...
Run Format

Source file src/cmd/internal/goobj/read.go

Documentation: cmd/internal/goobj

  // Copyright 2013 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.
  
  // Package goobj implements reading of Go object files and archives.
  //
  // TODO(rsc): Decide where this package should live. (golang.org/issue/6932)
  // TODO(rsc): Decide the appropriate integer types for various fields.
  // TODO(rsc): Write tests. (File format still up in the air a little.)
  package goobj
  
  import (
  	"bufio"
  	"bytes"
  	"cmd/internal/objabi"
  	"errors"
  	"fmt"
  	"io"
  	"strconv"
  	"strings"
  )
  
  // A Sym is a named symbol in an object file.
  type Sym struct {
  	SymID                // symbol identifier (name and version)
  	Kind  objabi.SymKind // kind of symbol
  	DupOK bool           // are duplicate definitions okay?
  	Size  int            // size of corresponding data
  	Type  SymID          // symbol for Go type information
  	Data  Data           // memory image of symbol
  	Reloc []Reloc        // relocations to apply to Data
  	Func  *Func          // additional data for functions
  }
  
  // A SymID - the combination of Name and Version - uniquely identifies
  // a symbol within a package.
  type SymID struct {
  	// Name is the name of a symbol.
  	Name string
  
  	// Version is zero for symbols with global visibility.
  	// Symbols with only file visibility (such as file-level static
  	// declarations in C) have a non-zero version distinguishing
  	// a symbol in one file from a symbol of the same name
  	// in another file
  	Version int
  }
  
  func (s SymID) String() string {
  	if s.Version == 0 {
  		return s.Name
  	}
  	return fmt.Sprintf("%s<%d>", s.Name, s.Version)
  }
  
  // A Data is a reference to data stored in an object file.
  // It records the offset and size of the data, so that a client can
  // read the data only if necessary.
  type Data struct {
  	Offset int64
  	Size   int64
  }
  
  // A Reloc describes a relocation applied to a memory image to refer
  // to an address within a particular symbol.
  type Reloc struct {
  	// The bytes at [Offset, Offset+Size) within the containing Sym
  	// should be updated to refer to the address Add bytes after the start
  	// of the symbol Sym.
  	Offset int
  	Size   int
  	Sym    SymID
  	Add    int
  
  	// The Type records the form of address expected in the bytes
  	// described by the previous fields: absolute, PC-relative, and so on.
  	// TODO(rsc): The interpretation of Type is not exposed by this package.
  	Type objabi.RelocType
  }
  
  // A Var describes a variable in a function stack frame: a declared
  // local variable, an input argument, or an output result.
  type Var struct {
  	// The combination of Name, Kind, and Offset uniquely
  	// identifies a variable in a function stack frame.
  	// Using fewer of these - in particular, using only Name - does not.
  	Name   string // Name of variable.
  	Kind   int    // TODO(rsc): Define meaning.
  	Offset int    // Frame offset. TODO(rsc): Define meaning.
  
  	Type SymID // Go type for variable.
  }
  
  // Func contains additional per-symbol information specific to functions.
  type Func struct {
  	Args     int        // size in bytes of argument frame: inputs and outputs
  	Frame    int        // size in bytes of local variable frame
  	Leaf     bool       // function omits save of link register (ARM)
  	NoSplit  bool       // function omits stack split prologue
  	Var      []Var      // detail about local variables
  	PCSP     Data       // PC → SP offset map
  	PCFile   Data       // PC → file number map (index into File)
  	PCLine   Data       // PC → line number map
  	PCInline Data       // PC → inline tree index map
  	PCData   []Data     // PC → runtime support data map
  	FuncData []FuncData // non-PC-specific runtime support data
  	File     []string   // paths indexed by PCFile
  	InlTree  []InlinedCall
  }
  
  // TODO: Add PCData []byte and PCDataIter (similar to liblink).
  
