internal/goobj/read.go

// 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.
package goobj

import (
	"bufio"
	"bytes"
	"cmd/internal/objabi"
	"errors"
	"fmt"
	"io"
	"os"
	"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  int64          // 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 int64
}

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 int64
	Size   int64
	Sym    SymID
	Add    int64

	// 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   int64  // TODO(rsc): Define meaning.
	Offset int64  // 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     int64      // size in bytes of argument frame: inputs and outputs
	Frame    int64      // 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 int64
	File   string
	Line   int64
	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 int64           // maximum Version in any SymID in Syms
	Arch       string          // architecture
	Native     []*NativeReader // native object data (e.g. ELF)
}

type NativeReader struct {
	Name string
	io.ReaderAt
}

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          *os.File
	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 *os.File, 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
}

// peek returns the next n bytes without advancing the reader.
func (r *objReader) peek(n int) ([]byte, error) {
	if r.err != nil {
		return nil, r.err
	}
	if r.offset >= r.limit {
		r.error(io.ErrUnexpectedEOF)
		return nil, r.err
	}
	b, err := r.b.Peek(n)
	if err != nil {
		if err != bufio.ErrBufferFull {
			r.error(err)
		}
	}
	return b, 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() int64 {
	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
		}
	}

	return int64(u>>1) ^ (int64(u) << 63 >> 63)
}

// 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: n}
	r.dataOffset += 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 f,
// assuming that its import path is pkgpath.
func Parse(f *os.File, pkgpath string) (*Package, error) {
	if pkgpath == "" {
		pkgpath = `""`
	}
	p := new(Package)
	p.ImportPath = pkgpath

	var rd objReader
	rd.init(f, 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.
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

			p, err := r.peek(8)
			if err != nil {
				return err
			}
			if bytes.Equal(p, goobjHeader) {
				if err := r.parseObject(nil); err != nil {
					return fmt.Errorf("parsing archive member %q: %v", name, err)
				}
			} else {
				r.p.Native = append(r.p.Native, &NativeReader{
					Name:     name,
					ReaderAt: io.NewSectionReader(r.f, r.offset, size),
				})
			}

			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(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 - int64(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
}