buildinfo.go

     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  	"encoding/binary"
    19  	"errors"
    20  	"fmt"
    21  	"internal/xcoff"
    22  	"io"
    23  	"io/fs"
    24  	"os"
    25  	"runtime/debug"
    26  )
    27  
    28  // Type alias for build info. We cannot move the types here, since
    29  // runtime/debug would need to import this package, which would make it
    30  // a much larger dependency.
    31  type BuildInfo = debug.BuildInfo
    32  
    33  var (
    34  	// errUnrecognizedFormat is returned when a given executable file doesn't
    35  	// appear to be in a known format, or it breaks the rules of that format,
    36  	// or when there are I/O errors reading the file.
    37  	errUnrecognizedFormat = errors.New("unrecognized file format")
    38  
    39  	// errNotGoExe is returned when a given executable file is valid but does
    40  	// not contain Go build information.
    41  	errNotGoExe = errors.New("not a Go executable")
    42  
    43  	// The build info blob left by the linker is identified by
    44  	// a 16-byte header, consisting of buildInfoMagic (14 bytes),
    45  	// the binary's pointer size (1 byte),
    46  	// and whether the binary is big endian (1 byte).
    47  	buildInfoMagic = []byte("\xff Go buildinf:")
    48  )
    49  
    50  // ReadFile returns build information embedded in a Go binary
    51  // file at the given path. Most information is only available for binaries built
    52  // with module support.
    53  func ReadFile(name string) (info *BuildInfo, err error) {
    54  	defer func() {
    55  		if pathErr := (*fs.PathError)(nil); errors.As(err, &pathErr) {
    56  			err = fmt.Errorf("could not read Go build info: %w", err)
    57  		} else if err != nil {
    58  			err = fmt.Errorf("could not read Go build info from %s: %w", name, err)
    59  		}
    60  	}()
    61  
    62  	f, err := os.Open(name)
    63  	if err != nil {
    64  		return nil, err
    65  	}
    66  	defer f.Close()
    67  	return Read(f)
    68  }
    69  
    70  // Read returns build information embedded in a Go binary file
    71  // accessed through the given ReaderAt. Most information is only available for
    72  // binaries built with module support.
    73  func Read(r io.ReaderAt) (*BuildInfo, error) {
    74  	vers, mod, err := readRawBuildInfo(r)
    75  	if err != nil {
    76  		return nil, err
    77  	}
    78  	bi, err := debug.ParseBuildInfo(mod)
    79  	if err != nil {
    80  		return nil, err
    81  	}
    82  	bi.GoVersion = vers
    83  	return bi, nil
    84  }
    85  
    86  type exe interface {
    87  	// ReadData reads and returns up to size bytes starting at virtual address addr.
    88  	ReadData(addr, size uint64) ([]byte, error)
    89  
    90  	// DataStart returns the virtual address of the segment or section that
    91  	// should contain build information. This is either a specially named section
    92  	// or the first writable non-zero data segment.
    93  	DataStart() uint64
    94  }
    95  
    96  // readRawBuildInfo extracts the Go toolchain version and module information
    97  // strings from a Go binary. On success, vers should be non-empty. mod
    98  // is empty if the binary was not built with modules enabled.
    99  func readRawBuildInfo(r io.ReaderAt) (vers, mod string, err error) {
   100  	// Read the first bytes of the file to identify the format, then delegate to
   101  	// a format-specific function to load segment and section headers.
