Skip to contentSkip to conversation

x86/vdso: sgx: Bug fixes

PATCH Superseded Sean Christopherson 3 participants 12 messages Oct 11, 2019 – Oct 16, 2019

Subsystems (5)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover x86/vdso: sgx: Bug fixes

Fix gwo bugs that were introduced in the refactoring, and tweak the callback prototype to make it more readable.

Please don't apply patches 3/5 -> 5/5 until Cedric has weighed in. The CFI patch may or may not be correct, and the prototype change is a continuation of a discussion (sending a patch seemed like the easiest way to move forward).

Sean Christopherson (5):
x86/vdso: sgx: Fix misaligned stack bug when invoking exit handler selftests/x86/sgx: Add check to verify exit handler stack alignment x86/vdso: sgx: Fix unwinder support
x86/vdso: sgx: Reorder params to callback to improve readability selftests/x86/sgx: Update the callbacks function parameters

 arch/x86/entry/vdso/vsgx_enter_enclave.S  | 18 ++++++++---------
 arch/x86/include/uapi/asm/sgx.h           | 10 +++++-----
 tools/testing/selftests/x86/sgx/defines.h |  1 +
 tools/testing/selftests/x86/sgx/main.c    | 24 ++++++++++++++++++-----
 4 files changed, 34 insertions(+), 19 deletions(-)
Signature · 1 line
2.22.0
#1 lore raw

PATCH 1/5 x86/vdso: sgx: Fix misaligned stack bug when invoking exit handler

Fix a recently introduced bug where an odd number of 8-byte parameters are pushed on the stack prior to invoking the userspace callback, which causes the CALL to execute with an unaligned stack and violate the x86_64 ABI.

---
 arch/x86/entry/vdso/vsgx_enter_enclave.S | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/entry/vdso/vsgx_enter_enclave.S +3 -1
index fc5622dcd2fa..b63091818df1 100644
@@ -135,10 +135,12 @@ ENTRY(__vdso_sgx_enter_enclave)
 .Linvoke_userspace_handler:
 	/*
 	 * Align stack per x86_64 ABI. Save the original %rsp in %rbx to be
-	 * restored after the callback returns.
+	 * restored after the callback returns.  Note, %rsp needs to be 16-byte
+	 * aligned _after_ pushing the three parameters on the stack.
 	 */
 	mov	%rsp, %rbx
 	and	$-0x10, %rsp
+	sub	$0x8, %rsp
 
 	/* Push @e, u_rsp and @tcs as parameters to the callback. */
 	push	0x18(%rbp)
Signature · 1 line
2.22.0
#2 lore raw

PATCH 2/5 selftests/x86/sgx: Add check to verify exit handler stack alignment

In reply to Sean Christopherson

Add an assembly trampoline to the basic exit handler to snapshot the pre-CALL %rsp in order to verify that the stack is 16-byte aligned as required by the x86_64 ABI.

---
 tools/testing/selftests/x86/sgx/defines.h |  1 +
 tools/testing/selftests/x86/sgx/main.c    | 20 +++++++++++++++++---
 2 files changed, 18 insertions(+), 3 deletions(-)
tools/testing/selftests/x86/sgx/defines.h +1 -0
index 199a830e198a..8ddd3cfd1070 100644
@@ -15,6 +15,7 @@ typedef uint64_t u64;
 
 #define __aligned(x) __attribute__((__aligned__(x)))
 #define __packed __attribute__((packed))
+#define __used __attribute__((__used__))
 
 /* Derived from asm-generic/bitsperlong.h. */
 #if __x86_64__
tools/testing/selftests/x86/sgx/main.c +17 -3
index 029502d81ac9..f46e5c8fdac4 100644
@@ -331,14 +331,28 @@ static void test_vdso_no_exit_handler(struct sgx_secs *secs)
 	ASSERT_EQ(exception.leaf, ENCLU_EENTER);
 }
 
