ldpe.go

     1  // Copyright 2010 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // Package loadpe implements a PE/COFF file reader.
     6  package loadpe
     7  
     8  import (
     9  	"bytes"
    10  	"cmd/internal/bio"
    11  	"cmd/internal/objabi"
    12  	"cmd/internal/sys"
    13  	"cmd/link/internal/loader"
    14  	"cmd/link/internal/sym"
    15  	"debug/pe"
    16  	"encoding/binary"
    17  	"errors"
    18  	"fmt"
    19  	"io"
    20  	"strconv"
    21  	"strings"
    22  )
    23  
    24  const (
    25  	IMAGE_SYM_UNDEFINED              = 0
    26  	IMAGE_SYM_ABSOLUTE               = -1
    27  	IMAGE_SYM_DEBUG                  = -2
    28  	IMAGE_SYM_TYPE_NULL              = 0
    29  	IMAGE_SYM_TYPE_VOID              = 1
    30  	IMAGE_SYM_TYPE_CHAR              = 2
    31  	IMAGE_SYM_TYPE_SHORT             = 3
    32  	IMAGE_SYM_TYPE_INT               = 4
    33  	IMAGE_SYM_TYPE_LONG              = 5
    34  	IMAGE_SYM_TYPE_FLOAT             = 6
    35  	IMAGE_SYM_TYPE_DOUBLE            = 7
    36  	IMAGE_SYM_TYPE_STRUCT            = 8
    37  	IMAGE_SYM_TYPE_UNION             = 9
    38  	IMAGE_SYM_TYPE_ENUM              = 10
    39  	IMAGE_SYM_TYPE_MOE               = 11
    40  	IMAGE_SYM_TYPE_BYTE              = 12
    41  	IMAGE_SYM_TYPE_WORD              = 13
    42  	IMAGE_SYM_TYPE_UINT              = 14
    43  	IMAGE_SYM_TYPE_DWORD             = 15
    44  	IMAGE_SYM_TYPE_PCODE             = 32768
    45  	IMAGE_SYM_DTYPE_NULL             = 0
    46  	IMAGE_SYM_DTYPE_POINTER          = 1
    47  	IMAGE_SYM_DTYPE_FUNCTION         = 2
    48  	IMAGE_SYM_DTYPE_ARRAY            = 3
    49  	IMAGE_SYM_CLASS_END_OF_FUNCTION  = -1
    50  	IMAGE_SYM_CLASS_NULL             = 0
    51  	IMAGE_SYM_CLASS_AUTOMATIC        = 1
    52  	IMAGE_SYM_CLASS_EXTERNAL         = 2
    53  	IMAGE_SYM_CLASS_STATIC           = 3
    54  	IMAGE_SYM_CLASS_REGISTER         = 4
    55  	IMAGE_SYM_CLASS_EXTERNAL_DEF     = 5
    56  	IMAGE_SYM_CLASS_LABEL            = 6
    57  	IMAGE_SYM_CLASS_UNDEFINED_LABEL  = 7
    58  	IMAGE_SYM_CLASS_MEMBER_OF_STRUCT = 8
    59  	IMAGE_SYM_CLASS_ARGUMENT         = 9
    60  	IMAGE_SYM_CLASS_STRUCT_TAG       = 10
    61  	IMAGE_SYM_CLASS_MEMBER_OF_UNION  = 11
    62  	IMAGE_SYM_CLASS_UNION_TAG        = 12
    63  	IMAGE_SYM_CLASS_TYPE_DEFINITION  = 13
    64  	IMAGE_SYM_CLASS_UNDEFINED_STATIC = 14
    65  	IMAGE_SYM_CLASS_ENUM_TAG         = 15
    66  	IMAGE_SYM_CLASS_MEMBER_OF_ENUM   = 16
    67  	IMAGE_SYM_CLASS_REGISTER_PARAM   = 17
    68  	IMAGE_SYM_CLASS_BIT_FIELD        = 18
    69  	IMAGE_SYM_CLASS_FAR_EXTERNAL     = 68 /* Not in PECOFF v8 spec */
    70  	IMAGE_SYM_CLASS_BLOCK            = 100
    71  	IMAGE_SYM_CLASS_FUNCTION         = 101
    72  	IMAGE_SYM_CLASS_END_OF_STRUCT    = 102
    73  	IMAGE_SYM_CLASS_FILE             = 103
    74  	IMAGE_SYM_CLASS_SECTION          = 104
    75  	IMAGE_SYM_CLASS_WEAK_EXTERNAL    = 105
    76  	IMAGE_SYM_CLASS_CLR_TOKEN        = 107
    77  	IMAGE_REL_I386_ABSOLUTE          = 0x0000
    78  	IMAGE_REL_I386_DIR16             = 0x0001
    79  	IMAGE_REL_I386_REL16             = 0x0002
    80  	IMAGE_REL_I386_DIR32             = 0x0006
    81  	IMAGE_REL_I386_DIR32NB           = 0x0007
    82  	IMAGE_REL_I386_SEG12             = 0x0009
    83  	IMAGE_REL_I386_SECTION           = 0x000A
    84  	IMAGE_REL_I386_SECREL            = 0x000B
