string.go

     1  // Copyright 2014 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 runtime
     6  
     7  import (
     8  	"internal/abi"
     9  	"internal/bytealg"
    10  	"internal/goarch"
    11  	"unsafe"
    12  )
    13  
    14  // The constant is known to the compiler.
    15  // There is no fundamental theory behind this number.
    16  const tmpStringBufSize = 32
    17  
    18  type tmpBuf [tmpStringBufSize]byte
    19  
    20  // concatstrings implements a Go string concatenation x+y+z+...
    21  // The operands are passed in the slice a.
    22  // If buf != nil, the compiler has determined that the result does not
    23  // escape the calling function, so the string data can be stored in buf
    24  // if small enough.
    25  func concatstrings(buf *tmpBuf, a []string) string {
    26  	idx := 0
    27  	l := 0
    28  	count := 0
    29  	for i, x := range a {
    30  		n := len(x)
    31  		if n == 0 {
    32  			continue
    33  		}
    34  		if l+n < l {
    35  			throw("string concatenation too long")
    36  		}
    37  		l += n
    38  		count++
    39  		idx = i
    40  	}
    41  	if count == 0 {
    42  		return ""
    43  	}
    44  
    45  	// If there is just one string and either it is not on the stack
    46  	// or our result does not escape the calling frame (buf != nil),
    47  	// then we can return that string directly.
    48  	if count == 1 && (buf != nil || !stringDataOnStack(a[idx])) {
    49  		return a[idx]
    50  	}
    51  	s, b := rawstringtmp(buf, l)
    52  	for _, x := range a {
    53  		copy(b, x)
    54  		b = b[len(x):]
    55  	}
    56  	return s
    57  }
    58  
    59  func concatstring2(buf *tmpBuf, a0, a1 string) string {
    60  	return concatstrings(buf, []string{a0, a1})
    61  }
    62  
    63  func concatstring3(buf *tmpBuf, a0, a1, a2 string) string {
    64  	return concatstrings(buf, []string{a0, a1, a2})
    65  }
    66  
    67  func concatstring4(buf *tmpBuf, a0, a1, a2, a3 string) string {
    68  	return concatstrings(buf, []string{a0, a1, a2, a3})
    69  }
    70  
    71  func concatstring5(buf *tmpBuf, a0, a1, a2, a3, a4 string) string {
    72  	return concatstrings(buf, []string{a0, a1, a2, a3, a4})
    73  }
    74  
    75  // slicebytetostring converts a byte slice to a string.
    76  // It is inserted by the compiler into generated code.
    77  // ptr is a pointer to the first element of the slice;
    78  // n is the length of the slice.
    79  // Buf is a fixed-size buffer for the result,
    80  // it is not nil if the result does not escape.
    81  func slicebytetostring(buf *tmpBuf, ptr *byte, n int) string {
    82  	if n == 0 {
    83  		// Turns out to be a relatively common case.
    84  		// Consider that you want to parse out data between parens in "foo()bar",
    85  		// you find the indices and convert the subslice to string.
    86  		return ""
    87  	}
    88  	if raceenabled {
    89  		racereadrangepc(unsafe.Pointer(ptr),
    90  			uintptr(n),
    91  			getcallerpc(),
    92  			abi.FuncPCABIInternal(slicebytetostring))
    93  	}
    94  	if msanenabled {
    95  		msanread(unsafe.Pointer(ptr), uintptr(n))
    96  	}
    97  	if asanenabled {
    98  		asanread(unsafe.Pointer(ptr), uintptr(n))
    99  	}
   100  	if n == 1 {
   101  		p := unsafe.Pointer(&staticuint64s[*ptr])
   102  		if goarch.BigEndian {
   103  			p = add(p, 7)
   104  		}
   105  		return unsafe.String((*byte)(p), 1)
   106  	}
   107  
   108  	var p unsafe.Pointer
   109  	if buf != nil && n <= len(buf) {
   110  		p = unsafe.Pointer(buf)
   111  	} else {
   112  		p = mallocgc(uintptr(n), nil, false)
   113  	}
   114  	memmove(p, unsafe.Pointer(ptr), uintptr(n))
   115  	return unsafe.String((*byte)(p), n)
   116  }
   117  
   118  // stringDataOnStack reports whether the string's data is
   119  // stored on the current goroutine's stack.
