cmd/link: handle grouped resource sections

The Go PE linker does not support enough generalized PE logic to
properly handle .rsrc sections gracefully. Instead a few things are
special cased for these. The linker also does not support PE's "grouped
sections" features, in which input objects have several named sections
that are sorted, merged, and renamed in the output file. In the past,
more sophisticated support for resources or for PE features like grouped
sections have not been necessary, as Go's own object formats are pretty
vanilla, and GNU binutils also produces pretty vanilla objects where all
sections are already merged.

However, GNU binutils is lagging with arm support, and here LLVM has
picked up the slack. In particular, LLVM has its own rc/cvtres combo,
which are glued together in mingw LLVM distributions as windres, a
command line compatible tool with binutils' windres, which supports arm
and arm64. But there's a key difference between binutils' windres and
LLVM's windres: the LLVM one uses proper grouped sections.

So, this commit adds grouped sections support for resource sections to
the linker. We don't attempt to plumb generic support for grouped
sections, just as there isn't generic support already for what resources
require. Instead we augment the resource handling logic to deal with
standard two-section resource objects.

We also add a test for this, akin to the current test for more vanilla
binutils resource objects, and make sure that the rsrc tests are always
performed.

Fixes #42866.
Fixes #43182.

Change-Id: I059450021405cdf2ef1c195ddbab3960764ad711
Reviewed-on: https://go-review.googlesource.com/c/go/+/268337
Run-TryBot: Jason A. Donenfeld <Jason@zx2c4.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Trust: Alex Brainman <alex.brainman@gmail.com>
Trust: Jason A. Donenfeld <Jason@zx2c4.com>

src/cmd/link/internal/ld/lib.go

@@ -1820,7 +1820,7 @@ func ldobj(ctxt *Link, f *bio.Reader, lib *sym.Library, length int64, pn string,
 				Errorf(nil, "%v", err)
 				return
 			}
-			if rsrc != 0 {
+			if len(rsrc) != 0 {
 				setpersrc(ctxt, rsrc)
 			}
 			ctxt.Textp = append(ctxt.Textp, textp...)

src/cmd/link/internal/ld/pe.go

@@ -253,7 +253,7 @@ type Dll struct {
 }
 
 var (
-	rsrcsym     loader.Sym
+	rsrcsyms    []loader.Sym
 	PESECTHEADR int32
 	PEFILEHEADR int32
 	pe64        int
@@ -1508,46 +1508,56 @@ func (ctxt *Link) dope() {
 	initdynexport(ctxt)
 }
 
-func setpersrc(ctxt *Link, sym loader.Sym) {
-	if rsrcsym != 0 {
+func setpersrc(ctxt *Link, syms []loader.Sym) {
+	if len(rsrcsyms) != 0 {
 		Errorf(nil, "too many .rsrc sections")
 	}
-
-	rsrcsym = sym
+	rsrcsyms = syms
 }
 
 func addpersrc(ctxt *Link) {
-	if rsrcsym == 0 {
+	if len(rsrcsyms) == 0 {
 		return
 	}
 
-	data := ctxt.loader.Data(rsrcsym)
-	size := len(data)
-	h := pefile.addSection(".rsrc", size, size)
+	var size int64
+	for _, rsrcsym := range rsrcsyms {
+		size += ctxt.loader.SymSize(rsrcsym)
+	}
+	h := pefile.addSection(".rsrc", int(size), int(size))
 	h.characteristics = IMAGE_SCN_MEM_READ | IMAGE_SCN_CNT_INITIALIZED_DATA
 	h.checkOffset(ctxt.Out.Offset())
 
