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  	"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  	_ "unsafe" // for linkname
    29  )
    30  
    31  // Type alias for build info. We cannot move the types here, since
    32  // runtime/debug would need to import this package, which would make it
    33  // a much larger dependency.
    34  type BuildInfo = debug.BuildInfo
    35  
    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  var 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  //
    44  // errNotGoExe should be an internal detail,
    45  // but widely used packages access it using linkname.
    46  // Notable members of the hall of shame include:
    47  //   - github.com/quay/claircore
    48  //
    49  // Do not remove or change the type signature.
    50  // See go.dev/issue/67401.
    51  //
    52  //go:linkname errNotGoExe
    53  var errNotGoExe = errors.New("not a Go executable")
    54  
    55  // The build info blob left by the linker is identified by
    56  // a 16-byte header, consisting of buildInfoMagic (14 bytes),
    57  // the binary's pointer size (1 byte),
    58  // and whether the binary is big endian (1 byte).
    59  var buildInfoMagic = []byte("\xff Go buildinf:")
    60  
    61  // ReadFile returns build information embedded in a Go binary
    62  // file at the given path. Most information is only available for binaries built
    63  // with module support.
    64  func ReadFile(name string) (info *BuildInfo, err error) {
    65  	defer func() {
    66  		if pathErr := (*fs.PathError)(nil); errors.As(err, &pathErr) {
    67  			err = fmt.Errorf("could not read Go build info: %w", err)
    68  		} else if err != nil {
    69  			err = fmt.Errorf("could not read Go build info from %s: %w", name, err)
    70  		}
    71  	}()
    72  
    73  	f, err := os.Open(name)
    74  	if err != nil {
    75  		return nil, err
    76  	}
    77  	defer f.Close()
    78  	return Read(f)
    79  }
    80  
    81  // Read returns build information embedded in a Go binary file
    82  // accessed through the given ReaderAt. Most information is only available for
    83  // binaries built with module support.
    84  func Read(r io.ReaderAt) (*BuildInfo, error) {
    85  	vers, mod, err := readRawBuildInfo(r)
    86  	if err != nil {
    87  		return nil, err
    88  	}
    89  	bi, err := debug.ParseBuildInfo(mod)
    90  	if err != nil {
    91  		return nil, err
    92  	}
    93  	bi.GoVersion = vers
    94  	return bi, nil
    95  }
    96  
    97  type exe interface {
    98  	// ReadData reads and returns up to size bytes starting at virtual address addr.
    99  	ReadData(addr, size uint64) ([]byte, error)
   100  
   101  	// DataStart returns the virtual address and size of the segment or section that
   102  	// should contain build information. This is either a specially named section
   103  	// or the first writable non-zero data segment.
   104  	DataStart() (uint64, uint64)
   105  }
   106  
   107  // readRawBuildInfo extracts the Go toolchain version and module information
   108  // strings from a Go binary. On success, vers should be non-empty. mod
   109  // is empty if the binary was not built with modules enabled.
   110  func readRawBuildInfo(r io.ReaderAt) (vers, mod string, err error) {
   111  	// Read the first bytes of the file to identify the format, then delegate to
   112  	// a format-specific function to load segment and section headers.
   113  	ident := make([]byte, 16)
   114  	if n, err := r.ReadAt(ident, 0); n < len(ident) || err != nil {
   115  		return "", "", errUnrecognizedFormat
   116  	}
   117  
   118  	var x exe
   119  	switch {
   120  	case bytes.HasPrefix(ident, []byte("\x7FELF")):
   121  		f, err := elf.NewFile(r)
   122  		if err != nil {
   123  			return "", "", errUnrecognizedFormat
   124  		}
   125  		x = &elfExe{f}
   126  	case bytes.HasPrefix(ident, []byte("MZ")):
   127  		f, err := pe.NewFile(r)
   128  		if err != nil {
   129  			return "", "", errUnrecognizedFormat
   130  		}
   131  		x = &peExe{f}
   132  	case bytes.HasPrefix(ident, []byte("\xFE\xED\xFA")) || bytes.HasPrefix(ident[1:], []byte("\xFA\xED\xFE")):
   133  		f, err := macho.NewFile(r)
   134  		if err != nil {
   135  			return "", "", errUnrecognizedFormat
   136  		}
   137  		x = &machoExe{f}
   138  	case bytes.HasPrefix(ident, []byte("\xCA\xFE\xBA\xBE")) || bytes.HasPrefix(ident, []byte("\xCA\xFE\xBA\xBF")):
   139  		f, err := macho.NewFatFile(r)
   140  		if err != nil || len(f.Arches) == 0 {
   141  			return "", "", errUnrecognizedFormat
   142  		}
   143  		x = &machoExe{f.Arches[0].File}
   144  	case bytes.HasPrefix(ident, []byte{0x01, 0xDF}) || bytes.HasPrefix(ident, []byte{0x01, 0xF7}):
   145  		f, err := xcoff.NewFile(r)
   146  		if err != nil {
   147  			return "", "", errUnrecognizedFormat
   148  		}
   149  		x = &xcoffExe{f}
   150  	case hasPlan9Magic(ident):
   151  		f, err := plan9obj.NewFile(r)
   152  		if err != nil {
   153  			return "", "", errUnrecognizedFormat
   154  		}
   155  		x = &plan9objExe{f}
   156  	default:
   157  		return "", "", errUnrecognizedFormat
   158  	}
   159  
   160  	// Read segment or section to find the build info blob.