    85  	IMAGE_REL_I386_TOKEN             = 0x000C
    86  	IMAGE_REL_I386_SECREL7           = 0x000D
    87  	IMAGE_REL_I386_REL32             = 0x0014
    88  	IMAGE_REL_AMD64_ABSOLUTE         = 0x0000
    89  	IMAGE_REL_AMD64_ADDR64           = 0x0001
    90  	IMAGE_REL_AMD64_ADDR32           = 0x0002
    91  	IMAGE_REL_AMD64_ADDR32NB         = 0x0003
    92  	IMAGE_REL_AMD64_REL32            = 0x0004
    93  	IMAGE_REL_AMD64_REL32_1          = 0x0005
    94  	IMAGE_REL_AMD64_REL32_2          = 0x0006
    95  	IMAGE_REL_AMD64_REL32_3          = 0x0007
    96  	IMAGE_REL_AMD64_REL32_4          = 0x0008
    97  	IMAGE_REL_AMD64_REL32_5          = 0x0009
    98  	IMAGE_REL_AMD64_SECTION          = 0x000A
    99  	IMAGE_REL_AMD64_SECREL           = 0x000B
   100  	IMAGE_REL_AMD64_SECREL7          = 0x000C
   101  	IMAGE_REL_AMD64_TOKEN            = 0x000D
   102  	IMAGE_REL_AMD64_SREL32           = 0x000E
   103  	IMAGE_REL_AMD64_PAIR             = 0x000F
   104  	IMAGE_REL_AMD64_SSPAN32          = 0x0010
   105  	IMAGE_REL_ARM_ABSOLUTE           = 0x0000
   106  	IMAGE_REL_ARM_ADDR32             = 0x0001
   107  	IMAGE_REL_ARM_ADDR32NB           = 0x0002
   108  	IMAGE_REL_ARM_BRANCH24           = 0x0003
   109  	IMAGE_REL_ARM_BRANCH11           = 0x0004
   110  	IMAGE_REL_ARM_SECTION            = 0x000E
   111  	IMAGE_REL_ARM_SECREL             = 0x000F
   112  	IMAGE_REL_ARM_MOV32              = 0x0010
   113  	IMAGE_REL_THUMB_MOV32            = 0x0011
   114  	IMAGE_REL_THUMB_BRANCH20         = 0x0012
   115  	IMAGE_REL_THUMB_BRANCH24         = 0x0014
   116  	IMAGE_REL_THUMB_BLX23            = 0x0015
   117  	IMAGE_REL_ARM_PAIR               = 0x0016
   118  	IMAGE_REL_ARM64_ABSOLUTE         = 0x0000
   119  	IMAGE_REL_ARM64_ADDR32           = 0x0001
   120  	IMAGE_REL_ARM64_ADDR32NB         = 0x0002
   121  	IMAGE_REL_ARM64_BRANCH26         = 0x0003
   122  	IMAGE_REL_ARM64_PAGEBASE_REL21   = 0x0004
   123  	IMAGE_REL_ARM64_REL21            = 0x0005
   124  	IMAGE_REL_ARM64_PAGEOFFSET_12A   = 0x0006
   125  	IMAGE_REL_ARM64_PAGEOFFSET_12L   = 0x0007
   126  	IMAGE_REL_ARM64_SECREL           = 0x0008
   127  	IMAGE_REL_ARM64_SECREL_LOW12A    = 0x0009
   128  	IMAGE_REL_ARM64_SECREL_HIGH12A   = 0x000A
   129  	IMAGE_REL_ARM64_SECREL_LOW12L    = 0x000B
   130  	IMAGE_REL_ARM64_TOKEN            = 0x000C
   131  	IMAGE_REL_ARM64_SECTION          = 0x000D
   132  	IMAGE_REL_ARM64_ADDR64           = 0x000E
   133  	IMAGE_REL_ARM64_BRANCH19         = 0x000F
   134  	IMAGE_REL_ARM64_BRANCH14         = 0x0010
   135  	IMAGE_REL_ARM64_REL32            = 0x0011
   136  )
   137  
   138  const (
   139  	// When stored into the PLT value for a symbol, this token tells
   140  	// windynrelocsym to redirect direct references to this symbol to a stub
   141  	// that loads from the corresponding import symbol and then does
   142  	// a jump to the loaded value.
   143  	CreateImportStubPltToken = -2
   144  
   145  	// When stored into the GOT value for an import symbol __imp_X this
   146  	// token tells windynrelocsym to redirect references to the
   147  	// underlying DYNIMPORT symbol X.
   148  	RedirectToDynImportGotToken = -2
   149  )
   150  
   151  // TODO(brainman): maybe just add ReadAt method to bio.Reader instead of creating peBiobuf
   152  
   153  // peBiobuf makes bio.Reader look like io.ReaderAt.
