Source file src/internal/poll/fd_windows.go

     1  // Copyright 2017 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 poll
     6  
     7  import (
     8  	"errors"
     9  	"internal/race"
    10  	"internal/syscall/windows"
    11  	"io"
    12  	"runtime"
    13  	"sync"
    14  	"syscall"
    15  	"unicode/utf16"
    16  	"unicode/utf8"
    17  	"unsafe"
    18  )
    19  
    20  var (
    21  	initErr error
    22  	ioSync  uint64
    23  )
    24  
    25  // ifsHandlesOnly returns true if the system only has IFS handles for TCP sockets.
    26  // See https://support.microsoft.com/kb/2568167 for details.
    27  var ifsHandlesOnly = sync.OnceValue(func() bool {
    28  	protos := [2]int32{syscall.IPPROTO_TCP, 0}
    29  	var buf [32]syscall.WSAProtocolInfo
    30  	len := uint32(unsafe.Sizeof(buf))
    31  	n, err := syscall.WSAEnumProtocols(&protos[0], &buf[0], &len)
    32  	if err != nil {
    33  		return false
    34  	}
    35  	for i := range n {
    36  		if buf[i].ServiceFlags1&syscall.XP1_IFS_HANDLES == 0 {
    37  			return false
    38  		}
    39  	}
    40  	return true
    41  })
    42  
    43  // canSkipCompletionPortOnSuccess returns true if we use FILE_SKIP_COMPLETION_PORT_ON_SUCCESS for the given handle.
    44  // See https://support.microsoft.com/kb/2568167 for details.
    45  func canSkipCompletionPortOnSuccess(h syscall.Handle, isSocket bool) bool {
    46  	if !isSocket {
    47  		// Non-socket handles can use SetFileCompletionNotificationModes without problems.
    48  		return true
    49  	}
    50  	if ifsHandlesOnly() {
    51  		// If the system only has IFS handles for TCP sockets, then there is nothing else to check.
    52  		return true
    53  	}
    54  	var info syscall.WSAProtocolInfo
    55  	size := int32(unsafe.Sizeof(info))
    56  	if syscall.Getsockopt(h, syscall.SOL_SOCKET, windows.SO_PROTOCOL_INFOW, (*byte)(unsafe.Pointer(&info)), &size) != nil {
    57  		return false
    58  	}
    59  	return info.ServiceFlags1&syscall.XP1_IFS_HANDLES != 0
    60  }
    61  
    62  // InitWSA initiates the use of the Winsock DLL by the current process.
    63  // It is called from the net package at init time to avoid
    64  // loading ws2_32.dll when net is not used.
    65  var InitWSA = sync.OnceFunc(func() {
    66  	var d syscall.WSAData
    67  	e := syscall.WSAStartup(uint32(0x202), &d)
    68  	if e != nil {
    69  		initErr = e
    70  	}
    71  })
    72  
    73  // operation contains superset of data necessary to perform all async IO.
    74  type operation struct {
    75  	// Used by IOCP interface, it must be first field
    76  	// of the struct, as our code relies on it.
    77  	o syscall.Overlapped
    78  
    79  	// fields used by runtime.netpoll
    80  	runtimeCtx uintptr
    81  	mode       int32
    82  }
    83  
    84  func (o *operation) setOffset(off int64) {
    85  	o.o.OffsetHigh = uint32(off >> 32)
    86  	o.o.Offset = uint32(off)
    87  }
    88  
    89  func (fd *FD) overlapped(o *operation) *syscall.Overlapped {
    90  	if fd.isBlocking {
    91  		// Don't return the overlapped object if the file handle
    92  		// doesn't use overlapped I/O. It could be used, but
    93  		// that would then use the file pointer stored in the
    94  		// overlapped object rather than the real file pointer.
    95  		return nil
    96  	}
    97  	return &o.o
    98  }
    99  
   100  func newWsaBuf(b []byte) *syscall.WSABuf {
   101  	return &syscall.WSABuf{Buf: unsafe.SliceData(b), Len: uint32(len(b))}
   102  }
   103  
   104  var wsaBufsPool = sync.Pool{
   105  	New: func() any {
   106  		buf := make([]syscall.WSABuf, 0, 16)
   107  		return &buf
   108  	},
   109  }
   110  
   111  func newWSABufs(buf *[][]byte) *[]syscall.WSABuf {
   112  	bufsPtr := wsaBufsPool.Get().(*[]syscall.WSABuf)
   113  	*bufsPtr = (*bufsPtr)[:0]
   114  	for _, b := range *buf {
   115  		if len(b) == 0 {
   116  			*bufsPtr = append(*bufsPtr, syscall.WSABuf{})
   117  			continue
   118  		}
   119  		for len(b) > maxRW {
   120  			*bufsPtr = append(*bufsPtr, syscall.WSABuf{Len: maxRW, Buf: &b[0]})
   121  			b = b[maxRW:]
   122  		}
   123  		if len(b) > 0 {
   124  			*bufsPtr = append(*bufsPtr, syscall.WSABuf{Len: uint32(len(b)), Buf: &b[0]})
   125  		}
   126  	}
   127  	return bufsPtr
   128  }
   129  
   130  func freeWSABufs(bufsPtr *[]syscall.WSABuf) {
   131  	// Clear pointers to buffers so they can be released by garbage collector.
   132  	bufs := *bufsPtr
   133  	for i := range bufs {
   134  		bufs[i].Buf = nil
   135  	}
   136  	// Proper usage of a sync.Pool requires each entry to have approximately
   137  	// the same memory cost. To obtain this property when the stored type
   138  	// contains a variably-sized buffer, we add a hard limit on the maximum buffer
   139  	// to place back in the pool.
   140  	//
   141  	// See https://go.dev/issue/23199
   142  	if cap(*bufsPtr) > 128 {
   143  		*bufsPtr = nil
   144  	}
   145  	wsaBufsPool.Put(bufsPtr)
   146  }
   147  
   148  // wsaMsgPool is a pool of WSAMsg structures that can only hold a single WSABuf.
   149  var wsaMsgPool = sync.Pool{
   150  	New: func() any {
   151  		return &windows.WSAMsg{
   152  			Buffers:     &syscall.WSABuf{},
   153  			BufferCount: 1,
   154  		}
   155  	},
   156  }
   157  
   158  // newWSAMsg creates a new WSAMsg with the provided parameters.
   159  // Use [freeWSAMsg] to free it.
   160  func newWSAMsg(p []byte, oob []byte, flags int, rsa *wsaRsa) *windows.WSAMsg {
   161  	// The returned object can't be allocated in the stack because it is accessed asynchronously
   162  	// by Windows in between several system calls. If the stack frame is moved while that happens,
   163  	// then Windows may access invalid memory.
   164  
   165  	// Use a pool to reuse allocations.
   166  	msg := wsaMsgPool.Get().(*windows.WSAMsg)
   167  	msg.Buffers.Len = uint32(len(p))
   168  	msg.Buffers.Buf = unsafe.SliceData(p)
   169  	if len(oob) > 0 {
   170  		msg.Control = syscall.WSABuf{
   171  			Len: uint32(len(oob)),
   172  			Buf: unsafe.SliceData(oob),
   173  		}
   174  	}
   175  	msg.Flags = uint32(flags)
   176  	if rsa != nil {
   177  		msg.Name = &rsa.name
   178  		msg.Namelen = rsa.namelen
   179  	}
   180  	return msg
   181  }
   182  
   183  func freeWSAMsg(msg *windows.WSAMsg) {
   184  	// Clear pointers to buffers so they can be released by garbage collector.
   185  	msg.Name = nil
   186  	msg.Namelen = 0
   187  	msg.Buffers.Len = 0
   188  	msg.Buffers.Buf = nil
   189  	msg.Control.Len = 0
   190  	msg.Control.Buf = nil
   191  	wsaMsgPool.Put(msg)
   192  }
   193  
   194  // wsaRsa bundles a [syscall.RawSockaddrAny] with its length for efficient caching.
   195  //
   196  // When used by WSARecvFrom, wsaRsa must be on the heap. See
   197  // https://learn.microsoft.com/en-us/windows/win32/api/winsock2/nf-winsock2-wsarecvfrom.
   198  type wsaRsa struct {
   199  	name    syscall.RawSockaddrAny
   200  	namelen int32
   201  }
   202  
   203  var wsaRsaPool = sync.Pool{
   204  	New: func() any {
   205  		return new(wsaRsa)
   206  	},
   207  }
   208  
   209  func newWSARsa() *wsaRsa {
   210  	rsa := wsaRsaPool.Get().(*wsaRsa)
   211  	rsa.name = syscall.RawSockaddrAny{}
   212  	rsa.namelen = int32(unsafe.Sizeof(syscall.RawSockaddrAny{}))
   213  	return rsa
   214  }
   215  
   216  var operationPool = sync.Pool{
   217  	New: func() any {
   218  		return new(operation)
   219  	},
   220  }
   221  
   222  // waitIO waits for the IO operation to complete,
   223  // handling cancellation if necessary.
