newlib-cygwin/winsup/cygwin/syscalls.cc

/* syscalls.cc: syscalls

This file is part of Cygwin.

This software is a copyrighted work licensed under the terms of the
Cygwin license.  Please consult the file "CYGWIN_LICENSE" for
details. */

#define fstat __FOOfstat__
#define lstat __FOOlstat__
//#define stat __FOOstat__
#define _close __FOO_close__
#define _lseek __FOO_lseek__
#define _open __FOO_open__
#define _read __FOO_read__
#define _write __FOO_write__
#define _open64 __FOO_open64__
#define _lseek64 __FOO_lseek64__
#define _fstat64 __FOO_fstat64__
#define pread __FOO_pread
#define pwrite __FOO_pwrite

#include "winsup.h"
#include "miscfuncs.h"
#include <sys/stat.h>
#include <sys/vfs.h> /* needed for statfs */
#include <sys/statvfs.h> /* needed for statvfs */
#include <stdlib.h>
#include <stdio.h>
#include <process.h>
#include <utmp.h>
#include <utmpx.h>
#include <sys/uio.h>
#include <ctype.h>
#include <wctype.h>
#include <unistd.h>
#include <sys/wait.h>
#include <dirent.h>
#include <ntsecapi.h>
#include "ntdll.h"

#undef fstat
#undef lstat
//#undef stat
#undef pread
#undef pwrite

#include <cygwin/version.h>
#include "cygerrno.h"
#include "perprocess.h"
#include "security.h"
#include "path.h"
#include "fhandler.h"
#include "dtable.h"
#include "sigproc.h"
#include "pinfo.h"
#include "shared_info.h"
#include "cygheap.h"
#include "registry.h"
#include "environ.h"
#include "tls_pbuf.h"
#include "sync.h"
#include "child_info.h"
#include <cygwin/fs.h>  /* needed for RENAME_NOREPLACE */

#undef _close
#undef _lseek
#undef _open
#undef _read
#undef _write
#undef _open64
#undef _lseek64
#undef _fstat64

static int __stdcall mknod_worker (const char *, mode_t, mode_t, _major_t,
				   _minor_t);

/* Close all files and process any queued deletions.
   Lots of unix style applications will open a tmp file, unlink it,
   but never call close.  This function is called by _exit to
   ensure we don't leave any such files lying around.  */

void __stdcall
close_all_files (bool norelease)
{
  cygheap->fdtab.lock ();

  semaphore::terminate ();

  HANDLE h = NULL;

  for (int i = 0; i < (int) cygheap->fdtab.size; i++)
    {
      cygheap_fdget cfd (i, false, false);
      if (cfd >= 0)
	{
	  debug_only_printf ("closing fd %d", i);
	  if (i == 2)
	    DuplicateHandle (GetCurrentProcess (), cfd->get_output_handle (),
			     GetCurrentProcess (), &h,
			     0, false, DUPLICATE_SAME_ACCESS);
	  cfd->close_with_arch ();
	  if (!norelease)
	    cfd.release ();
	}
    }

  if (!have_execed && cygheap->ctty)
    cygheap->close_ctty ();

  fhandler_base_overlapped::flush_all_async_io ();
  if (h)
    SetStdHandle (STD_ERROR_HANDLE, h);
  cygheap->fdtab.unlock ();
}

extern "C" int
dup (int fd)
{
  int res;
  cygheap_fdnew newfd;
  if (newfd < 0)
    res = -1;
  else
    res = cygheap->fdtab.dup3 (fd, newfd, 0);
  syscall_printf ("%R = dup(%d)", res, fd);
  return res;
}

inline int
dup_finish (int oldfd, int newfd, int flags)
{
  int res;
  if ((res = cygheap->fdtab.dup3 (oldfd, newfd, flags | O_EXCL)) == newfd)
    {
      cygheap_fdget (newfd)->inc_refcnt ();
      cygheap->fdtab.unlock ();	/* dup3 exits with lock set on success */
    }
  return res;
}

extern "C" int
dup2 (int oldfd, int newfd)
{
  int res;
  if (newfd >= OPEN_MAX_MAX || newfd < 0)
    {
      set_errno (EBADF);
      res = -1;
    }
  else if (newfd == oldfd)
    {
      cygheap_fdget cfd (oldfd);
      res = (cfd >= 0) ? oldfd : -1;
    }
  else
    res = dup_finish (oldfd, newfd, 0);

  syscall_printf ("%R = dup2(%d, %d)", res, oldfd, newfd);
  return res;
}

extern "C" int
dup3 (int oldfd, int newfd, int flags)
{
  int res;
  if (newfd >= OPEN_MAX_MAX)
    {
      set_errno (EBADF);
      res = -1;
    }
  else if (newfd == oldfd)
    {
      cygheap_fdget cfd (oldfd, false, false);
      set_errno (cfd < 0 ? EBADF : EINVAL);
      res = -1;
    }
  else
    res = dup_finish (oldfd, newfd, flags);

  syscall_printf ("%R = dup3(%d, %d, %y)", res, oldfd, newfd, flags);
  return res;
}

