From 2540738537b58250b79123668ee763768a1dfe84 Mon Sep 17 00:00:00 2001 From: Andrew Gerrand Date: Mon, 2 Jun 2014 13:04:43 +1000 Subject: [release-branch.go1.3] remove package debug/goobj and copy functionality to cmd/nm https://codereview.appspot.com/103760043/ --- src/cmd/nm/debug_goobj.go | 670 +++++++++++++++++++++++++++++++++++++++ src/cmd/nm/goobj.go | 21 +- src/pkg/debug/goobj/read.go | 664 -------------------------------------- src/pkg/debug/goobj/read_test.go | 28 -- 4 files changed, 680 insertions(+), 703 deletions(-) create mode 100644 src/cmd/nm/debug_goobj.go delete mode 100644 src/pkg/debug/goobj/read.go delete mode 100644 src/pkg/debug/goobj/read_test.go diff --git a/src/cmd/nm/debug_goobj.go b/src/cmd/nm/debug_goobj.go new file mode 100644 index 000000000..9a067e2b9 --- /dev/null +++ b/src/cmd/nm/debug_goobj.go @@ -0,0 +1,670 @@ +// DO NOT EDIT. Generated by code.google.com/p/rsc/cmd/bundle +// bundle -p main -x goobj_ debug/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. +// TODO(rsc): Write tests. (File format still up in the air a little.) + +package main + +import ( + "bufio" + "bytes" + "errors" + "fmt" + "io" + "strconv" + "strings" +) + +// A SymKind describes the kind of memory represented by a symbol. +type goobj_SymKind int + +// This list is taken from include/link.h. + +// Defined SymKind values. +// TODO(rsc): Give idiomatic Go names. +// TODO(rsc): Reduce the number of symbol types in the object files. +const ( + _ goobj_SymKind = iota + + // readonly, executable + goobj_STEXT + goobj_SELFRXSECT + + // readonly, non-executable + goobj_STYPE + goobj_SSTRING + goobj_SGOSTRING + goobj_SGOFUNC + goobj_SRODATA + goobj_SFUNCTAB + goobj_STYPELINK + goobj_SSYMTAB // TODO: move to unmapped section + goobj_SPCLNTAB + goobj_SELFROSECT + + // writable, non-executable + goobj_SMACHOPLT + goobj_SELFSECT + goobj_SMACHO // Mach-O __nl_symbol_ptr + goobj_SMACHOGOT + goobj_SNOPTRDATA + goobj_SINITARR + goobj_SDATA + goobj_SWINDOWS + goobj_SBSS + goobj_SNOPTRBSS + goobj_STLSBSS + + // not mapped + goobj_SXREF + goobj_SMACHOSYMSTR + goobj_SMACHOSYMTAB + goobj_SMACHOINDIRECTPLT + goobj_SMACHOINDIRECTGOT + goobj_SFILE + goobj_SFILEPATH + goobj_SCONST + goobj_SDYNIMPORT + goobj_SHOSTOBJ +) + +var goobj_symKindStrings = []string{ + goobj_SBSS: "SBSS", + goobj_SCONST: "SCONST", + goobj_SDATA: "SDATA", + goobj_SDYNIMPORT: "SDYNIMPORT", + goobj_SELFROSECT: "SELFROSECT", + goobj_SELFRXSECT: "SELFRXSECT", + goobj_SELFSECT: "SELFSECT", + goobj_SFILE: "SFILE", + goobj_SFILEPATH: "SFILEPATH", + goobj_SFUNCTAB: "SFUNCTAB", + goobj_SGOFUNC: "SGOFUNC", + goobj_SGOSTRING: "SGOSTRING", + goobj_SHOSTOBJ: "SHOSTOBJ", + goobj_SINITARR: "SINITARR", + goobj_SMACHO: "SMACHO", + goobj_SMACHOGOT: "SMACHOGOT", + goobj_SMACHOINDIRECTGOT: "SMACHOINDIRECTGOT", + goobj_SMACHOINDIRECTPLT: "SMACHOINDIRECTPLT", + goobj_SMACHOPLT: "SMACHOPLT", + goobj_SMACHOSYMSTR: "SMACHOSYMSTR", + goobj_SMACHOSYMTAB: "SMACHOSYMTAB", + goobj_SNOPTRBSS: "SNOPTRBSS", + goobj_SNOPTRDATA: "SNOPTRDATA", + goobj_SPCLNTAB: "SPCLNTAB", + goobj_SRODATA: "SRODATA", + goobj_SSTRING: "SSTRING", + goobj_SSYMTAB: "SSYMTAB", + goobj_STEXT: "STEXT", + goobj_STLSBSS: "STLSBSS", + goobj_STYPE: "STYPE", + goobj_STYPELINK: "STYPELINK", + goobj_SWINDOWS: "SWINDOWS", + goobj_SXREF: "SXREF", +} + +func (k goobj_SymKind) String() string { + if k < 0 || int(k) >= len(goobj_symKindStrings) { + return fmt.Sprintf("SymKind(%d)", k) + } + return goobj_symKindStrings[k] +} + +// A Sym is a named symbol in an object file. +type goobj_Sym struct { + goobj_SymID // symbol identifier (name and version) + Kind goobj_SymKind // kind of symbol + DupOK bool // are