   224  func (fd *FD) waitIO(o *operation) error {
   225  	if o.o.HEvent != 0 {
   226  		// Close may have tried to cancel I/O before this request was
   227  		// submitted. Retry cancellation now that the request is pending.
   228  		if fd.kind == kindPipe && fd.closing() {
   229  			if err := syscall.CancelIoEx(fd.Sysfd, &o.o); err != nil && err != syscall.ERROR_NOT_FOUND {
   230  				// TODO: maybe do something else, but panic.
   231  				panic(err)
   232  			}
   233  		}
   234  		// The overlapped handle is not added to the runtime poller,
   235  		// the only way to wait for the IO to complete is block until
   236  		// the overlapped event is signaled.
   237  		_, err := syscall.WaitForSingleObject(o.o.HEvent, syscall.INFINITE)
   238  		return err
   239  	}
   240  	// Wait for our request to complete.
   241  	err := fd.pd.wait(int(o.mode), fd.isFile)
   242  	switch err {
   243  	case nil:
   244  		// IO completed successfully.
   245  	case ErrNetClosing, ErrFileClosing, ErrDeadlineExceeded:
   246  		// IO interrupted by "close" or "timeout", cancel our request.
   247  		// ERROR_NOT_FOUND can be returned when the request succeded
   248  		// between the time wait returned and CancelIoEx was executed.
   249  		if err := syscall.CancelIoEx(fd.Sysfd, &o.o); err != nil && err != syscall.ERROR_NOT_FOUND {
   250  			// TODO(brainman): maybe do something else, but panic.
   251  			panic(err)
   252  		}
   253  		fd.pd.waitCanceled(int(o.mode))
   254  	default:
   255  		// No other error is expected.
   256  		panic("unexpected runtime.netpoll error: " + err.Error())
   257  	}
   258  	return err
   259  }
   260  
   261  // execIO executes a single IO operation o.
   262  // It supports both synchronous and asynchronous IO.
   263  // pinPtrs is a list of pointers that will be pinned to a fixed location in memory
   264  // during the lifetime of the operation.
   265  func (fd *FD) execIO(
   266  	mode int,
   267  	submit func(o *operation) (uint32, error),
   268  	pinPtrs ...any,
   269  ) (int, error) {
   270  	if err := fd.ensureInit(); err != nil {
   271  		return 0, err
   272  	}
   273  	// Notify runtime netpoll about starting IO.
   274  	err := fd.pd.prepare(mode, fd.isFile)
   275  	if err != nil {
   276  		return 0, err
   277  	}
   278  	o := operationPool.Get().(*operation)
   279  	defer operationPool.Put(o)
   280  	*o = operation{
   281  		runtimeCtx: fd.pd.runtimeCtx,
   282  		mode:       int32(mode),
   283  	}
   284  	if !fd.isBlocking {
   285  		var pinner *runtime.Pinner
   286  		if mode == 'r' {
   287  			pinner = &fd.readPinner
   288  		} else {
   289  			pinner = &fd.writePinner
   290  		}
   291  		defer pinner.Unpin()
   292  
   293  		pinner.Pin(o)
   294  		for _, ptr := range pinPtrs {
   295  			pinner.Pin(ptr)
   296  		}
   297  
   298  		if !fd.associated {
   299  			// If the handle is opened for overlapped IO but we can't
   300  			// use the runtime poller, then we need to use an
   301  			// event to wait for the IO to complete.
   302  			h, err := windows.CreateEvent(nil, 0, 0, nil)
   303  			if err != nil {
   304  				// This shouldn't happen when all CreateEvent arguments are zero.
   305  				panic(err)
   306  			}
   307  			// Set the low bit so that the external IOCP doesn't receive the completion packet.
   308  			o.o.HEvent = h | 1
   309  			defer syscall.CloseHandle(h)
   310  		}
   311  	}
   312  	// Start IO.
   313  	qty, err := submit(o)
   314  	var waitErr error
   315  	// An event-backed operation that succeeds inline is already complete.
   316  	// Only IOCP may require waiting for a completion packet on success.
   317  	waitOnSuccess := o.o.HEvent == 0 && fd.waitOnSuccess
   318  	// Blocking operations shouldn't return ERROR_IO_PENDING.
   319  	// Continue without waiting if that happens.
   320  	if !fd.isBlocking && (err == syscall.ERROR_IO_PENDING || (err == nil && waitOnSuccess)) {
   321  		// IO started asynchronously or completed synchronously but
   322  		// an IOCP completion packet is expected. Wait for completion.
   323  		waitErr = fd.waitIO(o)
   324  		if fd.isFile {
   325  			err = windows.GetOverlappedResult(fd.Sysfd, &o.o, &qty, false)
   326  		} else {
   327  			var flags uint32
   328  			err = windows.WSAGetOverlappedResult(fd.Sysfd, &o.o, &qty, false, &flags)
   329  		}
   330  	}
   331  	switch err {
   332  	case syscall.ERROR_OPERATION_ABORTED:
   333  		// ERROR_OPERATION_ABORTED may have been caused by us. In that case,
   334  		// map it to our own error. Don't do more than that, each submitted
   335  		// function may have its own meaning for each error.
   336  		if waitErr != nil {
   337  			// IO canceled by the poller while waiting for completion.
   338  			err = waitErr
   339  		} else if fd.kind == kindPipe && fd.closing() {
   340  			// Close uses CancelIoEx to interrupt concurrent I/O for pipes.
   341  			// If the fd is a pipe and the I/O was interrupted by CancelIoEx,
   342  			// we assume it is interrupted by Close.
   343  			err = errClosing(fd.isFile)
   344  		}
   345  	case windows.ERROR_IO_INCOMPLETE:
   346  		// waitIO couldn't wait for the IO to complete.
   347  		if waitErr != nil {
   348  			// The wait error will be more informative.
   349  			err = waitErr
   350  		}
   351  	}
   352  	return int(qty), err
   353  }
   354  
   355  // FD is a file descriptor. The net and os packages embed this type in
   356  // a larger type representing a network connection or OS file.
   357  type FD struct {
   358  	// Lock sysfd and serialize access to Read and Write methods.
   359  	fdmu fdMutex
   360  
   361  	// System file descriptor. Immutable until Close.
   362  	Sysfd syscall.Handle
   363  
   364  	// I/O poller.
   365  	pd pollDesc
   366  
   367  	// lazyInit is set by Init before the FD is made available to callers.
   368  	// initOnce serializes first use; initMu protects initialization against
   369  	// Close and DisassociateIOCP. Close cancels I/O before waiting for initMu,
   370  	// and DisassociateIOCP only tries to lock it.
   371  	lazyInit bool
   372  	initOnce sync.Once
   373  	initMu   sync.Mutex
   374  	skipIOCP bool // protected by initMu
   375  
   376  	// The file offset for the next read or write.
   377  	// Overlapped IO operations don't use the real file pointer,
   378  	// so we need to keep track of the offset ourselves.
   379  	// Read and Write only use this for kindFile.
   380  	// Protected by both the read and write locks.
   381  	offset int64
   382  
   383  	// For console I/O.
   384  	lastbits       []byte   // first few bytes of the last incomplete rune in last write
   385  	readuint16     []uint16 // buffer to hold uint16s obtained with ReadConsole
   386  	readbyte       []byte   // buffer to hold decoding of readuint16 from utf16 to utf8
   387  	readbyteOffset int      // readbyte[readOffset:] is yet to be consumed with file.Read
   388  
   389  	// Semaphore signaled when file is closed.
   390  	csema uint32
   391  
   392  	// Whether to wait for an IOCP completion packet for operations that
   393  	// complete synchronously. Only used while associated is true.
   394  	waitOnSuccess bool
   395  
   396  	// Whether this is a streaming descriptor, as opposed to a
   397  	// packet-based descriptor like a UDP socket.
   398  	IsStream bool
   399  
   400  	// Whether a zero byte read indicates EOF. This is false for a
   401  	// message based socket connection.
   402  	ZeroReadIsEOF bool
   403  
   404  	// KeepFileCompletionModes prevents Init from changing the file object's
   405  	// completion notification modes.
   406  	KeepFileCompletionModes bool
   407  
   408  	// Whether the handle is owned by os.File.
   409  	isFile bool
   410  
   411  	// The kind of this file.
   412  	kind fileKind
   413  
   414  	// Whether FILE_FLAG_OVERLAPPED was not set when opening the file.
   415  	isBlocking bool
   416  
   417  	// Whether the handle is currently associated with the IOCP.
   418  	associated bool
   419  
   420  	// readPinner and writePinner are automatically unpinned
   421  	// before execIO returns.