/* Define macro to simplify checking for a transactional error code. */
#define NT_TRANSACTIONAL_ERROR(s)	\
		(((ULONG)(s) >= (ULONG)STATUS_TRANSACTIONAL_CONFLICT) \
		 && ((ULONG)(s) <= (ULONG)STATUS_TRANSACTION_NOT_ENLISTED))

static inline void
start_transaction (HANDLE &old_trans, HANDLE &trans)
{
  NTSTATUS status = NtCreateTransaction (&trans,
				SYNCHRONIZE | TRANSACTION_ALL_ACCESS,
				NULL, NULL, NULL, 0, 0, 0, NULL, NULL);
  if (NT_SUCCESS (status))
    {
      old_trans = RtlGetCurrentTransaction ();
      RtlSetCurrentTransaction (trans);
    }
  else
    {
      debug_printf ("NtCreateTransaction failed, %y", status);
      old_trans = trans = NULL;
    }
}

static inline NTSTATUS
stop_transaction (NTSTATUS status, HANDLE old_trans, HANDLE &trans)
{
  RtlSetCurrentTransaction (old_trans);
  if (NT_SUCCESS (status))
    status = NtCommitTransaction (trans, TRUE);
  else
    status = NtRollbackTransaction (trans, TRUE);
  NtClose (trans);
  trans = NULL;
  return status;
}

static const char desktop_ini[] =
  "[.ShellClassInfo]\r\n"
  "CLSID={645FF040-5081-101B-9F08-00AA002F954E}\r\n"
  "LocalizedResourceName=@%SystemRoot%\\system32\\shell32.dll,-8964\r\n";

enum bin_status
{
  dont_move,
  move_to_bin,
  has_been_moved,
  dir_not_empty
};

static bin_status
try_to_bin (path_conv &pc, HANDLE &fh, ACCESS_MASK access, ULONG flags)
{
  bin_status bin_stat = move_to_bin;
  NTSTATUS status;
  OBJECT_ATTRIBUTES attr;
  IO_STATUS_BLOCK io;
  HANDLE rootdir = NULL, recyclerdir = NULL, tmp_fh = NULL;
  USHORT recycler_base_len = 0, recycler_user_len = 0;
  UNICODE_STRING root, recycler, fname;
  WCHAR recyclerbuf[NAME_MAX + 1]; /* Enough for recycler + SID + filename */
  PFILE_NAME_INFORMATION pfni;
  PFILE_INTERNAL_INFORMATION pfii;
  PFILE_RENAME_INFORMATION pfri;
  ULONG frisiz;
  FILE_DISPOSITION_INFORMATION disp = { TRUE };
  bool fs_has_per_user_recycler = pc.fs_is_ntfs () || pc.fs_is_refs ();

  tmp_pathbuf tp;
  PBYTE infobuf = (PBYTE) tp.w_get ();

  pfni = (PFILE_NAME_INFORMATION) infobuf;
  status = NtQueryInformationFile (fh, &io, pfni, 65536, FileNameInformation);
  if (!NT_SUCCESS (status))
    {
      debug_printf ("NtQueryInformationFile (%S, FileNameInformation) "
		    "failed, status = %y", pc.get_nt_native_path (), status);
      goto out;
    }
  /* The filename could change, the parent dir not.  So we split both paths
     and take the prefix.  However, there are two special cases:
     - The handle refers to the root dir of the volume.
     - The handle refers to the recycler or a subdir.
     Both cases are handled by just returning and not even trying to move
     them into the recycler. */
  if (pfni->FileNameLength == 2) /* root dir. */
    goto out;
  /* The recycler name is $Recycler.Bin by default.  If the recycler dir
     disappeared for some reason, the shell32.dll recreates the directory in
     all upper case.  So, we never know beforehand if the dir is written in
     mixed case or in all upper case.  That's a problem when using
     casesensitivity.  If the file handle given to FileRenameInformation
     has been opened casesensitive, the call also handles the path to the
     target dir casesensitive.  Rather than trying to find the right name
     of the recycler, we just reopen the file to move with OBJ_CASE_INSENSITIVE,
     so the subsequent FileRenameInformation works caseinsensitive in terms of
     the recycler directory name, too. */
  if (!pc.objcaseinsensitive ())
    {
      InitializeObjectAttributes (&attr, &ro_u_empty, OBJ_CASE_INSENSITIVE,
				  fh, NULL);
      status = NtOpenFile (&tmp_fh, access, &attr, &io, FILE_SHARE_VALID_FLAGS,
			   flags);
      if (!NT_SUCCESS (status))
	debug_printf ("NtOpenFile (%S) for reopening caseinsensitive failed, "
		      "status = %y", pc.get_nt_native_path (), status);
      else
	{
	  NtClose (fh);
	  fh = tmp_fh;
	}
    }
  /* Initialize recycler path. */
  RtlInitEmptyUnicodeString (&recycler, recyclerbuf, sizeof recyclerbuf);
  if (!pc.isremote ())
    {
      RtlAppendUnicodeToString (&recycler, L"\\$Recycle.Bin\\");
      RtlInitCountedUnicodeString(&fname, pfni->FileName, pfni->FileNameLength);
      /* Is the file a subdir of the recycler? */
      if (RtlEqualUnicodePathPrefix (&fname, &recycler, TRUE))
	goto out;
      /* Is fname the recycler?  Temporarily hide trailing backslash. */
      recycler.Length -= sizeof (WCHAR);
      if (RtlEqualUnicodeString (&fname, &recycler, TRUE))
	goto out;
      /* Is fname really a subcomponent of the full path?  If not, there's
	 a high probability we're acessing the file via a virtual drive
	 created with "subst".  Check and accommodate it.  Note that we
	 only get here if the virtual drive is really pointing to a local
	 drive.  Otherwise pc.isremote () returns "true". */
      if (!RtlEqualUnicodePathSuffix (pc.get_nt_native_path (), &fname, TRUE))
	{
	  WCHAR drive[3] = { pc.get_nt_native_path ()->Buffer[4], ':', '\0' };
	  PWCHAR tgt = tp.w_get ();