   154  type peBiobuf bio.Reader
   155  
   156  func (f *peBiobuf) ReadAt(p []byte, off int64) (int, error) {
   157  	ret := ((*bio.Reader)(f)).MustSeek(off, 0)
   158  	if ret < 0 {
   159  		return 0, errors.New("fail to seek")
   160  	}
   161  	n, err := f.Read(p)
   162  	if err != nil {
   163  		return 0, err
   164  	}
   165  	return n, nil
   166  }
   167  
   168  // makeUpdater creates a loader.SymbolBuilder if one hasn't been created previously.
   169  // We use this to lazily make SymbolBuilders as we don't always need a builder, and creating them for all symbols might be an error.
   170  func makeUpdater(l *loader.Loader, bld *loader.SymbolBuilder, s loader.Sym) *loader.SymbolBuilder {
   171  	if bld != nil {
   172  		return bld
   173  	}
   174  	bld = l.MakeSymbolUpdater(s)
   175  	return bld
   176  }
   177  
   178  // peImportSymsState tracks the set of DLL import symbols we've seen
   179  // while reading host objects. We create a singleton instance of this
   180  // type, which will persist across multiple host objects.
   181  type peImportSymsState struct {
   182  
   183  	// Text and non-text sections read in by the host object loader.
   184  	secSyms []loader.Sym
   185  
   186  	// Loader and arch, for use in postprocessing.
   187  	l    *loader.Loader
   188  	arch *sys.Arch
   189  }
   190  
   191  var importSymsState *peImportSymsState
   192  
   193  func createImportSymsState(l *loader.Loader, arch *sys.Arch) {
   194  	if importSymsState != nil {
   195  		return
   196  	}
   197  	importSymsState = &peImportSymsState{
   198  		l:    l,
   199  		arch: arch,
   200  	}
   201  }
   202  
   203  // peLoaderState holds various bits of useful state information needed
   204  // while loading a single PE object file.
   205  type peLoaderState struct {
   206  	l               *loader.Loader
   207  	arch            *sys.Arch
   208  	f               *pe.File
   209  	pn              string
   210  	sectsyms        map[*pe.Section]loader.Sym
   211  	comdats         map[uint16]int64 // key is section index, val is size
   212  	sectdata        map[*pe.Section][]byte
   213  	localSymVersion int
   214  }
   215  
   216  // comdatDefinitions records the names of symbols for which we've
   217  // previously seen a definition in COMDAT. Key is symbol name, value
   218  // is symbol size (or -1 if we're using the "any" strategy).
   219  var comdatDefinitions map[string]int64
   220  
   221  // Symbols contains the symbols that can be loaded from a PE file.
   222  type Symbols struct {
   223  	Textp     []loader.Sym // text symbols
   224  	Resources []loader.Sym // .rsrc section or set of .rsrc$xx sections
   225  	PData     loader.Sym
   226  	XData     loader.Sym
   227  }
   228  
   229  // Load loads the PE file pn from input.
   230  // Symbols from the object file are created via the loader 'l'.
   231  func Load(l *loader.Loader, arch *sys.Arch, localSymVersion int, input *bio.Reader, pkg string, length int64, pn string) (*Symbols, error) {
   232  	state := &peLoaderState{
   233  		l:               l,
   234  		arch:            arch,
   235  		sectsyms:        make(map[*pe.Section]loader.Sym),
   236  		sectdata:        make(map[*pe.Section][]byte),
   237  		localSymVersion: localSymVersion,
   238  		pn:              pn,
   239  	}
   240  	createImportSymsState(state.l, state.arch)
   241  	if comdatDefinitions == nil {
   242  		comdatDefinitions = make(map[string]int64)
   243  	}
   244  
   245  	// Some input files are archives containing multiple of
   246  	// object files, and pe.NewFile seeks to the start of
   247  	// input file and get confused. Create section reader
   248  	// to stop pe.NewFile looking before current position.
   249  	sr := io.NewSectionReader((*peBiobuf)(input), input.Offset(), 1<<63-1)
   250  
   251  	// TODO: replace pe.NewFile with pe.Load (grep for "add Load function" in debug/pe for details)
   252  	f, err := pe.NewFile(sr)
   253  	if err != nil {
   254  		return nil, err
   255  	}
   256  	defer f.Close()
   257  	state.f = f
   258  
   259  	var ls Symbols
   260  
   261  	// TODO return error if found .cormeta
   262  
   263  	// create symbols for mapped sections
   264  	for _, sect := range f.Sections {
   265  		if sect.Characteristics&pe.IMAGE_SCN_MEM_DISCARDABLE != 0 {
   266  			continue
   267  		}
   268  
   269  		if sect.Characteristics&(pe.IMAGE_SCN_CNT_CODE|pe.IMAGE_SCN_CNT_INITIALIZED_DATA|pe.IMAGE_SCN_CNT_UNINITIALIZED_DATA) == 0 {
   270  			// This has been seen for .idata sections, which we
   271  			// want to ignore. See issues 5106 and 5273.