  // A FuncData is a single function-specific data value.
  type FuncData struct {
  	Sym    SymID // symbol holding data
  	Offset int64 // offset into symbol for funcdata pointer
  }
  
  // An InlinedCall is a node in an InlTree.
  // See cmd/internal/obj.InlTree for details.
  type InlinedCall struct {
  	Parent int
  	File   string
  	Line   int
  	Func   SymID
  }
  
  // A Package is a parsed Go object file or archive defining a Go package.
  type Package struct {
  	ImportPath string   // import path denoting this package
  	Imports    []string // packages imported by this package
  	SymRefs    []SymID  // list of symbol names and versions referred to by this pack
  	Syms       []*Sym   // symbols defined by this package
  	MaxVersion int      // maximum Version in any SymID in Syms
  	Arch       string   // architecture
  }
  
  var (
  	archiveHeader = []byte("!<arch>\n")
  	archiveMagic  = []byte("`\n")
  	goobjHeader   = []byte("go objec") // truncated to size of archiveHeader
  
  	errCorruptArchive   = errors.New("corrupt archive")
  	errTruncatedArchive = errors.New("truncated archive")
  	errCorruptObject    = errors.New("corrupt object file")
  	errNotObject        = errors.New("unrecognized object file format")
  )
  
  // An objReader is an object file reader.
  type objReader struct {
  	p          *Package
  	b          *bufio.Reader
  	f          io.ReadSeeker
  	err        error
  	offset     int64
  	dataOffset int64
  	limit      int64
  	tmp        [256]byte
  	pkgprefix  string
  }
  
  // init initializes r to read package p from f.
  func (r *objReader) init(f io.ReadSeeker, p *Package) {
  	r.f = f
  	r.p = p
  	r.offset, _ = f.Seek(0, io.SeekCurrent)
  	r.limit, _ = f.Seek(0, io.SeekEnd)
  	f.Seek(r.offset, io.SeekStart)
  	r.b = bufio.NewReader(f)
  	r.pkgprefix = objabi.PathToPrefix(p.ImportPath) + "."
  }
  
  // error records that an error occurred.
  // It returns only the first error, so that an error
  // caused by an earlier error does not discard information
  // about the earlier error.
  func (r *objReader) error(err error) error {
  	if r.err == nil {
  		if err == io.EOF {
  			err = io.ErrUnexpectedEOF
  		}
  		r.err = err
  	}
  	// panic("corrupt") // useful for debugging
  	return r.err
  }
  
  // readByte reads and returns a byte from the input file.
  // On I/O error or EOF, it records the error but returns byte 0.
  // A sequence of 0 bytes will eventually terminate any
  // parsing state in the object file. In particular, it ends the
  // reading of a varint.
  func (r *objReader) readByte() byte {
  	if r.err != nil {
  		return 0
  	}
  	if r.offset >= r.limit {
  		r.error(io.ErrUnexpectedEOF)
  		return 0
  	}
  	b, err := r.b.ReadByte()
  	if err != nil {
  		if err == io.EOF {
  			err = io.ErrUnexpectedEOF
  		}
  		r.error(err)
  		b = 0
  	} else {
  		r.offset++
  	}
  	return b
  }
  
  // read reads exactly len(b) bytes from the input file.
  // If an error occurs, read returns the error but also
  // records it, so it is safe for callers to ignore the result
  // as long as delaying the report is not a problem.
  func (r *objReader) readFull(b []byte) error {
  	if r.err != nil {
  		return r.err
  	}
  	if r.offset+int64(len(b)) > r.limit {
  		return r.error(io.ErrUnexpectedEOF)
  	}
  	n, err := io.ReadFull(r.b, b)
  	r.offset += int64(n)
  	if err != nil {
  		return r.error(err)
  	}
  	return nil
  }
  