   422  	readPinner  runtime.Pinner
   423  	writePinner runtime.Pinner
   424  }
   425  
   426  // setOffset sets the offset fields of the overlapped object
   427  // to the given offset. The fd read/write lock must be held.
   428  //
   429  // Overlapped IO operations don't update the offset fields
   430  // of the overlapped object nor the file pointer automatically,
   431  // so we do that manually here.
   432  // Note that this is a best effort that only works if the file
   433  // pointer is completely owned by this operation. We could
   434  // call seek to allow other processes or other operations on the
   435  // same file to see the updated offset. That would be inefficient
   436  // and won't work for concurrent operations anyway. If concurrent
   437  // operations are needed, then the caller should serialize them
   438  // using an external mechanism.
   439  func (fd *FD) setOffset(off int64) {
   440  	fd.offset = off
   441  }
   442  
   443  // addOffset adds the given offset to the current offset.
   444  func (fd *FD) addOffset(off int) {
   445  	fd.offset += int64(off)
   446  }
   447  
   448  // fileKind describes the kind of file.
   449  type fileKind byte
   450  
   451  const (
   452  	kindNet fileKind = iota
   453  	kindFile
   454  	kindConsole
   455  	kindPipe
   456  )
   457  
   458  // Init initializes the FD. The Sysfd field should already be set.
   459  // This can be called multiple times on a single FD.
   460  // The net argument is a network name from the net package (e.g., "tcp"),
   461  // or "file", "console", or "pipe".
   462  // The overlapped argument reports whether the handle was opened for overlapped I/O.
   463  // Such handles use the runtime poller when possible, or explicit events otherwise.
   464  // If overlapped is nil, mode detection and poller initialization are deferred
   465  // until first use.
   466  func (fd *FD) Init(net string, overlapped *bool) error {
   467  	if initErr != nil {
   468  		return initErr
   469  	}
   470  
   471  	switch net {
   472  	case "file":
   473  		fd.kind = kindFile
   474  	case "console":
   475  		fd.kind = kindConsole
   476  	case "pipe":
   477  		fd.kind = kindPipe
   478  	default:
   479  		// We don't actually care about the various network types.
   480  		fd.kind = kindNet
   481  	}
   482  	fd.isFile = fd.kind != kindNet
   483  	fd.lazyInit = overlapped == nil
   484  	fd.isBlocking = overlapped == nil || !*overlapped
   485  
   486  	if fd.isBlocking {
   487  		return nil
   488  	}
   489  	return fd.initIOCP()
   490  }
   491  
   492  // ensureInit resolves the I/O mode and attempts poller initialization on first
   493  // use. The caller must hold an FD reference throughout this call and its I/O.
   494  func (fd *FD) ensureInit() error {
   495  	if !fd.lazyInit {
   496  		return nil
   497  	}
   498  	fd.initOnce.Do(func() {
   499  		fd.initMu.Lock()
   500  		defer fd.initMu.Unlock()
   501  		if fd.closing() {
   502  			return
   503  		}
   504  
   505  		// This query may block behind synchronous I/O, including I/O in
   506  		// another process. Close cancels I/O before waiting for initMu,
   507  		// and raw-handle access does not wait for it.
   508  		overlapped, _ := windows.IsNonblock(fd.Sysfd)
   509  		fd.isBlocking = !overlapped
   510  		if overlapped && !fd.skipIOCP && !fd.closing() {
   511  			// Like os.NewFile's eager initialization, association failures
   512  			// fall back to event-backed I/O rather than failing the operation.
   513  			_ = fd.initIOCP()
   514  		}
   515  	})
   516  	if fd.closing() {
   517  		return errClosing(fd.isFile)
   518  	}
   519  	return nil
   520  }
   521  
   522  // initIOCP sets up the runtime poller and completion notification modes for an
   523  // overlapped handle. If the existing notification modes cannot be determined,
   524  // it leaves the handle unassociated for event-backed I/O.
   525  func (fd *FD) initIOCP() error {
   526  	var modes uint32
   527  	if fd.KeepFileCompletionModes {
   528  		// Query before associating: we must know whether inline success
   529  		// queues a completion packet before using the runtime poller.
   530  		var err error
   531  		modes, err = fd.getFileCompletionModes()
   532  		if err != nil {
   533  			// Without knowing the modes, neither waiting for a completion
   534  			// packet on success nor skipping it is safe. Leave the handle
   535  			// unassociated and use explicit events for pending I/O instead.
   536  			// Inline success needs no wait, and deadlines are unavailable.
   537  			return nil
   538  		}
   539  	}
   540  
   541  	// It is safe to add overlapped handles that also perform I/O
   542  	// outside of the runtime poller. The runtime poller will ignore
   543  	// I/O completion notifications not initiated by us.
   544  	if err := fd.pd.init(fd); err != nil {
   545  		return err
   546  	}
   547  	fd.associated = true
   548  
   549  	if !fd.KeepFileCompletionModes {
   550  		// Only change notification modes after association succeeds.
   551  		modes = fd.setFileCompletionModes()
   552  	}
   553  	fd.waitOnSuccess = modes&syscall.FILE_SKIP_COMPLETION_PORT_ON_SUCCESS == 0
   554  	return nil
   555  }
   556  
   557  // getFileCompletionModes queries the file object's completion notification modes.
   558  func (fd *FD) getFileCompletionModes() (uint32, error) {
   559  	var info windows.FILE_IO_COMPLETION_NOTIFICATION_INFORMATION
   560  	err := windows.NtQueryInformationFile(fd.Sysfd, &windows.IO_STATUS_BLOCK{},
   561  		unsafe.Pointer(&info), uint32(unsafe.Sizeof(info)), windows.FileIoCompletionNotificationInformation)
   562  	return info.Flags, err
   563  }
   564  
   565  // setFileCompletionModes enables completion notification optimizations and returns
   566  // the requested modes on success, or zero if the request fails.
   567  func (fd *FD) setFileCompletionModes() uint32 {
   568  	// Suppressing the file object's event saves work for the I/O manager.
   569  	// Explicit per-operation events are still signaled.
   570  	// See https://devblogs.microsoft.com/oldnewthing/20200221-00/?p=103466.
   571  	modes := uint8(syscall.FILE_SKIP_SET_EVENT_ON_HANDLE)
   572  	if canSkipCompletionPortOnSuccess(fd.Sysfd, fd.kind == kindNet) {
   573  		modes |= syscall.FILE_SKIP_COMPLETION_PORT_ON_SUCCESS
   574  	}
   575  	if syscall.SetFileCompletionNotificationModes(fd.Sysfd, modes) != nil {
   576  		// Retain the default policy of waiting for an inline-success packet
   577  		// unless skip-success was enabled successfully.
   578  		return 0
   579  	}
   580  	return uint32(modes)
   581  }
   582  
   583  // DisassociateIOCP disassociates the file handle from the IOCP.
   584  // The disassociate operation will not succeed if there is any
   585  // in-progress I/O operation on the file handle.
   586  func (fd *FD) DisassociateIOCP() error {
   587  	// Hold both I/O locks while changing the completion mechanism. Don't wait
   588  	// for them, since an I/O operation might block indefinitely.
   589  	// NtSetInformationFile also rejects handles with pending I/O outside this FD.
   590  	if ok, err := fd.tryReadWriteLock(); err != nil || !ok {
   591  		if err == nil {
   592  			err = errors.New("can't disassociate the handle while there is in-progress I/O")
   593  		}
   594  		return err
   595  	}
   596  	defer fd.readWriteUnlock()
   597  
   598  	if fd.lazyInit {
   599  		if !fd.initMu.TryLock() {
   600  			// A deadline setter can initialize without holding the I/O locks.
   601  			return errors.New("can't disassociate the handle while initialization is in progress")
   602  		}
   603  		defer fd.initMu.Unlock()
   604  		// If initialization has not started, suppress future association
   605  		// without querying the handle's mode.
   606  		fd.skipIOCP = true
   607  	}
   608  	if !fd.associated {
   609  		// Nothing to disassociate.
   610  		return nil
   611  	}
   612  
   613  	info := windows.FILE_COMPLETION_INFORMATION{}
   614  	if err := windows.NtSetInformationFile(fd.Sysfd, &windows.IO_STATUS_BLOCK{}, unsafe.Pointer(&info), uint32(unsafe.Sizeof(info)), windows.FileReplaceCompletionInformation); err != nil {
   615  		return err
   616  	}
   617  	// tryReadWriteLock means we have exclusive access to fd.
   618  	fd.associated = false
   619  	// Don't call fd.pd.close(), it would be too racy.
   620  	// There is no harm on leaving fd.pd open until Close is called.
   621  	return nil
   622  }
   623  
   624  func (fd *FD) destroy() error {
   625  	if fd.Sysfd == syscall.InvalidHandle {
   626  		return syscall.EINVAL
   627  	}
   628  	// Poller may want to unregister fd in readiness notification mechanism,
   629  	// so this must be executed before fd.CloseFunc.