	  if (QueryDosDeviceW (drive, tgt, NT_MAX_PATH)
	      && !wcsncmp (tgt, L"\\??\\", 4))
	    {
	      UNICODE_STRING new_path;

	      wcsncpy (tgt + 6, fname.Buffer, fname.Length / sizeof (WCHAR));
	      RtlInitCountedUnicodeString(&new_path, tgt,
					  6 * sizeof (WCHAR) + fname.Length);
	      pc.set_nt_native_path (&new_path);
	    }
	}
      /* Create root dir path from file name information. */
      RtlSplitUnicodePath (&fname, &fname, NULL);
      RtlSplitUnicodePath (pc.get_nt_native_path (), &root, NULL);
      root.Length -= fname.Length - sizeof (WCHAR);

      /* Open root directory.  All recycler bin ops are caseinsensitive. */
      InitializeObjectAttributes (&attr, &root, OBJ_CASE_INSENSITIVE,
				  NULL, NULL);
      status = NtOpenFile (&rootdir, FILE_TRAVERSE, &attr, &io,
			   FILE_SHARE_VALID_FLAGS, FILE_OPEN_FOR_BACKUP_INTENT);
      if (!NT_SUCCESS (status))
	{
	  debug_printf ("NtOpenFile (%S) failed, status = %y", &root, status);
	  goto out;
	}

      /* Strip leading backslash */
      ++recycler.Buffer;
      recycler.Length -= sizeof (WCHAR);
      /* Store length of recycler base dir, if it's necessary to create it. */
      recycler_base_len = recycler.Length;
      /* On NTFS or ReFS the recycler dir contains user specific subdirs, which
	 are the actual recycle bins per user.  The name if this dir is the
	 string representation of the user SID. */
      if (fs_has_per_user_recycler)
	{
	  UNICODE_STRING sid;
	  WCHAR sidbuf[128];
	  /* Unhide trailing backslash. */
	  recycler.Length += sizeof (WCHAR);
	  RtlInitEmptyUnicodeString (&sid, sidbuf, sizeof sidbuf);
	  RtlConvertSidToUnicodeString (&sid, cygheap->user.sid (), FALSE);
	  RtlAppendUnicodeStringToString (&recycler, &sid);
	  recycler_user_len = recycler.Length;
	}
      RtlAppendUnicodeToString (&recycler, L"\\");
    }
  /* Create hopefully unique filename.
     Since we have to stick to the current directory on remote shares, make
     the new filename at least very unlikely to match by accident.  It starts
     with ".cyg", with "cyg" transposed into the Unicode low surrogate area
     starting at U+dc00.  Use plain ASCII chars on filesystems not supporting
     Unicode.  The rest of the filename is the inode number in hex encoding
     and a hash of the full NT path in hex.  The combination allows to remove
     multiple hardlinks to the same file.  Samba doesn't like the transposed
     names. */
  RtlAppendUnicodeToString (&recycler,
			    (pc.fs_flags () & FILE_UNICODE_ON_DISK
			     && !pc.fs_is_samba ())
			    ? L".\xdc63\xdc79\xdc67" : L".cyg");
  pfii = (PFILE_INTERNAL_INFORMATION) infobuf;
  /* Note: Modern Samba versions apparently don't like buffer sizes of more
     than 65535 in some NtQueryInformationFile/NtSetInformationFile calls.
     Therefore we better use exact buffer sizes from now on. */
  status = NtQueryInformationFile (fh, &io, pfii, sizeof *pfii,
				   FileInternalInformation);
  if (!NT_SUCCESS (status))
    {
      debug_printf ("NtQueryInformationFile (%S, FileInternalInformation) "
		    "failed, status = %y", pc.get_nt_native_path (), status);
      goto out;
    }
  RtlInt64ToHexUnicodeString (pfii->IndexNumber.QuadPart, &recycler, TRUE);
  RtlInt64ToHexUnicodeString (hash_path_name (0, pc.get_nt_native_path ()),
			      &recycler, TRUE);
  /* Shoot. */
  pfri = (PFILE_RENAME_INFORMATION) infobuf;
  pfri->ReplaceIfExists = TRUE;
  pfri->RootDirectory = rootdir;
  pfri->FileNameLength = recycler.Length;
  memcpy (pfri->FileName, recycler.Buffer, recycler.Length);
  frisiz = sizeof *pfri + pfri->FileNameLength - sizeof (WCHAR);

