[COFF] Make the DOS stub a real DOS program

[hans]

Summary

It only adds a few bytes and is nice for backward compatibility :-)

[hans]

COFF/Writer.cpp

45

Grateful for ideas on how to format/do this better.

700

Is there a guarantee that the buffer is zero-initialized? Otherwise maybe we should memset DOS here, because it's important that the un-set fields are really zero if this is going to be loaded.

[ruiu]

COFF/Writer.cpp

64

Maybe it's easier to make it always 64-byte long and use sizeof(DOSProgram) instead of alignTo()

static unsigned char DOSProgram[64] = { ... };

692–696

It's perhaps better to expand this comment a bit so that the thing we are doing here is not important to understand the linker itself -- it's just a small DOS program at beginning of every PE/COFF executable to help DOS 2.0 (!) users that the program needs Windows.

700

The buffer is a new mmap'ed file, so it is guaranteed to be zero-initialized.

(The situation is different for ELF executable sections because we write INT3 or equivalent before writing anything to that regions, but that's not relevant here because this is COFF header.)

[hans]

COFF/Writer.cpp

64

That seems wasteful since the program doesn't need 64 bytes (and it seems fragile in case it would ever change and require more than 64 bytes).

A better idea I think would be to put an

align   8, db 0

directive in the asm above to pad the program itself to a multiple of 8. That way we're not wasting bytes, it won't break if someone changes the program, and it simplifies the lld code a little. What do you think?

[ruiu]

LGTM

[hans]

In D44018#1030009, @ruiu wrote:

LGTM

Thanks! I'll make the change and commit.

COFF/Writer.cpp

692–696

I've expanded the comment to explain.

COFF/Writer.cpp

 using namespace llvm;
 using namespace llvm::COFF;
 using namespace llvm::object;
 using namespace llvm::support;
 using namespace llvm::support::endian;
 using namespace lld;
 using namespace lld::coff;
 
+/* To re-generate DOSProgram:
+$ cat > /tmp/DOSProgram.asm
+org 0
+        ; Copy cs to ds.
+        push cs
+        pop ds
+        ; Point ds:dx at the $-terminated string.
+        mov dx, str
+        ; Int 21/AH=09h: Write string to standard output.
+        mov ah, 0x9
+        int 0x21
+        ; Int 21/AH=4Ch: Exit with return code (in AL).
+        mov ax, 0x4C01
+        int 0x21
+str:
+        db 'This program cannot be run in DOS mode.$'
+align 8, db 0
+$ nasm -fbin /tmp/DOSProgram.asm -o /tmp/DOSProgram.bin
+$ xxd -i /tmp/DOSProgram.bin
+*/
+static unsigned char DOSProgram[] = {
+  0x0e, 0x1f, 0xba, 0x0e, 0x00, 0xb4, 0x09, 0xcd, 0x21, 0xb8, 0x01, 0x4c,
+  0xcd, 0x21, 0x54, 0x68, 0x69, 0x73, 0x20, 0x70, 0x72, 0x6f, 0x67, 0x72,
+  0x61, 0x6d, 0x20, 0x63, 0x61, 0x6e, 0x6e, 0x6f, 0x74, 0x20, 0x62, 0x65,
+  0x20, 0x72, 0x75, 0x6e, 0x20, 0x69, 0x6e, 0x20, 0x44, 0x4f, 0x53, 0x20,
+  0x6d, 0x6f, 0x64, 0x65, 0x2e, 0x24, 0x00, 0x00
+};
+static_assert(sizeof(DOSProgram) % 8 == 0,
+              "DOSProgram size must be multiple of 8");
+
 static const int SectorSize = 512;
-static const int DOSStubSize = 64;
+static const int DOSStubSize = sizeof(dos_header) + sizeof(DOSProgram);
+static_assert(DOSStubSize % 8 == 0, "DOSStub size must be multiple of 8");
+
 static const int NumberfOfDataDirectory = 16;
 
 namespace {
 
 class DebugDirectoryChunk : public Chunk {
 public:
   DebugDirectoryChunk(const std::vector<Chunk *> &R) : Records(R) {}
 
for (OutputSection *Sec : OutputSections) {
     Sec->setFileOffset(FileSize);
     RVA += alignTo(Sec->getVirtualSize(), PageSize);
     FileSize += alignTo(Sec->getRawSize(), SectorSize);
   }
   SizeOfImage = alignTo(RVA, PageSize);
 }
 
