Hi Jarkko, Reinette,
> -----Original Message-----
> From: Jarkko Sakkinen <jarkko@xxxxxxxxxx>
> Sent: Tuesday, August 16, 2022 4:34 PM
> To: Chatre, Reinette <reinette.chatre@xxxxxxxxx>
> Cc: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>; linux-
> sgx@xxxxxxxxxxxxxxx; Dhanraj, Vijay <vijay.dhanraj@xxxxxxxxx>; Shuah Khan
> <shuah@xxxxxxxxxx>; open list:KERNEL SELFTEST FRAMEWORK <linux-
> kselftest@xxxxxxxxxxxxxxx>; open list <linux-kernel@xxxxxxxxxxxxxxx>
> Subject: Re: [PATCH 2/2] selftests/sgx: Add SGX selftest
> augment_via_eaccept_long
>
> On Tue, Aug 16, 2022 at 09:26:40AM -0700, Reinette Chatre wrote:
> > Hi Vijay,
> >
> > Thank you very much for digging into this. A few comments below.
> >
> > On 8/15/2022 4:39 PM, Jarkko Sakkinen wrote:
> > > From: Vijay Dhanraj <vijay.dhanraj@xxxxxxxxx>
> > >
> > > Add a new test case which is same as augment_via_eaccept but adds a
> > > larger number of EPC pages to stress test EAUG via EACCEPT.
> > >
> > > Signed-off-by: Vijay Dhanraj <vijay.dhanraj@xxxxxxxxx>
> > > Reviewed-by: Jarkko Sakkinen <jarkko@xxxxxxxxxx>
> > > Signed-off-by: Jarkko Sakkinen <jarkko@xxxxxxxxxx>
> > > ---
> > > I removed Githubisms (hyphens), added missing subsystem tag, and
> > > cleaned up the commit message a bit.
> > > tools/testing/selftests/sgx/load.c | 5 +-
> > > tools/testing/selftests/sgx/main.c | 120
> +++++++++++++++++++++++-
> > > tools/testing/selftests/sgx/main.h | 3 +-
> > > tools/testing/selftests/sgx/sigstruct.c | 2 +-
> > > 4 files changed, 125 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/tools/testing/selftests/sgx/load.c
> > > b/tools/testing/selftests/sgx/load.c
> > > index 94bdeac1cf04..7de1b15c90b1 100644
> > > --- a/tools/testing/selftests/sgx/load.c
> > > +++ b/tools/testing/selftests/sgx/load.c
> > > @@ -171,7 +171,8 @@ uint64_t encl_get_entry(struct encl *encl, const
> char *symbol)
> > > return 0;
> > > }
> > >
> > > -bool encl_load(const char *path, struct encl *encl, unsigned long
> > > heap_size)
> > > +bool encl_load(const char *path, struct encl *encl, unsigned long
> heap_size,
> > > + unsigned long edmm_size)
> > > {
> > > const char device_path[] = "/dev/sgx_enclave";
> > > struct encl_segment *seg;
> > > @@ -300,7 +301,7 @@ bool encl_load(const char *path, struct encl
> > > *encl, unsigned long heap_size)
> > >
> > > encl->src_size = encl->segment_tbl[j].offset +
> > > encl->segment_tbl[j].size;
> > >
> > > - for (encl->encl_size = 4096; encl->encl_size < encl->src_size; )
> > > + for (encl->encl_size = 4096; encl->encl_size < encl->src_size +
> > > +edmm_size;)
> > > encl->encl_size <<= 1;
> > >
> >
> > This seems to create the hardcoded 8GB larger enclave for all (SGX1
> > and SGX2) tests, not just the test introduced with this commit (and the only
> user of this extra space).
> > Is this intended? This can be done without impacting all the other tests.
>
> It's a valid point. I can adjust the patch.
Thanks Jarkko.
>
> >
> > > return true;
> > > diff --git a/tools/testing/selftests/sgx/main.c
> > > b/tools/testing/selftests/sgx/main.c
> > > index 9820b3809c69..65e79682f75e 100644
> > > --- a/tools/testing/selftests/sgx/main.c
> > > +++ b/tools/testing/selftests/sgx/main.c
> > > @@ -25,6 +25,8 @@ static const uint64_t MAGIC =
> > > 0x1122334455667788ULL; static const uint64_t MAGIC2 =
> > > 0x8877665544332211ULL; vdso_sgx_enter_enclave_t
> > > vdso_sgx_enter_enclave;
> > >
> > > +static const unsigned long edmm_size = 8589934592; //8G
> > > +
> >
> > Could you please elaborate how this constant was chosen? I understand
> > that this test helped to uncover a bug and it is useful to add to the
> > kernel. When doing so this test will be run on systems with a variety
> > of SGX memory sizes, could you please elaborate (and add a
> > snippet) how 8GB is the right value for all systems?