-static int basic_exit_handler(long rdi, long rsi, long rdx, int ret,
-			      long r8, long r9, void *tcs, long ursp,
-			      struct sgx_enclave_exception *e)
+static int __used __basic_exit_handler(long rdi, long rsi, long rdx, int ret,
+				       long r8, long r9, void *tcs, long ursp,
+				       struct sgx_enclave_exception *e)
 {
+	TEST_ASSERT(!(r9 & 0xf), "Pre-CALL RSP not 16-byte aligned: %lx\n", r9);
 	ASSERT_EQ(ret, 0);
 	return 0;
 }
 
+extern void *basic_exit_handler;
+
+static void __used basic_exit_handler_trampoline(void)
+{
+	/* Load the pre-CALL %rsp into %r9 to verify correct alignment. */
+	asm volatile("1:\n\t"
+		     "lea 0x8(%%rsp), %%r9\n\t"
+		     "jmp __basic_exit_handler\n\t"
+		     "basic_exit_handler: .quad 1b\n\t"
+		     ".global basic_exit_handler"
+		     ::: "memory");
+}
+
 static int nr_page_faults;
 
 static int mprotect_exit_handler(long rdi, long rsi, long rdx, int ret,
Signature · 1 line
2.22.0
#3 lore raw

Re: [PATCH for_v23 2/5] selftests/x86/sgx: Add check to verify exit handler stack alignment

Quoted 5 lines · Sean Christopherson

Add an assembly trampoline to the basic exit handler to snapshot the pre-CALL %rsp in order to verify that the stack is 16-byte aligned as required by the x86_64 ABI.

I'll rework the existing selftest commit in my tree to use the "non-VDSO" (aka plain EENTER) call path. Please then provide me commits for vDSO selftests. These commits will be their own commits in the patch set. I'll move my selftest commit earlier in the patch set before the vDSO commits.

Should be more than obvious why this makes sene, so I skip explaining it but can provide rationale on demand basis.

I'll work this change out tomorrow.

/Jarkko

#4 lore raw

Re: [PATCH for_v23 2/5] selftests/x86/sgx: Add check to verify exit handler stack alignment

Quoted 13 lines · Jarkko Sakkinen
Quoted 5 lines · Sean Christopherson

Add an assembly trampoline to the basic exit handler to snapshot the pre-CALL %rsp in order to verify that the stack is 16-byte aligned as required by the x86_64 ABI.

I'll rework the existing selftest commit in my tree to use the "non-VDSO" (aka plain EENTER) call path. Please then provide me commits for vDSO selftests. These commits will be their own commits in the patch set. I'll move my selftest commit earlier in the patch set before the vDSO commits.

Quoted 4 lines

Should be more than obvious why this makes sene, so I skip explaining it but can provide rationale on demand basis.

I'll work this change out tomorrow.

I'd be also happy to get more clean commit messages for the vDSO commits. It is more important than refining documentation because commit log is immutable. I've already stated what is wrong ATM but I'll explain it once more shortly.

When writing a commit message you should always go below your expectations what reviewers already know. You want to make them easily accessible for maintainers who have to deal with dozens of patches on a daily basis.

/Jarkko

#5 lore raw

PATCH 3/5 x86/vdso: sgx: Fix unwinder support

In reply to Sean Christopherson

Relocate the .cfi_endproc directive to the RET that actually returns from the vDSO function. During recent refactoring, it was inadvertantly left next to the RET used for the retpoline.

---
 arch/x86/entry/vdso/vsgx_enter_enclave.S | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/entry/vdso/vsgx_enter_enclave.S +1 -1
index b63091818df1..3dd22780b7ef 100644
@@ -113,6 +113,7 @@ ENTRY(__vdso_sgx_enter_enclave)
 	leave
 	.cfi_def_cfa		%rsp, 8
 	ret
+	.cfi_endproc
 
 .Linvalid_leaf:
 	mov	$(-EINVAL), %eax
@@ -176,7 +177,6 @@ ENTRY(__vdso_sgx_enter_enclave)
 	jmp	1b
 2:	mov	%rax, (%rsp)
 	ret
-	.cfi_endproc
 
 _ASM_VDSO_EXTABLE_HANDLE(.Lenclu_eenter_eresume, .Lhandle_exception)
 
Signature · 1 line
2.22.0
#6 lore raw

Re: [PATCH for_v23 3/5] x86/vdso: sgx: Fix unwinder support

Quoted 15 lines · Sean Christopherson

Relocate the .cfi_endproc directive to the RET that actually returns from the vDSO function. During recent refactoring, it was inadvertantly left next to the RET used for the retpoline.