  // readInt reads a zigzag varint from the input file.
  func (r *objReader) readInt() int {
  	var u uint64
  
  	for shift := uint(0); ; shift += 7 {
  		if shift >= 64 {
  			r.error(errCorruptObject)
  			return 0
  		}
  		c := r.readByte()
  		u |= uint64(c&0x7F) << shift
  		if c&0x80 == 0 {
  			break
  		}
  	}
  
  	v := int64(u>>1) ^ (int64(u) << 63 >> 63)
  	if int64(int(v)) != v {
  		r.error(errCorruptObject) // TODO
  		return 0
  	}
  	return int(v)
  }
  
  // readString reads a length-delimited string from the input file.
  func (r *objReader) readString() string {
  	n := r.readInt()
  	buf := make([]byte, n)
  	r.readFull(buf)
  	return string(buf)
  }
  
  // readSymID reads a SymID from the input file.
  func (r *objReader) readSymID() SymID {
  	i := r.readInt()
  	return r.p.SymRefs[i]
  }
  
  func (r *objReader) readRef() {
  	name, vers := r.readString(), r.readInt()
  
  	// In a symbol name in an object file, "". denotes the
  	// prefix for the package in which the object file has been found.
  	// Expand it.
  	name = strings.Replace(name, `"".`, r.pkgprefix, -1)
  
  	// An individual object file only records version 0 (extern) or 1 (static).
  	// To make static symbols unique across all files being read, we
  	// replace version 1 with the version corresponding to the current
  	// file number. The number is incremented on each call to parseObject.
  	if vers != 0 {
  		vers = r.p.MaxVersion
  	}
  	r.p.SymRefs = append(r.p.SymRefs, SymID{name, vers})
  }
  
  // readData reads a data reference from the input file.
  func (r *objReader) readData() Data {
  	n := r.readInt()
  	d := Data{Offset: r.dataOffset, Size: int64(n)}
  	r.dataOffset += int64(n)
  	return d
  }
  
  // skip skips n bytes in the input.
  func (r *objReader) skip(n int64) {
  	if n < 0 {
  		r.error(fmt.Errorf("debug/goobj: internal error: misuse of skip"))
  	}
  	if n < int64(len(r.tmp)) {
  		// Since the data is so small, a just reading from the buffered
  		// reader is better than flushing the buffer and seeking.
  		r.readFull(r.tmp[:n])
  	} else if n <= int64(r.b.Buffered()) {
  		// Even though the data is not small, it has already been read.
  		// Advance the buffer instead of seeking.
  		for n > int64(len(r.tmp)) {
  			r.readFull(r.tmp[:])
  			n -= int64(len(r.tmp))
  		}
  		r.readFull(r.tmp[:n])
  	} else {
  		// Seek, giving up buffered data.
  		_, err := r.f.Seek(r.offset+n, io.SeekStart)
  		if err != nil {
  			r.error(err)
  		}
  		r.offset += n
  		r.b.Reset(r.f)
  	}
  }
  
  // Parse parses an object file or archive from r,
  // assuming that its import path is pkgpath.
  func Parse(r io.ReadSeeker, pkgpath string) (*Package, error) {
  	if pkgpath == "" {
  		pkgpath = `""`
  	}
  	p := new(Package)
  	p.ImportPath = pkgpath
  
  	var rd objReader
  	rd.init(r, p)
  	err := rd.readFull(rd.tmp[:8])
  	if err != nil {
  		if err == io.EOF {
  			err = io.ErrUnexpectedEOF
  		}
  		return nil, err
  	}
  
  	switch {
  	default:
  		return nil, errNotObject
  
  	case bytes.Equal(rd.tmp[:8], archiveHeader):
  		if err := rd.parseArchive(); err != nil {
  			return nil, err
  		}
  	case bytes.Equal(rd.tmp[:8], goobjHeader):
  		if err := rd.parseObject(goobjHeader); err != nil {
  			return nil, err
  		}
  	}
  
  	return p, nil
  }
  
  // trimSpace removes trailing spaces from b and returns the corresponding string.
  // This effectively parses the form used in archive headers.
  func trimSpace(b []byte) string {
  	return string(bytes.TrimRight(b, " "))
  }
  
