buildinfo.go

Documentation: debug/buildinfo

     1  // Copyright 2021 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 buildinfo provides access to information embedded in a Go binary
     6  // about how it was built. This includes the Go toolchain version, and the
     7  // set of modules used (for binaries built in module mode).
     8  //
     9  // Build information is available for the currently running binary in
    10  // runtime/debug.ReadBuildInfo.
    11  package buildinfo
    12  
    13  import (
    14  	"bytes"
    15  	"debug/elf"
    16  	"debug/macho"
    17  	"debug/pe"
    18  	"debug/plan9obj"
    19  	"encoding/binary"
    20  	"errors"
    21  	"fmt"
    22  	"internal/saferio"
    23  	"internal/xcoff"
    24  	"io"
    25  	"io/fs"
    26  	"os"
    27  	"runtime/debug"
    28  )
    29  
    30  // Type alias for build info. We cannot move the types here, since
    31  // runtime/debug would need to import this package, which would make it
    32  // a much larger dependency.
    33  type BuildInfo = debug.BuildInfo
    34  
    35  var (
    36  	// errUnrecognizedFormat is returned when a given executable file doesn't
    37  	// appear to be in a known format, or it breaks the rules of that format,
    38  	// or when there are I/O errors reading the file.
    39  	errUnrecognizedFormat = errors.New("unrecognized file format")
    40  
    41  	// errNotGoExe is returned when a given executable file is valid but does
    42  	// not contain Go build information.
    43  	errNotGoExe = errors.New("not a Go executable")
    44  
    45  	// The build info blob left by the linker is identified by
    46  	// a 16-byte header, consisting of buildInfoMagic (14 bytes),
    47  	// the binary's pointer size (1 byte),
    48  	// and whether the binary is big endian (1 byte).
    49  	buildInfoMagic = []byte("\xff Go buildinf:")
    50  )
    51  
    52  // ReadFile returns build information embedded in a Go binary
    53  // file at the given path. Most information is only available for binaries built
    54  // with module support.
    55  func ReadFile(name string) (info *BuildInfo, err error) {
    56  	defer func() {
    57  		if pathErr := (*fs.PathError)(nil); errors.As(err, &pathErr) {
    58  			err = fmt.Errorf("could not read Go build info: %w", err)
    59  		} else if err != nil {
    60  			err = fmt.Errorf("could not read Go build info from %s: %w", name, err)
    61  		}
    62  	}()
    63  
    64  	f, err := os.Open(name)
    65  	if err != nil {
    66  		return nil, err
    67  	}
    68  	defer f.Close()
    69  	return Read(f)
    70  }
    71  
    72  // Read returns build information embedded in a Go binary file
    73  // accessed through the given ReaderAt. Most information is only available for
    74  // binaries built with module support.
    75  func Read(r io.ReaderAt) (*BuildInfo, error) {
    76  	vers, mod, err := readRawBuildInfo(r)
    77  	if err != nil {
    78  		return nil, err
    79  	}
    80  	bi, err := debug.ParseBuildInfo(mod)
    81  	if err != nil {
    82  		return nil, err
    83  	}
    84  	bi.GoVersion = vers
    85  	return bi, nil
    86  }
    87  
    88  type exe interface {
    89  	// ReadData reads and returns up to size bytes starting at virtual address addr.
    90  	ReadData(addr, size uint64) ([]byte, error)
    91  
    92  	// DataStart returns the virtual address and size of the segment or section that
    93  	// should contain build information. This is either a specially named section
    94  	// or the first writable non-zero data segment.
    95  	DataStart() (uint64, uint64)
    96  }
    97  
    98  // readRawBuildInfo extracts the Go toolchain version and module information
    99  // strings from a Go binary. On success, vers should be non-empty. mod
   100  // is empty if the binary was not built with modules enabled.
   101  func readRawBuildInfo(r io.ReaderAt) (vers, mod string, err error) {
   102  	// Read the first bytes of the file to identify the format, then delegate to
   103  	// a format-specific function to load segment and section headers.
