Adding the VerifyModule function (and helpers) to safe browsing.

chrome/browser/safe_browsing/module_integrity_verifier.cc

Index: chrome/browser/safe_browsing/module_integrity_verifier.cc
diff --git a/chrome/browser/safe_browsing/module_integrity_verifier.cc b/chrome/browser/safe_browsing/module_integrity_verifier.cc
new file mode 100644
index 0000000000000000000000000000000000000000..a144f2c03bc352a202261903c84a0cc7cfccbfc3
--- /dev/null
+++ b/chrome/browser/safe_browsing/module_integrity_verifier.cc
@@ -0,0 +1,175 @@
+// Copyright 2014 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "chrome/browser/safe_browsing/module_integrity_verifier.h"
+
+#include <windows.h>
+#include <psapi.h>
+
+#include "base/files/file_path.h"
+#include "base/files/memory_mapped_file.h"
+#include "base/scoped_native_library.h"
+#include "base/win/pe_image.h"
+
+namespace safe_browsing {
+
+namespace {
+
+// The number of bytes made to agree (between memory and disk) when the
+// relocations are accounted for.
+int bytes_corrected_by_reloc = 0;

[grt (UTC plus 2), 2014/07/29 19:32:22]

instead of using global state (which makes concurrent verifications impossible),
please define a struct for this data and create it on the stack in VerifyModule
and pass a pointer to it via |cookie|. for example:

struct RelocEnumerationState {
  explicit RelocEnumerationState(HMODULE hModule);
  ~RelocEnumerationState();

  base::win::PEImageAsData disk_peimage;
  int bytes_corrected_by_reloc;
  bool unknown_reloc_type;

  DISALLOW_COPY_AND_ASSIGN(RelocEnumerationState);
};

RelocEnumerationState::RelocEnumerationState(HMODULE hModule)
    : disk_peimage(hModule),
      bytes_corrected_by_reloc(),
      unknown_reloc_type() {
}

RelocEnumerationState::~RelocEnumerationState() {
}

and then instantiate this on line 147 and pass a pointer to it through the call
to EnumRelocs.

[krstnmnlsn, 2014/07/30 14:34:35]

Okay yes, I was asking earlier if I should do this.  Done.

+
+// This is true if the relocation table contains a reloc of type we don't
+// currently handle.
+bool unknown_reloc_type = false;
+
+int CountBytesDiffInMemory(BYTE* disk_code_start,
+                           BYTE* memory_code_start,
+                           SIZE_T code_size) {
+  int counter = 0;
+  for (int i = 0; i < static_cast<int>(code_size); ++i) {
+    if (*(disk_code_start + i) != *(memory_code_start + i))
+      ++counter;
+  }
+  return counter;
+}
+
+bool AddrIsInCodeSection(void* address, BYTE* code_addr, SIZE_T code_size) {
+  return (code_addr <= address && address < code_addr + code_size);
+}
+
+bool EnumRelocsCallback(const base::win::PEImage& mem_peimage,
+                        WORD type,
+                        void* address,
+                        void* cookie) {
+  base::win::PEImageAsData disk_peimage =
+      *reinterpret_cast<base::win::PEImageAsData*>(cookie);
+  BYTE* mem_code_addr = NULL;
+  BYTE* disk_code_addr = NULL;
+  SIZE_T code_size = 0;
+  if (!GetCodeAddrsAndSize(mem_peimage, disk_peimage,
+                           &mem_code_addr, &disk_code_addr, &code_size))
+    return false;
+
+  switch (type) {
+    case IMAGE_REL_BASED_HIGHLOW: {
+      // If not in the code section return true to continue to the next reloc.
+      if (!AddrIsInCodeSection(address, mem_code_addr, code_size))
+        return true;
+
+      BYTE* preferred_image_base = reinterpret_cast<BYTE*>(
+          disk_peimage.GetNTHeaders()->OptionalHeader.ImageBase);
+      intptr_t delta =
+          preferred_image_base - reinterpret_cast<BYTE*>(mem_peimage.module());
+      BYTE* new_value = (*reinterpret_cast<BYTE**>(address)) + delta;
+      int bytes_corrected =
+          CountBytesDiffInPtr(reinterpret_cast<intptr_t>(new_value),
+                              *reinterpret_cast<intptr_t*>(address));
+
+      // Check that the adding delta did correct the value to agree with disk.
+      BYTE** disk_address = reinterpret_cast<BYTE**>(
+          reinterpret_cast<BYTE*>(address) - mem_code_addr + disk_code_addr);
+      if (new_value == *disk_address) {
+        bytes_corrected_by_reloc += bytes_corrected;
+      }
+      break;
+    }
+    case IMAGE_REL_BASED_ABSOLUTE:
+      break;
+    default: {
+      // TODO(krstnmnlsn): Find a reliable description of the behaviour of the
+      // remaining types of relocation and handle them.
+      unknown_reloc_type = true;
+      break;
+    }
+  }
+  return true;
+}
+
+}  // namespace
+
+bool GetCodeAddrsAndSize(const base::win::PEImage& mem_peimage,
+                         const base::win::PEImageAsData& disk_peimage,
+                         BYTE** mem_code_addr,
+                         BYTE** disk_code_addr,
+                         SIZE_T* code_size) {
+  DWORD base_of_code = mem_peimage.GetNTHeaders()->OptionalHeader.BaseOfCode;
+
+  // Get the address and size of the code section in the loaded module image.
+  PIMAGE_SECTION_HEADER mem_code_header =
+      mem_peimage.GetImageSectionFromAddr(mem_peimage.RVAToAddr(base_of_code));
+  if (mem_code_header == NULL)
+    return false;
+  *mem_code_addr = reinterpret_cast<BYTE*>(
+      mem_peimage.RVAToAddr(mem_code_header->VirtualAddress));
+  *code_size = mem_code_header->Misc.VirtualSize;
+
+  // Get the address of the code section in the module mapped as data from disk.
+  // Section size will be the same as in the loaded module.

