[Sanitizers] Don't inline unpoisoning of small stacks when inlining disabled #75555
Conversation
…isabled When ASan.MaxInlinePoisoningSize == 0 , it means that no shadow memory operations should be made via inlined instrumentation code, but only via calls to shadow setting functions. This change fixes one violation of this, which happened when the function allocas count was small, i.e. less than 5 - in the code modifying the shadow just before ret instruction. We now explicitly check ASan.MaxInlinePoisoningSize , and if it's 0 then we disallow inlining. It is required for the instrumentation emitting code suitable for handling by ABI implementation. rdar://119513720
|
@llvm/pr-subscribers-compiler-rt-sanitizer @llvm/pr-subscribers-llvm-transforms Author: Mariusz Borsa (wrotki) ChangesWhen ASan.MaxInlinePoisoningSize == 0 , it means that no shadow memory operations should be made via inlined instrumentation code, rdar://119513720 Full diff: https://github.com/llvm/llvm-project/pull/75555.diff 3 Files Affected:
diff --git a/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp b/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp
index b175e6f93f3e8f..6468d07b4f4f45 100644
--- a/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp
+++ b/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp
@@ -3505,7 +3505,7 @@ void FunctionStackPoisoner::processStaticAllocas() {
SplitBlockAndInsertIfThenElse(Cmp, Ret, &ThenTerm, &ElseTerm);
IRBuilder<> IRBPoison(ThenTerm);
- if (StackMallocIdx <= 4) {
+ if (ASan.MaxInlinePoisoningSize != 0 && StackMallocIdx <= 4) {
int ClassSize = kMinStackMallocSize << StackMallocIdx;
ShadowAfterReturn.resize(ClassSize / L.Granularity,
kAsanStackUseAfterReturnMagic);
diff --git a/llvm/test/Instrumentation/AddressSanitizer/calls-only-smallfn.ll b/llvm/test/Instrumentation/AddressSanitizer/calls-only-smallfn.ll
new file mode 100644
index 00000000000000..3d67778049430f
--- /dev/null
+++ b/llvm/test/Instrumentation/AddressSanitizer/calls-only-smallfn.ll
@@ -0,0 +1,28 @@
+; RUN: opt < %s -passes=asan -asan-max-inline-poisoning-size=0 -asan-stack-dynamic-alloca=0 -S | FileCheck --check-prefix=OUTLINE %s
+; RUN: opt < %s -passes=asan -asan-max-inline-poisoning-size=999 -asan-stack-dynamic-alloca=0 -S | FileCheck --check-prefix=INLINE %s
+
+target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
+target triple = "arm64-apple-macosx13.0.0"
+
+; Function Attrs: noinline nounwind optnone sanitize_address ssp uwtable(sync)
+define void @foo() #0 {
+entry:
+ %array01 = alloca [1 x i8], align 1
+ %array02 = alloca [2 x i8], align 1
+; OUTLINE: call void @__asan_set_shadow_f1(i64 %23, i64 4)
+; OUTLINE: call void @__asan_set_shadow_01(i64 %24, i64 1)
+; OUTLINE: call void @__asan_set_shadow_f2(i64 %25, i64 1)
+; OUTLINE: call void @__asan_set_shadow_02(i64 %26, i64 1)
+; OUTLINE: call void @__asan_set_shadow_f3(i64 %27, i64 1)
+; OUTLINE: call void @__asan_stack_free_0(i64 %7, i64 64)
+; OUTLINE: call void @__asan_set_shadow_00(i64 %55, i64 8)
+; INLINE: store i64 -935919682371587599, ptr %24, align 1
+; INLINE: store i64 -723401728380766731, ptr %52, align 1
+ %arrayidx = getelementptr inbounds [1 x i8], ptr %array01, i64 0, i64 1
+ store i8 1, ptr %arrayidx, align 1
+ %arrayidx1 = getelementptr inbounds [2 x i8], ptr %array02, i64 0, i64 2
+ store i8 2, ptr %arrayidx1, align 1
+ ret void
+}
+attributes #0 = { noinline nounwind optnone sanitize_address ssp uwtable(sync) "frame-pointer"="non-leaf" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-cpu"="apple-m1" "target-features"="+aes,+crc,+crypto,+dotprod,+fp-armv8,+fp16fml,+fullfp16,+lse,+neon,+ras,+rcpc,+rdm,+sha2,+sha3,+sm4,+v8.1a,+v8.2a,+v8.3a,+v8.4a,+v8.5a,+v8a,+zcm,+zcz" }
+
diff --git a/llvm/test/Instrumentation/AddressSanitizer/calls-only.ll b/llvm/test/Instrumentation/AddressSanitizer/calls-only.ll
index 2cf0070cf08623..fa491105e017d7 100644
--- a/llvm/test/Instrumentation/AddressSanitizer/calls-only.ll
+++ b/llvm/test/Instrumentation/AddressSanitizer/calls-only.ll