   104  	ident := make([]byte, 16)
   105  	if n, err := r.ReadAt(ident, 0); n < len(ident) || err != nil {
   106  		return "", "", errUnrecognizedFormat
   107  	}
   108  
   109  	var x exe
   110  	switch {
   111  	case bytes.HasPrefix(ident, []byte("\x7FELF")):
   112  		f, err := elf.NewFile(r)
   113  		if err != nil {
   114  			return "", "", errUnrecognizedFormat
   115  		}
   116  		x = &elfExe{f}
   117  	case bytes.HasPrefix(ident, []byte("MZ")):
   118  		f, err := pe.NewFile(r)
   119  		if err != nil {
   120  			return "", "", errUnrecognizedFormat
   121  		}
   122  		x = &peExe{f}
   123  	case bytes.HasPrefix(ident, []byte("\xFE\xED\xFA")) || bytes.HasPrefix(ident[1:], []byte("\xFA\xED\xFE")):
   124  		f, err := macho.NewFile(r)
   125  		if err != nil {
   126  			return "", "", errUnrecognizedFormat
   127  		}
   128  		x = &machoExe{f}
   129  	case bytes.HasPrefix(ident, []byte("\xCA\xFE\xBA\xBE")) || bytes.HasPrefix(ident, []byte("\xCA\xFE\xBA\xBF")):
   130  		f, err := macho.NewFatFile(r)
   131  		if err != nil || len(f.Arches) == 0 {
   132  			return "", "", errUnrecognizedFormat
   133  		}
   134  		x = &machoExe{f.Arches[0].File}
   135  	case bytes.HasPrefix(ident, []byte{0x01, 0xDF}) || bytes.HasPrefix(ident, []byte{0x01, 0xF7}):
   136  		f, err := xcoff.NewFile(r)
   137  		if err != nil {
   138  			return "", "", errUnrecognizedFormat
   139  		}
   140  		x = &xcoffExe{f}
   141  	case hasPlan9Magic(ident):
   142  		f, err := plan9obj.NewFile(r)
   143  		if err != nil {
   144  			return "", "", errUnrecognizedFormat
   145  		}
   146  		x = &plan9objExe{f}
   147  	default:
   148  		return "", "", errUnrecognizedFormat
   149  	}
   150  
   151  	// Read segment or section to find the build info blob.
   152  	// On some platforms, the blob will be in its own section, and DataStart
   153  	// returns the address of that section. On others, it's somewhere in the
   154  	// data segment; the linker puts it near the beginning.
   155  	// See cmd/link/internal/ld.Link.buildinfo.
   156  	dataAddr, dataSize := x.DataStart()
   157  	if dataSize == 0 {
   158  		return "", "", errNotGoExe
   159  	}
   160  	data, err := x.ReadData(dataAddr, dataSize)
   161  	if err != nil {
   162  		return "", "", err
   163  	}
   164  	const (
   165  		buildInfoAlign = 16
   166  		buildInfoSize  = 32
   167  	)
   168  	for {
   169  		i := bytes.Index(data, buildInfoMagic)
   170  		if i < 0 || len(data)-i < buildInfoSize {
   171  			return "", "", errNotGoExe
   172  		}
   173  		if i%buildInfoAlign == 0 && len(data)-i >= buildInfoSize {
   174  			data = data[i:]
   175  			break
   176  		}
   177  		data = data[(i+buildInfoAlign-1)&^(buildInfoAlign-1):]
   178  	}
   179  
   180  	// Decode the blob.
   181  	// The first 14 bytes are buildInfoMagic.
   182  	// The next two bytes indicate pointer size in bytes (4 or 8) and endianness
   183  	// (0 for little, 1 for big).
   184  	// Two virtual addresses to Go strings follow that: runtime.buildVersion,
   185  	// and runtime.modinfo.
   186  	// On 32-bit platforms, the last 8 bytes are unused.
   187  	// If the endianness has the 2 bit set, then the pointers are zero
   188  	// and the 32-byte header is followed by varint-prefixed string data
   189  	// for the two string values we care about.
   190  	ptrSize := int(data[14])
   191  	if data[15]&2 != 0 {
   192  		vers, data = decodeString(data[32:])
   193  		mod, data = decodeString(data)
   194  	} else {
   195  		bigEndian := data[15] != 0
   196  		var bo binary.ByteOrder
   197  		if bigEndian {
   198  			bo = binary.BigEndian
   199  		} else {
   200  			bo = binary.LittleEndian
   201  		}
   202  		var readPtr func([]byte) uint64
   203  		if ptrSize == 4 {
   204  			readPtr = func(b []byte) uint64 { return uint64(bo.Uint32(b)) }
   205  		} else if ptrSize == 8 {
   206  			readPtr = bo.Uint64
   207  		} else {
   208  			return "", "", errNotGoExe
   209  		}
   210  		vers = readString(x, ptrSize, readPtr, readPtr(data[16:]))
   211  		mod = readString(x, ptrSize, readPtr, readPtr(data[16+ptrSize:]))
   212  	}
   213  	if vers == "" {
   214  		return "", "", errNotGoExe
   215  	}
   216  	if len(mod) >= 33 && mod[len(mod)-17] == '\n' {
   217  		// Strip module framing: sentinel strings delimiting the module info.