  status = NtSetInformationFile (fh, &io, pfri, frisiz, FileRenameInformation);
  if (status == STATUS_OBJECT_PATH_NOT_FOUND && !pc.isremote ())
    {
      /* Ok, so the recycler and/or the recycler/SID directory don't exist.
	 First reopen root dir with permission to create subdirs. */
      NtClose (rootdir);
      InitializeObjectAttributes (&attr, &root, OBJ_CASE_INSENSITIVE,
				  NULL, NULL);
      status = NtOpenFile (&rootdir, FILE_ADD_SUBDIRECTORY, &attr, &io,
			   FILE_SHARE_VALID_FLAGS, FILE_OPEN_FOR_BACKUP_INTENT);
      if (!NT_SUCCESS (status))
	{
	  debug_printf ("NtOpenFile (%S) failed, status = %y",
	  		&recycler, status);
	  goto out;
	}
      /* Then check if recycler exists by opening and potentially creating it.
	 Yes, we can really do that.  Typically the recycle bin is created
	 by the first user actually using the bin. */
      InitializeObjectAttributes (&attr, &recycler, OBJ_CASE_INSENSITIVE,
				  rootdir, recycler_sd (true, true));
      recycler.Length = recycler_base_len;
      status = NtCreateFile (&recyclerdir,
			     READ_CONTROL
			     | (fs_has_per_user_recycler ? 0 : FILE_ADD_FILE),
			     &attr, &io, NULL,
			     FILE_ATTRIBUTE_DIRECTORY
			     | FILE_ATTRIBUTE_SYSTEM
			     | FILE_ATTRIBUTE_HIDDEN,
			     FILE_SHARE_VALID_FLAGS, FILE_OPEN_IF,
			     FILE_DIRECTORY_FILE, NULL, 0);
      if (!NT_SUCCESS (status))
	{
	  debug_printf ("NtCreateFile (%S) failed, status = %y",
	  		&recycler, status);
	  goto out;
	}
      /* Next, if necessary, check if the recycler/SID dir exists and
	 create it if not. */
      if (fs_has_per_user_recycler)
	{
	  NtClose (recyclerdir);
	  recycler.Length = recycler_user_len;
	  InitializeObjectAttributes (&attr, &recycler, OBJ_CASE_INSENSITIVE,
				      rootdir, recycler_sd (false, true));
	  status = NtCreateFile (&recyclerdir, READ_CONTROL | FILE_ADD_FILE,
				 &attr, &io, NULL, FILE_ATTRIBUTE_DIRECTORY
						   | FILE_ATTRIBUTE_SYSTEM
						   | FILE_ATTRIBUTE_HIDDEN,
				 FILE_SHARE_VALID_FLAGS, FILE_OPEN_IF,
				 FILE_DIRECTORY_FILE, NULL, 0);
	  if (!NT_SUCCESS (status))
	    {
	      debug_printf ("NtCreateFile (%S) failed, status = %y",
			    &recycler, status);
	      goto out;
	    }
	}
      /* The desktop.ini file is expected by Windows Explorer.  Otherwise,
         the created bin is treated as corrupted */
      if (io.Information == FILE_CREATED)
	{
	  RtlInitUnicodeString (&fname, L"desktop.ini");
	  InitializeObjectAttributes (&attr, &fname, OBJ_CASE_INSENSITIVE,
				      recyclerdir, recycler_sd (false, false));
	  status = NtCreateFile (&tmp_fh, FILE_GENERIC_WRITE, &attr, &io, NULL,
				 FILE_ATTRIBUTE_SYSTEM | FILE_ATTRIBUTE_HIDDEN,
				 FILE_SHARE_VALID_FLAGS, FILE_CREATE,
				 FILE_SYNCHRONOUS_IO_NONALERT
				 | FILE_NON_DIRECTORY_FILE, NULL, 0);
	  if (!NT_SUCCESS (status))
	    debug_printf ("NtCreateFile (%S) failed, status = %y",
			  &recycler, status);
	  else
	    {
	      status = NtWriteFile (tmp_fh, NULL, NULL, NULL, &io,
				    (PVOID) desktop_ini, sizeof desktop_ini - 1,
				    NULL, NULL);
	      if (!NT_SUCCESS (status))
		debug_printf ("NtWriteFile (%S) failed, status = %y",
			      &fname, status);
	      NtClose (tmp_fh);
	    }
	}
      NtClose (recyclerdir);
      /* Shoot again. */
      status = NtSetInformationFile (fh, &io, pfri, frisiz,
				     FileRenameInformation);
    }
  if (!NT_SUCCESS (status))
    {
      debug_printf ("Move %S to %S failed, status = %y",
		    pc.get_nt_native_path (), &recycler, status);
      goto out;
    }
  /* Moving to the bin worked. */
  bin_stat = has_been_moved;
  /* Now we try to set the delete disposition.  If that worked, we're done.
     We try this here first, as long as we still have the open handle.
     Otherwise the below code closes the handle to allow replacing the file. */
  status = NtSetInformationFile (fh, &io, &disp, sizeof disp,
				 FileDispositionInformation);
  switch (status)
    {
    case STATUS_SUCCESS:
      break;
    case STATUS_DIRECTORY_NOT_EMPTY:
      /* Uh oh!  This was supposed to be avoided by the check_dir_not_empty
	 test in unlink_nt, but given that the test isn't atomic, this *can*
	 happen.  Try to move the dir back ASAP. */
      pfri->RootDirectory = NULL;
      pfri->FileNameLength = pc.get_nt_native_path ()->Length;