   102  	ident := make([]byte, 16)
   103  	if n, err := r.ReadAt(ident, 0); n < len(ident) || err != nil {
   104  		return "", "", errUnrecognizedFormat
   105  	}
   106  
   107  	var x exe
   108  	switch {
   109  	case bytes.HasPrefix(ident, []byte("\x7FELF")):
   110  		f, err := elf.NewFile(r)
   111  		if err != nil {
   112  			return "", "", errUnrecognizedFormat
   113  		}
   114  		x = &elfExe{f}
   115  	case bytes.HasPrefix(ident, []byte("MZ")):
   116  		f, err := pe.NewFile(r)
   117  		if err != nil {
   118  			return "", "", errUnrecognizedFormat
   119  		}
   120  		x = &peExe{f}
   121  	case bytes.HasPrefix(ident, []byte("\xFE\xED\xFA")) || bytes.HasPrefix(ident[1:], []byte("\xFA\xED\xFE")):
   122  		f, err := macho.NewFile(r)
   123  		if err != nil {
   124  			return "", "", errUnrecognizedFormat
   125  		}
   126  		x = &machoExe{f}
   127  	case bytes.HasPrefix(ident, []byte{0x01, 0xDF}) || bytes.HasPrefix(ident, []byte{0x01, 0xF7}):
   128  		f, err := xcoff.NewFile(r)
   129  		if err != nil {
   130  			return "", "", errUnrecognizedFormat
   131  		}
   132  		x = &xcoffExe{f}
   133  	default:
   134  		return "", "", errUnrecognizedFormat
   135  	}
   136  
   137  	// Read the first 64kB of dataAddr to find the build info blob.
   138  	// On some platforms, the blob will be in its own section, and DataStart
   139  	// returns the address of that section. On others, it's somewhere in the
   140  	// data segment; the linker puts it near the beginning.
   141  	// See cmd/link/internal/ld.Link.buildinfo.
   142  	dataAddr := x.DataStart()
   143  	data, err := x.ReadData(dataAddr, 64*1024)
   144  	if err != nil {
   145  		return "", "", err
   146  	}
   147  	const (
   148  		buildInfoAlign = 16
   149  		buildInfoSize  = 32
   150  	)
   151  	for {
   152  		i := bytes.Index(data, buildInfoMagic)
   153  		if i < 0 || len(data)-i < buildInfoSize {
   154  			return "", "", errNotGoExe
   155  		}
   156  		if i%buildInfoAlign == 0 && len(data)-i >= buildInfoSize {
   157  			data = data[i:]
   158  			break
   159  		}
   160  		data = data[(i+buildInfoAlign-1)&^buildInfoAlign:]
   161  	}
   162  
   163  	// Decode the blob.
   164  	// The first 14 bytes are buildInfoMagic.
   165  	// The next two bytes indicate pointer size in bytes (4 or 8) and endianness
   166  	// (0 for little, 1 for big).
   167  	// Two virtual addresses to Go strings follow that: runtime.buildVersion,
   168  	// and runtime.modinfo.
   169  	// On 32-bit platforms, the last 8 bytes are unused.
   170  	// If the endianness has the 2 bit set, then the pointers are zero
   171  	// and the 32-byte header is followed by varint-prefixed string data
   172  	// for the two string values we care about.
   173  	ptrSize := int(data[14])
   174  	if data[15]&2 != 0 {
   175  		vers, data = decodeString(data[32:])
   176  		mod, data = decodeString(data)
   177  	} else {
   178  		bigEndian := data[15] != 0
   179  		var bo binary.ByteOrder
   180  		if bigEndian {
   181  			bo = binary.BigEndian
   182  		} else {
   183  			bo = binary.LittleEndian
   184  		}
   185  		var readPtr func([]byte) uint64
   186  		if ptrSize == 4 {
   187  			readPtr = func(b []byte) uint64 { return uint64(bo.Uint32(b)) }
   188  		} else {
   189  			readPtr = bo.Uint64
   190  		}
   191  		vers = readString(x, ptrSize, readPtr, readPtr(data[16:]))
   192  		mod = readString(x, ptrSize, readPtr, readPtr(data[16+ptrSize:]))
   193  	}
   194  	if vers == "" {
   195  		return "", "", errNotGoExe
   196  	}
   197  	if len(mod) >= 33 && mod[len(mod)-17] == '\n' {
   198  		// Strip module framing: sentinel strings delimiting the module info.
   199  		// These are cmd/go/internal/modload.infoStart and infoEnd.