 template <typename PEHeaderTy> void Writer::writeHeader() {
-  // Write DOS stub
+  // Write DOS header. For backwards compatibility, the first part of a PE/COFF
+  // executable consists of an MS-DOS MZ executable. If the executable is run
+  // under DOS, that program gets run (usually to just print an error message).
+  // When run under Windows, the loader looks at AddressOfNewExeHeader and uses
+  // the PE header instead.
   uint8_t *Buf = Buffer->getBufferStart();
   auto *DOS = reinterpret_cast<dos_header *>(Buf);
-  Buf += DOSStubSize;
+  Buf += sizeof(dos_header);
   DOS->Magic[0] = 'M';
   DOS->Magic[1] = 'Z';
+  DOS->UsedBytesInTheLastPage = DOSStubSize % 512;
+  DOS->FileSizeInPages = divideCeil(DOSStubSize, 512);
+  DOS->HeaderSizeInParagraphs = sizeof(dos_header) / 16;
+
   DOS->AddressOfRelocationTable = sizeof(dos_header);
   DOS->AddressOfNewExeHeader = DOSStubSize;
 
+  // Write DOS program.
+  memcpy(Buf, DOSProgram, sizeof(DOSProgram));
+  Buf += sizeof(DOSProgram);
+
   // Write PE magic
   memcpy(Buf, PEMagic, sizeof(PEMagic));
   Buf += sizeof(PEMagic);
 
   // Write COFF header
   auto *COFF = reinterpret_cast<coff_file_header *>(Buf);
   Buf += sizeof(*COFF);
   COFF->Machine = Config->Machine;

test/COFF/hello32.test

 HEADER-NEXT:   FileAlignment: 512
 HEADER-NEXT:   MajorOperatingSystemVersion: 6
 HEADER-NEXT:   MinorOperatingSystemVersion: 0
 HEADER-NEXT:   MajorImageVersion: 0
 HEADER-NEXT:   MinorImageVersion: 0
 HEADER-NEXT:   MajorSubsystemVersion: 6
 HEADER-NEXT:   MinorSubsystemVersion: 0
 HEADER-NEXT:   SizeOfImage: 20480
-HEADER-NEXT:   SizeOfHeaders: 512
+HEADER-NEXT:   SizeOfHeaders: 1024
 HEADER-NEXT:   Subsystem: IMAGE_SUBSYSTEM_WINDOWS_CUI (0x3)
 HEADER-NEXT:   Characteristics [ (0x9540)
 HEADER-NEXT:     IMAGE_DLL_CHARACTERISTICS_APPCONTAINER (0x1000)
 HEADER-NEXT:     IMAGE_DLL_CHARACTERISTICS_DYNAMIC_BASE (0x40)
 HEADER-NEXT:     IMAGE_DLL_CHARACTERISTICS_NO_SEH (0x400)
 HEADER-NEXT:     IMAGE_DLL_CHARACTERISTICS_NX_COMPAT (0x100)
 HEADER-NEXT:     IMAGE_DLL_CHARACTERISTICS_TERMINAL_SERVER_AWARE (0x8000)
 HEADER-NEXT:   ]

 HEADER-NEXT:     CLRRuntimeHeaderRVA: 0x0
 HEADER-NEXT:     CLRRuntimeHeaderSize: 0x0
 HEADER-NEXT:     ReservedRVA: 0x0
 HEADER-NEXT:     ReservedSize: 0x0
 HEADER-NEXT:   }
 HEADER-NEXT: }
 HEADER-NEXT: DOSHeader {
 HEADER-NEXT:   Magic: MZ
-HEADER-NEXT:   UsedBytesInTheLastPage: 0
-HEADER-NEXT:   FileSizeInPages: 0
+HEADER-NEXT:   UsedBytesInTheLastPage: 120
+HEADER-NEXT:   FileSizeInPages: 1
 HEADER-NEXT:   NumberOfRelocationItems: 0
-HEADER-NEXT:   HeaderSizeInParagraphs: 0
+HEADER-NEXT:   HeaderSizeInParagraphs: 4
 HEADER-NEXT:   MinimumExtraParagraphs: 0
 HEADER-NEXT:   MaximumExtraParagraphs: 0
 HEADER-NEXT:   InitialRelativeSS: 0
 HEADER-NEXT:   InitialSP: 0
 HEADER-NEXT:   Checksum: 0
 HEADER-NEXT:   InitialIP: 0
 HEADER-NEXT:   InitialRelativeCS: 0
 HEADER-NEXT:   AddressOfRelocationTable: 64
 HEADER-NEXT:   OverlayNumber: 0
 HEADER-NEXT:   OEMid: 0
 HEADER-NEXT:   OEMinfo: 0
-HEADER-NEXT:   AddressOfNewExeHeader: 64
+HEADER-NEXT:   AddressOfNewExeHeader: 120
 HEADER-NEXT: }
 
 IMPORTS: Format: COFF-i386
 IMPORTS: Arch: i386
 IMPORTS: AddressSize: 32bit
 IMPORTS: Import {
 IMPORTS:   Name: std32.dll
 IMPORTS:   ImportLookupTableRVA: 0x3028