-	// relocation
+	for _, rsrcsym := range rsrcsyms {
+		// A split resource happens when the actual resource data and its relocations are
+		// split across multiple sections, denoted by a $01 or $02 at the end of the .rsrc
+		// section name.
+		splitResources := strings.Contains(ctxt.loader.SymName(rsrcsym), ".rsrc$")
 		relocs := ctxt.loader.Relocs(rsrcsym)
-	for i := 0; i < relocs.Count(); i++ {
-		r := relocs.At(i)
+		data := ctxt.loader.Data(rsrcsym)
+		for ri := 0; ri < relocs.Count(); ri++ {
+			r := relocs.At(ri)
 			p := data[r.Off():]
 			val := uint32(int64(h.virtualAddress) + r.Add())
-
-		// 32-bit little-endian
-		p[0] = byte(val)
-
-		p[1] = byte(val >> 8)
-		p[2] = byte(val >> 16)
-		p[3] = byte(val >> 24)
+			if splitResources {
+				// If we're a split resource section, and that section has relocation
+				// symbols, then the data that it points to doesn't actually begin at
+				// the virtual address listed in this current section, but rather
+				// begins at the section immediately after this one. So, in order to
+				// calculate the proper virtual address of the data it's pointing to,
+				// we have to add the length of this section to the virtual address.
+				// This works because .rsrc sections are divided into two (but not more)
+				// of these sections.
+				val += uint32(len(data))
+			}
+			binary.LittleEndian.PutUint32(p, val)
 		}
-
 		ctxt.Out.Write(data)
+	}
 	h.pad(ctxt.Out, uint32(size))
 
 	// update data directory
 	pefile.dataDirectory[pe.IMAGE_DIRECTORY_ENTRY_RESOURCE].VirtualAddress = h.virtualAddress
-
 	pefile.dataDirectory[pe.IMAGE_DIRECTORY_ENTRY_RESOURCE].Size = h.virtualSize
 }
 

src/cmd/link/internal/loadpe/ldpe.go

@@ -157,8 +157,9 @@ func makeUpdater(l *loader.Loader, bld *loader.SymbolBuilder, s loader.Sym) *loa
 
 // Load loads the PE file pn from input.
 // Symbols are written into syms, and a slice of the text symbols is returned.
-// If an .rsrc section is found, its symbol is returned as rsrc.
-func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Reader, pkg string, length int64, pn string) (textp []loader.Sym, rsrc loader.Sym, err error) {
+// If an .rsrc section or set of .rsrc$xx sections is found, its symbols are
+// returned as rsrc.
+func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Reader, pkg string, length int64, pn string) (textp []loader.Sym, rsrc []loader.Sym, err error) {
 	lookup := func(name string, version int) (*loader.SymbolBuilder, loader.Sym) {
 		s := l.LookupOrCreateSym(name, version)
 		sb := l.MakeSymbolUpdater(s)
@@ -176,7 +177,7 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 	// TODO: replace pe.NewFile with pe.Load (grep for "add Load function" in debug/pe for details)
 	f, err := pe.NewFile(sr)
 	if err != nil {
-		return nil, 0, err
+		return nil, nil, err
 	}
 	defer f.Close()
 
@@ -211,21 +212,21 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 			bld.SetType(sym.STEXT)
 
 		default:
-			return nil, 0, fmt.Errorf("unexpected flags %#06x for PE section %s", sect.Characteristics, sect.Name)
+			return nil, nil, fmt.Errorf("unexpected flags %#06x for PE section %s", sect.Characteristics, sect.Name)
 		}
 
 		if bld.Type() != sym.SNOPTRBSS {
 			data, err := sect.Data()
 			if err != nil {
-				return nil, 0, err
+				return nil, nil, err
 			}
 			sectdata[sect] = data
 			bld.SetData(data)
 		}
 		bld.SetSize(int64(sect.Size))
 		sectsyms[sect] = s
-		if sect.Name == ".rsrc" {
-			rsrc = s
+		if sect.Name == ".rsrc" || strings.HasPrefix(sect.Name, ".rsrc$") {
+			rsrc = append(rsrc, s)
 		}
 	}
 
@@ -246,22 +247,23 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 			continue
 		}
 
+		splitResources := strings.HasPrefix(rsect.Name, ".rsrc$")
 		sb := l.MakeSymbolUpdater(sectsyms[rsect])
 		for j, r := range rsect.Relocs {
 			if int(r.SymbolTableIndex) >= len(f.COFFSymbols) {
-				return nil, 0, fmt.Errorf("relocation number %d symbol index idx=%d cannot be large then number of symbols %d", j, r.SymbolTableIndex, len(f.COFFSymbols))
+				return nil, nil, fmt.Errorf("relocation number %d symbol index idx=%d cannot be large then number of symbols %d", j, r.SymbolTableIndex, len(f.COFFSymbols))
 			}
 			pesym := &f.COFFSymbols[r.SymbolTableIndex]
 			_, gosym, err := readpesym(l, arch, l.LookupOrCreateSym, f, pesym, sectsyms, localSymVersion)
 			if err != nil {
-				return nil, 0, err
+				return nil, nil, err
 			}
 			if gosym == 0 {
 				name, err := pesym.FullName(f.StringTable)
 				if err != nil {
 					name = string(pesym.Name[:])
 				}
-				return nil, 0, fmt.Errorf("reloc of invalid sym %s idx=%d type=%d", name, r.SymbolTableIndex, pesym.Type)
+				return nil, nil, fmt.Errorf("reloc of invalid sym %s idx=%d type=%d", name, r.SymbolTableIndex, pesym.Type)
 			}
 
