On Wed, 2022-04-13 at 14:10 -0700, Reinette Chatre wrote:
> struct sgx_encl should be protected with the mutex
> sgx_encl->lock. One exception is sgx_encl->page_cnt that
> is incremented (in sgx_encl_grow()) when an enclave page
> is added to the enclave. The reason the mutex is not held
> is to allow the reclaimer to be called directly if there are
> no EPC pages (in support of a new VA page) available at the time.
>
> Incrementing sgx_encl->page_cnt without sgc_encl->lock held
> is currently (before SGX2) safe from concurrent updates because
> all paths in which sgx_encl_grow() is called occur before
> enclave initialization and are protected with an atomic
> operation on SGX_ENCL_IOCTL.
>
> SGX2 includes support for dynamically adding pages after
> enclave initialization where the protection of SGX_ENCL_IOCTL
> is not available.
>
> Make direct reclaim of EPC pages optional when new VA pages
> are added to the enclave. Essentially the existing "reclaim"
> flag used when regular EPC pages are added to an enclave
> becomes available to the caller when used to allocate VA pages
> instead of always being "true".
>
> When adding pages without invoking the reclaimer it is possible
> to do so with sgx_encl->lock held, gaining its protection against
> concurrent updates to sgx_encl->page_cnt after enclave
> initialization.
>
> No functional change.
>
> Reported-by: Haitao Huang <[email protected]>
> Tested-by: Haitao Huang <[email protected]>
> Signed-off-by: Reinette Chatre <[email protected]>
Reviewed-by: Jarkko Sakkinen <[email protected]>
Nit: I don't think tested-by is in the right patch here. Maybe
Haitao's tested-by should be moved into patch that actually adds
support for EAUG? Not something I would NAK this patch, just
wondering...
> ---
> Changes since V3:
> - New patch prompted by Haitao encountering the
> WARN_ON_ONCE(encl->page_cnt % SGX_VA_SLOT_COUNT)
> within sgx_encl_grow() during his SGX2 multi-threaded
> unit tests.
>
> arch/x86/kernel/cpu/sgx/encl.c | 6 ++++--
> arch/x86/kernel/cpu/sgx/encl.h | 4 ++--
> arch/x86/kernel/cpu/sgx/ioctl.c | 8 ++++----
> 3 files changed, 10 insertions(+), 8 deletions(-)
>
> diff --git a/arch/x86/kernel/cpu/sgx/encl.c b/arch/x86/kernel/cpu/sgx/encl.c
> index 546423753e4c..92516aeca405 100644
> --- a/arch/x86/kernel/cpu/sgx/encl.c
> +++ b/arch/x86/kernel/cpu/sgx/encl.c
> @@ -869,6 +869,8 @@ void sgx_zap_enclave_ptes(struct sgx_encl *encl, unsigned long addr)
>
> /**
> * sgx_alloc_va_page() - Allocate a Version Array (VA) page
> + * @reclaim: Reclaim EPC pages directly if none available. Enclave
> + * mutex should not be held if this is set.
> *
> * Allocate a free EPC page and convert it to a Version Array (VA) page.
> *
> @@ -876,12 +878,12 @@ void sgx_zap_enclave_ptes(struct sgx_encl *encl, unsigned long addr)
> * a VA page,
> * -errno otherwise
> */
> -struct sgx_epc_page *sgx_alloc_va_page(void)
> +struct sgx_epc_page *sgx_alloc_va_page(bool reclaim)
> {
> struct sgx_epc_page *epc_page;
> int ret;
>
> - epc_page = sgx_alloc_epc_page(NULL, true);
> + epc_page = sgx_alloc_epc_page(NULL, reclaim);
> if (IS_ERR(epc_page))
> return ERR_CAST(epc_page);
>
> diff --git a/arch/x86/kernel/cpu/sgx/encl.h b/arch/x86/kernel/cpu/sgx/encl.h
> index 253ebdd1c5be..66adb8faec45 100644
> --- a/arch/x86/kernel/cpu/sgx/encl.h
> +++ b/arch/x86/kernel/cpu/sgx/encl.h
> @@ -116,14 +116,14 @@ struct sgx_encl_page *sgx_encl_page_alloc(struct sgx_encl *encl,
> unsigned long offset,
> u64 secinfo_flags);
> void sgx_zap_enclave_ptes(struct sgx_encl *encl, unsigned long addr);
> -struct sgx_epc_page *sgx_alloc_va_page(void);
> +struct sgx_epc_page *sgx_alloc_va_page(bool reclaim);
> unsigned int sgx_alloc_va_slot(struct sgx_va_page *va_page);
> void sgx_free_va_slot(struct sgx_va_page *va_page, unsigned int offset);
> bool sgx_va_page_full(struct sgx_va_page *va_page);