   120  func stringDataOnStack(s string) bool {
   121  	ptr := uintptr(unsafe.Pointer(unsafe.StringData(s)))
   122  	stk := getg().stack
   123  	return stk.lo <= ptr && ptr < stk.hi
   124  }
   125  
   126  func rawstringtmp(buf *tmpBuf, l int) (s string, b []byte) {
   127  	if buf != nil && l <= len(buf) {
   128  		b = buf[:l]
   129  		s = slicebytetostringtmp(&b[0], len(b))
   130  	} else {
   131  		s, b = rawstring(l)
   132  	}
   133  	return
   134  }
   135  
   136  // slicebytetostringtmp returns a "string" referring to the actual []byte bytes.
   137  //
   138  // Callers need to ensure that the returned string will not be used after
   139  // the calling goroutine modifies the original slice or synchronizes with
   140  // another goroutine.
   141  //
   142  // The function is only called when instrumenting
   143  // and otherwise intrinsified by the compiler.
   144  //
   145  // Some internal compiler optimizations use this function.
   146  //   - Used for m[T1{... Tn{..., string(k), ...} ...}] and m[string(k)]
   147  //     where k is []byte, T1 to Tn is a nesting of struct and array literals.
   148  //   - Used for "<"+string(b)+">" concatenation where b is []byte.
   149  //   - Used for string(b)=="foo" comparison where b is []byte.
   150  func slicebytetostringtmp(ptr *byte, n int) string {
   151  	if raceenabled && n > 0 {
   152  		racereadrangepc(unsafe.Pointer(ptr),
   153  			uintptr(n),
   154  			getcallerpc(),
   155  			abi.FuncPCABIInternal(slicebytetostringtmp))
   156  	}
   157  	if msanenabled && n > 0 {
   158  		msanread(unsafe.Pointer(ptr), uintptr(n))
   159  	}
   160  	if asanenabled && n > 0 {
   161  		asanread(unsafe.Pointer(ptr), uintptr(n))
   162  	}
   163  	return unsafe.String(ptr, n)
   164  }
   165  
   166  func stringtoslicebyte(buf *tmpBuf, s string) []byte {
   167  	var b []byte
   168  	if buf != nil && len(s) <= len(buf) {
   169  		*buf = tmpBuf{}
   170  		b = buf[:len(s)]
   171  	} else {
   172  		b = rawbyteslice(len(s))
   173  	}
   174  	copy(b, s)
   175  	return b
   176  }
   177  
   178  func stringtoslicerune(buf *[tmpStringBufSize]rune, s string) []rune {
   179  	// two passes.
   180  	// unlike slicerunetostring, no race because strings are immutable.
   181  	n := 0
   182  	for range s {
   183  		n++
   184  	}
   185  
   186  	var a []rune
   187  	if buf != nil && n <= len(buf) {
   188  		*buf = [tmpStringBufSize]rune{}
   189  		a = buf[:n]
   190  	} else {
   191  		a = rawruneslice(n)
   192  	}
   193  
   194  	n = 0
   195  	for _, r := range s {
   196  		a[n] = r
   197  		n++
   198  	}
   199  	return a
   200  }
   201  
   202  func slicerunetostring(buf *tmpBuf, a []rune) string {
   203  	if raceenabled && len(a) > 0 {
   204  		racereadrangepc(unsafe.Pointer(&a[0]),
   205  			uintptr(len(a))*unsafe.Sizeof(a[0]),
   206  			getcallerpc(),
   207  			abi.FuncPCABIInternal(slicerunetostring))
   208  	}
   209  	if msanenabled && len(a) > 0 {
   210  		msanread(unsafe.Pointer(&a[0]), uintptr(len(a))*unsafe.Sizeof(a[0]))
   211  	}
   212  	if asanenabled && len(a) > 0 {
   213  		asanread(unsafe.Pointer(&a[0]), uintptr(len(a))*unsafe.Sizeof(a[0]))
   214  	}
   215  	var dum [4]byte
   216  	size1 := 0
   217  	for _, r := range a {
   218  		size1 += encoderune(dum[:], r)
   219  	}
   220  	s, b := rawstringtmp(buf, size1+3)
   221  	size2 := 0
   222  	for _, r := range a {
   223  		// check for race
   224  		if size2 >= size1 {
   225  			break
   226  		}
   227  		size2 += encoderune(b[size2:], r)
   228  	}
   229  	return s[:size2]
   230  }
   231  
   232  type stringStruct struct {
   233  	str unsafe.Pointer
   234  	len int
   235  }
   236  
   237  // Variant with *byte pointer type for DWARF debugging.
