compiler-rt: decommit asan memory for unmaps in fuchsia.

[charco]

Summary

This CL allows asan allocator in fuchsia to decommit shadow memory
for memory allocated using mmap.

Big allocations in asan end up being allocated via mmap and freed with
munmap. However, when that memory is freed, asan returns the
corresponding shadow memory back to the OS via a call to
ReleaseMemoryPagesToOs.

In fuchsia, ReleaseMemoryPagesToOs is a no-op: to be able to free
memory back to the OS, you have to hold a handle to the vmo you want to
modify, which is tricky at the ReleaseMemoryPagesToOs level as that
function is not exclusively used for shadow memory.

The function __sanitizer_fill_shadow fills a given shadow memory range
with a specific value, and if that value is 0 (unpoison) and the memory
range is bigger than a threshold parameter, it will decommit that memory
if it is all zeroes.

This CL modifies the FlushUnneededASanShadowMemory function in
asan_poisoning.cpp to add a call to __sanitizer_fill_shadow with
value and threshold = 0. This way, all the unneeded shadow memory gets
returned back to the OS.

A test for this behavior can be found in fxrev.dev/391974

[mcgrathr]

compiler-rt/lib/asan/asan_poisoning.cpp

70

It would more consistent with how most of the OS support is handled to define this function separately in OS-specific files instead. I don't have strong opinions about separate files vs #if in this codebase, but its maintainers might.

If we're going to use #if let's put the other call in the #else branch instead of commenting that it's a no-op.
There's no reason to think it should necessarily be a no-op forever if it has other uses. The reason to call Fuchsia's API here is that Fuchsia offers an optimal and explicit API for doing exactly what needs to be done here, not that the way it's implemented on other systems hasn't been implemented for Fuchsia.

[charco]

Move FlushUnneededASanShadowMemory into asan_fuchsia.cpp

[charco]

compiler-rt/lib/asan/asan_poisoning.cpp

70

Moved the code to asan_fuchsia.cpp.

[vitalybuka]

compiler-rt/lib/asan/asan_fuchsia.cpp

199 ↗

(On Diff #265531)

Can you override empty ReleaseMemoryPagesToOS?

[charco]

compiler-rt/lib/asan/asan_fuchsia.cpp

199 ↗

(On Diff #265531)

In Fuchsia, you can't return memory back to the OS if you don't have the corresponding handle of the object that you are releasing. We only hold the handle to the shadow VMO, so "ReleaseMemoryPagesToOS" would only work with the shadow memory.

If we use sanitizer_fill_shadow in ReleaseMemoryPagesToOs, we would need to go back from asan shadow memory to regular memory so we can call sanitizer_fill_shadow.

[vitalybuka]

compiler-rt/lib/asan/asan_poisoning.cpp

65

Can you do that without #if
like AsanApplyToGlobals

[charco]

Create platform-specific versions of FlushUnneededASanShadowMemory.

[vitalybuka]

do you need empty version in _fuchsia.cpp ?

[charco]

• Move FlushUnneededASanShadowMemory into asan_fuchsia.cpp
• Fork FlushUnneededASanShadowMemory for all platforms.

[charco]

In D80355#2116122, @vitalybuka wrote:

do you need empty version in _fuchsia.cpp ?

I'm not very good at using arcanist/phabricator. I think I screwed up in the last arc diff, but I think I fixed it now.

[charco]

I'm not sure what the harbormaster errors are about, is there a way to see the error logs?

[charco]

Rebase.

[charco]

Rebase

[vitalybuka]

Do you need someone to land it for you?

[charco]

In D80355#2148419, @vitalybuka wrote:

Do you need someone to land it for you?

I have access, I'm trying to understand why it is failing the build... Harbormaster says that it is failing to apply the patch, even though I rebased it recently.

[charco]

Rebase... Third try.

[charco]

Rebase.

[vitalybuka]

protoent.cpp?
it's definitely unrelated

[charco]

• Move FlushUnneededASanShadowMemory into asan_fuchsia.cpp
• Fork FlushUnneededASanShadowMemory for all platforms.

compiler-rt/lib/asan/asan_linux.cpp

#endif
 
   return shadow_start;
 }
 
 void AsanApplyToGlobals(globals_op_fptr op, const void *needle) {
   UNIMPLEMENTED();
 }
 
