[scudo] Check if MADV_DONTNEED zeroes memory

[vitalybuka]

Summary

QEMU just ignores MADV_DONTNEED
https://github.com/qemu/qemu/blob/b1cffefa1b163bce9aebc3416f562c1d3886eeaa/linux-user/syscall.c#L11941

Depends on D100998.

[vitalybuka]

remove unneeded include

[vitalybuka]

reorder includes

[vitalybuka]

comment

[cryptoad]

Thank you for finding this!
LGTM with a couple of nits

compiler-rt/lib/scudo/standalone/linux.cpp

95

getPageSizeCached() maybe

113

Maybe for clarity some enum to have descriptive names associated to 0/1/2

[vitalybuka]

add enum

[vitalybuka]

getPageSizeCached

[vitalybuka]

compiler-rt/lib/scudo/standalone/linux.cpp

123

I believe in most cases of NoFill it's not needed.
However I could not tell if anything internal, like zeroing header, will not be broken without memset. I don't see issues other then FillContentsMode tests.

Having that we don't expect this branch running without QEMU, I guess it's fine to keep as is.

[pcc]

Have you reported this bug to QEMU? Unfortunately the comment you linked to is inaccurate about madvise being a hint on Linux [1].

[1] https://www.youtube.com/watch?v=bg6-LVCHmGM&t=3548s

[vitalybuka]

In D101031#2710494, @pcc wrote:

Have you reported this bug to QEMU? Unfortunately the comment you linked to is inaccurate about madvise being a hint on Linux [1].

[1] https://www.youtube.com/watch?v=bg6-LVCHmGM&t=3548s

I agree that this does not match strange but documented behavior.
Unfortunately there is a lot of comments about that online but no new/closed bug on the QEMU tracker.
Created one https://bugs.launchpad.net/qemu/+bug/1926521

compiler-rt/lib/scudo/standalone/linux.cpp

 //===-- linux.cpp -----------------------------------------------*- C++ -*-===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
+#include "atomic_helpers.h"
 #include "platform.h"
 
 #if SCUDO_LINUX
 
 #include "common.h"
 #include "linux.h"
 #include "mutex.h"
 #include "string_utils.h"
 
 #include <errno.h>
 #include <fcntl.h>
 #include <linux/futex.h>
 #include <sched.h>
+#include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <sys/mman.h>
 #include <sys/stat.h>
 #include <sys/syscall.h>
 #include <sys/time.h>
 #include <time.h>
 #include <unistd.h>

 
 void setMemoryPermission(uptr Addr, uptr Size, uptr Flags,
                          UNUSED MapPlatformData *Data) {
   int Prot = (Flags & MAP_NOACCESS) ? PROT_NONE : (PROT_READ | PROT_WRITE);
   if (mprotect(reinterpret_cast<void *>(Addr), Size, Prot) != 0)
     dieOnMapUnmapError();
 }
 
+static bool madviseNeedsMemset() {
+  uptr Size = getPageSize();
+  char *P = (char *)mmap(0, Size, PROT_READ | PROT_WRITE,
+                         MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+  if (!P)
+    dieOnMapUnmapError(errno == ENOMEM);
+  *P = -1;
+  while (madvise(P, Size, MADV_DONTNEED) == -1 && errno == EAGAIN) {
+  }
+  bool R = (*P != 0);
+  if (munmap(P, Size) != 0)
+    dieOnMapUnmapError();
+  return R;
+}
+
+static bool madviseNeedsMemsetCached() {
+  static atomic_u8 NeedsMemset;
+  u8 Z = atomic_load_relaxed(&NeedsMemset);
+  if (!Z) {
+    Z = madviseNeedsMemset() ? 1 : 2;
+    atomic_store_relaxed(&NeedsMemset, Z);
+  }
+  return Z == 1;
+}
+
 void releasePagesToOS(uptr BaseAddress, uptr Offset, uptr Size,
                       UNUSED MapPlatformData *Data) {
   void *Addr = reinterpret_cast<void *>(BaseAddress + Offset);
+  if (madviseNeedsMemsetCached()) {
+    // Workaround for QEMU-user which ignores MADV_DONTNEED.
+    memset(Addr, 0, Size);
+  }
   while (madvise(Addr, Size, MADV_DONTNEED) == -1 && errno == EAGAIN) {
   }
 }
 
 // Calling getenv should be fine (c)(tm) at any time.
 const char *getEnv(const char *Name) { return getenv(Name); }
 
 namespace {