   218  		// These are cmd/go/internal/modload.infoStart and infoEnd.
   219  		mod = mod[16 : len(mod)-16]
   220  	} else {
   221  		mod = ""
   222  	}
   223  
   224  	return vers, mod, nil
   225  }
   226  
   227  func hasPlan9Magic(magic []byte) bool {
   228  	if len(magic) >= 4 {
   229  		m := binary.BigEndian.Uint32(magic)
   230  		switch m {
   231  		case plan9obj.Magic386, plan9obj.MagicAMD64, plan9obj.MagicARM:
   232  			return true
   233  		}
   234  	}
   235  	return false
   236  }
   237  
   238  func decodeString(data []byte) (s string, rest []byte) {
   239  	u, n := binary.Uvarint(data)
   240  	if n <= 0 || u > uint64(len(data)-n) {
   241  		return "", nil
   242  	}
   243  	return string(data[n : uint64(n)+u]), data[uint64(n)+u:]
   244  }
   245  
   246  // readString returns the string at address addr in the executable x.
   247  func readString(x exe, ptrSize int, readPtr func([]byte) uint64, addr uint64) string {
   248  	hdr, err := x.ReadData(addr, uint64(2*ptrSize))
   249  	if err != nil || len(hdr) < 2*ptrSize {
   250  		return ""
   251  	}
   252  	dataAddr := readPtr(hdr)
   253  	dataLen := readPtr(hdr[ptrSize:])
   254  	data, err := x.ReadData(dataAddr, dataLen)
   255  	if err != nil || uint64(len(data)) < dataLen {
   256  		return ""
   257  	}
   258  	return string(data)
   259  }
   260  
   261  // elfExe is the ELF implementation of the exe interface.
   262  type elfExe struct {
   263  	f *elf.File
   264  }
   265  
   266  func (x *elfExe) ReadData(addr, size uint64) ([]byte, error) {
   267  	for _, prog := range x.f.Progs {
   268  		if prog.Vaddr <= addr && addr <= prog.Vaddr+prog.Filesz-1 {
   269  			n := prog.Vaddr + prog.Filesz - addr
   270  			if n > size {
   271  				n = size
   272  			}
   273  			return saferio.ReadDataAt(prog, n, int64(addr-prog.Vaddr))
   274  		}
   275  	}
   276  	return nil, errUnrecognizedFormat
   277  }
   278  
   279  func (x *elfExe) DataStart() (uint64, uint64) {
   280  	for _, s := range x.f.Sections {
   281  		if s.Name == ".go.buildinfo" {
   282  			return s.Addr, s.Size
   283  		}
   284  	}
   285  	for _, p := range x.f.Progs {
   286  		if p.Type == elf.PT_LOAD && p.Flags&(elf.PF_X|elf.PF_W) == elf.PF_W {
   287  			return p.Vaddr, p.Memsz
   288  		}
   289  	}
   290  	return 0, 0
   291  }
   292  
   293  // peExe is the PE (Windows Portable Executable) implementation of the exe interface.
   294  type peExe struct {
   295  	f *pe.File
   296  }
   297  
   298  func (x *peExe) imageBase() uint64 {
   299  	switch oh := x.f.OptionalHeader.(type) {
   300  	case *pe.OptionalHeader32:
   301  		return uint64(oh.ImageBase)
   302  	case *pe.OptionalHeader64:
   303  		return oh.ImageBase
   304  	}
   305  	return 0
   306  }
   307  
   308  func (x *peExe) ReadData(addr, size uint64) ([]byte, error) {
   309  	addr -= x.imageBase()
   310  	for _, sect := range x.f.Sections {
   311  		if uint64(sect.VirtualAddress) <= addr && addr <= uint64(sect.VirtualAddress+sect.Size-1) {
   312  			n := uint64(sect.VirtualAddress+sect.Size) - addr
   313  			if n > size {
   314  				n = size
   315  			}
   316  			return saferio.ReadDataAt(sect, n, int64(addr-uint64(sect.VirtualAddress)))
   317  		}
   318  	}
   319  	return nil, errUnrecognizedFormat
   320  }
   321  
   322  func (x *peExe) DataStart() (uint64, uint64) {
   323  	// Assume data is first writable section.