   630  	fd.pd.close()
   631  	var err error
   632  	switch fd.kind {
   633  	case kindNet:
   634  		// The net package uses the CloseFunc variable for testing.
   635  		err = CloseFunc(fd.Sysfd)
   636  	default:
   637  		err = syscall.CloseHandle(fd.Sysfd)
   638  	}
   639  	fd.Sysfd = syscall.InvalidHandle
   640  	runtime_Semrelease(&fd.csema)
   641  	return err
   642  }
   643  
   644  // Close closes the FD. The underlying file descriptor is closed by
   645  // the destroy method when there are no remaining references.
   646  func (fd *FD) Close() error {
   647  	if !fd.fdmu.increfAndClose() {
   648  		return errClosing(fd.isFile)
   649  	}
   650  
   651  	if fd.kind == kindPipe {
   652  		syscall.CancelIoEx(fd.Sysfd, nil)
   653  	}
   654  	// unblock pending reader and writer
   655  	if fd.lazyInit {
   656  		// Cancel before waiting for initialization, then evict its descriptor.
   657  		fd.initMu.Lock()
   658  		fd.pd.evict()
   659  		fd.initMu.Unlock()
   660  	} else {
   661  		fd.pd.evict()
   662  	}
   663  	err := fd.decref()
   664  	// Wait until the descriptor is closed. If this was the only
   665  	// reference, it is already closed.
   666  	runtime_Semacquire(&fd.csema)
   667  	return err
   668  }
   669  
   670  // Windows ReadFile and WSARecv use DWORD (uint32) parameter to pass buffer length.
   671  // This prevents us reading blocks larger than 4GB.
   672  // See golang.org/issue/26923.
   673  const maxRW = 1 << 30 // 1GB is large enough and keeps subsequent reads aligned
   674  
   675  func pinPtrsFromBuf(buf []byte) []any {
   676  	if len(buf) == 0 {
   677  		return nil
   678  	}
   679  	return []any{unsafe.SliceData(buf)}
   680  }
   681  
   682  // Read implements io.Reader.
   683  func (fd *FD) Read(buf []byte) (int, error) {
   684  	if fd.kind == kindFile {
   685  		if err := fd.readWriteLock(); err != nil {
   686  			return 0, err
   687  		}
   688  		defer fd.readWriteUnlock()
   689  	} else {
   690  		if err := fd.readLock(); err != nil {
   691  			return 0, err
   692  		}
   693  		defer fd.readUnlock()
   694  	}
   695  
   696  	if len(buf) > maxRW {
   697  		buf = buf[:maxRW]
   698  	}
   699  
   700  	var n int
   701  	var err error
   702  	switch fd.kind {
   703  	case kindConsole:
   704  		n, err = fd.readConsole(buf)
   705  	case kindFile, kindPipe:
   706  		n, err = fd.execIO('r', func(o *operation) (qty uint32, err error) {
   707  			if fd.kind == kindFile {
   708  				o.setOffset(fd.offset)
   709  			}
   710  			err = syscall.ReadFile(fd.Sysfd, buf, &qty, fd.overlapped(o))
   711  			return qty, err
   712  		}, pinPtrsFromBuf(buf)...)
   713  		if fd.kind == kindFile {
   714  			fd.addOffset(n)
   715  		}
   716  		switch err {
   717  		case syscall.ERROR_HANDLE_EOF:
   718  			err = io.EOF
   719  		case syscall.ERROR_BROKEN_PIPE:
   720  			// ReadFile only documents ERROR_BROKEN_PIPE for pipes.
   721  			if fd.kind == kindPipe {
   722  				err = io.EOF
   723  			}
   724  		}
   725  	case kindNet:
   726  		n, err = fd.execIO('r', func(o *operation) (qty uint32, err error) {
   727  			var flags uint32
   728  			err = syscall.WSARecv(fd.Sysfd, newWsaBuf(buf), 1, &qty, &flags, &o.o, nil)
   729  			return qty, err
   730  		}, pinPtrsFromBuf(buf)...)
   731  		if race.Enabled {
   732  			race.Acquire(unsafe.Pointer(&ioSync))
   733  		}
   734  	}
   735  	if len(buf) != 0 {
   736  		err = fd.eofError(n, err)
   737  	}
   738  	return n, err
   739  }
   740  
   741  var ReadConsole = syscall.ReadConsole // changed for testing
   742  
   743  // readConsole reads utf16 characters from console File,
   744  // encodes them into utf8 and stores them in buffer b.
   745  // It returns the number of utf8 bytes read and an error, if any.
   746  func (fd *FD) readConsole(b []byte) (int, error) {
   747  	if len(b) == 0 {
   748  		return 0, nil
   749  	}
   750  
   751  	if fd.readuint16 == nil {
   752  		// Note: syscall.ReadConsole fails for very large buffers.
   753  		// The limit is somewhere around (but not exactly) 16384.
   754  		// Stay well below.
   755  		fd.readuint16 = make([]uint16, 0, 10000)
   756  		fd.readbyte = make([]byte, 0, 4*cap(fd.readuint16))
   757  	}
   758  
   759  	for fd.readbyteOffset >= len(fd.readbyte) {
   760  		n := cap(fd.readuint16) - len(fd.readuint16)
   761  		if n > len(b) {
   762  			n = len(b)
   763  		}
   764  		var nw uint32
   765  		err := ReadConsole(fd.Sysfd, &fd.readuint16[:len(fd.readuint16)+1][len(fd.readuint16)], uint32(n), &nw, nil)
   766  		if err != nil {
   767  			return 0, err
   768  		}
   769  		uint16s := fd.readuint16[:len(fd.readuint16)+int(nw)]
   770  		fd.readuint16 = fd.readuint16[:0]
   771  		buf := fd.readbyte[:0]
   772  		for i := 0; i < len(uint16s); i++ {
   773  			r := rune(uint16s[i])
   774  			if utf16.IsSurrogate(r) {
   775  				if i+1 == len(uint16s) {
   776  					if nw > 0 {
   777  						// Save half surrogate pair for next time.
   778  						fd.readuint16 = fd.readuint16[:1]
   779  						fd.readuint16[0] = uint16(r)
   780  						break
   781  					}
   782  					r = utf8.RuneError
   783  				} else {
   784  					r = utf16.DecodeRune(r, rune(uint16s[i+1]))
   785  					if r != utf8.RuneError {
   786  						i++
   787  					}
   788  				}
   789  			}
   790  			buf = utf8.AppendRune(buf, r)
   791  		}
   792  		fd.readbyte = buf
   793  		fd.readbyteOffset = 0
   794  		if nw == 0 {
   795  			break
   796  		}
   797  	}
   798  
   799  	src := fd.readbyte[fd.readbyteOffset:]
   800  	var i int
   801  	for i = 0; i < len(src) && i < len(b); i++ {
   802  		x := src[i]
   803  		if x == 0x1A { // Ctrl-Z
   804  			if i == 0 {
   805  				fd.readbyteOffset++
   806  			}
   807  			break
   808  		}
   809  		b[i] = x
   810  	}
   811  	fd.readbyteOffset += i
   812  	return i, nil
   813  }
   814  
   815  // Pread emulates the Unix pread system call.
   816  func (fd *FD) Pread(buf []byte, off int64) (int, error) {
   817  	if fd.kind == kindPipe {
   818  		// Pread does not work with pipes
   819  		return 0, syscall.ESPIPE
   820  	}
   821  
   822  	if err := fd.readWriteLock(); err != nil {
   823  		return 0, err
   824  	}
   825  	defer fd.readWriteUnlock()
   826  
   827  	if len(buf) > maxRW {
   828  		buf = buf[:maxRW]
   829  	}
   830  
   831  	n, err := fd.execIO('r', func(o *operation) (qty uint32, err error) {
   832  		// Overlapped handles don't have the file pointer updated
   833  		// when performing I/O operations, so there is no need to
   834  		// call Seek to reset the file pointer.
   835  		// Also, some overlapped file handles don't support seeking.
   836  		// See https://go.dev/issues/74951.
   837  		if fd.isBlocking {
   838  			curoffset, err := syscall.Seek(fd.Sysfd, 0, io.SeekCurrent)
   839  			if err != nil {
   840  				return 0, err
   841  			}
   842  			defer syscall.Seek(fd.Sysfd, curoffset, io.SeekStart)
   843  		}
   844  		o.setOffset(off)
   845  
   846  		err = syscall.ReadFile(fd.Sysfd, buf, &qty, &o.o)
   847  		return qty, err
   848  	}, pinPtrsFromBuf(buf)...)
   849  	if err == syscall.ERROR_HANDLE_EOF {
   850  		err = io.EOF
   851  	}
   852  	if len(buf) != 0 {
   853  		err = fd.eofError(n, err)
   854  	}
   855  	return n, err
   856  }
   857  
   858  // ReadFrom wraps the recvfrom network call.
