diff options
Diffstat (limited to 'src/cmd/compile/internal/staticdata')
| -rw-r--r-- | src/cmd/compile/internal/staticdata/data.go | 296 | ||||
| -rw-r--r-- | src/cmd/compile/internal/staticdata/embed.go | 193 |
2 files changed, 489 insertions, 0 deletions
diff --git a/src/cmd/compile/internal/staticdata/data.go b/src/cmd/compile/internal/staticdata/data.go new file mode 100644 index 0000000000..7627aaa11a --- /dev/null +++ b/src/cmd/compile/internal/staticdata/data.go @@ -0,0 +1,296 @@ +// Copyright 2009 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 staticdata + +import ( + "crypto/sha256" + "fmt" + "io" + "io/ioutil" + "os" + "sort" + "strconv" + "sync" + + "cmd/compile/internal/base" + "cmd/compile/internal/ir" + "cmd/compile/internal/objw" + "cmd/compile/internal/typecheck" + "cmd/compile/internal/types" + "cmd/internal/obj" + "cmd/internal/objabi" + "cmd/internal/src" +) + +// InitAddr writes the static address of a to n. a must be an ONAME. +// Neither n nor a is modified. +func InitAddr(n *ir.Name, noff int64, a *ir.Name, aoff int64) { + if n.Op() != ir.ONAME { + base.Fatalf("addrsym n op %v", n.Op()) + } + if n.Sym() == nil { + base.Fatalf("addrsym nil n sym") + } + if a.Op() != ir.ONAME { + base.Fatalf("addrsym a op %v", a.Op()) + } + s := n.Sym().Linksym() + s.WriteAddr(base.Ctxt, noff, types.PtrSize, a.Sym().Linksym(), aoff) +} + +// InitFunc writes the static address of f to n. f must be a global function. +// Neither n nor f is modified. +func InitFunc(n *ir.Name, noff int64, f *ir.Name) { + if n.Op() != ir.ONAME { + base.Fatalf("pfuncsym n op %v", n.Op()) + } + if n.Sym() == nil { + base.Fatalf("pfuncsym nil n sym") + } + if f.Class_ != ir.PFUNC { + base.Fatalf("pfuncsym class not PFUNC %d", f.Class_) + } + s := n.Sym().Linksym() + s.WriteAddr(base.Ctxt, noff, types.PtrSize, FuncSym(f.Sym()).Linksym(), 0) +} + +// InitSlice writes a static slice symbol {&arr, lencap, lencap} to n+noff. +// InitSlice does not modify n. +func InitSlice(n *ir.Name, noff int64, arr *ir.Name, lencap int64) { + s := n.Sym().Linksym() + if arr.Op() != ir.ONAME { + base.Fatalf("slicesym non-name arr %v", arr) + } + s.WriteAddr(base.Ctxt, noff, types.PtrSize, arr.Sym().Linksym(), 0) + s.WriteInt(base.Ctxt, noff+types.SliceLenOffset, types.PtrSize, lencap) + s.WriteInt(base.Ctxt, noff+types.SliceCapOffset, types.PtrSize, lencap) +} + +func InitSliceBytes(nam *ir.Name, off int64, s string) { + if nam.Op() != ir.ONAME { + base.Fatalf("slicebytes %v", nam) + } + InitSlice(nam, off, slicedata(nam.Pos(), s), int64(len(s))) +} + +const ( + stringSymPrefix = "go.string." + stringSymPattern = ".gostring.%d.%x" +) + +// StringSym returns a symbol containing the string s. +// The symbol contains the string data, not a string header. +func StringSym(pos src.XPos, s string) (data *obj.LSym) { + var symname string + if len(s) > 100 { + // Huge strings are hashed to avoid long names in object files. + // Indulge in some paranoia by writing the length of s, too, + // as protection against length extension attacks. + // Same pattern is known to fileStringSym below. + h := sha256.New() + io.WriteString(h, s) + symname = fmt.Sprintf(stringSymPattern, len(s), h.Sum(nil)) + } else { + // Small strings get named directly by their contents. + symname = strconv.Quote(s) + } + + symdata := base.Ctxt.Lookup(stringSymPrefix + symname) + if !symdata.OnList() { + off := dstringdata(symdata, 0, s, pos, "string") + objw.Global(symdata, int32(off), obj.DUPOK|obj.RODATA|obj.LOCAL) + symdata.Set(obj.AttrContentAddressable, true) + } + + return symdata +} + +// fileStringSym returns a symbol for the contents and the size of file. +// If readonly is true, the symbol shares storage with any literal string +// or other file with the same content and is placed in a read-only section. +// If readonly is false, the symbol is a read-write copy separate from any other, +// for use as the backing store of a []byte. +// The content hash of file is copied into hash. (If hash is nil, nothing is copied.) +// The returned symbol contains the data itself, not a string header. +func fileStringSym(pos src.XPos, file string, readonly bool, hash []byte) (*obj.LSym, int64, error) { + f, err := os.Open(file) + if err != nil { + return nil, 0, err + } + defer f.Close() + info, err := f.Stat() + if err != nil { + return nil, 0, err + } + if !info.Mode().IsRegular() { + return nil, 0, fmt.Errorf("not a regular file") + } + size := info.Size() + if size <= 1*1024 { + data, err := ioutil.ReadAll(f) + if err != nil { + return nil, 0, err + } + if int64(len(data)) != size { + return nil, 0, fmt.Errorf("file changed between reads") + } + var sym *obj.LSym + if readonly { + sym = StringSym(pos, string(data)) + } else { + sym = slicedata(pos, string(data)).Sym().Linksym() + } + if len(hash) > 0 { + sum := sha256.Sum256(data) + copy(hash, sum[:]) + } + return sym, size, nil + } + if size > 2e9 { + // ggloblsym takes an int32, + // and probably the rest of the toolchain + // can't handle such big symbols either. + // See golang.org/issue/9862. + return nil, 0, fmt.Errorf("file too large") + } + + // File is too big to read and keep in memory. + // Compute hash if needed for read-only content hashing or if the caller wants it. + var sum []byte + if readonly || len(hash) > 0 { + h := sha256.New() + n, err := io.Copy(h, f) + if err != nil { + return nil, 0, err + } + if n != size { + return nil, 0, fmt.Errorf("file changed between reads") + } + sum = h.Sum(nil) + copy(hash, sum) + } + + var symdata *obj.LSym + if readonly { + symname := fmt.Sprintf(stringSymPattern, size, sum) + symdata = base.Ctxt.Lookup(stringSymPrefix + symname) + if !symdata.OnList() { + info := symdata.NewFileInfo() + info.Name = file + info.Size = size + objw.Global(symdata, int32(size), obj.DUPOK|obj.RODATA|obj.LOCAL) + // Note: AttrContentAddressable cannot be set here, + // because the content-addressable-handling code + // does not know about file symbols. + } + } else { + // Emit a zero-length data symbol + // and then fix up length and content to use file. + symdata = slicedata(pos, "").Sym().Linksym() + symdata.Size = size + symdata.Type = objabi.SNOPTRDATA + info := symdata.NewFileInfo() + info.Name = file + info.Size = size + } + + return symdata, size, nil +} + +var slicedataGen int + +func slicedata(pos src.XPos, s string) *ir.Name { + slicedataGen++ + symname := fmt.Sprintf(".gobytes.%d", slicedataGen) + sym := types.LocalPkg.Lookup(symname) + symnode := typecheck.NewName(sym) + sym.Def = symnode + + lsym := sym.Linksym() + off := dstringdata(lsym, 0, s, pos, "slice") + objw.Global(lsym, int32(off), obj.NOPTR|obj.LOCAL) + + return symnode +} + +func dstringdata(s *obj.LSym, off int, t string, pos src.XPos, what string) int { + // Objects that are too large will cause the data section to overflow right away, + // causing a cryptic error message by the linker. Check for oversize objects here + // and provide a useful error message instead. + if int64(len(t)) > 2e9 { + base.ErrorfAt(pos, "%v with length %v is too big", what, len(t)) + return 0 + } + + s.WriteString(base.Ctxt, int64(off), len(t), t) + return off + len(t) +} + +var ( + funcsymsmu sync.Mutex // protects funcsyms and associated package lookups (see func funcsym) + funcsyms []*types.Sym +) + +// FuncSym returns s·f. +func FuncSym(s *types.Sym) *types.Sym { + // funcsymsmu here serves to protect not just mutations of funcsyms (below), + // but also the package lookup of the func sym name, + // since this function gets called concurrently from the backend. + // There are no other concurrent package lookups in the backend, + // except for the types package, which is protected separately. + // Reusing funcsymsmu to also cover this package lookup + // avoids a general, broader, expensive package lookup mutex. + // Note makefuncsym also does package look-up