   238  type stringStructDWARF struct {
   239  	str *byte
   240  	len int
   241  }
   242  
   243  func stringStructOf(sp *string) *stringStruct {
   244  	return (*stringStruct)(unsafe.Pointer(sp))
   245  }
   246  
   247  func intstring(buf *[4]byte, v int64) (s string) {
   248  	var b []byte
   249  	if buf != nil {
   250  		b = buf[:]
   251  		s = slicebytetostringtmp(&b[0], len(b))
   252  	} else {
   253  		s, b = rawstring(4)
   254  	}
   255  	if int64(rune(v)) != v {
   256  		v = runeError
   257  	}
   258  	n := encoderune(b, rune(v))
   259  	return s[:n]
   260  }
   261  
   262  // rawstring allocates storage for a new string. The returned
   263  // string and byte slice both refer to the same storage.
   264  // The storage is not zeroed. Callers should use
   265  // b to set the string contents and then drop b.
   266  func rawstring(size int) (s string, b []byte) {
   267  	p := mallocgc(uintptr(size), nil, false)
   268  	return unsafe.String((*byte)(p), size), unsafe.Slice((*byte)(p), size)
   269  }
   270  
   271  // rawbyteslice allocates a new byte slice. The byte slice is not zeroed.
   272  func rawbyteslice(size int) (b []byte) {
   273  	cap := roundupsize(uintptr(size), true)
   274  	p := mallocgc(cap, nil, false)
   275  	if cap != uintptr(size) {
   276  		memclrNoHeapPointers(add(p, uintptr(size)), cap-uintptr(size))
   277  	}
   278  
   279  	*(*slice)(unsafe.Pointer(&b)) = slice{p, size, int(cap)}
   280  	return
   281  }
   282  
   283  // rawruneslice allocates a new rune slice. The rune slice is not zeroed.
   284  func rawruneslice(size int) (b []rune) {
   285  	if uintptr(size) > maxAlloc/4 {
   286  		throw("out of memory")
   287  	}
   288  	mem := roundupsize(uintptr(size)*4, true)
   289  	p := mallocgc(mem, nil, false)
   290  	if mem != uintptr(size)*4 {
   291  		memclrNoHeapPointers(add(p, uintptr(size)*4), mem-uintptr(size)*4)
   292  	}
   293  
   294  	*(*slice)(unsafe.Pointer(&b)) = slice{p, size, int(mem / 4)}
   295  	return
   296  }
   297  
   298  // used by cmd/cgo
   299  func gobytes(p *byte, n int) (b []byte) {
   300  	if n == 0 {
   301  		return make([]byte, 0)
   302  	}
   303  
   304  	if n < 0 || uintptr(n) > maxAlloc {
   305  		panic(errorString("gobytes: length out of range"))
   306  	}
   307  
   308  	bp := mallocgc(uintptr(n), nil, false)
   309  	memmove(bp, unsafe.Pointer(p), uintptr(n))
   310  
   311  	*(*slice)(unsafe.Pointer(&b)) = slice{bp, n, n}
   312  	return
   313  }
   314  
   315  // This is exported via linkname to assembly in syscall (for Plan9) and cgo.
   316  //
   317  //go:linkname gostring
   318  func gostring(p *byte) string {
   319  	l := findnull(p)
   320  	if l == 0 {
   321  		return ""
   322  	}
   323  	s, b := rawstring(l)
   324  	memmove(unsafe.Pointer(&b[0]), unsafe.Pointer(p), uintptr(l))
   325  	return s
   326  }
   327  
   328  // internal_syscall_gostring is a version of gostring for internal/syscall/unix.