 			rSym := gosym
@@ -271,11 +273,11 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 			var rType objabi.RelocType
 			switch arch.Family {
 			default:
-				return nil, 0, fmt.Errorf("%s: unsupported arch %v", pn, arch.Family)
+				return nil, nil, fmt.Errorf("%s: unsupported arch %v", pn, arch.Family)
 			case sys.I386, sys.AMD64:
 				switch r.Type {
 				default:
-					return nil, 0, fmt.Errorf("%s: %v: unknown relocation type %v", pn, sectsyms[rsect], r.Type)
+					return nil, nil, fmt.Errorf("%s: %v: unknown relocation type %v", pn, sectsyms[rsect], r.Type)
 
 				case IMAGE_REL_I386_REL32, IMAGE_REL_AMD64_REL32,
 					IMAGE_REL_AMD64_ADDR32, // R_X86_64_PC32
@@ -302,7 +304,7 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 			case sys.ARM:
 				switch r.Type {
 				default:
-					return nil, 0, fmt.Errorf("%s: %v: unknown ARM relocation type %v", pn, sectsyms[rsect], r.Type)
+					return nil, nil, fmt.Errorf("%s: %v: unknown ARM relocation type %v", pn, sectsyms[rsect], r.Type)
 
 				case IMAGE_REL_ARM_SECREL:
 					rType = objabi.R_PCREL
@@ -323,8 +325,9 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 
 			// ld -r could generate multiple section symbols for the
 			// same section but with different values, we have to take
-			// that into account
-			if issect(pesym) {
+			// that into account, or in the case of split resources,
+			// the section and its symbols are split into two sections.
+			if issect(pesym) || splitResources {
 				rAdd += int64(pesym.Value)
 			}
 
@@ -346,7 +349,7 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 
 		name, err := pesym.FullName(f.StringTable)
 		if err != nil {
-			return nil, 0, err
+			return nil, nil, err
 		}
 		if name == "" {
 			continue
@@ -384,7 +387,7 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 
 		bld, s, err := readpesym(l, arch, l.LookupOrCreateSym, f, pesym, sectsyms, localSymVersion)
 		if err != nil {
-			return nil, 0, err
+			return nil, nil, err
 		}
 
 		if pesym.SectionNumber == 0 { // extern
@@ -402,14 +405,14 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 		} else if pesym.SectionNumber > 0 && int(pesym.SectionNumber) <= len(f.Sections) {
 			sect = f.Sections[pesym.SectionNumber-1]
 			if _, found := sectsyms[sect]; !found {
-				return nil, 0, fmt.Errorf("%s: %v: missing sect.sym", pn, s)
+				return nil, nil, fmt.Errorf("%s: %v: missing sect.sym", pn, s)
 			}
 		} else {
-			return nil, 0, fmt.Errorf("%s: %v: sectnum < 0!", pn, s)
+			return nil, nil, fmt.Errorf("%s: %v: sectnum < 0!", pn, s)
 		}
 
 		if sect == nil {
-			return nil, 0, nil
+			return nil, nil, nil
 		}
 
 		if l.OuterSym(s) != 0 {
@@ -418,7 +421,7 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 			}
 			outerName := l.SymName(l.OuterSym(s))
 			sectName := l.SymName(sectsyms[sect])
-			return nil, 0, fmt.Errorf("%s: duplicate symbol reference: %s in both %s and %s", pn, l.SymName(s), outerName, sectName)
+			return nil, nil, fmt.Errorf("%s: duplicate symbol reference: %s in both %s and %s", pn, l.SymName(s), outerName, sectName)
 		}
 