   324  	const (
   325  		IMAGE_SCN_CNT_CODE               = 0x00000020
   326  		IMAGE_SCN_CNT_INITIALIZED_DATA   = 0x00000040
   327  		IMAGE_SCN_CNT_UNINITIALIZED_DATA = 0x00000080
   328  		IMAGE_SCN_MEM_EXECUTE            = 0x20000000
   329  		IMAGE_SCN_MEM_READ               = 0x40000000
   330  		IMAGE_SCN_MEM_WRITE              = 0x80000000
   331  		IMAGE_SCN_MEM_DISCARDABLE        = 0x2000000
   332  		IMAGE_SCN_LNK_NRELOC_OVFL        = 0x1000000
   333  		IMAGE_SCN_ALIGN_32BYTES          = 0x600000
   334  	)
   335  	for _, sect := range x.f.Sections {
   336  		if sect.VirtualAddress != 0 && sect.Size != 0 &&
   337  			sect.Characteristics&^IMAGE_SCN_ALIGN_32BYTES == IMAGE_SCN_CNT_INITIALIZED_DATA|IMAGE_SCN_MEM_READ|IMAGE_SCN_MEM_WRITE {
   338  			return uint64(sect.VirtualAddress) + x.imageBase(), uint64(sect.VirtualSize)
   339  		}
   340  	}
   341  	return 0, 0
   342  }
   343  
   344  // machoExe is the Mach-O (Apple macOS/iOS) implementation of the exe interface.
   345  type machoExe struct {
   346  	f *macho.File
   347  }
   348  
   349  func (x *machoExe) ReadData(addr, size uint64) ([]byte, error) {
   350  	for _, load := range x.f.Loads {
   351  		seg, ok := load.(*macho.Segment)
   352  		if !ok {
   353  			continue
   354  		}
   355  		if seg.Addr <= addr && addr <= seg.Addr+seg.Filesz-1 {
   356  			if seg.Name == "__PAGEZERO" {
   357  				continue
   358  			}
   359  			n := seg.Addr + seg.Filesz - addr
   360  			if n > size {
   361  				n = size
   362  			}
   363  			return saferio.ReadDataAt(seg, n, int64(addr-seg.Addr))
   364  		}
   365  	}
   366  	return nil, errUnrecognizedFormat
   367  }
   368  
   369  func (x *machoExe) DataStart() (uint64, uint64) {
   370  	// Look for section named "__go_buildinfo".
   371  	for _, sec := range x.f.Sections {
   372  		if sec.Name == "__go_buildinfo" {
   373  			return sec.Addr, sec.Size
   374  		}
   375  	}
   376  	// Try the first non-empty writable segment.
   377  	const RW = 3
   378  	for _, load := range x.f.Loads {
   379  		seg, ok := load.(*macho.Segment)
   380  		if ok && seg.Addr != 0 && seg.Filesz != 0 && seg.Prot == RW && seg.Maxprot == RW {
   381  			return seg.Addr, seg.Memsz
   382  		}
   383  	}
   384  	return 0, 0
   385  }
   386  
   387  // xcoffExe is the XCOFF (AIX eXtended COFF) implementation of the exe interface.
   388  type xcoffExe struct {
   389  	f *xcoff.File
   390  }
   391  
   392  func (x *xcoffExe) ReadData(addr, size uint64) ([]byte, error) {
   393  	for _, sect := range x.f.Sections {
   394  		if sect.VirtualAddress <= addr && addr <= sect.VirtualAddress+sect.Size-1 {
   395  			n := sect.VirtualAddress + sect.Size - addr
   396  			if n > size {
   397  				n = size
   398  			}
   399  			return saferio.ReadDataAt(sect, n, int64(addr-sect.VirtualAddress))
   400  		}
   401  	}
   402  	return nil, errors.New("address not mapped")
   403  }
   404  
   405  func (x *xcoffExe) DataStart() (uint64, uint64) {
   406  	if s := x.f.SectionByType(xcoff.STYP_DATA); s != nil {
   407  		return s.VirtualAddress, s.Size
   408  	}
   409  	return 0, 0
   410  }
   411  
   412  // plan9objExe is the Plan 9 a.out implementation of the exe interface.
   413  type plan9objExe struct {
   414  	f *plan9obj.File
   415  }
   416  
   417  func (x *plan9objExe) DataStart() (uint64, uint64) {
   418  	if s := x.f.Section("data"); s != nil {
   419  		return uint64(s.Offset), uint64(s.Size)
   420  	}
   421  	return 0, 0
   422  }
   423  
   424  func (x *plan9objExe) ReadData(addr, size uint64) ([]byte, error) {
   425  	for _, sect := range x.f.Sections {
   426  		if uint64(sect.Offset) <= addr && addr <= uint64(sect.Offset+sect.Size-1) {
   427  			n := uint64(sect.Offset+sect.Size) - addr
   428  			if n > size {
   429  				n = size
   430  			}
   431  			return saferio.ReadDataAt(sect, n, int64(addr-uint64(sect.Offset)))
   432  		}
   433  	}
   434  	return nil, errors.New("address not mapped")
   435  }
   436
View as plain text