   329  //
   330  //go:linkname internal_syscall_gostring internal/syscall/unix.gostring
   331  func internal_syscall_gostring(p *byte) string {
   332  	return gostring(p)
   333  }
   334  
   335  func gostringn(p *byte, l int) string {
   336  	if l == 0 {
   337  		return ""
   338  	}
   339  	s, b := rawstring(l)
   340  	memmove(unsafe.Pointer(&b[0]), unsafe.Pointer(p), uintptr(l))
   341  	return s
   342  }
   343  
   344  const (
   345  	maxUint64 = ^uint64(0)
   346  	maxInt64  = int64(maxUint64 >> 1)
   347  )
   348  
   349  // atoi64 parses an int64 from a string s.
   350  // The bool result reports whether s is a number
   351  // representable by a value of type int64.
   352  func atoi64(s string) (int64, bool) {
   353  	if s == "" {
   354  		return 0, false
   355  	}
   356  
   357  	neg := false
   358  	if s[0] == '-' {
   359  		neg = true
   360  		s = s[1:]
   361  	}
   362  
   363  	un := uint64(0)
   364  	for i := 0; i < len(s); i++ {
   365  		c := s[i]
   366  		if c < '0' || c > '9' {
   367  			return 0, false
   368  		}
   369  		if un > maxUint64/10 {
   370  			// overflow
   371  			return 0, false
   372  		}
   373  		un *= 10
   374  		un1 := un + uint64(c) - '0'
   375  		if un1 < un {
   376  			// overflow
   377  			return 0, false
   378  		}
   379  		un = un1
   380  	}
   381  
   382  	if !neg && un > uint64(maxInt64) {
   383  		return 0, false
   384  	}
   385  	if neg && un > uint64(maxInt64)+1 {
   386  		return 0, false
   387  	}
   388  
   389  	n := int64(un)
   390  	if neg {
   391  		n = -n
   392  	}
   393  
   394  	return n, true
   395  }
   396  
   397  // atoi is like atoi64 but for integers
   398  // that fit into an int.
   399  func atoi(s string) (int, bool) {
   400  	if n, ok := atoi64(s); n == int64(int(n)) {
   401  		return int(n), ok
   402  	}
   403  	return 0, false
   404  }
   405  
   406  // atoi32 is like atoi but for integers
   407  // that fit into an int32.
   408  func atoi32(s string) (int32, bool) {
   409  	if n, ok := atoi64(s); n == int64(int32(n)) {
   410  		return int32(n), ok
   411  	}
   412  	return 0, false
   413  }
   414  
   415  // parseByteCount parses a string that represents a count of bytes.
   416  //
   417  // s must match the following regular expression:
   418  //
   419  //	^[0-9]+(([KMGT]i)?B)?$
   420  //
   421  // In other words, an integer byte count with an optional unit
   422  // suffix. Acceptable suffixes include one of
   423  // - KiB, MiB, GiB, TiB which represent binary IEC/ISO 80000 units, or
   424  // - B, which just represents bytes.
   425  //
   426  // Returns an int64 because that's what its callers want and receive,
   427  // but the result is always non-negative.
   428  func parseByteCount(s string) (int64, bool) {
   429  	// The empty string is not valid.
   430  	if s == "" {
   431  		return 0, false
   432  	}
   433  	// Handle the easy non-suffix case.
   434  	last := s[len(s)-1]
   435  	if last >= '0' && last <= '9' {
   436  		n, ok := atoi64(s)
   437  		if !ok || n < 0 {
   438  			return 0, false
   439  		}
   440  		return n, ok
   441  	}
   442  	// Failing a trailing digit, this must always end in 'B'.
   443  	// Also at this point there must be at least one digit before
   444  	// that B.
   445  	if last != 'B' || len(s) < 2 {
   446  		return 0, false
   447  	}
   448  	// The one before that must always be a digit or 'i'.
   449  	if c := s[len(s)-2]; c >= '0' && c <= '9' {
   450  		// Trivial 'B' suffix.
   451  		n, ok := atoi64(s[:len(s)-1])
   452  		if !ok || n < 0 {
   453  			return 0, false
   454  		}
   455  		return n, ok
   456  	} else if c != 'i' {
   457  		return 0, false
   458  	}
   459  	// Finally, we need at least 4 characters now, for the unit
   460  	// prefix and at least one digit.