duplicate definitions okay? + Size int // size of corresponding data + Type goobj_SymID // symbol for Go type information + Data goobj_Data // memory image of symbol + Reloc []goobj_Reloc // relocations to apply to Data + Func *goobj_Func // additional data for functions +} + +// A SymID - the combination of Name and Version - uniquely identifies +// a symbol within a package. +type goobj_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 goobj_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 goobj_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 goobj_Reloc struct { + // The bytes at [Offset, Offset+Size) within the memory image + // should be updated to refer to the address Add bytes after the start + // of the symbol Sym. + Offset int + Size int + Sym goobj_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 int +} + +// A Var describes a variable in a function stack frame: a declared +// local variable, an input argument, or an output result. +type goobj_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 goobj_SymID // Go type for variable. +} + +// Func contains additional per-symbol information specific to functions. +type goobj_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 []goobj_Var // detail about local variables + PCSP goobj_Data // PC → SP offset map + PCFile goobj_Data // PC → file number map (index into File) + PCLine goobj_Data // PC → line number map + PCData []goobj_Data // PC → runtime support data map + FuncData []goobj_FuncData // non-PC-specific runtime support data + File []string // paths indexed by PCFile +} + +// TODO: Add PCData []byte and PCDataIter (similar to liblink). + +// A FuncData is a single function-specific data value. +type goobj_FuncData struct { + Sym goobj_SymID // symbol holding data + Offset int64 // offset into symbol for funcdata pointer +} + +// A Package is a parsed Go object file or archive defining a Go package. +type goobj_Package struct { + ImportPath string // import path denoting this package + Imports []string // packages imported by this package + Syms []*goobj_Sym // symbols defined by this package + MaxVersion int // maximum Version in any SymID in Syms +} + +var ( + goobj_archiveHeader = []byte("!\n") + goobj_archiveMagic = []byte("`\n") + goobj_goobjHeader = []byte("go objec") // truncated to size of archiveHeader + + goobj_errCorruptArchive = errors.New("corrupt archive") + goobj_errTruncatedArchive = errors.New("truncated archive") + goobj_errNotArchive = errors.New("unrecognized archive format") + + goobj_errCorruptObject = errors.New("corrupt object file") + goobj_errTruncatedObject = errors.New("truncated object file") + goobj_errNotObject = errors.New("unrecognized object file format") +) + +// An objReader is an object file reader. +type goobj_objReader struct { + p *goobj_Package + b *bufio.Reader + f io.ReadSeeker + err error + offset int64 + limit int64 + tmp [256]byte + pkg string + pkgprefix string +} + +// importPathToPrefix returns the prefix that will be used in the +// final symbol table for the given import path. +// We escape '%', '"', all control characters and non-ASCII bytes, +// and any '.' after the final slash. +// +// See ../../../cmd/ld/lib.c:/^pathtoprefix and +// ../../../cmd/gc/subr.c:/^pathtoprefix. +func goobj_importPathToPrefix(s string) string { + // find index of last slash, if any, or else -1. + // used for determining whether an index is after the last slash. + slash := strings.LastIndex(s, "/") + + // check for chars that need escaping + n := 0 + for r := 0; r < len(s); r++ { + if c := s[r]; c <= ' ' || (c == '.' && r > slash) || c == '%' || c == '"' || c >= 0x7F { + n++ + } + } + + // quick exit + if n == 0 { + return s + } + + // escape + const hex = "0123456789abcdef" + p := make([]byte, 0, len(s)+2*n) + for r := 0; r < len(s); r++ { + if c := s[r]; c <= ' ' || (c == '.' && r > slash) || c == '%' || c == '"' || c >= 0x7F { + p = append(p, '%', hex[c>>4], hex[c&0xF]) + } else { + p = append(p, c) + } + } + + return string(p) +} + +// init initializes r to read package p from f. +func (r *goobj_objReader) init(f io.ReadSeeker, p *goobj_Package) { + r.f = f + r.p = p + r.offset, _ = f.Seek(0, 1) + r.limit, _ = f.Seek(0, 2) + f.Seek(r.offset, 0) + r.b = bufio.NewReader(f) + r.pkgprefix = goobj_importPathToPrefix(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 *goobj_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 *goobj_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 *goobj_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 *goobj_objReader) readInt() int { + var u uint64 + + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + r.error(goobj_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(goobj_errCorruptObject) // TODO + return 0 + } + return int(v) +} + +// readString reads a length-delimited string from the input file. +func (r *goobj_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 *goobj_objReader) readSymID() goobj_SymID { + 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 + } + + return goobj_SymID{name, vers} +} + +// readData reads a data reference from the input file. +func (r *goobj_objReader) readData() goobj_Data { + n := r.readInt() + d := goobj_Data{Offset: r.offset, Size: int64(n)} + r.skip(int64(n)) + return d +} + +// skip skips n bytes in the input. +func (r *goobj_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, 0) + 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 goobj_Parse(r io.ReadSeeker, pkgpath string) (*goobj_Package, error) { + if pkgpath == "" { + pkgpath = `""` + } + p := new(goobj_Package) + p.ImportPath = pkgpath + + var rd goobj_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, goobj_errNotObject + + case bytes.Equal(rd.tmp[:8], goobj_archiveHeader): + if err := rd.parseArchive(); err != nil { + return nil, err + } + case bytes.Equal(rd.tmp[:8], goobj_goobjHeader): + if err := rd.parseObject(goobj_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 goobj_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 *goobj_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 goobj_errTruncatedArchive + } + if !bytes.Equal(data[58:60], goobj_archiveMagic) { + return goobj_errCorruptArchive + } + name := goobj_trimSpace(data[0:16]) + size, err := strconv.ParseInt(goobj_trimSpace(data[48:58]), 10, 64) + if err != nil { + return goobj_errCorruptArchive + } + data = data[60:] + fsize := size + size&1 + if fsize < 0 || fsize < size { + return goobj_errCorruptArchive + } + switch name { + case "__.SYMDEF", "__.GOSYMDEF", "__.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 *goobj_objReader) parseObject(prefix []byte) error { + // TODO(rsc): Maybe use prefix and the initial input to + // record the header line from the file, which would + // give the architecture and other version information. + + r.p.MaxVersion++ + var c1, c2, c3 byte + for { + c1, c2, c3 = c2, c3, r.readByte() + if c3 == 0 { // NUL or EOF, either is bad + return goobj_errCorruptObject + } + if c1 == '\n' && c2 == '!' && c3 == '\n' { + break + } + } + + r.readFull(r.tmp[:8]) + if !bytes.Equal(r.tmp[:8], []byte("\x00\x00go13ld")) { + return r.error(goobj_errCorruptObject) + } + + b := r.readByte() + if b != 1 { + return r.error(goobj_errCorruptObject) + } + + // Direct package dependencies. + for { + s := r.readString() + if