[grt (UTC plus 2), 2014/07/29 19:32:22]

is this necessarily true? VirtualSize (size of the section when mapped into
memory) may be greater than or less than SizeOfRawData. in the former case
(VirtualSize is bigger), the section is zero padded when mapped. in the latter
case (VirtualSize is smaller), SizeOfRawData has been rounded up to the nearest
OptionalHeader.FileAlignment boundary.

[krstnmnlsn, 2014/07/30 14:34:35]

Ah, I misunderstood that.  It should be safe/sufficient to take the minimum of
the two then yes?

[grt (UTC plus 2), 2014/07/31 17:03:33]

I believe so, yes.

[krstnmnlsn, 2014/08/04 15:18:14]

Done.

+  DWORD disk_code_offset = 0;
+  if (!mem_peimage.ImageAddrToOnDiskOffset(
+          reinterpret_cast<void*>(*mem_code_addr), &disk_code_offset))
+    return false;
+  *disk_code_addr =
+      reinterpret_cast<BYTE*>(disk_peimage.module()) + disk_code_offset;
+  return true;
+}
+
+int CountBytesDiffInPtr(intptr_t num_a, intptr_t num_b) {
+  int num_bytes = 0;
+  for (int i = 0; i < sizeof(num_a); ++i) {
+    if ((num_a & 0xFF) != (num_b & 0xFF))
+      ++num_bytes;
+    num_a >>= 8;
+    num_b >>= 8;
+  }
+  return num_bytes;
+}
+
+ModuleState VerifyModule(const wchar_t* module_name) {
+  bytes_corrected_by_reloc = 0;
+  unknown_reloc_type = false;
+
+  HMODULE module_handle = NULL;
+  if (!GetModuleHandleEx(0, module_name, &module_handle))
+    return MODULE_STATE_UNKNOWN;
+  base::ScopedNativeLibrary native_library(module_handle);
+
+  WCHAR module_path[MAX_PATH] = {};
+  DWORD length =
+      GetModuleFileName(module_handle, module_path, arraysize(module_path));
+  if (length == arraysize(module_path))

[grt (UTC plus 2), 2014/07/29 19:32:22]

this function returns 0 on failure, so:
  if (!length || length == arraysize(module_path))

[krstnmnlsn, 2014/07/30 14:34:35]

Done.

+    return MODULE_STATE_UNKNOWN;
+
+  base::MemoryMappedFile mapped_module;
+  mapped_module.Initialize(base::FilePath(module_path));

[grt (UTC plus 2), 2014/07/29 19:32:22]

returns false if the file cannot be mapped

[krstnmnlsn, 2014/07/30 14:34:35]

Done.

+  base::win::PEImageAsData disk_peimage(
+      reinterpret_cast<HMODULE>(const_cast<uint8*>(mapped_module.data())));
+
+  base::win::PEImage mem_peimage(module_handle);
+  if (!mem_peimage.VerifyMagic() || !disk_peimage.VerifyMagic())
+    return MODULE_STATE_UNKNOWN;
+
+  BYTE* mem_code_addr = NULL;
+  BYTE* disk_code_addr = NULL;
+  SIZE_T code_size = 0;
+  if (!GetCodeAddrsAndSize(mem_peimage,
+                           disk_peimage,
+                           &mem_code_addr,
+                           &disk_code_addr,
+                           &code_size))
+    return MODULE_STATE_UNKNOWN;
+
+  int num_bytes_different =
+      CountBytesDiffInMemory(disk_code_addr, mem_code_addr, code_size);
+  mem_peimage.EnumRelocs(EnumRelocsCallback, &disk_peimage);
+  if (unknown_reloc_type)
+    return MODULE_STATE_UNKNOWN;
+
+  if (num_bytes_different == bytes_corrected_by_reloc)
+    return MODULE_STATE_UNMODIFIED;
+  return MODULE_STATE_MODIFIED;
+}
+
+}  // namespace safe_browsing