>
> It is the only constant I know for sure that some people (Vijay and Haitao)
> have been able to reproduce the bug.
>
> Unless someone can show that the same bug reproduces with a smaller
> constant, changing it would make the whole test irrelevant.
I tried with 2GB and it always succeed and with 4GB was able to repro sporadically. But with 8GB failure was consistent. One thing to note is even with 8GB Haitao couldn't reproduce this every time. So not sure if it good for all the systems but on my ICX system, I was able to consistently repro with this value.
>
> >
> > /on page to be added/on every page to be added/ ?
> >
> > > + */
> > > +#define TIMEOUT_LONG 900 /* seconds */ TEST_F_TIMEOUT(enclave,
> > > +augment_via_eaccept_long, TIMEOUT_LONG) {
> > > + struct encl_op_get_from_addr get_addr_op;
> > > + struct encl_op_put_to_addr put_addr_op;
> > > + struct encl_op_eaccept eaccept_op;
> > > + size_t total_size = 0;
> > > + void *addr;
> > > + unsigned long i;
> >
> > (reverse fir tree order)
>
> I would just change this to "int i" instead.
I think changing it to "int i" will cause a buffer overflow with edmm_size being 8GB.
>
> >
> > > +
> > > + if (!sgx2_supported())
> > > + SKIP(return, "SGX2 not supported");
> > > +
> > > + ASSERT_TRUE(setup_test_encl(ENCL_HEAP_SIZE_DEFAULT, &self-
> >encl,
> > > +_metadata));
> > > +
> > > + memset(&self->run, 0, sizeof(self->run));
> > > + self->run.tcs = self->encl.encl_base;
> > > +
> > > + for (i = 0; i < self->encl.nr_segments; i++) {
> > > + struct encl_segment *seg = &self->encl.segment_tbl[i];
> > > +
> > > + total_size += seg->size;
> > > + TH_LOG("test enclave: total_size = %ld, seg->size = %ld",
> total_size, seg->size);
> > > + }
> > > +
> > > + /*
> > > + * Actual enclave size is expected to be larger than the loaded
> > > + * test enclave since enclave size must be a power of 2 in bytes while
> > > + * test_encl does not consume it all.
> > > + */
> > > + EXPECT_LT(total_size + edmm_size, self->encl.encl_size);
> >
> > Will this test ever fail?
>
> With a *quick* look: no.
>
> Vijay, what was the point of this check?
Yes we can remove this check. I tried to copy from `augment_via_eaccept` and just changed the request size.
>
> > > +
> > > + /*
> > > + * mmap() a page at end of existing enclave to be used for dynamic
> > > + * EPC page.
> >
> > copy&paste line still refers to single page
> >
> > > + *
> > > + * Kernel will allow new mapping using any permissions if it
> > > + * falls into the enclave's address range but not backed
> > > + * by existing enclave pages.
> > > + */
> > > + TH_LOG("mmaping pages at end of enclave...");
> > > + addr = mmap((void *)self->encl.encl_base + total_size, edmm_size,
> > > + PROT_READ | PROT_WRITE | PROT_EXEC,
> MAP_SHARED | MAP_FIXED,
> > > + self->encl.fd, 0);
> > > + EXPECT_NE(addr, MAP_FAILED);
> > > +
> > > + self->run.exception_vector = 0;
> > > + self->run.exception_error_code = 0;
> > > + self->run.exception_addr = 0;
> > > +
> > > + /*
> > > + * Run EACCEPT on new page to trigger the #PF->EAUG-
> >EACCEPT(again
> > > + * without a #PF). All should be transparent to userspace.
> > > + */
> >
> > copy&paste from single page test referring to one page
> >
> > > + TH_LOG("Entering enclave to run EACCEPT for each page of %zd
> bytes may take a while ...",
> > > + edmm_size);
> > > + eaccept_op.flags = SGX_SECINFO_R | SGX_SECINFO_W |
> SGX_SECINFO_REG | SGX_SECINFO_PENDING;
> > > + eaccept_op.ret = 0;
> > > + eaccept_op.header.type = ENCL_OP_EACCEPT;
> > > +
> > > + for (i = 0; i < edmm_size; i += 4096) {
> > > + eaccept_op.epc_addr = (uint64_t)(addr + i);
> > > +
> > > + EXPECT_EQ(ENCL_CALL(&eaccept_op, &self->run, true), 0);
> > > + if (self->run.exception_vector == 14 &&
> > > + self->run.exception_error_code == 4 &&
> > > + self->run.exception_addr == self->encl.encl_base) {
> > > + munmap(addr, edmm_size);
> > > + SKIP(return, "Kernel does not support adding pages
> to initialized enclave");
> > > + }
> > > +
> > > + EXPECT_EQ(self->run.exception_vector, 0);
> > > + EXPECT_EQ(self->run.exception_error_code, 0);
> > > + EXPECT_EQ(self->run.exception_addr, 0);
> > > + ASSERT_EQ(eaccept_op.ret, 0);
> > > + ASSERT_EQ(self->run.function, EEXIT);
> > > + }
> > > +
> > > + /*
> > > + * New page should be accessible from within enclave - attempt to
> > > + * write to it.