   859  func (fd *FD) ReadFrom(buf []byte) (int, syscall.Sockaddr, error) {
   860  	if len(buf) == 0 {
   861  		return 0, nil, nil
   862  	}
   863  	if len(buf) > maxRW {
   864  		buf = buf[:maxRW]
   865  	}
   866  	if err := fd.readLock(); err != nil {
   867  		return 0, nil, err
   868  	}
   869  	defer fd.readUnlock()
   870  
   871  	rsa := newWSARsa()
   872  	defer wsaRsaPool.Put(rsa)
   873  	n, err := fd.execIO('r', func(o *operation) (qty uint32, err error) {
   874  		var flags uint32
   875  		err = syscall.WSARecvFrom(fd.Sysfd, newWsaBuf(buf), 1, &qty, &flags, &rsa.name, &rsa.namelen, &o.o, nil)
   876  		return qty, err
   877  	}, unsafe.SliceData(buf), rsa)
   878  	err = fd.eofError(n, err)
   879  	if err != nil {
   880  		return n, nil, err
   881  	}
   882  	sa, _ := rsa.name.Sockaddr()
   883  	return n, sa, nil
   884  }
   885  
   886  // ReadFromInet4 wraps the recvfrom network call for IPv4.
   887  func (fd *FD) ReadFromInet4(buf []byte, sa4 *syscall.SockaddrInet4) (int, error) {
   888  	if len(buf) == 0 {
   889  		return 0, nil
   890  	}
   891  	if len(buf) > maxRW {
   892  		buf = buf[:maxRW]
   893  	}
   894  	if err := fd.readLock(); err != nil {
   895  		return 0, err
   896  	}
   897  	defer fd.readUnlock()
   898  
   899  	rsa := newWSARsa()
   900  	defer wsaRsaPool.Put(rsa)
   901  	n, err := fd.execIO('r', func(o *operation) (qty uint32, err error) {
   902  		var flags uint32
   903  		err = syscall.WSARecvFrom(fd.Sysfd, newWsaBuf(buf), 1, &qty, &flags, &rsa.name, &rsa.namelen, &o.o, nil)
   904  		return qty, err
   905  	}, unsafe.SliceData(buf), rsa)
   906  	err = fd.eofError(n, err)
   907  	if err != nil {
   908  		return n, err
   909  	}
   910  	rawToSockaddrInet4(&rsa.name, sa4)
   911  	return n, err
   912  }
   913  
   914  // ReadFromInet6 wraps the recvfrom network call for IPv6.
   915  func (fd *FD) ReadFromInet6(buf []byte, sa6 *syscall.SockaddrInet6) (int, error) {
   916  	if len(buf) == 0 {
   917  		return 0, nil
   918  	}
   919  	if len(buf) > maxRW {
   920  		buf = buf[:maxRW]
   921  	}
   922  	if err := fd.readLock(); err != nil {
   923  		return 0, err
   924  	}
   925  	defer fd.readUnlock()
   926  
   927  	rsa := newWSARsa()
   928  	defer wsaRsaPool.Put(rsa)
   929  	n, err := fd.execIO('r', func(o *operation) (qty uint32, err error) {
   930  		var flags uint32
   931  		err = syscall.WSARecvFrom(fd.Sysfd, newWsaBuf(buf), 1, &qty, &flags, &rsa.name, &rsa.namelen, &o.o, nil)
   932  		return qty, err
   933  	}, unsafe.SliceData(buf), rsa)
   934  	err = fd.eofError(n, err)
   935  	if err != nil {
   936  		return n, err
   937  	}
   938  	rawToSockaddrInet6(&rsa.name, sa6)
   939  	return n, err
   940  }
   941  
   942  // Write implements io.Writer.
   943  func (fd *FD) Write(buf []byte) (int, error) {
   944  	if fd.kind == kindFile {
   945  		if err := fd.readWriteLock(); err != nil {
   946  			return 0, err
   947  		}
   948  		defer fd.readWriteUnlock()
   949  	} else {
   950  		if err := fd.writeLock(); err != nil {
   951  			return 0, err
   952  		}
   953  		defer fd.writeUnlock()
   954  	}
   955  
   956  	var ntotal int
   957  	for {
   958  		max := len(buf)
   959  		if max-ntotal > maxRW {
   960  			max = ntotal + maxRW
   961  		}
   962  		b := buf[ntotal:max]
   963  		var n int
   964  		var err error
   965  		switch fd.kind {
   966  		case kindConsole:
   967  			n, err = fd.writeConsole(b)
   968  		case kindPipe, kindFile:
   969  			n, err = fd.execIO('w', func(o *operation) (qty uint32, err error) {
   970  				if fd.kind == kindFile {
   971  					o.setOffset(fd.offset)
   972  				}
   973  				err = syscall.WriteFile(fd.Sysfd, b, &qty, fd.overlapped(o))
   974  				return qty, err
   975  			}, pinPtrsFromBuf(b)...)
   976  			if fd.kind == kindFile {
   977  				fd.addOffset(n)
   978  			}
   979  		case kindNet:
   980  			if race.Enabled {
   981  				race.ReleaseMerge(unsafe.Pointer(&ioSync))
   982  			}
   983  			n, err = fd.execIO('w', func(o *operation) (qty uint32, err error) {
   984  				err = syscall.WSASend(fd.Sysfd, newWsaBuf(b), 1, &qty, 0, &o.o, nil)
   985  				return qty, err
   986  			}, pinPtrsFromBuf(b)...)
   987  		}
   988  		ntotal += n
   989  		if ntotal == len(buf) || err != nil {
   990  			return ntotal, err
   991  		}
   992  		if n == 0 {
   993  			return ntotal, io.ErrUnexpectedEOF
   994  		}
   995  	}
   996  }
   997  
   998  // writeConsole writes len(b) bytes to the console File.
   999  // It returns the number of bytes written and an error, if any.
  1000  func (fd *FD) writeConsole(b []byte) (int, error) {
  1001  	n := len(b)
  1002  	runes := make([]rune, 0, 256)
  1003  	if len(fd.lastbits) > 0 {
  1004  		b = append(fd.lastbits, b...)
  1005  		fd.lastbits = nil
  1006  
  1007  	}
  1008  	for len(b) >= utf8.UTFMax || utf8.FullRune(b) {
  1009  		r, l := utf8.DecodeRune(b)
  1010  		runes = append(runes, r)
  1011  		b = b[l:]
  1012  	}
  1013  	if len(b) > 0 {
  1014  		fd.lastbits = make([]byte, len(b))
  1015  		copy(fd.lastbits, b)
  1016  	}
  1017  	// syscall.WriteConsole seems to fail, if given large buffer.
  1018  	// So limit the buffer to 16000 characters. This number was
  1019  	// discovered by experimenting with syscall.WriteConsole.
  1020  	const maxWrite = 16000
  1021  	for len(runes) > 0 {
  1022  		m := len(runes)
  1023  		if m > maxWrite {
  1024  			m = maxWrite
  1025  		}
  1026  		chunk := runes[:m]
  1027  		runes = runes[m:]
  1028  		uint16s := utf16.Encode(chunk)
  1029  		for len(uint16s) > 0 {
  1030  			var written uint32
  1031  			err := syscall.WriteConsole(fd.Sysfd, &uint16s[0], uint32(len(uint16s)), &written, nil)
  1032  			if err != nil {
  1033  				return 0, err
  1034  			}
  1035  			uint16s = uint16s[written:]
  1036  		}
  1037  	}
  1038  	return n, nil
  1039  }
  1040  
  1041  // Pwrite emulates the Unix pwrite system call.
  1042  func (fd *FD) Pwrite(buf []byte, off int64) (int, error) {
  1043  	if fd.kind == kindPipe {
  1044  		// Pwrite does not work with pipes
  1045  		return 0, syscall.ESPIPE
  1046  	}
  1047  
  1048  	if err := fd.readWriteLock(); err != nil {
  1049  		return 0, err
  1050  	}
  1051  	defer fd.readWriteUnlock()
  1052  
  1053  	var ntotal int
  1054  	for {
  1055  		max := len(buf)
  1056  		if max-ntotal > maxRW {
  1057  			max = ntotal + maxRW
  1058  		}
  1059  		b := buf[ntotal:max]
  1060  		n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1061  			// Overlapped handles don't have the file pointer updated
  1062  			// when performing I/O operations, so there is no need to
  1063  			// call Seek to reset the file pointer.
  1064  			// Also, some overlapped file handles don't support seeking.
  1065  			// See https://go.dev/issues/74951.