 		bld = makeUpdater(l, bld, s)
@@ -429,7 +432,7 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 		bld.SetSize(4)
 		if l.SymType(sectsym) == sym.STEXT {
 			if bld.External() && !bld.DuplicateOK() {
-				return nil, 0, fmt.Errorf("%s: duplicate symbol definition", l.SymName(s))
+				return nil, nil, fmt.Errorf("%s: duplicate symbol definition", l.SymName(s))
 			}
 			bld.SetExternal(true)
 		}
@@ -446,7 +449,7 @@ func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Read
 		if l.SymType(s) == sym.STEXT {
 			for ; s != 0; s = l.SubSym(s) {
 				if l.AttrOnList(s) {
-					return nil, 0, fmt.Errorf("symbol %s listed multiple times", l.SymName(s))
+					return nil, nil, fmt.Errorf("symbol %s listed multiple times", l.SymName(s))
 				}
 				l.SetAttrOnList(s, true)
 				textp = append(textp, s)

src/cmd/link/link_test.go

@@ -786,6 +786,25 @@ func TestPErsrc(t *testing.T) {
 	if !bytes.Contains(b, []byte("Hello Gophers!")) {
 		t.Fatalf("binary does not contain expected content")
 	}
+
+	pkgdir = filepath.Join("testdata", "testPErsrc-complex")
+	exe = filepath.Join(tmpdir, "a.exe")
+	cmd = exec.Command(testenv.GoToolPath(t), "build", "-o", exe)
+	cmd.Dir = pkgdir
+	// cmd.Env = append(os.Environ(), "GOOS=windows", "GOARCH=amd64") // uncomment if debugging in a cross-compiling environment
+	out, err = cmd.CombinedOutput()
+	if err != nil {
+		t.Fatalf("building failed: %v, output:\n%s", err, out)
+	}
+
+	// Check that the binary contains the rsrc data
+	b, err = ioutil.ReadFile(exe)
+	if err != nil {
+		t.Fatalf("reading output failed: %v", err)
+	}
+	if !bytes.Contains(b, []byte("resname RCDATA a.rc")) {
+		t.Fatalf("binary does not contain expected content")
+	}
 }
 
 func TestContentAddressableSymbols(t *testing.T) {

src/cmd/link/testdata/testPErsrc-complex/main.go

@@ -0,0 +1,43 @@
+// Copyright 2020 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Test that a PE rsrc section is handled correctly, when the object files
+// have been created by llvm-rc or msvc's rc.exe, which means there's the
+// @feat.00 symbol as well as split .rsrc$00 and .rsrc$01 section to deal with.
+//
+// rsrc.syso is created with:
+//    windres -i a.rc -o rsrc.syso -O coff
+// where this windres calls into llvm-rc and llvm-cvtres. The source file,
+// a.rc, simply contains a reference to its own bytes:
+//
+//    resname RCDATA a.rc
+//
+// Object dumping the resultant rsrc.syso, we can see the split sections and
+// the @feat.00 SEH symbol:
+//
+//     rsrc.syso:      file format coff-x86-64
+//
+//     architecture: x86_64
+//     start address: 0x0000000000000000
+//
+//     Export Table:
+//     Sections:
+//     Idx Name          Size     VMA              Type
+//       0 .rsrc$01      00000068 0000000000000000 DATA
+//       1 .rsrc$02      00000018 0000000000000000 DATA
+//
+//     SYMBOL TABLE:
+//     [ 0](sec -1)(fl 0x00)(ty   0)(scl   3) (nx 0) 0x00000011 @feat.00
+//     [ 1](sec  1)(fl 0x00)(ty   0)(scl   3) (nx 1) 0x00000000 .rsrc$01
+//     AUX scnlen 0x68 nreloc 1 nlnno 0 checksum 0x0 assoc 0 comdat 0
+//     [ 3](sec  2)(fl 0x00)(ty   0)(scl   3) (nx 1) 0x00000000 .rsrc$02
+//     AUX scnlen 0x18 nreloc 0 nlnno 0 checksum 0x0 assoc 0 comdat 0
+//     [ 5](sec  2)(fl 0x00)(ty   0)(scl   3) (nx 0) 0x00000000 $R000000
+//     RELOCATION RECORDS FOR [.rsrc$01]:
+//     OFFSET           TYPE                     VALUE
+//     0000000000000048 IMAGE_REL_AMD64_ADDR32NB $R000000
+
+package main
+
+func main() {}