      memcpy (pfri->FileName, pc.get_nt_native_path ()->Buffer,
			      pc.get_nt_native_path ()->Length);
      frisiz = sizeof *pfri + pfri->FileNameLength - sizeof (WCHAR);
      if (NT_SUCCESS (NtSetInformationFile (fh, &io, pfri, frisiz,
					    FileRenameInformation)))
	{
	  /* Give notice to unlink_nt and leave immediately.  This avoids
	     closing the handle, which might still be used if called from
	     the rm -r workaround code. */
	  bin_stat = dir_not_empty;
	  goto out;
	}
      debug_printf ("Renaming dir %S back to %S failed, status = %y",
		    &recycler, pc.get_nt_native_path (), status);
      break;
    case STATUS_FILE_RENAMED:
      /* On NFS, the subsequent request to set the delete disposition fails
	 with STATUS_FILE_RENAMED.  We have to reopen the file, close the
	 original handle, and set the delete disposition on the reopened
	 handle to make sure setting delete disposition works. */
      InitializeObjectAttributes (&attr, &ro_u_empty, 0, fh, NULL);
      status = NtOpenFile (&tmp_fh, access, &attr, &io,
			   FILE_SHARE_VALID_FLAGS, flags);
      if (!NT_SUCCESS (status))
	debug_printf ("NtOpenFile (%S) for reopening in renamed case failed, "
		      "status = %y", pc.get_nt_native_path (), status);
      else
	{
	  NtClose (fh);
	  fh = tmp_fh;
	  status = NtSetInformationFile (fh, &io, &disp, sizeof disp,
					 FileDispositionInformation);
	  if (!NT_SUCCESS (status))
	    debug_printf ("Setting delete disposition %S (%S) in renamed "
			  "case failed, status = %y",
			  &recycler, pc.get_nt_native_path (), status);
	}
      break;
    default:
      debug_printf ("Setting delete disposition on %S (%S) failed, status = %y",
		    &recycler, pc.get_nt_native_path (), status);
      break;
    }
  /* In case of success, restore R/O attribute to accommodate hardlinks.
     That leaves potentially hardlinks around with the R/O bit suddenly
     off if setting the delete disposition failed, but please, keep in
     mind this is really a border case only. */
  if ((access & FILE_WRITE_ATTRIBUTES) && NT_SUCCESS (status) && !pc.isdir ())
    NtSetAttributesFile (fh, pc.file_attributes ());
  NtClose (fh);
  fh = NULL; /* So unlink_nt doesn't close the handle twice. */
  /* On success or when trying to unlink a directory we just return here.
     The below code only works for files.  It also fails on NFS. */
  if (NT_SUCCESS (status) || pc.isdir () || pc.fs_is_nfs ())
    goto out;
  /* The final trick.  We create a temporary file with delete-on-close
     semantic and rename that file to the file just moved to the bin.
     This typically overwrites the original file and we get rid of it,
     even if neither setting the delete dispostion, nor setting
     delete-on-close on the original file succeeds.  There are still
     cases in which this fails, for instance, when trying to delete a
     hardlink to a DLL used by the unlinking application itself. */
  if (pc.isremote ())
    {
      /* In the remote case we need the full path, but recycler is only
	 a relative path.  Convert to absolute path. */
      RtlInitEmptyUnicodeString (&fname, (PCWSTR) tp.w_get (),
				 (NT_MAX_PATH - 1) * sizeof (WCHAR));
      RtlCopyUnicodeString (&fname, pc.get_nt_native_path ());
      RtlSplitUnicodePath (&fname, &fname, NULL);
      /* Reset max length, overwritten by RtlSplitUnicodePath. */
      fname.MaximumLength = (NT_MAX_PATH - 1) * sizeof (WCHAR); /* reset */
      RtlAppendUnicodeStringToString (&fname, &recycler);
    }
  else
    fname = recycler;
  RtlAppendUnicodeToString (&fname, L"X");
  InitializeObjectAttributes (&attr, &fname, 0, rootdir, NULL);
  status = NtCreateFile (&tmp_fh, DELETE, &attr, &io, NULL,
			 FILE_ATTRIBUTE_NORMAL, 0, FILE_SUPERSEDE,
			 FILE_NON_DIRECTORY_FILE | FILE_DELETE_ON_CLOSE,
			 NULL, 0);
  if (!NT_SUCCESS (status))
    {
      debug_printf ("Creating file %S for overwriting %S (%S) failed, "
		    "status = %y", &fname, &recycler, pc.get_nt_native_path (),
		    status);
      goto out;
    }
  status = NtSetInformationFile (tmp_fh, &io, pfri, frisiz,
				 FileRenameInformation);
  NtClose (tmp_fh);
  if (!NT_SUCCESS (status))
    debug_printf ("Overwriting %S (%S) with %S failed, status = %y",
		  &recycler, pc.get_nt_native_path (), &fname, status);