  1066  			if fd.isBlocking {
  1067  				curoffset, err := syscall.Seek(fd.Sysfd, 0, io.SeekCurrent)
  1068  				if err != nil {
  1069  					return 0, err
  1070  				}
  1071  				defer syscall.Seek(fd.Sysfd, curoffset, io.SeekStart)
  1072  			}
  1073  			o.setOffset(off + int64(ntotal))
  1074  
  1075  			err = syscall.WriteFile(fd.Sysfd, b, &qty, &o.o)
  1076  			return qty, err
  1077  		}, pinPtrsFromBuf(b)...)
  1078  		if n > 0 {
  1079  			ntotal += n
  1080  		}
  1081  		if ntotal == len(buf) || err != nil {
  1082  			return ntotal, err
  1083  		}
  1084  		if n == 0 {
  1085  			return ntotal, io.ErrUnexpectedEOF
  1086  		}
  1087  	}
  1088  }
  1089  
  1090  // Writev emulates the Unix writev system call.
  1091  func (fd *FD) Writev(buf *[][]byte) (int64, error) {
  1092  	if len(*buf) == 0 {
  1093  		return 0, nil
  1094  	}
  1095  	if err := fd.writeLock(); err != nil {
  1096  		return 0, err
  1097  	}
  1098  	defer fd.writeUnlock()
  1099  	if race.Enabled {
  1100  		race.ReleaseMerge(unsafe.Pointer(&ioSync))
  1101  	}
  1102  	bufs := newWSABufs(buf)
  1103  	defer freeWSABufs(bufs)
  1104  	n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1105  		err = syscall.WSASend(fd.Sysfd, &(*bufs)[0], uint32(len(*bufs)), &qty, 0, &o.o, nil)
  1106  		return qty, err
  1107  	})
  1108  	TestHookDidWritev(n)
  1109  	consume(buf, int64(n))
  1110  	return int64(n), err
  1111  }
  1112  
  1113  // WriteTo wraps the sendto network call.
  1114  func (fd *FD) WriteTo(buf []byte, sa syscall.Sockaddr) (int, error) {
  1115  	if err := fd.writeLock(); err != nil {
  1116  		return 0, err
  1117  	}
  1118  	defer fd.writeUnlock()
  1119  
  1120  	if len(buf) == 0 {
  1121  		// handle zero-byte payload
  1122  		n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1123  			err = syscall.WSASendto(fd.Sysfd, &syscall.WSABuf{}, 1, &qty, 0, sa, &o.o, nil)
  1124  			return qty, err
  1125  		})
  1126  		return n, err
  1127  	}
  1128  
  1129  	ntotal := 0
  1130  	for len(buf) > 0 {
  1131  		b := buf
  1132  		if len(b) > maxRW {
  1133  			b = b[:maxRW]
  1134  		}
  1135  		n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1136  			err = syscall.WSASendto(fd.Sysfd, newWsaBuf(b), 1, &qty, 0, sa, &o.o, nil)
  1137  			return qty, err
  1138  		}, unsafe.SliceData(b))
  1139  		ntotal += int(n)
  1140  		if err != nil {
  1141  			return ntotal, err
  1142  		}
  1143  		buf = buf[n:]
  1144  	}
  1145  	return ntotal, nil
  1146  }
  1147  
  1148  // WriteToInet4 is WriteTo, specialized for syscall.SockaddrInet4.
  1149  func (fd *FD) WriteToInet4(buf []byte, sa4 *syscall.SockaddrInet4) (int, error) {
  1150  	if err := fd.writeLock(); err != nil {
  1151  		return 0, err
  1152  	}
  1153  	defer fd.writeUnlock()
  1154  
  1155  	if len(buf) == 0 {
  1156  		// handle zero-byte payload
  1157  		n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1158  			err = windows.WSASendtoInet4(fd.Sysfd, &syscall.WSABuf{}, 1, &qty, 0, sa4, &o.o, nil)
  1159  			return qty, err
  1160  		})
  1161  		return n, err
  1162  	}
  1163  
  1164  	ntotal := 0
  1165  	for len(buf) > 0 {
  1166  		b := buf
  1167  		if len(b) > maxRW {
  1168  			b = b[:maxRW]
  1169  		}
  1170  		n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1171  			err = windows.WSASendtoInet4(fd.Sysfd, newWsaBuf(b), 1, &qty, 0, sa4, &o.o, nil)
  1172  			return qty, err
  1173  		}, unsafe.SliceData(b))
  1174  		ntotal += int(n)
  1175  		if err != nil {
  1176  			return ntotal, err
  1177  		}
  1178  		buf = buf[n:]
  1179  	}
  1180  	return ntotal, nil
  1181  }
  1182  
  1183  // WriteToInet6 is WriteTo, specialized for syscall.SockaddrInet6.
  1184  func (fd *FD) WriteToInet6(buf []byte, sa6 *syscall.SockaddrInet6) (int, error) {
  1185  	if err := fd.writeLock(); err != nil {
  1186  		return 0, err
  1187  	}
  1188  	defer fd.writeUnlock()
  1189  
  1190  	if len(buf) == 0 {
  1191  		// handle zero-byte payload
  1192  		n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1193  			err = windows.WSASendtoInet6(fd.Sysfd, &syscall.WSABuf{}, 1, &qty, 0, sa6, &o.o, nil)
  1194  			return qty, err
  1195  		})
  1196  		return n, err
  1197  	}
  1198  
  1199  	ntotal := 0
  1200  	for len(buf) > 0 {
  1201  		b := buf
  1202  		if len(b) > maxRW {
  1203  			b = b[:maxRW]
  1204  		}
  1205  		n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1206  			err = windows.WSASendtoInet6(fd.Sysfd, newWsaBuf(b), 1, &qty, 0, sa6, &o.o, nil)
  1207  			return qty, err
  1208  		}, unsafe.SliceData(b))
  1209  		ntotal += int(n)
  1210  		if err != nil {
  1211  			return ntotal, err
  1212  		}
  1213  		buf = buf[n:]
  1214  	}
  1215  	return ntotal, nil
  1216  }
  1217  
  1218  // Call ConnectEx. This doesn't need any locking, since it is only
  1219  // called when the descriptor is first created. This is here rather
  1220  // than in the net package so that it can use fd.wop.
  1221  func (fd *FD) ConnectEx(ra syscall.Sockaddr) error {
  1222  	_, err := fd.execIO('w', func(o *operation) (uint32, error) {
  1223  		return 0, ConnectExFunc(fd.Sysfd, ra, nil, 0, nil, &o.o)
  1224  	})
  1225  	return err
  1226  }
  1227  
  1228  func (fd *FD) acceptOne(s syscall.Handle, rawsa []syscall.RawSockaddrAny) (string, error) {
  1229  	// Submit accept request.
  1230  	rsan := uint32(unsafe.Sizeof(rawsa[0]))
  1231  	_, err := fd.execIO('r', func(o *operation) (qty uint32, err error) {
  1232  		err = AcceptFunc(fd.Sysfd, s, (*byte)(unsafe.Pointer(&rawsa[0])), 0, rsan, rsan, &qty, &o.o)
  1233  		return qty, err
  1234  
  1235  	})
  1236  	if err != nil {
  1237  		CloseFunc(s)
  1238  		return "acceptex", err
  1239  	}
  1240  
  1241  	// Inherit properties of the listening socket.
  1242  	err = syscall.Setsockopt(s, syscall.SOL_SOCKET, syscall.SO_UPDATE_ACCEPT_CONTEXT, (*byte)(unsafe.Pointer(&fd.Sysfd)), int32(unsafe.Sizeof(fd.Sysfd)))
  1243  	if err != nil {
  1244  		CloseFunc(s)
  1245  		return "setsockopt", err
  1246  	}
  1247  
  1248  	return "", nil
  1249  }
  1250  
  1251  // Accept handles accepting a socket. The sysSocket parameter is used
  1252  // to allocate the net socket.
  1253  func (fd *FD) Accept(sysSocket func() (syscall.Handle, error)) (syscall.Handle, []syscall.RawSockaddrAny, uint32, string, error) {
  1254  	if err := fd.readLock(); err != nil {
  1255  		return syscall.InvalidHandle, nil, 0, "", err
  1256  	}
  1257  	defer fd.readUnlock()
  1258  
  1259  	var rawsa [2]syscall.RawSockaddrAny
  1260  	for {
  1261  		s, err := sysSocket()
  1262  		if err != nil {
  1263  			return syscall.InvalidHandle, nil, 0, "", err
  1264  		}
  1265  
  1266  		errcall, err := fd.acceptOne(s, rawsa[:])
  1267  		if err == nil {
  1268  			return s, rawsa[:], uint32(unsafe.Sizeof(rawsa[0])), "", nil
  1269  		}
  1270  
  1271  		// Sometimes we see WSAECONNRESET and ERROR_NETNAME_DELETED is
  1272  		// returned here. These happen if connection reset is received
  1273  		// before AcceptEx could complete. These errors relate to new
  1274  		// connection, not to AcceptEx, so ignore broken connection and
  1275  		// try AcceptEx again for more connections.