   200  		mod = mod[16 : len(mod)-16]
   201  	} else {
   202  		mod = ""
   203  	}
   204  
   205  	return vers, mod, nil
   206  }
   207  
   208  func decodeString(data []byte) (s string, rest []byte) {
   209  	u, n := binary.Uvarint(data)
   210  	if n <= 0 || u >= uint64(len(data)-n) {
   211  		return "", nil
   212  	}
   213  	return string(data[n : uint64(n)+u]), data[uint64(n)+u:]
   214  }
   215  
   216  // readString returns the string at address addr in the executable x.
   217  func readString(x exe, ptrSize int, readPtr func([]byte) uint64, addr uint64) string {
   218  	hdr, err := x.ReadData(addr, uint64(2*ptrSize))
   219  	if err != nil || len(hdr) < 2*ptrSize {
   220  		return ""
   221  	}
   222  	dataAddr := readPtr(hdr)
   223  	dataLen := readPtr(hdr[ptrSize:])
   224  	data, err := x.ReadData(dataAddr, dataLen)
   225  	if err != nil || uint64(len(data)) < dataLen {
   226  		return ""
   227  	}
   228  	return string(data)
   229  }
   230  
   231  // elfExe is the ELF implementation of the exe interface.
   232  type elfExe struct {
   233  	f *elf.File
   234  }
   235  
   236  func (x *elfExe) ReadData(addr, size uint64) ([]byte, error) {
   237  	for _, prog := range x.f.Progs {
   238  		if prog.Vaddr <= addr && addr <= prog.Vaddr+prog.Filesz-1 {
   239  			n := prog.Vaddr + prog.Filesz - addr
   240  			if n > size {
   241  				n = size
   242  			}
   243  			data := make([]byte, n)
   244  			_, err := prog.ReadAt(data, int64(addr-prog.Vaddr))
   245  			if err != nil {
   246  				return nil, err
   247  			}
   248  			return data, nil
   249  		}
   250  	}
   251  	return nil, errUnrecognizedFormat
   252  }
   253  
   254  func (x *elfExe) DataStart() uint64 {
   255  	for _, s := range x.f.Sections {
   256  		if s.Name == ".go.buildinfo" {
   257  			return s.Addr
   258  		}
   259  	}
   260  	for _, p := range x.f.Progs {
   261  		if p.Type == elf.PT_LOAD && p.Flags&(elf.PF_X|elf.PF_W) == elf.PF_W {
   262  			return p.Vaddr
   263  		}
   264  	}
   265  	return 0
   266  }
   267  
   268  // peExe is the PE (Windows Portable Executable) implementation of the exe interface.
   269  type peExe struct {
   270  	f *pe.File
   271  }
   272  
   273  func (x *peExe) imageBase() uint64 {
   274  	switch oh := x.f.OptionalHeader.(type) {
   275  	case *pe.OptionalHeader32:
   276  		return uint64(oh.ImageBase)
   277  	case *pe.OptionalHeader64:
   278  		return oh.ImageBase
   279  	}
   280  	return 0
   281  }
   282  
   283  func (x *peExe) ReadData(addr, size uint64) ([]byte, error) {
   284  	addr -= x.imageBase()
   285  	for _, sect := range x.f.Sections {
   286  		if uint64(sect.VirtualAddress) <= addr && addr <= uint64(sect.VirtualAddress+sect.Size-1) {
   287  			n := uint64(sect.VirtualAddress+sect.Size) - addr
   288  			if n > size {
   289  				n = size
   290  			}
   291  			data := make([]byte, n)
   292  			_, err := sect.ReadAt(data, int64(addr-uint64(sect.VirtualAddress)))
   293  			if err != nil {
   294  				return nil, errUnrecognizedFormat
   295  			}
   296  			return data, nil
   297  		}
   298  	}
   299  	return nil, errUnrecognizedFormat
   300  }
   301  
   302  func (x *peExe) DataStart() uint64 {
   303  	// Assume data is first writable section.