> void sgx_encl_free_epc_page(struct sgx_epc_page *page);
> struct sgx_encl_page *sgx_encl_load_page(struct sgx_encl *encl,
> unsigned long addr);
> -struct sgx_va_page *sgx_encl_grow(struct sgx_encl *encl);
> +struct sgx_va_page *sgx_encl_grow(struct sgx_encl *encl, bool reclaim);
> void sgx_encl_shrink(struct sgx_encl *encl, struct sgx_va_page *va_page);
>
> #endif /* _X86_ENCL_H */
> diff --git a/arch/x86/kernel/cpu/sgx/ioctl.c b/arch/x86/kernel/cpu/sgx/ioctl.c
> index bb8cdb2ad0d1..5d41aa204761 100644
> --- a/arch/x86/kernel/cpu/sgx/ioctl.c
> +++ b/arch/x86/kernel/cpu/sgx/ioctl.c
> @@ -17,7 +17,7 @@
> #include "encl.h"
> #include "encls.h"
>
> -struct sgx_va_page *sgx_encl_grow(struct sgx_encl *encl)
> +struct sgx_va_page *sgx_encl_grow(struct sgx_encl *encl, bool reclaim)
> {
> struct sgx_va_page *va_page = NULL;
> void *err;
> @@ -30,7 +30,7 @@ struct sgx_va_page *sgx_encl_grow(struct sgx_encl *encl)
> if (!va_page)
> return ERR_PTR(-ENOMEM);
>
> - va_page->epc_page = sgx_alloc_va_page();
> + va_page->epc_page = sgx_alloc_va_page(reclaim);
> if (IS_ERR(va_page->epc_page)) {
> err = ERR_CAST(va_page->epc_page);
> kfree(va_page);
> @@ -64,7 +64,7 @@ static int sgx_encl_create(struct sgx_encl *encl, struct sgx_secs *secs)
> struct file *backing;
> long ret;
>
> - va_page = sgx_encl_grow(encl);
> + va_page = sgx_encl_grow(encl, true);
> if (IS_ERR(va_page))
> return PTR_ERR(va_page);
> else if (va_page)
> @@ -275,7 +275,7 @@ static int sgx_encl_add_page(struct sgx_encl *encl, unsigned long src,
> return PTR_ERR(epc_page);
> }
>
> - va_page = sgx_encl_grow(encl);
> + va_page = sgx_encl_grow(encl, true);
> if (IS_ERR(va_page)) {
> ret = PTR_ERR(va_page);
> goto err_out_free;
BR, Jarkko
Hi Jarkko,
On 4/14/2022 4:18 AM, Jarkko Sakkinen wrote:
> On Wed, 2022-04-13 at 14:10 -0700, Reinette Chatre wrote:
>> struct sgx_encl should be protected with the mutex
>> sgx_encl->lock. One exception is sgx_encl->page_cnt that
>> is incremented (in sgx_encl_grow()) when an enclave page
>> is added to the enclave. The reason the mutex is not held
>> is to allow the reclaimer to be called directly if there are
>> no EPC pages (in support of a new VA page) available at the time.
>>
>> Incrementing sgx_encl->page_cnt without sgc_encl->lock held
>> is currently (before SGX2) safe from concurrent updates because
>> all paths in which sgx_encl_grow() is called occur before
>> enclave initialization and are protected with an atomic
>> operation on SGX_ENCL_IOCTL.
>>
>> SGX2 includes support for dynamically adding pages after
>> enclave initialization where the protection of SGX_ENCL_IOCTL
>> is not available.
>>
>> Make direct reclaim of EPC pages optional when new VA pages
>> are added to the enclave. Essentially the existing "reclaim"
>> flag used when regular EPC pages are added to an enclave
>> becomes available to the caller when used to allocate VA pages
>> instead of always being "true".
>>
>> When adding pages without invoking the reclaimer it is possible
>> to do so with sgx_encl->lock held, gaining its protection against
>> concurrent updates to sgx_encl->page_cnt after enclave
>> initialization.
>>
>> No functional change.
>>
>> Reported-by: Haitao Huang <[email protected]>
>> Tested-by: Haitao Huang <[email protected]>
>> Signed-off-by: Reinette Chatre <[email protected]>
>
> Reviewed-by: Jarkko Sakkinen <[email protected]>
Thank you.
>
> Nit: I don't think tested-by is in the right patch here. Maybe
> Haitao's tested-by should be moved into patch that actually adds
> support for EAUG? Not something I would NAK this patch, just
> wondering...
Yes, that is a good point. While this is the bulk of the fix where
the new API is introduced, the test is only applicable when this API
is used and that is in "x86/sgx: Support adding of pages to an
initialized enclave". I will move the "Tested-by" to that patch.
Reinette