> > > + */
> >
> > This portion below was also copied from previous test and by only
> > testing a write to the first page of the range the purpose is not
> > clear. Could you please elaborate if the intention is to only test
> > accessibility of the first page and why that is sufficient?
>
> It is sufficient because the test reproduces the bug. It would have to be
> rather elaborated why you would possibly want to do more than that.
>
> > > + put_addr_op.value = MAGIC;
> > > + put_addr_op.addr = (unsigned long)addr;
> > > + put_addr_op.header.type = ENCL_OP_PUT_TO_ADDRESS;
> > > +
> > > + EXPECT_EQ(ENCL_CALL(&put_addr_op, &self->run, true), 0);
> > > +
> > > + EXPECT_EEXIT(&self->run);
> > > + EXPECT_EQ(self->run.exception_vector, 0);
> > > + EXPECT_EQ(self->run.exception_error_code, 0);
> > > + EXPECT_EQ(self->run.exception_addr, 0);
> > > +
> > > + /*
> > > + * Read memory from newly added page that was just written to,
> > > + * confirming that data previously written (MAGIC) is present.
> > > + */
> > > + get_addr_op.value = 0;
> > > + get_addr_op.addr = (unsigned long)addr;
> > > + get_addr_op.header.type = ENCL_OP_GET_FROM_ADDRESS;
> > > +
> > > + EXPECT_EQ(ENCL_CALL(&get_addr_op, &self->run, true), 0);
> > > +
> > > + EXPECT_EQ(get_addr_op.value, MAGIC);
> > > + EXPECT_EEXIT(&self->run);
> > > + EXPECT_EQ(self->run.exception_vector, 0);
> > > + EXPECT_EQ(self->run.exception_error_code, 0);
> > > + EXPECT_EQ(self->run.exception_addr, 0);
> > > +
> > > + munmap(addr, edmm_size);
> > > +}
> > > +
> > > /*
> > > * SGX2 page type modification test in two phases:
> > > * Phase 1:
> > > diff --git a/tools/testing/selftests/sgx/main.h
> > > b/tools/testing/selftests/sgx/main.h
> > > index fc585be97e2f..fe5d39ac0e1e 100644
> > > --- a/tools/testing/selftests/sgx/main.h
> > > +++ b/tools/testing/selftests/sgx/main.h
> > > @@ -35,7 +35,8 @@ extern unsigned char sign_key[]; extern unsigned
> > > char sign_key_end[];
> > >
> > > void encl_delete(struct encl *ctx); -bool encl_load(const char
> > > *path, struct encl *encl, unsigned long heap_size);
> > > +bool encl_load(const char *path, struct encl *encl, unsigned long
> heap_size,
> > > + unsigned long edmm_size);
> > > bool encl_measure(struct encl *encl); bool encl_build(struct encl
> > > *encl); uint64_t encl_get_entry(struct encl *encl, const char
> > > *symbol); diff --git a/tools/testing/selftests/sgx/sigstruct.c
> > > b/tools/testing/selftests/sgx/sigstruct.c
> > > index 50c5ab1aa6fa..6000cf0e4975 100644
> > > --- a/tools/testing/selftests/sgx/sigstruct.c
> > > +++ b/tools/testing/selftests/sgx/sigstruct.c
> > > @@ -343,7 +343,7 @@ bool encl_measure(struct encl *encl)
> > > if (!ctx)
> > > goto err;
> > >
> > > - if (!mrenclave_ecreate(ctx, encl->src_size))
> > > + if (!mrenclave_ecreate(ctx, encl->encl_size))
> > > goto err;
> > >
> > > for (i = 0; i < encl->nr_segments; i++) {
> >
> >
> > Looking at mrenclave_ecreate() the above snippet seems separate from
> > this test and incomplete since it now obtains encl->encl_size but
> > continues to compute it again internally. Should this be a separate fix?
>
> I would remove this part completely but this also needs comment from Vijay.
If we restrict the large enclave size just for this test, then the above change can be reverted. Calling ` mrenclave_ecreate` with src_size esults in EINIT failure and I think the reason is because of incorrect MRenclave.
>
> > Reinette
>
>
> BR, Jarkko
Regards, Vijay