   161  	// On some platforms, the blob will be in its own section, and DataStart
   162  	// returns the address of that section. On others, it's somewhere in the
   163  	// data segment; the linker puts it near the beginning.
   164  	// See cmd/link/internal/ld.Link.buildinfo.
   165  	dataAddr, dataSize := x.DataStart()
   166  	if dataSize == 0 {
   167  		return "", "", errNotGoExe
   168  	}
   169  	data, err := x.ReadData(dataAddr, dataSize)
   170  	if err != nil {
   171  		return "", "", err
   172  	}
   173  	const (
   174  		buildInfoAlign = 16
   175  		buildInfoSize  = 32
   176  	)
   177  	for {
   178  		i := bytes.Index(data, buildInfoMagic)
   179  		if i < 0 || len(data)-i < buildInfoSize {
   180  			return "", "", errNotGoExe
   181  		}
   182  		if i%buildInfoAlign == 0 && len(data)-i >= buildInfoSize {
   183  			data = data[i:]
   184  			break
   185  		}
   186  		data = data[(i+buildInfoAlign-1)&^(buildInfoAlign-1):]
   187  	}
   188  
   189  	// Decode the blob.
   190  	// The first 14 bytes are buildInfoMagic.
   191  	// The next two bytes indicate pointer size in bytes (4 or 8) and endianness
   192  	// (0 for little, 1 for big).
   193  	// Two virtual addresses to Go strings follow that: runtime.buildVersion,
   194  	// and runtime.modinfo.
   195  	// On 32-bit platforms, the last 8 bytes are unused.
   196  	// If the endianness has the 2 bit set, then the pointers are zero
   197  	// and the 32-byte header is followed by varint-prefixed string data
   198  	// for the two string values we care about.
   199  	ptrSize := int(data[14])
   200  	if data[15]&2 != 0 {
   201  		vers, data = decodeString(data[32:])
   202  		mod, data = decodeString(data)
   203  	} else {
   204  		bigEndian := data[15] != 0
   205  		var bo binary.ByteOrder
   206  		if bigEndian {
   207  			bo = binary.BigEndian
   208  		} else {
   209  			bo = binary.LittleEndian
   210  		}
   211  		var readPtr func([]byte) uint64
   212  		if ptrSize == 4 {
   213  			readPtr = func(b []byte) uint64 { return uint64(bo.Uint32(b)) }
   214  		} else if ptrSize == 8 {
   215  			readPtr = bo.Uint64
   216  		} else {
   217  			return "", "", errNotGoExe
   218  		}
   219  		vers = readString(x, ptrSize, readPtr, readPtr(data[16:]))
   220  		mod = readString(x, ptrSize, readPtr, readPtr(data[16+ptrSize:]))
   221  	}
   222  	if vers == "" {
   223  		return "", "", errNotGoExe
   224  	}
   225  	if len(mod) >= 33 && mod[len(mod)-17] == '\n' {
   226  		// Strip module framing: sentinel strings delimiting the module info.
   227  		// These are cmd/go/internal/modload.infoStart and infoEnd.
