt_addr(epc_page));
> > > +       if (ret)
> > > +               goto err_out;
> > > +
> > > +       encl_page->encl = encl;
> > > +       encl_page->epc_page = epc_page;
> > > +       encl_page->type = SGX_PAGE_TYPE_REG;
> > > +       encl->secs_child_cnt++;
> > > +
> > > +       sgx_mark_page_reclaimable(encl_page->epc_page);
> > > +
> > > +       phys_addr = sgx_get_epc_phys_addr(epc_page);
> > > +       /*
> > > +        * Do not undo everything when creating PTE entry fails - next #PF
> > > +        * would find page ready for a PTE.
> > > +        */
> > > +       vmret = vmf_insert_pfn(vma, addr, PFN_DOWN(phys_addr));
> > > +       if (vmret != VM_FAULT_NOPAGE) {
> > > +               mutex_unlock(&encl->lock);
> > > +               return VM_FAULT_SIGBUS;
> > > +       }
> > > +       mutex_unlock(&encl->lock);
> > > +       return VM_FAULT_NOPAGE;
> > > +
> > > +err_out:
> > > +       xa_erase(&encl->page_array, PFN_DOWN(encl_page->desc));
> > > +
> > > +err_out_unlock:
> > > +       sgx_encl_shrink(encl, va_page);
> > > +       mutex_unlock(&encl->lock);
> > > +
> > > +err_out_free:
> > > +       sgx_encl_free_epc_page(epc_page);
> > > +       kfree(encl_page);
> > > +
> > > +       return VM_FAULT_SIGBUS;
> > > +}
> > > +
> > >  static vm_fault_t sgx_vma_fault(struct vm_fault *vmf)
> > >  {
> > >         unsigned long addr = (unsigned long)vmf->address;
> > > @@ -213,6 +326,17 @@ static vm_fault_t sgx_vma_fault(struct vm_fault *vmf)
> > >         if (unlikely(!encl))
> > >                 return VM_FAULT_SIGBUS;
> > >  
> > > +       /*
> > > +        * The page_array keeps track of all enclave pages, whether they
> > > +        * are swapped out or not. If there is no entry for this page and
> > > +        * the system supports SGX2 then it is possible to dynamically add
> > > +        * a new enclave page. This is only possible for an initialized
> > > +        * enclave that will be checked for right away.
> > > +        */
> > > +       if (cpu_feature_enabled(X86_FEATURE_SGX2) &&
> > > +           (!xa_load(&encl->page_array, PFN_DOWN(addr))))
> > > +               return sgx_encl_eaug_page(vma, encl, addr);
> > > +
> > >         mutex_lock(&encl->lock);
> > >  
> > >         entry = sgx_encl_load_page_in_vma(encl, addr, vma->vm_flags);
> > 
> > Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org>
> 
> Tested-by: Jarkko Sakkinen <jarkko@kernel.org>

For what is worth I also get a full pass with our test suite,
where the runtime is using EAUG together with EACCEPTCOPY:

    Finished test [unoptimized + debuginfo] target(s) in 0.26s
     Running unittests src/main.rs (target/debug/deps/enarx-ee7f422740eab404)

running 7 tests
test backend::sgx::attestation::tests::request_target_info ... ignored
test backend::sev::snp::tests::test_const_id_macro ... ok
test backend::sev::snp::firmware::test::test_vcek_url ... ok
test backend::sgx::ioctls::tests::restrict_permissions ... ok
test cli::snp::tests::test_empty_cache_path ... ok
test workldr::wasmldr::test::is_builtin ... ok
test cli::snp::tests::test_get_or_write ... ok

test result: ok. 6 passed; 0 failed; 1 ignored; 0 measured; 0 filtered out; finished in 0.20s

     Running tests/c_integration_tests.rs (target/debug/deps/c_integration_tests-f7a69c2274f59f90)

running 21 tests
test get_att ... ignored
test bind ... ok
test clock_gettime ... ok
test close ... ok
test exit_one ... ok
test getegid ... ok
test geteuid ... ok
test sgx_get_att_quote ... ignored
test sgx_get_att_quote_size ... ignored
test exit_zero ... ok
test getgid ... ok
test write_emsgsize ... ignored
test write_stderr ... ignored
test getuid ... ok
test listen ... ok
test read ... ok
test read_udp ... ok
test readv ... ok
test socket ... ok
test uname ... ok
test write_stdout ... ok

test result: ok. 16 passed; 0 failed; 5 ignored; 0 measured; 0 filtered out; finished in 18.46s

     Running tests/rust_integration_tests.rs (target/debug/deps/rust_integration_tests-0122fb231e20ea63)

running 6 tests
test rust_sev_attestation ... ignored
test echo ... ok
test cpuid ... ok
test memory_stress_test ... ok
test memspike ... ok
test unix_echo ... ok

test result: ok. 5 passed; 0 failed; 1 ignored; 0 measured; 0 filtered out; finished in 48.22s

     Running tests/wasmldr_tests.rs (target/debug/deps/wasmldr_tests-98b6ff656b9d815e)

running 9 tests
test check_tcp ... ok
test hello_wasi_snapshot1 ... ok
test memspike ... ok
test echo has been running for over 60 seconds
test memory_stress_test has been running for over 60 seconds
test no_export has been running for over 60 seconds
test return_1 has been running for over 60 seconds
test wasi_snapshot1 has been running for over 60 seconds
test memory_stress_test ... ok
Error: default export in '' is not a function
test no_export ... ok
test return_1 ... ok
test wasi_snapshot1 ... ok
test zerooneone ... ok
test echo ... ok

test result: ok. 9 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 102.75s

BR, Jarkko