  // parseArchive parses a Unix archive of Go object files.
  // TODO(rsc): Need to skip non-Go object files.
  // TODO(rsc): Maybe record table of contents in r.p so that
  // linker can avoid having code to parse archives too.
  func (r *objReader) parseArchive() error {
  	for r.offset < r.limit {
  		if err := r.readFull(r.tmp[:60]); err != nil {
  			return err
  		}
  		data := r.tmp[:60]
  
  		// Each file is preceded by this text header (slice indices in first column):
  		//	 0:16	name
  		//	16:28 date
  		//	28:34 uid
  		//	34:40 gid
  		//	40:48 mode
  		//	48:58 size
  		//	58:60 magic - `\n
  		// We only care about name, size, and magic.
  		// The fields are space-padded on the right.
  		// The size is in decimal.
  		// The file data - size bytes - follows the header.
  		// Headers are 2-byte aligned, so if size is odd, an extra padding
  		// byte sits between the file data and the next header.
  		// The file data that follows is padded to an even number of bytes:
  		// if size is odd, an extra padding byte is inserted betw the next header.
  		if len(data) < 60 {
  			return errTruncatedArchive
  		}
  		if !bytes.Equal(data[58:60], archiveMagic) {
  			return errCorruptArchive
  		}
  		name := trimSpace(data[0:16])
  		size, err := strconv.ParseInt(trimSpace(data[48:58]), 10, 64)
  		if err != nil {
  			return errCorruptArchive
  		}
  		data = data[60:]
  		fsize := size + size&1
  		if fsize < 0 || fsize < size {
  			return errCorruptArchive
  		}
  		switch name {
  		case "__.PKGDEF":
  			r.skip(size)
  		default:
  			oldLimit := r.limit
  			r.limit = r.offset + size
  			if err := r.parseObject(nil); err != nil {
  				return fmt.Errorf("parsing archive member %q: %v", name, err)
  			}
  			r.skip(r.limit - r.offset)
  			r.limit = oldLimit
  		}
  		if size&1 != 0 {
  			r.skip(1)
  		}
  	}
  	return nil
  }
  
  // parseObject parses a single Go object file.
  // The prefix is the bytes already read from the file,
  // typically in order to detect that this is an object file.
  // The object file consists of a textual header ending in "\n!\n"
  // and then the part we want to parse begins.
  // The format of that part is defined in a comment at the top
  // of src/liblink/objfile.c.
  func (r *objReader) parseObject(prefix []byte) error {
  	r.p.MaxVersion++
  	h := make([]byte, 0, 256)
  	h = append(h, prefix...)
  	var c1, c2, c3 byte
  	for {
  		c1, c2, c3 = c2, c3, r.readByte()
  		h = append(h, c3)
  		// The new export format can contain 0 bytes.
  		// Don't consider them errors, only look for r.err != nil.
  		if r.err != nil {
  			return errCorruptObject
  		}
  		if c1 == '\n' && c2 == '!' && c3 == '\n' {
  			break
  		}
  	}
  
  	hs := strings.Fields(string(h))
  	if len(hs) >= 4 {
  		r.p.Arch = hs[3]
  	}
  	// TODO: extract OS + build ID if/when we need it
  
  	r.readFull(r.tmp[:8])
  	if !bytes.Equal(r.tmp[:8], []byte("\x00\x00go19ld")) {
  		return r.error(errCorruptObject)
  	}
  
  	b := r.readByte()
  	if b != 1 {
  		return r.error(errCorruptObject)
  	}
  
  	// Direct package dependencies.
  	for {
  		s := r.readString()
  		if s == "" {
  			break
  		}
  		r.p.Imports = append(r.p.Imports, s)
  	}
  
  	r.p.SymRefs = []SymID{{"", 0}}
  	for {
  		if b := r.readByte(); b != 0xfe {
  			if b != 0xff {
  				return r.error(errCorruptObject)
  			}
  			break
  		}
  
  		r.readRef()
  	}
  
  	dataLength := r.readInt()
  	r.readInt() // n relocations - ignore
  	r.readInt() // n pcdata - ignore
  	r.readInt() // n autom - ignore
  	r.readInt() // n funcdata - ignore
  	r.readInt() // n files - ignore
  
  	r.dataOffset = r.offset
  	r.skip(int64(dataLength))
  