  1276  		errno, ok := err.(syscall.Errno)
  1277  		if !ok {
  1278  			return syscall.InvalidHandle, nil, 0, errcall, err
  1279  		}
  1280  		switch errno {
  1281  		case syscall.ERROR_NETNAME_DELETED, syscall.WSAECONNRESET:
  1282  			// ignore these and try again
  1283  		default:
  1284  			return syscall.InvalidHandle, nil, 0, errcall, err
  1285  		}
  1286  	}
  1287  }
  1288  
  1289  // Seek wraps syscall.Seek.
  1290  func (fd *FD) Seek(offset int64, whence int) (int64, error) {
  1291  	if fd.kind == kindPipe {
  1292  		return 0, syscall.ESPIPE
  1293  	}
  1294  	if err := fd.readWriteLock(); err != nil {
  1295  		return 0, err
  1296  	}
  1297  	defer fd.readWriteUnlock()
  1298  
  1299  	if err := fd.ensureInit(); err != nil {
  1300  		return 0, err
  1301  	}
  1302  	if !fd.isBlocking {
  1303  		// Windows doesn't use the file pointer for overlapped file handles,
  1304  		// there is no point on calling syscall.Seek.
  1305  		var newOffset int64
  1306  		switch whence {
  1307  		case io.SeekStart:
  1308  			newOffset = offset
  1309  		case io.SeekCurrent:
  1310  			newOffset = fd.offset + offset
  1311  		case io.SeekEnd:
  1312  			var size int64
  1313  			if err := windows.GetFileSizeEx(fd.Sysfd, &size); err != nil {
  1314  				return 0, err
  1315  			}
  1316  			newOffset = size + offset
  1317  		default:
  1318  			return 0, windows.ERROR_INVALID_PARAMETER
  1319  		}
  1320  		if newOffset < 0 {
  1321  			return 0, windows.ERROR_NEGATIVE_SEEK
  1322  		}
  1323  		fd.setOffset(newOffset)
  1324  		return newOffset, nil
  1325  	}
  1326  	n, err := syscall.Seek(fd.Sysfd, offset, whence)
  1327  	fd.setOffset(n)
  1328  	return n, err
  1329  }
  1330  
  1331  // Fchmod updates syscall.ByHandleFileInformation.Fileattributes when needed.
  1332  func (fd *FD) Fchmod(mode uint32) error {
  1333  	if err := fd.incref(); err != nil {
  1334  		return err
  1335  	}
  1336  	defer fd.decref()
  1337  
  1338  	var d syscall.ByHandleFileInformation
  1339  	if err := syscall.GetFileInformationByHandle(fd.Sysfd, &d); err != nil {
  1340  		return err
  1341  	}
  1342  	attrs := d.FileAttributes
  1343  	if mode&syscall.S_IWRITE != 0 {
  1344  		attrs &^= syscall.FILE_ATTRIBUTE_READONLY
  1345  	} else {
  1346  		attrs |= syscall.FILE_ATTRIBUTE_READONLY
  1347  	}
  1348  	if attrs == d.FileAttributes {
  1349  		return nil
  1350  	}
  1351  
  1352  	var du windows.FILE_BASIC_INFO
  1353  	du.FileAttributes = attrs
  1354  	return windows.SetFileInformationByHandle(fd.Sysfd, windows.FileBasicInfo, unsafe.Pointer(&du), uint32(unsafe.Sizeof(du)))
  1355  }
  1356  
  1357  // Fchdir wraps syscall.Fchdir.
  1358  func (fd *FD) Fchdir() error {
  1359  	if err := fd.incref(); err != nil {
  1360  		return err
  1361  	}
  1362  	defer fd.decref()
  1363  	return syscall.Fchdir(fd.Sysfd)
  1364  }
  1365  
  1366  // GetFileType wraps syscall.GetFileType.
  1367  func (fd *FD) GetFileType() (uint32, error) {
  1368  	if err := fd.incref(); err != nil {
  1369  		return 0, err
  1370  	}
  1371  	defer fd.decref()
  1372  	return syscall.GetFileType(fd.Sysfd)
  1373  }
  1374  
  1375  // GetFileInformationByHandle wraps GetFileInformationByHandle.
  1376  func (fd *FD) GetFileInformationByHandle(data *syscall.ByHandleFileInformation) error {
  1377  	if err := fd.incref(); err != nil {
  1378  		return err
  1379  	}
  1380  	defer fd.decref()
  1381  	return syscall.GetFileInformationByHandle(fd.Sysfd, data)
  1382  }
  1383  
  1384  // RawRead invokes the user-defined function f for a read operation.
  1385  func (fd *FD) RawRead(f func(uintptr) bool) error {
  1386  	if err := fd.readLock(); err != nil {
  1387  		return err
  1388  	}
  1389  	defer fd.readUnlock()
  1390  	for {
  1391  		if f(uintptr(fd.Sysfd)) {
  1392  			return nil
  1393  		}
  1394  
  1395  		// Use a zero-byte read as a way to get notified when this
  1396  		// socket is readable. h/t https://stackoverflow.com/a/42019668/332798
  1397  		_, err := fd.execIO('r', func(o *operation) (qty uint32, err error) {
  1398  			var flags uint32
  1399  			if !fd.IsStream {
  1400  				flags |= windows.MSG_PEEK
  1401  			}
  1402  			err = syscall.WSARecv(fd.Sysfd, &syscall.WSABuf{}, 1, &qty, &flags, &o.o, nil)
  1403  			return qty, err
  1404  		})
  1405  		if err == windows.WSAEMSGSIZE {
  1406  			// expected with a 0-byte peek, ignore.
  1407  		} else if err != nil {
  1408  			return err
  1409  		}
  1410  	}
  1411  }
  1412  
  1413  // RawWrite invokes the user-defined function f for a write operation.
  1414  func (fd *FD) RawWrite(f func(uintptr) bool) error {
  1415  	if err := fd.writeLock(); err != nil {
  1416  		return err
  1417  	}
  1418  	defer fd.writeUnlock()
  1419  
  1420  	if f(uintptr(fd.Sysfd)) {
  1421  		return nil
  1422  	}
  1423  
  1424  	// TODO(tmm1): find a way to detect socket writability
  1425  	return syscall.EWINDOWS
  1426  }
  1427  
  1428  func sockaddrInet4ToRaw(rsa *syscall.RawSockaddrAny, sa *syscall.SockaddrInet4) int32 {
  1429  	*rsa = syscall.RawSockaddrAny{}
  1430  	raw := (*syscall.RawSockaddrInet4)(unsafe.Pointer(rsa))
  1431  	raw.Family = syscall.AF_INET
  1432  	p := (*[2]byte)(unsafe.Pointer(&raw.Port))
  1433  	p[0] = byte(sa.Port >> 8)
  1434  	p[1] = byte(sa.Port)
  1435  	raw.Addr = sa.Addr
  1436  	return int32(unsafe.Sizeof(*raw))
  1437  }
  1438  
  1439  func sockaddrInet6ToRaw(rsa *syscall.RawSockaddrAny, sa *syscall.SockaddrInet6) int32 {
  1440  	*rsa = syscall.RawSockaddrAny{}
  1441  	raw := (*syscall.RawSockaddrInet6)(unsafe.Pointer(rsa))
  1442  	raw.Family = syscall.AF_INET6
  1443  	p := (*[2]byte)(unsafe.Pointer(&raw.Port))
  1444  	p[0] = byte(sa.Port >> 8)
  1445  	p[1] = byte(sa.Port)
  1446  	raw.Scope_id = sa.ZoneId
  1447  	raw.Addr = sa.Addr
  1448  	return int32(unsafe.Sizeof(*raw))
  1449  }
  1450  
  1451  func rawToSockaddrInet4(rsa *syscall.RawSockaddrAny, sa *syscall.SockaddrInet4) {
  1452  	pp := (*syscall.RawSockaddrInet4)(unsafe.Pointer(rsa))
  1453  	p := (*[2]byte)(unsafe.Pointer(&pp.Port))
  1454  	sa.Port = int(p[0])<<8 + int(p[1])
  1455  	sa.Addr = pp.Addr
  1456  }
  1457  
  1458  func rawToSockaddrInet6(rsa *syscall.RawSockaddrAny, sa *syscall.SockaddrInet6) {
  1459  	pp := (*syscall.RawSockaddrInet6)(unsafe.Pointer(rsa))
  1460  	p := (*[2]byte)(unsafe.Pointer(&pp.Port))
  1461  	sa.Port = int(p[0])<<8 + int(p[1])
  1462  	sa.ZoneId = pp.Scope_id
  1463  	sa.Addr = pp.Addr
  1464  }
  1465  
  1466  func sockaddrToRaw(rsa *syscall.RawSockaddrAny, sa syscall.Sockaddr) (int32, error) {
  1467  	switch sa := sa.(type) {
  1468  	case *syscall.SockaddrInet4:
  1469  		sz := sockaddrInet4ToRaw(rsa, sa)
  1470  		return sz, nil
  1471  	case *syscall.SockaddrInet6:
  1472  		sz := sockaddrInet6ToRaw(rsa, sa)
  1473  		return sz, nil
  1474  	default:
  1475  		return 0, syscall.EWINDOWS
  1476  	}
  1477  }
  1478  
  1479  // ReadMsg wraps the WSARecvMsg network call.