---
  arch/x86/entry/vdso/vsgx_enter_enclave.S | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/entry/vdso/vsgx_enter_enclave.S +0 -0
index b63091818df1..3dd22780b7ef 100644

arch/x86/entry/vdso/vsgx_enter_enclave.S ENTRY(__vdso_sgx_enter_enclave) · line 116

Quoted 4 lines
arch/x86/entry/vdso/vsgx_enter_enclave.S +1 -0
@@ -113,6 +113,7 @@ ENTRY(__vdso_sgx_enter_enclave)
  	leave
  	.cfi_def_cfa		%rsp, 8
  	ret
+	.cfi_endproc

As mentioned in my other email, the above won't work.

Quoted 12 lines
arch/x86/entry/vdso/vsgx_enter_enclave.S +0 -1
  .Linvalid_leaf:
  	mov	$(-EINVAL), %eax
@@ -176,7 +177,6 @@ ENTRY(__vdso_sgx_enter_enclave)
  	jmp	1b
  2:	mov	%rax, (%rsp)
  	ret
-	.cfi_endproc
 
  _ASM_VDSO_EXTABLE_HANDLE(.Lenclu_eenter_eresume, .Lhandle_exception)
 
#7 lore raw

PATCH 4/5 x86/vdso: sgx: Reorder params to callback to improve readability

In reply to Sean Christopherson

Swap @ret and @ursp in the callback prototype so that the output from the vDSO itself, @ret and @e, are grouped together. Having the first N parameters all share a type also makes the prototype easier to parse by (some) humans. And, passing @ursp via register saves one whole MOV instruction!

---
 arch/x86/entry/vdso/vsgx_enter_enclave.S | 16 +++++++---------
 arch/x86/include/uapi/asm/sgx.h          | 10 +++++-----
 2 files changed, 12 insertions(+), 14 deletions(-)
arch/x86/entry/vdso/vsgx_enter_enclave.S +7 -9
index 3dd22780b7ef..94f613b53b13 100644
@@ -134,23 +134,21 @@ ENTRY(__vdso_sgx_enter_enclave)
 	jmp	.Lhandle_exit
 
 .Linvoke_userspace_handler:
+	/* Pass the untrusted RSP (at exit) to the callback via %rcx. */
+	mov	%rsp, %rcx
+
 	/*
-	 * Align stack per x86_64 ABI. Save the original %rsp in %rbx to be
-	 * restored after the callback returns.  Note, %rsp needs to be 16-byte
-	 * aligned _after_ pushing the three parameters on the stack.
+	 * Align stack per x86_64 ABI. Note, %rsp needs to be 16-byte aligned
+	 * _after_ pushing the three parameters on the stack.
 	 */
-	mov	%rsp, %rbx
 	and	$-0x10, %rsp
 	sub	$0x8, %rsp
 
-	/* Push @e, u_rsp and @tcs as parameters to the callback. */
+	/* Push @e, the "return" value and @tcs as params to the callback. */
 	push	0x18(%rbp)
-	push	%rbx
+	push	%rax
 	push	0x10(%rbp)
 