   272  			continue
   273  		}
   274  
   275  		name := fmt.Sprintf("%s(%s)", pkg, sect.Name)
   276  		s := state.l.LookupOrCreateCgoExport(name, localSymVersion)
   277  		bld := l.MakeSymbolUpdater(s)
   278  
   279  		switch sect.Characteristics & (pe.IMAGE_SCN_CNT_UNINITIALIZED_DATA | pe.IMAGE_SCN_CNT_INITIALIZED_DATA | pe.IMAGE_SCN_MEM_READ | pe.IMAGE_SCN_MEM_WRITE | pe.IMAGE_SCN_CNT_CODE | pe.IMAGE_SCN_MEM_EXECUTE) {
   280  		case pe.IMAGE_SCN_CNT_INITIALIZED_DATA | pe.IMAGE_SCN_MEM_READ: //.rdata
   281  			if issehsect(arch, sect) {
   282  				bld.SetType(sym.SSEHSECT)
   283  				bld.SetAlign(4)
   284  			} else {
   285  				bld.SetType(sym.SRODATA)
   286  			}
   287  
   288  		case pe.IMAGE_SCN_CNT_UNINITIALIZED_DATA | pe.IMAGE_SCN_MEM_READ | pe.IMAGE_SCN_MEM_WRITE: //.bss
   289  			bld.SetType(sym.SNOPTRBSS)
   290  
   291  		case pe.IMAGE_SCN_CNT_INITIALIZED_DATA | pe.IMAGE_SCN_MEM_READ | pe.IMAGE_SCN_MEM_WRITE: //.data
   292  			bld.SetType(sym.SNOPTRDATA)
   293  
   294  		case pe.IMAGE_SCN_CNT_CODE | pe.IMAGE_SCN_MEM_EXECUTE | pe.IMAGE_SCN_MEM_READ: //.text
   295  			bld.SetType(sym.STEXT)
   296  
   297  		default:
   298  			return nil, fmt.Errorf("unexpected flags %#06x for PE section %s", sect.Characteristics, sect.Name)
   299  		}
   300  
   301  		if bld.Type() != sym.SNOPTRBSS {
   302  			data, err := sect.Data()
   303  			if err != nil {
   304  				return nil, err
   305  			}
   306  			state.sectdata[sect] = data
   307  			bld.SetData(data)
   308  		}
   309  		bld.SetSize(int64(sect.Size))
   310  		state.sectsyms[sect] = s
   311  		if sect.Name == ".rsrc" || strings.HasPrefix(sect.Name, ".rsrc$") {
   312  			ls.Resources = append(ls.Resources, s)
   313  		} else if bld.Type() == sym.SSEHSECT {
   314  			if sect.Name == ".pdata" {
   315  				ls.PData = s
   316  			} else if sect.Name == ".xdata" {
   317  				ls.XData = s
   318  			}
   319  		}
   320  	}
   321  
   322  	// Make a prepass over the symbols to collect info about COMDAT symbols.
   323  	if err := state.preprocessSymbols(); err != nil {
   324  		return nil, err
   325  	}
   326  
   327  	// load relocations
   328  	for _, rsect := range f.Sections {
   329  		if _, found := state.sectsyms[rsect]; !found {
   330  			continue
   331  		}
   332  		if rsect.NumberOfRelocations == 0 {
   333  			continue
   334  		}
   335  		if rsect.Characteristics&pe.IMAGE_SCN_MEM_DISCARDABLE != 0 {
   336  			continue
   337  		}
   338  		if rsect.Characteristics&(pe.IMAGE_SCN_CNT_CODE|pe.IMAGE_SCN_CNT_INITIALIZED_DATA|pe.IMAGE_SCN_CNT_UNINITIALIZED_DATA) == 0 {
   339  			// This has been seen for .idata sections, which we
   340  			// want to ignore. See issues 5106 and 5273.