  	// Symbols.
  	for {
  		if b := r.readByte(); b != 0xfe {
  			if b != 0xff {
  				return r.error(errCorruptObject)
  			}
  			break
  		}
  
  		typ := r.readByte()
  		s := &Sym{SymID: r.readSymID()}
  		r.p.Syms = append(r.p.Syms, s)
  		s.Kind = objabi.SymKind(typ)
  		flags := r.readInt()
  		s.DupOK = flags&1 != 0
  		s.Size = r.readInt()
  		s.Type = r.readSymID()
  		s.Data = r.readData()
  		s.Reloc = make([]Reloc, r.readInt())
  		for i := range s.Reloc {
  			rel := &s.Reloc[i]
  			rel.Offset = r.readInt()
  			rel.Size = r.readInt()
  			rel.Type = objabi.RelocType(r.readInt())
  			rel.Add = r.readInt()
  			rel.Sym = r.readSymID()
  		}
  
  		if s.Kind == objabi.STEXT {
  			f := new(Func)
  			s.Func = f
  			f.Args = r.readInt()
  			f.Frame = r.readInt()
  			flags := r.readInt()
  			f.Leaf = flags&(1<<0) != 0
  			f.NoSplit = r.readInt() != 0
  			f.Var = make([]Var, r.readInt())
  			for i := range f.Var {
  				v := &f.Var[i]
  				v.Name = r.readSymID().Name
  				v.Offset = r.readInt()
  				v.Kind = r.readInt()
  				v.Type = r.readSymID()
  			}
  
  			f.PCSP = r.readData()
  			f.PCFile = r.readData()
  			f.PCLine = r.readData()
  			f.PCInline = r.readData()
  			f.PCData = make([]Data, r.readInt())
  			for i := range f.PCData {
  				f.PCData[i] = r.readData()
  			}
  			f.FuncData = make([]FuncData, r.readInt())
  			for i := range f.FuncData {
  				f.FuncData[i].Sym = r.readSymID()
  			}
  			for i := range f.FuncData {
  				f.FuncData[i].Offset = int64(r.readInt()) // TODO
  			}
  			f.File = make([]string, r.readInt())
  			for i := range f.File {
  				f.File[i] = r.readSymID().Name
  			}
  			f.InlTree = make([]InlinedCall, r.readInt())
  			for i := range f.InlTree {
  				f.InlTree[i].Parent = r.readInt()
  				f.InlTree[i].File = r.readSymID().Name
  				f.InlTree[i].Line = r.readInt()
  				f.InlTree[i].Func = r.readSymID()
  			}
  		}
  	}
  
  	r.readFull(r.tmp[:7])
  	if !bytes.Equal(r.tmp[:7], []byte("\xffgo19ld")) {
  		return r.error(errCorruptObject)
  	}
  
  	return nil
  }
  
  func (r *Reloc) String(insnOffset uint64) string {
  	delta := r.Offset - int(insnOffset)
  	s := fmt.Sprintf("[%d:%d]%s", delta, delta+r.Size, r.Type)
  	if r.Sym.Name != "" {
  		if r.Add != 0 {
  			return fmt.Sprintf("%s:%s+%d", s, r.Sym.Name, r.Add)
  		}
  		return fmt.Sprintf("%s:%s", s, r.Sym.Name)
  	}
  	if r.Add != 0 {
  		return fmt.Sprintf("%s:%d", s, r.Add)
  	}
  	return s
  }
  

View as plain text