   228  		mod = mod[16 : len(mod)-16]
   229  	} else {
   230  		mod = ""
   231  	}
   232  
   233  	return vers, mod, nil
   234  }
   235  
   236  func hasPlan9Magic(magic []byte) bool {
   237  	if len(magic) >= 4 {
   238  		m := binary.BigEndian.Uint32(magic)
   239  		switch m {
   240  		case plan9obj.Magic386, plan9obj.MagicAMD64, plan9obj.MagicARM:
   241  			return true
   242  		}
   243  	}
   244  	return false
   245  }
   246  
   247  func decodeString(data []byte) (s string, rest []byte) {
   248  	u, n := binary.Uvarint(data)
   249  	if n <= 0 || u > uint64(len(data)-n) {
   250  		return "", nil
   251  	}
   252  	return string(data[n : uint64(n)+u]), data[uint64(n)+u:]
   253  }
   254  
   255  // readString returns the string at address addr in the executable x.
   256  func readString(x exe, ptrSize int, readPtr func([]byte) uint64, addr uint64) string {
   257  	hdr, err := x.ReadData(addr, uint64(2*ptrSize))
   258  	if err != nil || len(hdr) < 2*ptrSize {
   259  		return ""
   260  	}
   261  	dataAddr := readPtr(hdr)
   262  	dataLen := readPtr(hdr[ptrSize:])
   263  	data, err := x.ReadData(dataAddr, dataLen)
   264  	if err != nil || uint64(len(data)) < dataLen {
   265  		return ""
   266  	}
   267  	return string(data)
   268  }
   269  
   270  // elfExe is the ELF implementation of the exe interface.
   271  type elfExe struct {
   272  	f *elf.File
   273  }
   274  
   275  func (x *elfExe) ReadData(addr, size uint64) ([]byte, error) {
   276  	for _, prog := range x.f.Progs {
   277  		if prog.Vaddr <= addr && addr <= prog.Vaddr+prog.Filesz-1 {
   278  			n := prog.Vaddr + prog.Filesz - addr
   279  			if n > size {
   280  				n = size
   281  			}
   282  			return saferio.ReadDataAt(prog, n, int64(addr-prog.Vaddr))
   283  		}
   284  	}
   285  	return nil, errUnrecognizedFormat
   286  }
   287  
   288  func (x *elfExe) DataStart() (uint64, uint64) {
   289  	for _, s := range x.f.Sections {
   290  		if s.Name == ".go.buildinfo" {
   291  			return s.Addr, s.Size
   292  		}
   293  	}
   294  	for _, p := range x.f.Progs {
   295  		if p.Type == elf.PT_LOAD && p.Flags&(elf.PF_X|elf.PF_W) == elf.PF_W {
   296  			return p.Vaddr, p.Memsz
   297  		}
   298  	}
   299  	return 0, 0
   300  }
   301  
   302  // peExe is the PE (Windows Portable Executable) implementation of the exe interface.
   303  type peExe struct {
   304  	f *pe.File
   305  }
   306  
   307  func (x *peExe) imageBase() uint64 {
   308  	switch oh := x.f.OptionalHeader.(type) {
   309  	case *pe.OptionalHeader32:
   310  		return uint64(oh.ImageBase)
   311  	case *pe.OptionalHeader64:
   312  		return oh.ImageBase
   313  	}
   314  	return 0
   315  }
   316  
   317  func (x *peExe) ReadData(addr, size uint64) ([]byte, error) {
   318  	addr -= x.imageBase()
   319  	for _, sect := range x.f.Sections {
   320  		if uint64(sect.VirtualAddress) <= addr && addr <= uint64(sect.VirtualAddress+sect.Size-1) {
   321  			n := uint64(sect.VirtualAddress+sect.Size) - addr
   322  			if n > size {
   323  				n = size
   324  			}
   325  			return saferio.ReadDataAt(sect, n, int64(addr-uint64(sect.VirtualAddress)))
   326  		}
   327  	}
   328  	return nil, errUnrecognizedFormat
   329  }
   330  
   331  func (x *peExe) DataStart() (uint64, uint64) {
   332  	// Assume data is first writable section.