   341  			continue
   342  		}
   343  
   344  		splitResources := strings.HasPrefix(rsect.Name, ".rsrc$")
   345  		issehsect := issehsect(arch, rsect)
   346  		sb := l.MakeSymbolUpdater(state.sectsyms[rsect])
   347  		for j, r := range rsect.Relocs {
   348  			if int(r.SymbolTableIndex) >= len(f.COFFSymbols) {
   349  				return nil, fmt.Errorf("relocation number %d symbol index idx=%d cannot be large then number of symbols %d", j, r.SymbolTableIndex, len(f.COFFSymbols))
   350  			}
   351  			pesym := &f.COFFSymbols[r.SymbolTableIndex]
   352  			_, rSym, err := state.readpesym(pesym)
   353  			if err != nil {
   354  				return nil, err
   355  			}
   356  			if rSym == 0 {
   357  				name, err := pesym.FullName(f.StringTable)
   358  				if err != nil {
   359  					name = string(pesym.Name[:])
   360  				}
   361  				return nil, fmt.Errorf("reloc of invalid sym %s idx=%d type=%d", name, r.SymbolTableIndex, pesym.Type)
   362  			}
   363  
   364  			rSize := uint8(4)
   365  			rOff := int32(r.VirtualAddress)
   366  			var rType objabi.RelocType
   367  			switch arch.Family {
   368  			default:
   369  				return nil, fmt.Errorf("%s: unsupported arch %v", pn, arch.Family)
   370  			case sys.I386:
   371  				switch r.Type {
   372  				case IMAGE_REL_I386_REL32:
   373  					rType = objabi.R_PCREL
   374  				case IMAGE_REL_I386_DIR32:
   375  					rType = objabi.R_ADDR
   376  				case IMAGE_REL_I386_DIR32NB:
   377  					rType = objabi.R_PEIMAGEOFF
   378  				}
   379  			case sys.AMD64:
   380  				switch r.Type {
   381  				case IMAGE_REL_AMD64_REL32:
   382  					rType = objabi.R_PCREL
   383  				case IMAGE_REL_AMD64_ADDR32:
   384  					rType = objabi.R_ADDR
   385  				case IMAGE_REL_AMD64_ADDR64:
   386  					rType = objabi.R_ADDR
   387  					rSize = 8
   388  				case IMAGE_REL_AMD64_ADDR32NB:
   389  					rType = objabi.R_PEIMAGEOFF
   390  				}
   391  			case sys.ARM64:
   392  				switch r.Type {
   393  				case IMAGE_REL_ARM64_ADDR32:
   394  					rType = objabi.R_ADDR
   395  				case IMAGE_REL_ARM64_ADDR32NB:
   396  					rType = objabi.R_PEIMAGEOFF
   397  				}
   398  			}
   399  			if rType == 0 {
   400  				return nil, fmt.Errorf("%s: %v: unknown relocation type %v", pn, state.sectsyms[rsect], r.Type)
   401  			}
   402  			var rAdd int64
   403  			switch rSize {
   404  			default:
   405  				panic("unexpected relocation size " + strconv.Itoa(int(rSize)))
   406  			case 4:
   407  				rAdd = int64(int32(binary.LittleEndian.Uint32(state.sectdata[rsect][rOff:])))
   408  			case 8:
   409  				rAdd = int64(binary.LittleEndian.Uint64(state.sectdata[rsect][rOff:]))
   410  			}
   411  			// ld -r could generate multiple section symbols for the
   412  			// same section but with different values, we have to take
   413  			// that into account, or in the case of split resources,
   414  			// the section and its symbols are split into two sections.
   415  			if issect(pesym) || splitResources {
   416  				rAdd += int64(pesym.Value)
   417  			}
   418  			if issehsect {
   419  				// .pdata and .xdata sections can contain records
   420  				// associated to functions that won't be used in
   421  				// the final binary, in which case the relocation
   422  				// target symbol won't be reachable.
   423  				rType |= objabi.R_WEAK
   424  			}
   425  
   426  			rel, _ := sb.AddRel(rType)
   427  			rel.SetOff(rOff)
   428  			rel.SetSiz(rSize)
   429  			rel.SetSym(rSym)
   430  			rel.SetAdd(rAdd)
   431  
   432  		}
   433  
   434  		sb.SortRelocs()
   435  	}
   436  
   437  	// enter sub-symbols into symbol table.
   438  	for i, numaux := 0, 0; i < len(f.COFFSymbols); i += numaux + 1 {
   439  		pesym := &f.COFFSymbols[i]
   440  
   441  		numaux = int(pesym.NumberOfAuxSymbols)
   442  
   443  		name, err := pesym.FullName(f.StringTable)
   444  		if err != nil {
   445  			return nil, err
   446  		}
   447  		if name == "" {
   448  			continue
   449  		}
   450  		if issect(pesym) {
   451  			continue
   452  		}
   453  		if int(pesym.SectionNumber) > len(f.Sections) {
   454  			continue
   455  		}
   456  		if pesym.SectionNumber == IMAGE_SYM_DEBUG {
   457  			continue
   458  		}
   459  		if pesym.SectionNumber == IMAGE_SYM_ABSOLUTE && bytes.Equal(pesym.Name[:], []byte("@feat.00")) {
   460  			// The PE documentation says that, on x86 platforms, the absolute symbol named @feat.00
   461  			// is used to indicate that the COFF object supports SEH.
   462  			// Go doesn't support SEH on windows/386, so we can ignore this symbol.