-	/* Pass the "return" value to the callback via %rcx. */
-	mov	%eax, %ecx
-
 	/* Clear RFLAGS.DF per x86_64 ABI */
 	cld
 
arch/x86/include/uapi/asm/sgx.h +5 -5
index 255392f5054f..95f69f938b58 100644
@@ -88,16 +88,16 @@ struct sgx_enclave_exception {
  * @rdi:	RDI at the time of enclave exit
  * @rsi:	RSI at the time of enclave exit
  * @rdx:	RDX at the time of enclave exit
- * @ret:	0 on success (EEXIT), -EFAULT on an exception
+ * @ursp:	RSP at the time of enclave exit (untrusted stack)
  * @r8:		R8 at the time of enclave exit
  * @r9:		R9 at the time of enclave exit
  * @tcs:	Thread Control Structure used to enter enclave
- * @ursp:	RSP at the time of enclave exit
+ * @ret:	0 on success (EEXIT), -EFAULT on an exception
  * @e:		Pointer to struct sgx_enclave_exception (as provided by caller)
  */
-typedef int (*sgx_enclave_exit_handler_t)(long rdi, long rsi, long rdx, int ret,
-					  long r8, long r9, void *tcs,
-					  long ursp,
+typedef int (*sgx_enclave_exit_handler_t)(long rdi, long rsi, long rdx,
+					  long ursp, long r8, long r9,
+					  void *tcs, int ret,
 					  struct sgx_enclave_exception *e);
 
 #endif /* _UAPI_ASM_X86_SGX_H */
Signature · 1 line
2.22.0
#8 lore raw

Re: [PATCH for_v23 4/5] x86/vdso: sgx: Reorder params to callback to improve readability

Quoted 21 lines · Sean Christopherson

Swap @ret and @ursp in the callback prototype so that the output from the vDSO itself, @ret and @e, are grouped together. Having the first N parameters all share a type also makes the prototype easier to parse by (some) humans. And, passing @ursp via register saves one whole MOV instruction!

---
  arch/x86/entry/vdso/vsgx_enter_enclave.S | 16 +++++++---------
  arch/x86/include/uapi/asm/sgx.h          | 10 +++++-----
  2 files changed, 12 insertions(+), 14 deletions(-)
arch/x86/entry/vdso/vsgx_enter_enclave.S +1 -0
index 3dd22780b7ef..94f613b53b13 100644
@@ -134,23 +134,21 @@ ENTRY(__vdso_sgx_enter_enclave)
  	jmp	.Lhandle_exit
 
  .Linvoke_userspace_handler:
+	/* Pass the untrusted RSP (at exit) to the callback via %rcx. */

arch/x86/entry/vdso/vsgx_enter_enclave.S ENTRY(__vdso_sgx_enter_enclave) · line 142

Quoted 10 lines
arch/x86/entry/vdso/vsgx_enter_enclave.S +4 -4
+	mov	%rsp, %rcx
+
  	/*
-	 * Align stack per x86_64 ABI. Save the original %rsp in %rbx to be
-	 * restored after the callback returns.  Note, %rsp needs to be 16-byte
-	 * aligned _after_ pushing the three parameters on the stack.
+	 * Align stack per x86_64 ABI. Note, %rsp needs to be 16-byte aligned
+	 * _after_ pushing the three parameters on the stack.
  	 */
-	mov	%rsp, %rbx

Per x86_64 ABI, %rcx is _not_ preserved across function call. How are you going to restore the stack after callback returns?

arch/x86/entry/vdso/vsgx_enter_enclave.S ENTRY(__vdso_sgx_enter_enclave) · line 145

Quoted 2 lines
arch/x86/entry/vdso/vsgx_enter_enclave.S +0 -0
  	and	$-0x10, %rsp
  	sub	$0x8, %rsp

Usually compilers would just use a 1-byte "push" instead of a 4-byte "sub" instruction here.