out:
  if (rootdir)
    NtClose (rootdir);
  debug_printf ("%S, return bin_status %d", pc.get_nt_native_path (), bin_stat);
  return bin_stat;
}

static NTSTATUS
check_dir_not_empty (HANDLE dir, path_conv &pc)
{
  IO_STATUS_BLOCK io;
  const ULONG bufsiz = 3 * sizeof (FILE_NAMES_INFORMATION)
		       + 3 * NAME_MAX * sizeof (WCHAR);
  PFILE_NAMES_INFORMATION pfni = (PFILE_NAMES_INFORMATION)
				 alloca (bufsiz);
  NTSTATUS status = NtQueryDirectoryFile (dir, NULL, NULL, 0, &io, pfni,
					  bufsiz, FileNamesInformation,
					  FALSE, NULL, TRUE);
  if (!NT_SUCCESS (status))
    {
      debug_printf ("Checking if directory %S is empty failed, status = %y",
		    pc.get_nt_native_path (), status);
      return status;
    }
  int cnt = 1;
  do
    {
      while (pfni->NextEntryOffset)
	{
	  if (++cnt > 2)
	    {
	      UNICODE_STRING fname;
	      OBJECT_ATTRIBUTES attr;
	      FILE_BASIC_INFORMATION fbi;

	      pfni = (PFILE_NAMES_INFORMATION)
		     ((caddr_t) pfni + pfni->NextEntryOffset);
	      RtlInitCountedUnicodeString(&fname, pfni->FileName,
					  pfni->FileNameLength);
	      InitializeObjectAttributes (&attr, &fname, 0, dir, NULL);
	      status = NtQueryAttributesFile (&attr, &fbi);
	      /* Intensive testing shows that sometimes directories, for which
		 the delete disposition has already been set, and the deleting
		 handle is already closed, can linger in the parent dir for a
		 couple of ms for no apparent reason (Windows Defender or other
		 real-time scanners are suspect).

		 A fast rm -r is capable to exploit this problem.  Setting the
		 delete disposition of the parent dir then fails with
		 STATUS_DIRECTORY_NOT_EMPTY.  Examining the content of the
		 affected dir can then show either that the dir is empty, or it
		 can contain a lingering subdir.  Calling NtQueryAttributesFile
		 on that subdir returns with STATUS_DELETE_PENDING, or it
		 disappeared before that call.

		 That's what we do here.  If NtQueryAttributesFile succeeded,
		 or if the error code does not indicate an already deleted
		 entry, STATUS_DIRECTORY_NOT_EMPTY is returned.

		 Otherwise STATUS_SUCCESS is returned.  Read on in unlink_nt. */
	      if (status != STATUS_DELETE_PENDING
		  && status != STATUS_OBJECT_NAME_NOT_FOUND
		  && status != STATUS_OBJECT_PATH_NOT_FOUND)
		{
		  debug_printf ("Directory %S not empty, found file <%S>, "
				 "query status = %y",
				pc.get_nt_native_path (), &fname, status);
		  return STATUS_DIRECTORY_NOT_EMPTY;
		}
	    }
	  pfni = (PFILE_NAMES_INFORMATION) ((caddr_t) pfni + pfni->NextEntryOffset);
	}
    }
  while (NT_SUCCESS (NtQueryDirectoryFile (dir, NULL, NULL, 0, &io, pfni,
					   bufsiz, FileNamesInformation,
					   FALSE, NULL, FALSE)));
  return STATUS_SUCCESS;
}

NTSTATUS
unlink_nt (path_conv &pc)
{
  NTSTATUS status;
  HANDLE fh, fh_ro = NULL;
  OBJECT_ATTRIBUTES attr;
  IO_STATUS_BLOCK io;
  HANDLE old_trans = NULL, trans = NULL;
  ULONG num_links = 1;
  FILE_DISPOSITION_INFORMATION disp = { TRUE };
  int reopened = 0;

  bin_status bin_stat = dont_move;

  syscall_printf ("Trying to delete %S, isdir = %d",
		  pc.get_nt_native_path (), pc.isdir ());
  ACCESS_MASK access = DELETE;
  ULONG flags = FILE_OPEN_FOR_BACKUP_INTENT;
  /* Add the reparse point flag to native symlinks, otherwise we remove the
     target, not the symlink. */
  if (pc.is_rep_symlink ())
    flags |= FILE_OPEN_REPARSE_POINT;

  pc.get_object_attr (attr, sec_none_nih);
  /* If the R/O attribute is set, we have to open the file with
     FILE_WRITE_ATTRIBUTES to be able to remove this flags before trying
     to delete it.  We do this separately because there are filesystems
     out there (MVFS), which refuse a request to open a file for DELETE
     if the DOS R/O attribute is set for the file.  After removing the R/O
     attribute, just re-open the file for DELETE and go ahead. */
  if (pc.file_attributes () & FILE_ATTRIBUTE_READONLY)
    {
      FILE_STANDARD_INFORMATION fsi;