   463  			// See https://learn.microsoft.com/en-us/windows/win32/debug/pe-format#the-sxdata-section
   464  			continue
   465  		}
   466  		var sect *pe.Section
   467  		if pesym.SectionNumber > 0 {
   468  			sect = f.Sections[pesym.SectionNumber-1]
   469  			if _, found := state.sectsyms[sect]; !found {
   470  				continue
   471  			}
   472  		}
   473  
   474  		bld, s, err := state.readpesym(pesym)
   475  		if err != nil {
   476  			return nil, err
   477  		}
   478  
   479  		if pesym.SectionNumber == 0 { // extern
   480  			if l.SymType(s) == sym.SXREF && pesym.Value > 0 { // global data
   481  				bld = makeUpdater(l, bld, s)
   482  				bld.SetType(sym.SNOPTRDATA)
   483  				bld.SetSize(int64(pesym.Value))
   484  			}
   485  
   486  			continue
   487  		} else if pesym.SectionNumber > 0 && int(pesym.SectionNumber) <= len(f.Sections) {
   488  			sect = f.Sections[pesym.SectionNumber-1]
   489  			if _, found := state.sectsyms[sect]; !found {
   490  				return nil, fmt.Errorf("%s: %v: missing sect.sym", pn, s)
   491  			}
   492  		} else {
   493  			return nil, fmt.Errorf("%s: %v: sectnum < 0!", pn, s)
   494  		}
   495  
   496  		if sect == nil {
   497  			return nil, nil
   498  		}
   499  
   500  		// Check for COMDAT symbol.
   501  		if sz, ok1 := state.comdats[uint16(pesym.SectionNumber-1)]; ok1 {
   502  			if psz, ok2 := comdatDefinitions[l.SymName(s)]; ok2 {
   503  				if sz == psz {
   504  					//  OK to discard, we've seen an instance
   505  					// already.
   506  					continue
   507  				}
   508  			}
   509  		}
   510  		if l.OuterSym(s) != 0 {
   511  			if l.AttrDuplicateOK(s) {
   512  				continue
   513  			}
   514  			outerName := l.SymName(l.OuterSym(s))
   515  			sectName := l.SymName(state.sectsyms[sect])
   516  			return nil, fmt.Errorf("%s: duplicate symbol reference: %s in both %s and %s", pn, l.SymName(s), outerName, sectName)
   517  		}
   518  
   519  		bld = makeUpdater(l, bld, s)
   520  		sectsym := state.sectsyms[sect]
   521  		bld.SetType(l.SymType(sectsym))
   522  		l.AddInteriorSym(sectsym, s)
   523  		bld.SetValue(int64(pesym.Value))
   524  		bld.SetSize(4)
   525  		if l.SymType(sectsym).IsText() {
   526  			if bld.External() && !bld.DuplicateOK() {
   527  				return nil, fmt.Errorf("%s: duplicate symbol definition", l.SymName(s))
   528  			}
   529  			bld.SetExternal(true)
   530  		}
   531  		if sz, ok := state.comdats[uint16(pesym.SectionNumber-1)]; ok {
   532  			// This is a COMDAT definition. Record that we're picking
   533  			// this instance so that we can ignore future defs.
   534  			if _, ok := comdatDefinitions[l.SymName(s)]; ok {
   535  				return nil, fmt.Errorf("internal error: preexisting COMDAT definition for %q", name)
   536  			}
   537  			comdatDefinitions[l.SymName(s)] = sz
   538  		}
   539  	}
   540  
   541  	// Sort outer lists by address, adding to textp.
   542  	// This keeps textp in increasing address order.
   543  	for _, sect := range f.Sections {
   544  		s := state.sectsyms[sect]
   545  		if s == 0 {
   546  			continue
   547  		}
   548  		l.SortSub(s)
   549  		importSymsState.secSyms = append(importSymsState.secSyms, s)
   550  		if l.SymType(s).IsText() {
   551  			for ; s != 0; s = l.SubSym(s) {
   552  				if l.AttrOnList(s) {
   553  					return nil, fmt.Errorf("symbol %s listed multiple times", l.SymName(s))
   554  				}
   555  				l.SetAttrOnList(s, true)
   556  				ls.Textp = append(ls.Textp, s)
   557  			}
   558  		}
   559  	}
   560  
   561  	if ls.PData != 0 {
   562  		processSEH(l, arch, ls.PData, ls.XData)
   563  	}
   564  
   565  	return &ls, nil
   566  }
   567  
   568  // PostProcessImports works to resolve inconsistencies with DLL import
   569  // symbols; it is needed when building with more "modern" C compilers
   570  // with internal linkage.
   571  //
   572  // Background: DLL import symbols are data (SNOPTRDATA) symbols whose
   573  // name is of the form "__imp_XXX", which contain a pointer/reference
   574  // to symbol XXX. It's possible to have import symbols for both data
   575  // symbols ("__imp__fmode") and text symbols ("__imp_CreateEventA").
   576  // In some case import symbols are just references to some external
   577  // thing, and in other cases we see actual definitions of import
   578  // symbols when reading host objects.
   579  //
   580  // Previous versions of the linker would in most cases immediately
   581  // "forward" import symbol references, e.g. treat a references to
   582  // "__imp_XXX" a references to "XXX", however this doesn't work well
   583  // with more modern compilers, where you can sometimes see import
   584  // symbols that are defs (as opposed to external refs).