s == "" { + break + } + r.p.Imports = append(r.p.Imports, s) + } + + // Symbols. + for { + if b := r.readByte(); b != 0xfe { + if b != 0xff { + return r.error(goobj_errCorruptObject) + } + break + } + + typ := r.readInt() + s := &goobj_Sym{goobj_SymID: r.readSymID()} + r.p.Syms = append(r.p.Syms, s) + s.Kind = goobj_SymKind(typ) + s.DupOK = r.readInt() != 0 + s.Size = r.readInt() + s.Type = r.readSymID() + s.Data = r.readData() + s.Reloc = make([]goobj_Reloc, r.readInt()) + for i := range s.Reloc { + rel := &s.Reloc[i] + rel.Offset = r.readInt() + rel.Size = r.readInt() + rel.Type = r.readInt() + rel.Add = r.readInt() + r.readInt() // Xadd - ignored + rel.Sym = r.readSymID() + r.readSymID() // Xsym - ignored + } + + if s.Kind == goobj_STEXT { + f := new(goobj_Func) + s.Func = f + f.Args = r.readInt() + f.Frame = r.readInt() + f.Leaf = r.readInt() != 0 + f.NoSplit = r.readInt() != 0 + f.Var = make([]goobj_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.PCData = make([]goobj_Data, r.readInt()) + for i := range f.PCData { + f.PCData[i] = r.readData() + } + f.FuncData = make([]goobj_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 + } + } + } + + r.readFull(r.tmp[:7]) + if !bytes.Equal(r.tmp[:7], []byte("\xffgo13ld")) { + return r.error(goobj_errCorruptObject) + } + + return nil +} diff --git a/src/cmd/nm/goobj.go b/src/cmd/nm/goobj.go index b0de51db9..5e0817d95 100644 --- a/src/cmd/nm/goobj.go +++ b/src/cmd/nm/goobj.go @@ -7,12 +7,11 @@ package main import ( - "debug/goobj" "fmt" "os" ) -func goobjName(id goobj.SymID) string { +func goobjName(id goobj_SymID) string { if id.Version == 0 { return id.Name } @@ -20,28 +19,28 @@ func goobjName(id goobj.SymID) string { } func goobjSymbols(f *os.File) []Sym { - pkg, err := goobj.Parse(f, `""`) + pkg, err := goobj_Parse(f, `""`) if err != nil { errorf("parsing %s: %v", f.Name(), err) return nil } - seen := make(map[goobj.SymID]bool) + seen := make(map[goobj_SymID]bool) var syms []Sym for _, s := range pkg.Syms { - seen[s.SymID] = true - sym := Sym{Addr: uint64(s.Data.Offset), Name: goobjName(s.SymID), Size: int64(s.Size), Type: s.Type.Name, Code: '?'} + seen[s.goobj_SymID] = true + sym := Sym{Addr: uint64(s.Data.Offset), Name: goobjName(s.goobj_SymID), Size: int64(s.Size), Type: s.Type.Name, Code: '?'} switch s.Kind { - case goobj.STEXT, goobj.SELFRXSECT: + case goobj_STEXT, goobj_SELFRXSECT: sym.Code = 'T' - case goobj.STYPE, goobj.SSTRING, goobj.SGOSTRING, goobj.SGOFUNC, goobj.SRODATA, goobj.SFUNCTAB, goobj.STYPELINK, goobj.SSYMTAB, goobj.SPCLNTAB, goobj.SELFROSECT: + case goobj_STYPE, goobj_SSTRING, goobj_SGOSTRING, goobj_SGOFUNC, goobj_SRODATA, goobj_SFUNCTAB, goobj_STYPELINK, goobj_SSYMTAB, goobj_SPCLNTAB, goobj_SELFROSECT: sym.Code = 'R' - case goobj.SMACHOPLT, goobj.SELFSECT, goobj.SMACHO, goobj.SMACHOGOT, goobj.SNOPTRDATA, goobj.SINITARR, goobj.SDATA, goobj.SWINDOWS: + case goobj_SMACHOPLT, goobj_SELFSECT, goobj_SMACHO, goobj_SMACHOGOT, goobj_SNOPTRDATA, goobj_SINITARR, goobj_SDATA, goobj_SWINDOWS: sym.Code = 'D' - case goobj.SBSS, goobj.SNOPTRBSS, goobj.STLSBSS: + case goobj_SBSS, goobj_SNOPTRBSS, goobj_STLSBSS: sym.Code = 'B' - case goobj.SXREF, goobj.SMACHOSYMSTR, goobj.SMACHOSYMTAB, goobj.SMACHOINDIRECTPLT, goobj.SMACHOINDIRECTGOT, goobj.SFILE, goobj.SFILEPATH, goobj.SCONST, goobj.SDYNIMPORT, goobj.SHOSTOBJ: + case goobj_SXREF, goobj_SMACHOSYMSTR, goobj_SMACHOSYMTAB, goobj_SMACHOINDIRECTPLT, goobj_SMACHOINDIRECTGOT, goobj_SFILE, goobj_SFILEPATH, goobj_SCONST, goobj_SDYNIMPORT, goobj_SHOSTOBJ: sym.Code = 'X' // should not see } if s.Version != 0 { diff --git a/src/pkg/debug/goobj/read.go b/src/pkg/debug/goobj/read.go deleted file mode 