  1480  func (fd *FD) ReadMsg(p []byte, oob []byte, flags int) (int, int, int, syscall.Sockaddr, error) {
  1481  	if err := fd.readLock(); err != nil {
  1482  		return 0, 0, 0, nil, err
  1483  	}
  1484  	defer fd.readUnlock()
  1485  
  1486  	if len(p) > maxRW {
  1487  		p = p[:maxRW]
  1488  	}
  1489  
  1490  	rsa := newWSARsa()
  1491  	defer wsaRsaPool.Put(rsa)
  1492  	msg := newWSAMsg(p, oob, flags, rsa)
  1493  	defer freeWSAMsg(msg)
  1494  	n, err := fd.execIO('r', func(o *operation) (qty uint32, err error) {
  1495  		err = windows.WSARecvMsg(fd.Sysfd, msg, &qty, &o.o, nil)
  1496  		return qty, err
  1497  	}, rsa, msg)
  1498  	err = fd.eofError(n, err)
  1499  	var sa syscall.Sockaddr
  1500  	if err == nil {
  1501  		sa, err = msg.Name.Sockaddr()
  1502  	}
  1503  	return n, int(msg.Control.Len), int(msg.Flags), sa, err
  1504  }
  1505  
  1506  // ReadMsgInet4 is ReadMsg, but specialized to return a syscall.SockaddrInet4.
  1507  func (fd *FD) ReadMsgInet4(p []byte, oob []byte, flags int, sa4 *syscall.SockaddrInet4) (int, int, int, error) {
  1508  	if err := fd.readLock(); err != nil {
  1509  		return 0, 0, 0, err
  1510  	}
  1511  	defer fd.readUnlock()
  1512  
  1513  	if len(p) > maxRW {
  1514  		p = p[:maxRW]
  1515  	}
  1516  
  1517  	rsa := newWSARsa()
  1518  	defer wsaRsaPool.Put(rsa)
  1519  	msg := newWSAMsg(p, oob, flags, rsa)
  1520  	defer freeWSAMsg(msg)
  1521  	n, err := fd.execIO('r', func(o *operation) (qty uint32, err error) {
  1522  		err = windows.WSARecvMsg(fd.Sysfd, msg, &qty, &o.o, nil)
  1523  		return qty, err
  1524  	}, rsa, msg)
  1525  	err = fd.eofError(n, err)
  1526  	if err == nil {
  1527  		rawToSockaddrInet4(msg.Name, sa4)
  1528  	}
  1529  	return n, int(msg.Control.Len), int(msg.Flags), err
  1530  }
  1531  
  1532  // ReadMsgInet6 is ReadMsg, but specialized to return a syscall.SockaddrInet6.
  1533  func (fd *FD) ReadMsgInet6(p []byte, oob []byte, flags int, sa6 *syscall.SockaddrInet6) (int, int, int, error) {
  1534  	if err := fd.readLock(); err != nil {
  1535  		return 0, 0, 0, err
  1536  	}
  1537  	defer fd.readUnlock()
  1538  
  1539  	if len(p) > maxRW {
  1540  		p = p[:maxRW]
  1541  	}
  1542  
  1543  	rsa := newWSARsa()
  1544  	defer wsaRsaPool.Put(rsa)
  1545  	msg := newWSAMsg(p, oob, flags, rsa)
  1546  	defer freeWSAMsg(msg)
  1547  	n, err := fd.execIO('r', func(o *operation) (qty uint32, err error) {
  1548  		err = windows.WSARecvMsg(fd.Sysfd, msg, &qty, &o.o, nil)
  1549  		return qty, err
  1550  	}, rsa, msg)
  1551  	err = fd.eofError(n, err)
  1552  	if err == nil {
  1553  		rawToSockaddrInet6(msg.Name, sa6)
  1554  	}
  1555  	return n, int(msg.Control.Len), int(msg.Flags), err
  1556  }
  1557  
  1558  // WriteMsg wraps the WSASendMsg network call.
  1559  func (fd *FD) WriteMsg(p []byte, oob []byte, sa syscall.Sockaddr) (int, int, error) {
  1560  	if len(p) > maxRW {
  1561  		return 0, 0, errors.New("packet is too large (only 1GB is allowed)")
  1562  	}
  1563  
  1564  	if err := fd.writeLock(); err != nil {
  1565  		return 0, 0, err
  1566  	}
  1567  	defer fd.writeUnlock()
  1568  
  1569  	var rsa *wsaRsa
  1570  	if sa != nil {
  1571  		rsa = newWSARsa()
  1572  		defer wsaRsaPool.Put(rsa)
  1573  		var err error
  1574  		rsa.namelen, err = sockaddrToRaw(&rsa.name, sa)
  1575  		if err != nil {
  1576  			return 0, 0, err
  1577  		}
  1578  	}
  1579  	msg := newWSAMsg(p, oob, 0, rsa)
  1580  	defer freeWSAMsg(msg)
  1581  	n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1582  		err = windows.WSASendMsg(fd.Sysfd, msg, 0, nil, &o.o, nil)
  1583  		return qty, err
  1584  	}, rsa, msg)
  1585  	return n, int(msg.Control.Len), err
  1586  }
  1587  
  1588  // WriteMsgInet4 is WriteMsg specialized for syscall.SockaddrInet4.
  1589  func (fd *FD) WriteMsgInet4(p []byte, oob []byte, sa *syscall.SockaddrInet4) (int, int, error) {
  1590  	if len(p) > maxRW {
  1591  		return 0, 0, errors.New("packet is too large (only 1GB is allowed)")
  1592  	}
  1593  
  1594  	if err := fd.writeLock(); err != nil {
  1595  		return 0, 0, err
  1596  	}
  1597  	defer fd.writeUnlock()
  1598  
  1599  	var rsa *wsaRsa
  1600  	if sa != nil {
  1601  		rsa = newWSARsa()
  1602  		defer wsaRsaPool.Put(rsa)
  1603  		rsa.namelen = sockaddrInet4ToRaw(&rsa.name, sa)
  1604  	}
  1605  	msg := newWSAMsg(p, oob, 0, rsa)
  1606  	defer freeWSAMsg(msg)
  1607  	n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1608  		err = windows.WSASendMsg(fd.Sysfd, msg, 0, nil, &o.o, nil)
  1609  		return qty, err
  1610  	}, rsa, msg)
  1611  	return n, int(msg.Control.Len), err
  1612  }
  1613  
  1614  // WriteMsgInet6 is WriteMsg specialized for syscall.SockaddrInet6.
  1615  func (fd *FD) WriteMsgInet6(p []byte, oob []byte, sa *syscall.SockaddrInet6) (int, int, error) {
  1616  	if len(p) > maxRW {
  1617  		return 0, 0, errors.New("packet is too large (only 1GB is allowed)")
  1618  	}
  1619  
  1620  	if err := fd.writeLock(); err != nil {
  1621  		return 0, 0, err
  1622  	}
  1623  	defer fd.writeUnlock()
  1624  
  1625  	var rsa *wsaRsa
  1626  	if sa != nil {
  1627  		rsa = newWSARsa()
  1628  		defer wsaRsaPool.Put(rsa)
  1629  		rsa.namelen = sockaddrInet6ToRaw(&rsa.name, sa)
  1630  	}
  1631  	msg := newWSAMsg(p, oob, 0, rsa)
  1632  	defer freeWSAMsg(msg)
  1633  	n, err := fd.execIO('w', func(o *operation) (qty uint32, err error) {
  1634  		err = windows.WSASendMsg(fd.Sysfd, msg, 0, nil, &o.o, nil)
  1635  		return qty, err
  1636  	}, rsa, msg)
  1637  	return n, int(msg.Control.Len), err
  1638  }
  1639  
  1640  func DupCloseOnExec(fd int) (int, string, error) {
  1641  	proc, err := syscall.GetCurrentProcess()
  1642  	if err != nil {
  1643  		return 0, "GetCurrentProcess", err
  1644  	}
  1645  
  1646  	var nfd syscall.Handle
  1647  	const inherit = false // analogous to CLOEXEC
  1648  	if err := syscall.DuplicateHandle(proc, syscall.Handle(fd), proc, &nfd, 0, inherit, syscall.DUPLICATE_SAME_ACCESS); err != nil {
  1649  		return 0, "DuplicateHandle", err
  1650  	}
  1651  	return int(nfd), "", nil
  1652  }
  1653  

View as plain text