   585  //
   586  // The main actions taken below are to search for references to
   587  // SDYNIMPORT symbols in host object text/data sections and flag the
   588  // symbols for later fixup. When we see a reference to an import
   589  // symbol __imp_XYZ where XYZ corresponds to some SDYNIMPORT symbol,
   590  // we flag the symbol (via GOT setting) so that it can be redirected
   591  // to XYZ later in windynrelocsym. When we see a direct reference to
   592  // an SDYNIMPORT symbol XYZ, we also flag the symbol (via PLT setting)
   593  // to indicated that the reference will need to be redirected to a
   594  // stub.
   595  func PostProcessImports() error {
   596  	ldr := importSymsState.l
   597  	arch := importSymsState.arch
   598  	keeprelocneeded := make(map[loader.Sym]loader.Sym)
   599  	for _, s := range importSymsState.secSyms {
   600  		isText := ldr.SymType(s).IsText()
   601  		relocs := ldr.Relocs(s)
   602  		for i := 0; i < relocs.Count(); i++ {
   603  			r := relocs.At(i)
   604  			rs := r.Sym()
   605  			if ldr.SymType(rs) == sym.SDYNIMPORT {
   606  				// Tag the symbol for later stub generation.
   607  				ldr.SetPlt(rs, CreateImportStubPltToken)
   608  				continue
   609  			}
   610  			isym, err := LookupBaseFromImport(rs, ldr, arch)
   611  			if err != nil {
   612  				return err
   613  			}
   614  			if isym == 0 {
   615  				continue
   616  			}
   617  			if ldr.SymType(isym) != sym.SDYNIMPORT {
   618  				continue
   619  			}
   620  			// For non-text symbols, forward the reference from __imp_X to
   621  			// X immediately.
   622  			if !isText {
   623  				r.SetSym(isym)
   624  				continue
   625  			}
   626  			// Flag this imp symbol to be processed later in windynrelocsym.
   627  			ldr.SetGot(rs, RedirectToDynImportGotToken)
   628  			// Consistency check: should be no PLT token here.
   629  			splt := ldr.SymPlt(rs)
   630  			if splt != -1 {
   631  				return fmt.Errorf("internal error: import symbol %q has invalid PLT setting %d", ldr.SymName(rs), splt)
   632  			}
   633  			// Flag for dummy relocation.
   634  			keeprelocneeded[rs] = isym
   635  		}
   636  	}
   637  	for k, v := range keeprelocneeded {
   638  		sb := ldr.MakeSymbolUpdater(k)
   639  		r, _ := sb.AddRel(objabi.R_KEEP)
   640  		r.SetSym(v)
   641  	}
   642  	importSymsState = nil
   643  	return nil
   644  }
   645  
   646  func issehsect(arch *sys.Arch, s *pe.Section) bool {
   647  	return arch.Family == sys.AMD64 && (s.Name == ".pdata" || s.Name == ".xdata")
   648  }
   649  
   650  func issect(s *pe.COFFSymbol) bool {
   651  	return s.StorageClass == IMAGE_SYM_CLASS_STATIC && s.Type == 0 && s.Name[0] == '.'
   652  }
   653  
   654  func (state *peLoaderState) readpesym(pesym *pe.COFFSymbol) (*loader.SymbolBuilder, loader.Sym, error) {
   655  	symname, err := pesym.FullName(state.f.StringTable)
   656  	if err != nil {
   657  		return nil, 0, err
   658  	}
   659  	var name string
   660  	if issect(pesym) {
   661  		name = state.l.SymName(state.sectsyms[state.f.Sections[pesym.SectionNumber-1]])
   662  	} else {
   663  		name = symname
   664  		// A note on the "_main" exclusion below: the main routine
   665  		// defined by the Go runtime is named "_main", not "main", so
   666  		// when reading references to _main from a host object we want
   667  		// to avoid rewriting "_main" to "main" in this specific
   668  		// instance. See #issuecomment-1143698749 on #35006 for more
   669  		// details on this problem.
   670  		if state.arch.Family == sys.I386 && name[0] == '_' && name != "_main" && !strings.HasPrefix(name, "__imp_") {
   671  			name = name[1:] // _Name => Name
   672  		}
   673  	}
   674  
   675  	// remove last @XXX
   676  	if i := strings.LastIndex(name, "@"); i >= 0 {
   677  		name = name[:i]
   678  	}
   679  
   680  	var s loader.Sym
   681  	var bld *loader.SymbolBuilder
   682  	// Microsoft's PE documentation is contradictory. It says that the symbol's complex type
   683  	// is stored in the pesym.Type most significant byte, but MSVC, LLVM, and mingw store it
   684  	// in the 4 high bits of the less significant byte.