100644 index c95fe1e47..000000000 --- a/src/pkg/debug/goobj/read.go +++ /dev/null @@ -1,664 +0,0 @@ -// 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" - "errors" - "fmt" - "io" - "strconv" - "strings" -) - -// A SymKind describes the kind of memory represented by a symbol. -type SymKind int - -// This list is taken from include/link.h. - -// Defined SymKind values. -// TODO(rsc): Give idiomatic Go names. -// TODO(rsc): Reduce the number of symbol types in the object files. -const ( - _ SymKind = iota - - // readonly, executable - STEXT - SELFRXSECT - - // readonly, non-executable - STYPE - SSTRING - SGOSTRING - SGOFUNC - SRODATA - SFUNCTAB - STYPELINK - SSYMTAB // TODO: move to unmapped section - SPCLNTAB - SELFROSECT - - // writable, non-executable - SMACHOPLT - SELFSECT - SMACHO // Mach-O __nl_symbol_ptr - SMACHOGOT - SNOPTRDATA - SINITARR - SDATA - SWINDOWS - SBSS - SNOPTRBSS - STLSBSS - - // not mapped - SXREF - SMACHOSYMSTR - SMACHOSYMTAB - SMACHOINDIRECTPLT - SMACHOINDIRECTGOT - SFILE - SFILEPATH - SCONST - SDYNIMPORT - SHOSTOBJ -) - -var symKindStrings = []string{ - SBSS: "SBSS", - SCONST: "SCONST", - SDATA: "SDATA", - SDYNIMPORT: "SDYNIMPORT", - SELFROSECT: "SELFROSECT", - SELFRXSECT: "SELFRXSECT", - SELFSECT: "SELFSECT", - SFILE: "SFILE", - SFILEPATH: "SFILEPATH", - SFUNCTAB: "SFUNCTAB", - SGOFUNC: "SGOFUNC", - SGOSTRING: "SGOSTRING", - SHOSTOBJ: "SHOSTOBJ", - SINITARR: "SINITARR", - SMACHO: "SMACHO", - SMACHOGOT: "SMACHOGOT", - SMACHOINDIRECTGOT: "SMACHOINDIRECTGOT", - SMACHOINDIRECTPLT: "SMACHOINDIRECTPLT", - SMACHOPLT: "SMACHOPLT", - SMACHOSYMSTR: "SMACHOSYMSTR", - SMACHOSYMTAB: "SMACHOSYMTAB", - SNOPTRBSS: "SNOPTRBSS", - SNOPTRDATA: "SNOPTRDATA", - SPCLNTAB: "SPCLNTAB", - SRODATA: "SRODATA", - SSTRING: "SSTRING", - SSYMTAB: "SSYMTAB", - STEXT: "STEXT", - STLSBSS: "STLSBSS", - STYPE: "STYPE", - STYPELINK: "STYPELINK", - SWINDOWS: "SWINDOWS", - SXREF: "SXREF", -} - -func (k SymKind) String() string { - if k < 0 || int(k) >= len(symKindStrings) { - return fmt.Sprintf("SymKind(%d)", k) - } - return symKindStrings[k] -} - -// A Sym is a named symbol in an object file. -type Sym struct { - SymID // symbol identifier (name and version) - Kind 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 memory image - // 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 int -} - -// 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 - PCData []Data // PC → runtime support data map - FuncData []FuncData // non-PC-specific runtime support data - File []string // paths indexed by PCFile -} - -// 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 -} - -// 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 - Syms []*Sym // symbols defined by this package - MaxVersion int // maximum Version in any SymID in Syms -} - -var ( - archiveHeader = []byte("!\n") - archiveMagic = []byte("`\n") - goobjHeader = []byte("go objec") // truncated to size of archiveHeader - - errCorruptArchive = errors.New("corrupt archive") - errTruncatedArchive = errors.New("truncated archive") - errNotArchive = errors.New("unrecognized archive format") - - errCorruptObject = errors.New("corrupt object file") - errTruncatedObject = errors.New("truncated 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 - limit int64 - tmp [256]byte - pkg string - pkgprefix string -} - -// importPathToPrefix returns the prefix that will be used in the -// final symbol table for the given import path. -// We escape '%', '"', all control characters and non-ASCII bytes, -// and any '.' after the final slash. -// -// See ../../../cmd/ld/lib.c:/^pathtoprefix and -// ../../../cmd/gc/subr.c:/^pathtoprefix. -func importPathToPrefix(s