+void FlushUnneededASanShadowMemory(uptr p, uptr size) {
+  // Since asan's mapping is compacting, the shadow chunk may be
+  // not page-aligned, so we only flush the page-aligned portion.
+  ReleaseMemoryPagesToOS(MemToShadow(p), MemToShadow(p + size));
+}
+
 #if SANITIZER_ANDROID
 // FIXME: should we do anything for Android?
 void AsanCheckDynamicRTPrereqs() {}
 void AsanCheckIncompatibleRT() {}
 #else
 static int FindFirstDSOCallback(struct dl_phdr_info *info, size_t size,
                                 void *data) {
   VReport(2, "info->dlpi_name = %s\tinfo->dlpi_addr = %p\n",

compiler-rt/lib/asan/asan_mac.cpp

__asan_global *globals = (__asan_global *)getsectiondata(
       "__DATA", "__asan_globals",
       &size);
 
   if (!globals) return;
   if (size % sizeof(__asan_global) != 0) return;
   op(globals, size / sizeof(__asan_global));
 }
 
+void FlushUnneededASanShadowMemory(uptr p, uptr size) {
+  // Since asan's mapping is compacting, the shadow chunk may be
+  // not page-aligned, so we only flush the page-aligned portion.
+  ReleaseMemoryPagesToOS(MemToShadow(p), MemToShadow(p + size));
+}
+
 void ReadContextStack(void *context, uptr *stack, uptr *ssize) {
   UNIMPLEMENTED();
 }
 
 // Support for the following functions from libdispatch on Mac OS:
 //   dispatch_async_f()
 //   dispatch_async()
 //   dispatch_sync_f()

compiler-rt/lib/asan/asan_poisoning.cpp

struct ShadowSegmentEndpoint {
 
   explicit ShadowSegmentEndpoint(uptr address) {
     chunk = (u8*)MemToShadow(address);
     offset = address & (SHADOW_GRANULARITY - 1);
     value = *chunk;
   }
 };
 
-#if !SANITIZER_FUCHSIA
-void FlushUnneededASanShadowMemory(uptr p, uptr size) {
-  // Since asan's mapping is compacting, the shadow chunk may be
-  // not page-aligned, so we only flush the page-aligned portion.
-  ReleaseMemoryPagesToOS(MemToShadow(p), MemToShadow(p + size));
-}
-#endif  // !SANITIZER_FUCHSIA
-
 void AsanPoisonOrUnpoisonIntraObjectRedzone(uptr ptr, uptr size, bool poison) {
   uptr end = ptr + size;
   if (Verbosity()) {
     Printf("__asan_%spoison_intra_object_redzone [%p,%p) %zd\n",
            poison ? "" : "un", ptr, end, size);
     if (Verbosity() >= 2)
       PRINT_CURRENT_STACK();
   }
   CHECK(size);
   CHECK_LE(size, 4096);
   CHECK(IsAligned(end, SHADOW_GRANULARITY));
   if (!IsAligned(ptr, SHADOW_GRANULARITY)) {
     *(u8 *)MemToShadow(ptr) =
         poison ? static_cast<u8>(ptr % SHADOW_GRANULARITY) : 0;

compiler-rt/lib/asan/asan_rtems.cpp

void InitializeShadowMemory() {
 
   ResetShadowMemory();
 }
 
 void AsanApplyToGlobals(globals_op_fptr op, const void *needle) {
   UNIMPLEMENTED();
 }
 
+void FlushUnneededASanShadowMemory(uptr p, uptr size) {
+  // Since asan's mapping is compacting, the shadow chunk may be
+  // not page-aligned, so we only flush the page-aligned portion.
+  ReleaseMemoryPagesToOS(MemToShadow(p), MemToShadow(p + size));
+}
+
 void AsanCheckDynamicRTPrereqs() {}
 void AsanCheckIncompatibleRT() {}
 void InitializeAsanInterceptors() {}
 void InitializePlatformInterceptors() {}
 void InitializePlatformExceptionHandlers() {}
 
 // RTEMS only support static linking; it sufficies to return with no
 // error.

compiler-rt/lib/asan/asan_win.cpp

CHECK(::__interception::OverrideFunction("RtlRaiseException",
                                              (uptr *)&REAL(RtlRaiseException)));
   }
 }
 
 void AsanApplyToGlobals(globals_op_fptr op, const void *needle) {
   UNIMPLEMENTED();
 }
 
+void FlushUnneededASanShadowMemory(uptr p, uptr size) {
+  // Since asan's mapping is compacting, the shadow chunk may be
+  // not page-aligned, so we only flush the page-aligned portion.
+  ReleaseMemoryPagesToOS(MemToShadow(p), MemToShadow(p + size));
+}
+
 // ---------------------- TSD ---------------- {{{
 static bool tsd_key_inited = false;
 
 static __declspec(thread) void *fake_tsd = 0;
 
 // https://docs.microsoft.com/en-us/windows/desktop/api/winternl/ns-winternl-_teb
 // "[This structure may be altered in future versions of Windows. Applications
 // should use the alternate functions listed in this topic.]"