Quoted 41 lines
arch/x86/entry/vdso/vsgx_enter_enclave.S +2 -5
-	/* Push @e, u_rsp and @tcs as parameters to the callback. */
+	/* Push @e, the "return" value and @tcs as params to the callback. */
  	push	0x18(%rbp)
-	push	%rbx
+	push	%rax
  	push	0x10(%rbp)
 
-	/* Pass the "return" value to the callback via %rcx. */
-	mov	%eax, %ecx
-
  	/* Clear RFLAGS.DF per x86_64 ABI */
  	cld
 
arch/x86/include/uapi/asm/sgx.h +5 -5
index 255392f5054f..95f69f938b58 100644
@@ -88,16 +88,16 @@ struct sgx_enclave_exception {
   * @rdi:	RDI at the time of enclave exit
   * @rsi:	RSI at the time of enclave exit
   * @rdx:	RDX at the time of enclave exit
- * @ret:	0 on success (EEXIT), -EFAULT on an exception
+ * @ursp:	RSP at the time of enclave exit (untrusted stack)
   * @r8:		R8 at the time of enclave exit
   * @r9:		R9 at the time of enclave exit
   * @tcs:	Thread Control Structure used to enter enclave
- * @ursp:	RSP at the time of enclave exit
+ * @ret:	0 on success (EEXIT), -EFAULT on an exception
   * @e:		Pointer to struct sgx_enclave_exception (as provided by caller)
   */
-typedef int (*sgx_enclave_exit_handler_t)(long rdi, long rsi, long rdx, int ret,
-					  long r8, long r9, void *tcs,
-					  long ursp,
+typedef int (*sgx_enclave_exit_handler_t)(long rdi, long rsi, long rdx,
+					  long ursp, long r8, long r9,
+					  void *tcs, int ret,
  					  struct sgx_enclave_exception *e);
 
  #endif /* _UAPI_ASM_X86_SGX_H */
#9 lore raw

Re: [PATCH for_v23 4/5] x86/vdso: sgx: Reorder params to callback to improve readability

Quoted 26 lines · Cedric
Quoted 25 lines · Sean Christopherson

Swap @ret and @ursp in the callback prototype so that the output from the vDSO itself, @ret and @e, are grouped together. Having the first N parameters all share a type also makes the prototype easier to parse by (some) humans. And, passing @ursp via register saves one whole MOV instruction!

---
 arch/x86/entry/vdso/vsgx_enter_enclave.S | 16 +++++++---------
 arch/x86/include/uapi/asm/sgx.h          | 10 +++++-----
 2 files changed, 12 insertions(+), 14 deletions(-)
arch/x86/entry/vdso/vsgx_enter_enclave.S +3 -2
index 3dd22780b7ef..94f613b53b13 100644
@@ -134,23 +134,21 @@ ENTRY(__vdso_sgx_enter_enclave)
 	jmp	.Lhandle_exit
 .Linvoke_userspace_handler:
+	/* Pass the untrusted RSP (at exit) to the callback via %rcx. */
+	mov	%rsp, %rcx
+
 	/*
-	 * Align stack per x86_64 ABI. Save the original %rsp in %rbx to be
-	 * restored after the callback returns.  Note, %rsp needs to be 16-byte
Quoted 8 lines

arch/x86/entry/vdso/vsgx_enter_enclave.S ENTRY(__vdso_sgx_enter_enclave) · line 142

Quoted 5 lines
arch/x86/entry/vdso/vsgx_enter_enclave.S +2 -2
-	 * aligned _after_ pushing the three parameters on the stack.
+	 * Align stack per x86_64 ABI. Note, %rsp needs to be 16-byte aligned
+	 * _after_ pushing the three parameters on the stack.
 	 */
-	mov	%rsp, %rbx

Per x86_64 ABI, %rcx is _not_ preserved across function call. How are you going to restore the stack after callback returns?

Magic?

I completely spaced on restoring RSP, as evidenced by the fact that I didn't even change the restoration code to use RCX.

Quoted 6 lines

arch/x86/entry/vdso/vsgx_enter_enclave.S ENTRY(__vdso_sgx_enter_enclave) · line 145

Quoted 3 lines
arch/x86/entry/vdso/vsgx_enter_enclave.S +0 -0
 	and	$-0x10, %rsp
 	sub	$0x8, %rsp

Usually compilers would just use a 1-byte "push" instead of a 4-byte "sub" instruction here.

Ah! Took me a minute to understand 'N-byte' was referring to the opcode...

I'll update to use a push.