   333  	const (
   334  		IMAGE_SCN_CNT_CODE               = 0x00000020
   335  		IMAGE_SCN_CNT_INITIALIZED_DATA   = 0x00000040
   336  		IMAGE_SCN_CNT_UNINITIALIZED_DATA = 0x00000080
   337  		IMAGE_SCN_MEM_EXECUTE            = 0x20000000
   338  		IMAGE_SCN_MEM_READ               = 0x40000000
   339  		IMAGE_SCN_MEM_WRITE              = 0x80000000
   340  		IMAGE_SCN_MEM_DISCARDABLE        = 0x2000000
   341  		IMAGE_SCN_LNK_NRELOC_OVFL        = 0x1000000
   342  		IMAGE_SCN_ALIGN_32BYTES          = 0x600000
   343  	)
   344  	for _, sect := range x.f.Sections {
   345  		if sect.VirtualAddress != 0 && sect.Size != 0 &&
   346  			sect.Characteristics&^IMAGE_SCN_ALIGN_32BYTES == IMAGE_SCN_CNT_INITIALIZED_DATA|IMAGE_SCN_MEM_READ|IMAGE_SCN_MEM_WRITE {
   347  			return uint64(sect.VirtualAddress) + x.imageBase(), uint64(sect.VirtualSize)
   348  		}
   349  	}
   350  	return 0, 0
   351  }
   352  
   353  // machoExe is the Mach-O (Apple macOS/iOS) implementation of the exe interface.
   354  type machoExe struct {
   355  	f *macho.File
   356  }
   357  
   358  func (x *machoExe) ReadData(addr, size uint64) ([]byte, error) {
   359  	for _, load := range x.f.Loads {
   360  		seg, ok := load.(*macho.Segment)
   361  		if !ok {
   362  			continue
   363  		}
   364  		if seg.Addr <= addr && addr <= seg.Addr+seg.Filesz-1 {
   365  			if seg.Name == "__PAGEZERO" {
   366  				continue
   367  			}
   368  			n := seg.Addr + seg.Filesz - addr
   369  			if n > size {
   370  				n = size
   371  			}
   372  			return saferio.ReadDataAt(seg, n, int64(addr-seg.Addr))
   373  		}
   374  	}
   375  	return nil, errUnrecognizedFormat
   376  }
   377  
   378  func (x *machoExe) DataStart() (uint64, uint64) {
   379  	// Look for section named "__go_buildinfo".
   380  	for _, sec := range x.f.Sections {
   381  		if sec.Name == "__go_buildinfo" {
   382  			return sec.Addr, sec.Size
   383  		}
   384  	}
   385  	// Try the first non-empty writable segment.
   386  	const RW = 3
   387  	for _, load := range x.f.Loads {
   388  		seg, ok := load.(*macho.Segment)
   389  		if ok && seg.Addr != 0 && seg.Filesz != 0 && seg.Prot == RW && seg.Maxprot == RW {
   390  			return seg.Addr, seg.Memsz
   391  		}
   392  	}
   393  	return 0, 0
   394  }
   395  
   396  // xcoffExe is the XCOFF (AIX eXtended COFF) implementation of the exe interface.
   397  type xcoffExe struct {
   398  	f *xcoff.File
   399  }
   400  
   401  func (x *xcoffExe) ReadData(addr, size uint64) ([]byte, error) {
   402  	for _, sect := range x.f.Sections {
   403  		if sect.VirtualAddress <= addr && addr <= sect.VirtualAddress+sect.Size-1 {
   404  			n := sect.VirtualAddress + sect.Size - addr
   405  			if n > size {
   406  				n = size
   407  			}
   408  			return saferio.ReadDataAt(sect, n, int64(addr-sect.VirtualAddress))
   409  		}
   410  	}
   411  	return nil, errors.New("address not mapped")
   412  }
   413  
   414  func (x *xcoffExe) DataStart() (uint64, uint64) {
   415  	if s := x.f.SectionByType(xcoff.STYP_DATA); s != nil {
   416  		return s.VirtualAddress, s.Size
   417  	}
   418  	return 0, 0
   419  }
   420  
   421  // plan9objExe is the Plan 9 a.out implementation of the exe interface.
   422  type plan9objExe struct {
   423  	f *plan9obj.File
   424  }
   425  
   426  func (x *plan9objExe) DataStart() (uint64, uint64) {
   427  	if s := x.f.Section("data"); s != nil {
   428  		return uint64(s.Offset), uint64(s.Size)
   429  	}
   430  	return 0, 0
   431  }
   432  
   433  func (x *plan9objExe) ReadData(addr, size uint64) ([]byte, error) {
   434  	for _, sect := range x.f.Sections {
   435  		if uint64(sect.Offset) <= addr && addr <= uint64(sect.Offset+sect.Size-1) {
   436  			n := uint64(sect.Offset+sect.Size) - addr
   437  			if n > size {
   438  				n = size
   439  			}
   440  			return saferio.ReadDataAt(sect, n, int64(addr-uint64(sect.Offset)))
   441  		}
   442  	}
   443  	return nil, errors.New("address not mapped")
   444  }
   445
View as plain text