   685  	switch uint8(pesym.Type&0xf0) >> 4 {
   686  	default:
   687  		return nil, 0, fmt.Errorf("%s: invalid symbol type %d", symname, pesym.Type)
   688  
   689  	case IMAGE_SYM_DTYPE_FUNCTION, IMAGE_SYM_DTYPE_NULL:
   690  		switch pesym.StorageClass {
   691  		case IMAGE_SYM_CLASS_EXTERNAL: //global
   692  			s = state.l.LookupOrCreateCgoExport(name, 0)
   693  
   694  		case IMAGE_SYM_CLASS_NULL, IMAGE_SYM_CLASS_STATIC, IMAGE_SYM_CLASS_LABEL:
   695  			s = state.l.LookupOrCreateCgoExport(name, state.localSymVersion)
   696  			bld = makeUpdater(state.l, bld, s)
   697  			bld.SetDuplicateOK(true)
   698  
   699  		default:
   700  			return nil, 0, fmt.Errorf("%s: invalid symbol binding %d", symname, pesym.StorageClass)
   701  		}
   702  	}
   703  
   704  	if s != 0 && state.l.SymType(s) == 0 && (pesym.StorageClass != IMAGE_SYM_CLASS_STATIC || pesym.Value != 0) {
   705  		bld = makeUpdater(state.l, bld, s)
   706  		bld.SetType(sym.SXREF)
   707  	}
   708  
   709  	return bld, s, nil
   710  }
   711  
   712  // preprocessSymbols walks the COFF symbols for the PE file we're
   713  // reading and looks for cases where we have both a symbol definition
   714  // for "XXX" and an "__imp_XXX" symbol, recording these cases in a map
   715  // in the state struct. This information will be used in readpesym()
   716  // above to give such symbols special treatment. This function also
   717  // gathers information about COMDAT sections/symbols for later use
   718  // in readpesym().
   719  func (state *peLoaderState) preprocessSymbols() error {
   720  
   721  	// Locate comdat sections.
   722  	state.comdats = make(map[uint16]int64)
   723  	for i, s := range state.f.Sections {
   724  		if s.Characteristics&uint32(pe.IMAGE_SCN_LNK_COMDAT) != 0 {
   725  			state.comdats[uint16(i)] = int64(s.Size)
   726  		}
   727  	}
   728  
   729  	// Examine symbol defs.
   730  	for i, numaux := 0, 0; i < len(state.f.COFFSymbols); i += numaux + 1 {
   731  		pesym := &state.f.COFFSymbols[i]
   732  		numaux = int(pesym.NumberOfAuxSymbols)
   733  		if pesym.SectionNumber == 0 { // extern
   734  			continue
   735  		}
   736  		symname, err := pesym.FullName(state.f.StringTable)
   737  		if err != nil {
   738  			return err
   739  		}
   740  		if _, isc := state.comdats[uint16(pesym.SectionNumber-1)]; !isc {
   741  			continue
   742  		}
   743  		if pesym.StorageClass != uint8(IMAGE_SYM_CLASS_STATIC) {
   744  			continue
   745  		}
   746  		// This symbol corresponds to a COMDAT section. Read the
   747  		// aux data for it.
   748  		auxsymp, err := state.f.COFFSymbolReadSectionDefAux(i)
   749  		if err != nil {
   750  			return fmt.Errorf("unable to read aux info for section def symbol %d %s: pe.COFFSymbolReadComdatInfo returns %v", i, symname, err)
   751  		}
   752  		if auxsymp.Selection == pe.IMAGE_COMDAT_SELECT_SAME_SIZE {
   753  			// This is supported.
   754  		} else if auxsymp.Selection == pe.IMAGE_COMDAT_SELECT_ANY {
   755  			// Also supported.
   756  			state.comdats[uint16(pesym.SectionNumber-1)] = int64(-1)
   757  		} else {
   758  			// We don't support any of the other strategies at the
   759  			// moment. I suspect that we may need to also support
   760  			// "associative", we'll see.
   761  			return fmt.Errorf("internal error: unsupported COMDAT selection strategy found in path=%s sec=%d strategy=%d idx=%d, please file a bug", state.pn, auxsymp.SecNum, auxsymp.Selection, i)
   762  		}
   763  	}
   764  	return nil
   765  }
   766  
   767  // LookupBaseFromImport examines the symbol "s" to see if it
   768  // corresponds to an import symbol (name of the form "__imp_XYZ") and
   769  // if so, it looks up the underlying target of the import symbol and
   770  // returns it. An error is returned if the symbol is of the form
   771  // "__imp_XYZ" but no XYZ can be found.
   772  func LookupBaseFromImport(s loader.Sym, ldr *loader.Loader, arch *sys.Arch) (loader.Sym, error) {
   773  	sname := ldr.SymName(s)
   774  	if !strings.HasPrefix(sname, "__imp_") {
   775  		return 0, nil
   776  	}
   777  	basename := sname[len("__imp_"):]
   778  	if arch.Family == sys.I386 && basename[0] == '_' {
   779  		basename = basename[1:] // _Name => Name
   780  	}
   781  	isym := ldr.Lookup(basename, 0)
   782  	if isym == 0 {
   783  		return 0, fmt.Errorf("internal error: import symbol %q with no underlying sym", sname)
   784  	}
   785  	return isym, nil
   786  }
   787
View as plain text