   304  	const (
   305  		IMAGE_SCN_CNT_CODE               = 0x00000020
   306  		IMAGE_SCN_CNT_INITIALIZED_DATA   = 0x00000040
   307  		IMAGE_SCN_CNT_UNINITIALIZED_DATA = 0x00000080
   308  		IMAGE_SCN_MEM_EXECUTE            = 0x20000000
   309  		IMAGE_SCN_MEM_READ               = 0x40000000
   310  		IMAGE_SCN_MEM_WRITE              = 0x80000000
   311  		IMAGE_SCN_MEM_DISCARDABLE        = 0x2000000
   312  		IMAGE_SCN_LNK_NRELOC_OVFL        = 0x1000000
   313  		IMAGE_SCN_ALIGN_32BYTES          = 0x600000
   314  	)
   315  	for _, sect := range x.f.Sections {
   316  		if sect.VirtualAddress != 0 && sect.Size != 0 &&
   317  			sect.Characteristics&^IMAGE_SCN_ALIGN_32BYTES == IMAGE_SCN_CNT_INITIALIZED_DATA|IMAGE_SCN_MEM_READ|IMAGE_SCN_MEM_WRITE {
   318  			return uint64(sect.VirtualAddress) + x.imageBase()
   319  		}
   320  	}
   321  	return 0
   322  }
   323  
   324  // machoExe is the Mach-O (Apple macOS/iOS) implementation of the exe interface.
   325  type machoExe struct {
   326  	f *macho.File
   327  }
   328  
   329  func (x *machoExe) ReadData(addr, size uint64) ([]byte, error) {
   330  	for _, load := range x.f.Loads {
   331  		seg, ok := load.(*macho.Segment)
   332  		if !ok {
   333  			continue
   334  		}
   335  		if seg.Addr <= addr && addr <= seg.Addr+seg.Filesz-1 {
   336  			if seg.Name == "__PAGEZERO" {
   337  				continue
   338  			}
   339  			n := seg.Addr + seg.Filesz - addr
   340  			if n > size {
   341  				n = size
   342  			}
   343  			data := make([]byte, n)
   344  			_, err := seg.ReadAt(data, int64(addr-seg.Addr))
   345  			if err != nil {
   346  				return nil, err
   347  			}
   348  			return data, nil
   349  		}
   350  	}
   351  	return nil, errUnrecognizedFormat
   352  }
   353  
   354  func (x *machoExe) DataStart() uint64 {
   355  	// Look for section named "__go_buildinfo".
   356  	for _, sec := range x.f.Sections {
   357  		if sec.Name == "__go_buildinfo" {
   358  			return sec.Addr
   359  		}
   360  	}
   361  	// Try the first non-empty writable segment.
   362  	const RW = 3
   363  	for _, load := range x.f.Loads {
   364  		seg, ok := load.(*macho.Segment)
   365  		if ok && seg.Addr != 0 && seg.Filesz != 0 && seg.Prot == RW && seg.Maxprot == RW {
   366  			return seg.Addr
   367  		}
   368  	}
   369  	return 0
   370  }
   371  
   372  // xcoffExe is the XCOFF (AIX eXtended COFF) implementation of the exe interface.
   373  type xcoffExe struct {
   374  	f *xcoff.File
   375  }
   376  
   377  func (x *xcoffExe) ReadData(addr, size uint64) ([]byte, error) {
   378  	for _, sect := range x.f.Sections {
   379  		if uint64(sect.VirtualAddress) <= addr && addr <= uint64(sect.VirtualAddress+sect.Size-1) {
   380  			n := uint64(sect.VirtualAddress+sect.Size) - addr
   381  			if n > size {
   382  				n = size
   383  			}
   384  			data := make([]byte, n)
   385  			_, err := sect.ReadAt(data, int64(addr-uint64(sect.VirtualAddress)))
   386  			if err != nil {
   387  				return nil, err
   388  			}
   389  			return data, nil
   390  		}
   391  	}
   392  	return nil, fmt.Errorf("address not mapped")
   393  }
   394  
   395  func (x *xcoffExe) DataStart() uint64 {
   396  	return x.f.SectionByType(xcoff.STYP_DATA).VirtualAddress
   397  }
   398
View as plain text