      /* If possible, hide the non-atomicity of the "remove R/O flag, remove
	 link to file" operation behind a transaction. */
      if ((pc.fs_flags () & FILE_SUPPORTS_TRANSACTIONS))
	start_transaction (old_trans, trans);
retry_open:
      status = NtOpenFile (&fh_ro, FILE_WRITE_ATTRIBUTES, &attr, &io,
			   FILE_SHARE_VALID_FLAGS, flags);
      if (NT_SUCCESS (status))
	{
	  debug_printf ("Opening %S for removing R/O succeeded",
			pc.get_nt_native_path ());
	  NTSTATUS status2 = NtSetAttributesFile (fh_ro,
						  pc.file_attributes ()
						  & ~FILE_ATTRIBUTE_READONLY);
	  if (!NT_SUCCESS (status2))
	    debug_printf ("Removing R/O on %S failed, status = %y",
			  pc.get_nt_native_path (), status2);
	  pc.init_reopen_attr (attr, fh_ro);
	}
      else
	{
	  debug_printf ("Opening %S for removing R/O failed, status = %y",
			pc.get_nt_native_path (), status);
	  if (NT_TRANSACTIONAL_ERROR (status) && trans)
	    {
	      /* If NtOpenFile fails due to transactional problems, stop
		 transaction and go ahead without. */
	      stop_transaction (status, old_trans, trans);
	      debug_printf ("Transaction failure.  Retry open.");
	      goto retry_open;
	    }
	}
      if (pc.is_lnk_symlink ())
	{
	  status = NtQueryInformationFile (fh_ro, &io, &fsi, sizeof fsi,
					   FileStandardInformation);
	  if (NT_SUCCESS (status))
	    num_links = fsi.NumberOfLinks;
	}
      access |= FILE_WRITE_ATTRIBUTES;
    }
  /* First try to open the file with only allowing sharing for delete.  If
     the file has an open handle on it, other than just for deletion, this
     will fail.  That indicates that the file has to be moved to the recycle
     bin so that it actually disappears from its directory even though its
     in use.  Otherwise, if opening doesn't fail, the file is not in use and
     we can go straight to setting the delete disposition flag.

     NOTE: The missing sharing modes FILE_SHARE_READ and FILE_SHARE_WRITE do
	   NOT result in a STATUS_SHARING_VIOLATION, if another handle is
	   opened for reading/writing metadata only.  In other words, if
	   another handle is open, but does not have the file open with
	   FILE_READ_DATA or FILE_WRITE_DATA, the following NtOpenFile call
	   will succeed.  So, apparently there is no reliable way to find out
	   if a file is already open elsewhere for other purposes than
	   reading and writing data.  */
  status = NtOpenFile (&fh, access, &attr, &io, FILE_SHARE_DELETE, flags);
  /* STATUS_SHARING_VIOLATION is what we expect. STATUS_LOCK_NOT_GRANTED can
     be generated under not quite clear circumstances when trying to open a
     file on NFS with FILE_SHARE_DELETE only.  This has been observed with
     SFU 3.5 if the NFS share has been mounted under a drive letter.  It's
     not generated for all files, but only for some.  If it's generated once
     for a file, it will be generated all the time.  It looks as if wrong file
     state information is stored within the NFS client which never times out.
     Opening the file with FILE_SHARE_VALID_FLAGS will work, though, and it
     is then possible to delete the file quite normally. */
  if (status == STATUS_SHARING_VIOLATION || status == STATUS_LOCK_NOT_GRANTED)
    {
      debug_printf ("Sharing violation when opening %S",
		    pc.get_nt_native_path ());
      /* We never call try_to_bin on NetApp.  Netapp filesystems don't
	 understand the "move and delete" method at all and have all kinds
	 of weird effects.  Just setting the delete dispositon usually
	 works fine, though.

	 NFS implements its own mechanism to remove in-use files, which looks
	 quite similar to what we do in try_to_bin for remote files.  However,
	 apparently it doesn't work as desired in all cases.  This has been
	 observed when running the gawk 4.1.62++ testcase "testext.awk" under
	 Windows 10.  So for NFS we still call try_to_bin to rename the file,
	 at least to make room for subsequent creation of a file with the
	 same filename. */
      if (!pc.fs_is_netapp ())
	bin_stat = move_to_bin;
      /* If the file is not a directory, of if we didn't set the move_to_bin
	 flag, just proceed with the FILE_SHARE_VALID_FLAGS set. */
      if (!pc.isdir () || bin_stat == dont_move)
	status = NtOpenFile (&fh, access, &attr, &io,
			     FILE_SHARE_VALID_FLAGS, flags);
      else
	{
	  /* Otherwise it's getting tricky.  The directory is opened in some
	     process, so we're supposed to move it to the recycler and mark it
	     for deletion.  But what if the directory is not empty?  The move
	     will work, but the subsequent delete will fail.  So we would
	     have to move it back.  While we do that in try_to_bin, it's bad,
	     because the move results in a temporary inconsistent state.
	     So, we test first if the directory is empty.  If not, we bail
	     out with STATUS_DIRECTORY_NOT_EMPTY.  This avoids most of the
	     problems. */
	  status = NtOpenFile (&fh, access | FILE_LIST_DIRECTORY | SYNCHRONIZE,
			       &attr, &io, FILE_SHARE_VALID_FLAGS,
			       flags | FILE_SYNCHRONOUS_IO_NONALERT);
	  if (NT_SUCCESS (status))
	    {
	      status = check_dir_not_empty (fh, pc);
	      if (!NT_SUCCESS (status))
		{
		  NtClose (fh);
		  if (fh_ro)
		    NtClose (fh_ro);
		  goto out;
		}
	    }
	}
    }
  if (fh_ro)
    NtClose (fh_ro);
  if (!NT_SUCCESS (status))
    {
      if (status == STATUS_DELETE_PENDING)
	{
	  debug_printf ("Delete %S already pending", pc.get_nt_native_path ());
	  status = STATUS_SUCCESS;
	  goto out;
	}
      debug_printf ("Opening %S for delete failed, status = %y",
		    pc.get_nt_native_path (), status);
      goto out;
    }
  /* Try to move to bin if a sharing violation occured.  If that worked,
     we're done. */
  if (bin_stat == move_to_bin
      && (bin_stat = try_to_bin (pc, fh, access, flags)) >= has_been_moved)
    {
      if (bin_stat == has_been_moved)
	status = STATUS_SUCCESS;
      else
	{
	  status = STATUS_DIRECTORY_NOT_EMPTY;
	  NtClose (fh);
	}
      goto out;
    }