   461  	if len(s) < 4 {
   462  		return 0, false
   463  	}
   464  	power := 0
   465  	switch s[len(s)-3] {
   466  	case 'K':
   467  		power = 1
   468  	case 'M':
   469  		power = 2
   470  	case 'G':
   471  		power = 3
   472  	case 'T':
   473  		power = 4
   474  	default:
   475  		// Invalid suffix.
   476  		return 0, false
   477  	}
   478  	m := uint64(1)
   479  	for i := 0; i < power; i++ {
   480  		m *= 1024
   481  	}
   482  	n, ok := atoi64(s[:len(s)-3])
   483  	if !ok || n < 0 {
   484  		return 0, false
   485  	}
   486  	un := uint64(n)
   487  	if un > maxUint64/m {
   488  		// Overflow.
   489  		return 0, false
   490  	}
   491  	un *= m
   492  	if un > uint64(maxInt64) {
   493  		// Overflow.
   494  		return 0, false
   495  	}
   496  	return int64(un), true
   497  }
   498  
   499  //go:nosplit
   500  func findnull(s *byte) int {
   501  	if s == nil {
   502  		return 0
   503  	}
   504  
   505  	// Avoid IndexByteString on Plan 9 because it uses SSE instructions
   506  	// on x86 machines, and those are classified as floating point instructions,
   507  	// which are illegal in a note handler.
   508  	if GOOS == "plan9" {
   509  		p := (*[maxAlloc/2 - 1]byte)(unsafe.Pointer(s))
   510  		l := 0
   511  		for p[l] != 0 {
   512  			l++
   513  		}
   514  		return l
   515  	}
   516  
   517  	// pageSize is the unit we scan at a time looking for NULL.
   518  	// It must be the minimum page size for any architecture Go
   519  	// runs on. It's okay (just a minor performance loss) if the
   520  	// actual system page size is larger than this value.
   521  	const pageSize = 4096
   522  
   523  	offset := 0
   524  	ptr := unsafe.Pointer(s)
   525  	// IndexByteString uses wide reads, so we need to be careful
   526  	// with page boundaries. Call IndexByteString on
   527  	// [ptr, endOfPage) interval.
   528  	safeLen := int(pageSize - uintptr(ptr)%pageSize)
   529  
   530  	for {
   531  		t := *(*string)(unsafe.Pointer(&stringStruct{ptr, safeLen}))
   532  		// Check one page at a time.
   533  		if i := bytealg.IndexByteString(t, 0); i != -1 {
   534  			return offset + i
   535  		}
   536  		// Move to next page
   537  		ptr = unsafe.Pointer(uintptr(ptr) + uintptr(safeLen))
   538  		offset += safeLen
   539  		safeLen = pageSize
   540  	}
   541  }
   542  
   543  func findnullw(s *uint16) int {
   544  	if s == nil {
   545  		return 0
   546  	}
   547  	p := (*[maxAlloc/2/2 - 1]uint16)(unsafe.Pointer(s))
   548  	l := 0
   549  	for p[l] != 0 {
   550  		l++
   551  	}
   552  	return l
   553  }
   554  
   555  //go:nosplit
   556  func gostringnocopy(str *byte) string {
   557  	ss := stringStruct{str: unsafe.Pointer(str), len: findnull(str)}
   558  	s := *(*string)(unsafe.Pointer(&ss))
   559  	return s
   560  }
   561  
   562  func gostringw(strw *uint16) string {
   563  	var buf [8]byte
   564  	str := (*[maxAlloc/2/2 - 1]uint16)(unsafe.Pointer(strw))
   565  	n1 := 0
   566  	for i := 0; str[i] != 0; i++ {
   567  		n1 += encoderune(buf[:], rune(str[i]))
   568  	}
   569  	s, b := rawstring(n1 + 4)
   570  	n2 := 0
   571  	for i := 0; str[i] != 0; i++ {
   572  		// check for race
   573  		if n2 >= n1 {
   574  			break
   575  		}
   576  		n2 += encoderune(b[n2:], rune(str[i]))
   577  	}
   578  	b[n2] = 0 // for luck
   579  	return s[:n2]
   580  }
   581
View as plain text