of func sym names, + // but that it is only called serially, from the front end. + funcsymsmu.Lock() + sf, existed := s.Pkg.LookupOK(ir.FuncSymName(s)) + // Don't export s·f when compiling for dynamic linking. + // When dynamically linking, the necessary function + // symbols will be created explicitly with makefuncsym. + // See the makefuncsym comment for details. + if !base.Ctxt.Flag_dynlink && !existed { + funcsyms = append(funcsyms, s) + } + funcsymsmu.Unlock() + return sf +} + +// NeedFuncSym ensures that s·f is exported. +// It is only used with -dynlink. +// When not compiling for dynamic linking, +// the funcsyms are created as needed by +// the packages that use them. +// Normally we emit the s·f stubs as DUPOK syms, +// but DUPOK doesn't work across shared library boundaries. +// So instead, when dynamic linking, we only create +// the s·f stubs in s's package. +func NeedFuncSym(s *types.Sym) { + if !base.Ctxt.Flag_dynlink { + base.Fatalf("makefuncsym dynlink") + } + if s.IsBlank() { + return + } + if base.Flag.CompilingRuntime && (s.Name == "getg" || s.Name == "getclosureptr" || s.Name == "getcallerpc" || s.Name == "getcallersp") { + // runtime.getg(), getclosureptr(), getcallerpc(), and + // getcallersp() are not real functions and so do not + // get funcsyms. + return + } + if _, existed := s.Pkg.LookupOK(ir.FuncSymName(s)); !existed { + funcsyms = append(funcsyms, s) + } +} + +func WriteFuncSyms() { + sort.Slice(funcsyms, func(i, j int) bool { + return funcsyms[i].LinksymName() < funcsyms[j].LinksymName() + }) + for _, s := range funcsyms { + sf := s.Pkg.Lookup(ir.FuncSymName(s)).Linksym() + objw.SymPtr(sf, 0, s.Linksym(), 0) + objw.Global(sf, int32(types.PtrSize), obj.DUPOK|obj.RODATA) + } +} diff --git a/src/cmd/compile/internal/staticdata/embed.go b/src/cmd/compile/internal/staticdata/embed.go new file mode 100644 index 0000000000..55c9a3356e --- /dev/null +++ b/src/cmd/compile/internal/staticdata/embed.go @@ -0,0 +1,193 @@ +// Copyright 2020 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 staticdata + +import ( + "path" + "sort" + "strings" + + "cmd/compile/internal/base" + "cmd/compile/internal/ir" + "cmd/compile/internal/objw" + "cmd/compile/internal/types" + "cmd/internal/obj" +) + +const ( + embedUnknown = iota + embedBytes + embedString + embedFiles +) + +func embedFileList(v *ir.Name) []string { + kind := embedKind(v.Type()) + if kind == embedUnknown { + base.ErrorfAt(v.Pos(), "go:embed cannot apply to var of type %v", v.Type()) + return nil + } + + // Build list of files to store. + have := make(map[string]bool) + var list []string + for _, e := range *v.Embed { + for _, pattern := range e.Patterns { + files, ok := base.Flag.Cfg.Embed.Patterns[pattern] + if !ok { + base.ErrorfAt(e.Pos, "invalid go:embed: build system did not map pattern: %s", pattern) + } + for _, file := range files { + if base.Flag.Cfg.Embed.Files[file] == "" { + base.ErrorfAt(e.Pos, "invalid go:embed: build system did not map file: %s", file) + continue + } + if !have[file] { + have[file] = true + list = append(list, file) + } + if kind == embedFiles { + for dir := path.Dir(file); dir != "." && !have[dir]; dir = path.Dir(dir) { + have[dir] = true + list = append(list, dir+"/") + } + } + } + } + } + sort.Slice(list, func(i, j int) bool { + return embedFileLess(list[i], list[j]) + }) + + if kind == embedString || kind == embedBytes { + if len(list) > 1 { + base.ErrorfAt(v.Pos(), "invalid go:embed: multiple files for type %v", v.Type()) + return nil + } + } + + return list +} + +// embedKindApprox determines the kind of embedding variable, approximately. +// The match is approximate because we haven't done scope resolution yet and +// can't tell whether "string" and "byte" really mean "string" and "byte". +// The result must be confirmed later, after type checking, using embedKind. +func embedKindApprox(typ ir.Node) int { + if typ.Sym() != nil && typ.Sym().Name == "FS" && (typ.Sym().Pkg.Path == "embed" || (typ.Sym().Pkg == types.LocalPkg && base.Ctxt.Pkgpath == "embed")) { + return embedFiles + } + // These are not guaranteed to match only string and []byte - + // maybe the local package has redefined one of those words. + // But it's the best we can do now during the noder. + // The stricter check happens later, in initEmbed calling embedKind. + if typ.Sym() != nil && typ.Sym().Name == "string" && typ.Sym().Pkg == types.LocalPkg { + return embedString + } + if typ, ok := typ.