Quoted 38 lines
Quoted 38 lines
arch/x86/entry/vdso/vsgx_enter_enclave.S +2 -5
-	/* Push @e, u_rsp and @tcs as parameters to the callback. */
+	/* Push @e, the "return" value and @tcs as params to the callback. */
 	push	0x18(%rbp)
-	push	%rbx
+	push	%rax
 	push	0x10(%rbp)
-	/* Pass the "return" value to the callback via %rcx. */
-	mov	%eax, %ecx
-
 	/* Clear RFLAGS.DF per x86_64 ABI */
 	cld
arch/x86/include/uapi/asm/sgx.h +5 -5
index 255392f5054f..95f69f938b58 100644
@@ -88,16 +88,16 @@ struct sgx_enclave_exception {
  * @rdi:	RDI at the time of enclave exit
  * @rsi:	RSI at the time of enclave exit
  * @rdx:	RDX at the time of enclave exit
- * @ret:	0 on success (EEXIT), -EFAULT on an exception
+ * @ursp:	RSP at the time of enclave exit (untrusted stack)
  * @r8:		R8 at the time of enclave exit
  * @r9:		R9 at the time of enclave exit
  * @tcs:	Thread Control Structure used to enter enclave
- * @ursp:	RSP at the time of enclave exit
+ * @ret:	0 on success (EEXIT), -EFAULT on an exception
  * @e:		Pointer to struct sgx_enclave_exception (as provided by caller)
  */
-typedef int (*sgx_enclave_exit_handler_t)(long rdi, long rsi, long rdx, int ret,
-					  long r8, long r9, void *tcs,
-					  long ursp,
+typedef int (*sgx_enclave_exit_handler_t)(long rdi, long rsi, long rdx,
+					  long ursp, long r8, long r9,
+					  void *tcs, int ret,
 					  struct sgx_enclave_exception *e);
 #endif /* _UAPI_ASM_X86_SGX_H */
 
#10 lore raw

PATCH 5/5 selftests/x86/sgx: Update the callbacks function parameters

In reply to Sean Christopherson

Swap @ursp and @ret in the exit handler callbacks to match a recent kernel change.

---
 tools/testing/selftests/x86/sgx/main.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
tools/testing/selftests/x86/sgx/main.c +4 -4
index f46e5c8fdac4..b84ffbf6e1e2 100644
@@ -331,8 +331,8 @@ static void test_vdso_no_exit_handler(struct sgx_secs *secs)
 	ASSERT_EQ(exception.leaf, ENCLU_EENTER);
 }
 
-static int __used __basic_exit_handler(long rdi, long rsi, long rdx, int ret,
-				       long r8, long r9, void *tcs, long ursp,
+static int __used __basic_exit_handler(long rdi, long rsi, long rdx, long ursp,
+				       long r8, long r9, void *tcs, int ret,
 				       struct sgx_enclave_exception *e)
 {
 	TEST_ASSERT(!(r9 & 0xf), "Pre-CALL RSP not 16-byte aligned: %lx\n", r9);
@@ -355,8 +355,8 @@ static void __used basic_exit_handler_trampoline(void)
 
 static int nr_page_faults;
 
-static int mprotect_exit_handler(long rdi, long rsi, long rdx, int ret,
-				 long r8, long r9, void *tcs, long ursp,
+static int mprotect_exit_handler(long rdi, long rsi, long rdx, long ursp,
+				 long r8, long r9, void *tcs, int ret,
 				 struct sgx_enclave_exception *e)
 {
 	int prot, rc;
Signature · 1 line
2.22.0
#11 lore raw
In reply to Sean Christopherson
Quoted 7 lines · Sean Christopherson

Fix gwo bugs that were introduced in the refactoring, and tweak the callback prototype to make it more readable.

Please don't apply patches 3/5 -> 5/5 until Cedric has weighed in. The CFI patch may or may not be correct, and the prototype change is a continuation of a discussion (sending a patch seemed like the easiest way to move forward).

Cedric, Sean, thank you.

I merged everything except the selftest updates.

/Jarkko