string) string { - // find index of last slash, if any, or else -1. - // used for determining whether an index is after the last slash. - slash := strings.LastIndex(s, "/") - - // check for chars that need escaping - n := 0 - for r := 0; r < len(s); r++ { - if c := s[r]; c <= ' ' || (c == '.' && r > slash) || c == '%' || c == '"' || c >= 0x7F { - n++ - } - } - - // quick exit - if n == 0 { - return s - } - - // escape - const hex = "0123456789abcdef" - p := make([]byte, 0, len(s)+2*n) - for r := 0; r < len(s); r++ { - if c := s[r]; c <= ' ' || (c == '.' && r > slash) || c == '%' || c == '"' || c >= 0x7F { - p = append(p, '%', hex[c>>4], hex[c&0xF]) - } else { - p = append(p, c) - } - } - - return string(p) -} - -// 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, 1) - r.limit, _ = f.Seek(0, 2) - f.Seek(r.offset, 0) - r.b = bufio.NewReader(f) - r.pkgprefix = importPathToPrefix(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 { - 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 - } - - return SymID{name, vers} -} - -// readData reads a data reference from the input file. -func (r *objReader) readData() Data { - n := r.readInt() - d := Data{Offset: r.offset, Size: int64(n)} - r.skip(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, 0) - 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 "__.SYMDEF", "__.GOSYMDEF", "__.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 { - // TODO(rsc): Maybe use prefix and the initial input to - // record the header line from the file, which would - // give the architecture and other version information. - - r.p.MaxVersion++ - var c1, c2, c3 byte - for { - c1, c2, c3 = c2, c3, r.readByte() - if c3 == 0 { // NUL or EOF, either is bad - return errCorruptObject - } - if c1 == '\n' && c2 == '!' && c3 == '\n' { - break - } - } - - r.readFull(r.tmp[:8]) - if !bytes.Equal(r.tmp[:8], []byte("\x00\x00go13ld")) { - 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) - } - - // Symbols. - for { - if b := r.readByte(); b != 0xfe { - if b != 0xff { - return r.error(errCorruptObject) - } - break - } - - typ := r.readInt() - s := &Sym{SymID: r.readSymID()} - r.p.Syms = append(r.p.Syms, s) - s.Kind = SymKind(typ) - s.DupOK = r.readInt() != 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 = r.readInt() - rel.Add = r.readInt() - r.readInt() // Xadd - ignored - rel.Sym = r.readSymID() - r.readSymID() // Xsym - ignored - } - - if s.Kind == STEXT { - f := new(Func) - s.Func = f - f.Args = r.readInt() - f.Frame = r.readInt() - f.Leaf = r.readInt() != 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.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 - } - } - } - - r.readFull(r.tmp[:7]) - if !bytes.Equal(r.tmp[:7], []byte("\xffgo13ld")) { - return r.error(errCorruptObject) - } - - return nil -} diff --git a/src/pkg/debug/goobj/read_test.go b/src/pkg/debug/goobj/read_test.go deleted file mode 100644 index dee140533..000000000 --- a/src/pkg/debug/goobj/read_test.go +++ /dev/null @@ -1,28 +0,0 @@ -// 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 - -import "testing" - -var importPathToPrefixTests = []struct { - in string - out string -}{ - {"runtime", "runtime"}, - {"sync/atomic", "sync/atomic"}, - {"code.google.com/p/go.tools/godoc", "code.google.com/p/go.tools/godoc"}, - {"foo.bar/baz.quux", "foo.bar/baz%2equux"}, - {"", ""}, - {"%foo%bar", "%25foo%25bar"}, - {"\x01\x00\x7F☺", "%01%00%7f%e2%98%ba"}, -} - -func TestImportPathToPrefix(t *testing.T) { - for _, tt := range importPathToPrefixTests { - if out := importPathToPrefix(tt.in); out != tt.out { - t.Errorf("importPathToPrefix(%q) = %q, want %q", tt.in, out, tt.out) - } - } -} -- cgit v1.2.1