(*ir.SliceType); ok { + if sym := typ.Elem.Sym(); sym != nil && sym.Name == "byte" && sym.Pkg == types.LocalPkg { + return embedBytes + } + } + return embedUnknown +} + +// embedKind determines the kind of embedding variable. +func embedKind(typ *types.Type) int { + if typ.Sym() != nil && typ.Sym().Name == "FS" && (typ.Sym().Pkg.Path == "embed" || (typ.Sym().Pkg == types.LocalPkg && base.Ctxt.Pkgpath == "embed")) { + return embedFiles + } + if typ == types.Types[types.TSTRING] { + return embedString + } + if typ.Sym() == nil && typ.IsSlice() && typ.Elem() == types.ByteType { + return embedBytes + } + return embedUnknown +} + +func embedFileNameSplit(name string) (dir, elem string, isDir bool) { + if name[len(name)-1] == '/' { + isDir = true + name = name[:len(name)-1] + } + i := len(name) - 1 + for i >= 0 && name[i] != '/' { + i-- + } + if i < 0 { + return ".", name, isDir + } + return name[:i], name[i+1:], isDir +} + +// embedFileLess implements the sort order for a list of embedded files. +// See the comment inside ../../../../embed/embed.go's Files struct for rationale. +func embedFileLess(x, y string) bool { + xdir, xelem, _ := embedFileNameSplit(x) + ydir, yelem, _ := embedFileNameSplit(y) + return xdir < ydir || xdir == ydir && xelem < yelem +} + +// WriteEmbed emits the init data for a //go:embed variable, +// which is either a string, a []byte, or an embed.FS. +func WriteEmbed(v *ir.Name) { + files := embedFileList(v) + switch kind := embedKind(v.Type()); kind { + case embedUnknown: + base.ErrorfAt(v.Pos(), "go:embed cannot apply to var of type %v", v.Type()) + + case embedString, embedBytes: + file := files[0] + fsym, size, err := fileStringSym(v.Pos(), base.Flag.Cfg.Embed.Files[file], kind == embedString, nil) + if err != nil { + base.ErrorfAt(v.Pos(), "embed %s: %v", file, err) + } + sym := v.Sym().Linksym() + off := 0 + off = objw.SymPtr(sym, off, fsym, 0) // data string + off = objw.Uintptr(sym, off, uint64(size)) // len + if kind == embedBytes { + objw.Uintptr(sym, off, uint64(size)) // cap for slice + } + + case embedFiles: + slicedata := base.Ctxt.Lookup(`"".` + v.Sym().Name + `.files`) + off := 0 + // []files pointed at by Files + off = objw.SymPtr(slicedata, off, slicedata, 3*types.PtrSize) // []file, pointing just past slice + off = objw.Uintptr(slicedata, off, uint64(len(files))) + off = objw.Uintptr(slicedata, off, uint64(len(files))) + + // embed/embed.go type file is: + // name string + // data string + // hash [16]byte + // Emit one of these per file in the set. + const hashSize = 16 + hash := make([]byte, hashSize) + for _, file := range files { + off = objw.SymPtr(slicedata, off, StringSym(v.Pos(), file), 0) // file string + off = objw.Uintptr(slicedata, off, uint64(len(file))) + if strings.HasSuffix(file, "/") { + // entry for directory - no data + off = objw.Uintptr(slicedata, off, 0) + off = objw.Uintptr(slicedata, off, 0) + off += hashSize + } else { + fsym, size, err := fileStringSym(v.Pos(), base.Flag.Cfg.Embed.Files[file], true, hash) + if err != nil { + base.ErrorfAt(v.Pos(), "embed %s: %v", file, err) + } + off = objw.SymPtr(slicedata, off, fsym, 0) // data string + off = objw.Uintptr(slicedata, off, uint64(size)) + off = int(slicedata.WriteBytes(base.Ctxt, int64(off), hash)) + } + } + objw.Global(slicedata, int32(off), obj.RODATA|obj.LOCAL) + sym := v.Sym().Linksym() + objw.SymPtr(sym, 0, slicedata, 0) + } +} |