@@ -29,8 +29,8 @@ entry:
; OUTLINE: call void @__asan_set_shadow_f2(i64 %45, i64 3)
; OUTLINE: call void @__asan_set_shadow_07(i64 %46, i64 1)
; OUTLINE: call void @__asan_set_shadow_f3(i64 %47, i64 3)
-; OUTLINE: call void @__asan_set_shadow_f5(i64 %134, i64 32)
-; OUTLINE: call void @__asan_set_shadow_00(i64 %140, i64 24)
+; OUTLINE: call void @__asan_stack_free_2(i64 %7, i64 192)
+; OUTLINE: call void @__asan_set_shadow_00(i64 %135, i64 24)
; INLINE: store i64 -1007977276409515535, ptr %34, align 1
; INLINE: store i64 -940423264817843709, ptr %36, align 1
; INLINE: store i64 -868083087686045178, ptr %38, align 1
|
| @@ -3505,7 +3505,7 @@ void FunctionStackPoisoner::processStaticAllocas() { | |||
| SplitBlockAndInsertIfThenElse(Cmp, Ret, &ThenTerm, &ElseTerm); | |||
|
|
|||
| IRBuilder<> IRBPoison(ThenTerm); | |||
| if (StackMallocIdx <= 4) { | |||
| if (ASan.MaxInlinePoisoningSize != 0 && StackMallocIdx <= 4) { | |||
There was a problem hiding this comment.
Should it be rather
int ClassSize = kMinStackMallocSize << StackMallocIdx;
if (ASan.MaxInlinePoisoningSize < ClassSize / L.Granularity && StackMallocIdx <= 4) {
ShadowAfterReturn.resize(ClassSize / L.Granularity,
kAsanStackUseAfterReturnMagic);
There was a problem hiding this comment.
Not sure if I understand - minimum ClassSize is going to be 64 (kMinStackMallocSize << StackMallocIdx), L.Granularity = 8 , so if ASan.MaxInlinePoisoningSize == 0, the whole condition will be true, so inlining will happen. ASan.MaxInlinePoisoningSize == 0 should guarantee not inlining. So I'd rather keep it explicitly the way I had it, unless I misunderstood your suggestion?
There was a problem hiding this comment.
My comment is about the case when MaxInlinePoisoningSize > 0
There is no point in copyToShadow when the frame is larger then MaxInlinePoisoningSize
There was a problem hiding this comment.
Perhaps I used wrong setting for this (MaxInlinePoisoningSize), which caused misunderstanding. I want no inline shadow accesses at all - it is absolutely critical for my scenario. So this optimization StackMallocIdx <= 4) for small frames should be turned off (use code under else) when ClInstrumentationWithCallsThreshold() == 0. Would that work / be more explicit when I change it to be this way?
There was a problem hiding this comment.
Or, ASAN.InstrumentationWithCallsThreshold as it is available in this code today
There was a problem hiding this comment.
Or, ASAN.InstrumentationWithCallsThreshold as it is available in this code today
InstrumentationWithCallsThreshold is for checks
MaxInlinePoisoningSize for poisoning, like the code in question
There was a problem hiding this comment.
This LGTM @wrotki . I see what @vitalybuka is saying but it might not be worth the added risk of the change at this time.
…isabled (llvm#75555) When ASan.MaxInlinePoisoningSize == 0 , it means that no shadow memory operations should be made via inlined instrumentation code, but only via calls to shadow setting functions. This change fixes one violation of this, which happened when the function allocas count was small, i.e. less than 5 - in the code modifying the shadow just before ret instruction. We now explicitly check ASan.MaxInlinePoisoningSize , and if it's 0 then we disallow inlining. It is required for the instrumentation emitting code suitable for handling by ABI implementation. rdar://119513720 Co-authored-by: Mariusz Borsa <m_borsa@apple.com> (cherry picked from commit 7e4ae28)
When ASan.MaxInlinePoisoningSize == 0 , it means that no shadow memory operations should be made via inlined instrumentation code,
but only via calls to shadow setting functions. This change fixes one violation of this, which happened when the function allocas count
was small, i.e. less than 5 - in the code modifying the shadow just before ret instruction.
We now explicitly check ASan.MaxInlinePoisoningSize , and if it's 0 then we disallow inlining. It is required for the instrumentation
emitting code suitable for handling by ABI implementation.
rdar://119513720