try_again:
  /* Try to set delete disposition. */
  status = NtSetInformationFile (fh, &io, &disp, sizeof disp,
				 FileDispositionInformation);
  if (!NT_SUCCESS (status))
    {
      debug_printf ("Setting delete disposition on %S failed, status = %y",
		    pc.get_nt_native_path (), status);
      if (status == STATUS_DIRECTORY_NOT_EMPTY)
	{
	  NTSTATUS status2 = STATUS_SUCCESS;

	  if (!reopened)
	    {
	      /* Have to close and reopen the file from scratch, otherwise
		 we collide with the delete-only sharing mode. */
	      pc.get_object_attr (attr, sec_none_nih);
	      NtClose (fh);
	      status2 = NtOpenFile (&fh, access | FILE_LIST_DIRECTORY
					 | SYNCHRONIZE,
				    &attr, &io, FILE_SHARE_VALID_FLAGS,
				    flags | FILE_SYNCHRONOUS_IO_NONALERT);
	    }
	  if (NT_SUCCESS (status2) && reopened < 20)
	    {
	      /* Workaround rm -r problem:

		 Sometimes a deleted directory lingers in its parent dir
		 after the deleting handle has already been closed.  This
		 can break deleting the parent dir.  See the comment in
		 check_dir_not_empty for more information.

		 What we do here is this:  If check_dir_not_empty returns
		 STATUS_SUCCESS, the dir is either empty, or only inhabited
		 by already deleted entries.  If so, we try to move the dir
		 into the bin.  This usually works.

		 However, if we're on a filesystem which doesn't support
		 the try_to_bin method, or if moving to the bin doesn't work
		 for some reason, just try to delete the directory again,
		 with a very short grace period to free the CPU for a while.
		 This gives the OS time to clean up.  5ms is enough in my
		 testing to make sure that we don't have to try more than
		 once in practically all cases.
		 While this is an extrem bordercase, we don't want to hang
		 infinitely in case a file in the directory is in the "delete
		 pending" state but an application holds an open handle to it
		 for a longer time.  So we don't try this more than 20 times,
		 which means a process time of 100-120ms. */
	      if (check_dir_not_empty (fh, pc) == STATUS_SUCCESS)
		{
		  if (bin_stat == dont_move)
		    {
		      bin_stat = move_to_bin;
		      if (!pc.fs_is_nfs () && !pc.fs_is_netapp ())
			{
			  debug_printf ("Try-to-bin %S",
					pc.get_nt_native_path ());
			  bin_stat = try_to_bin (pc, fh, access, flags);
			}
		    }
		  /* Do NOT handle bin_stat == dir_not_empty here! */
		  if (bin_stat == has_been_moved)
		    status = STATUS_SUCCESS;
		  else
		    {
		      debug_printf ("Try %S again", pc.get_nt_native_path ());
		      ++reopened;
		      Sleep (5L);
		      goto try_again;
		    }
		}
	    }
	  else if (status2 != STATUS_OBJECT_PATH_NOT_FOUND
		   && status2 !=  STATUS_OBJECT_NAME_NOT_FOUND)
	    {
	      fh = NULL;
	      debug_printf ("Opening dir %S for check_dir_not_empty failed, "
			    "status = %y", pc.get_nt_native_path (), status2);
	    }
	  else /* Directory disappeared between NtClose and NtOpenFile. */
	    status = STATUS_SUCCESS;
	}
      /* Trying to delete a hardlink to a file in use by the system in some
	 way (for instance, font files) by setting the delete disposition fails
	 with STATUS_CANNOT_DELETE.  Strange enough, deleting these hardlinks
	 using delete-on-close semantic works... most of the time.

	 Don't use delete-on-close on remote shares.  If two processes
	 have open handles on a file and one of them calls unlink, the
	 file is removed from the remote share even though the other
	 process still has an open handle.  That process than gets Win32
	 error 59, ERROR_UNEXP_NET_ERR when trying to access the file.
	 Microsoft KB 837665 describes this problem as a bug in 2K3, but
	 I have reproduced it on other systems. */
      else if (status == STATUS_CANNOT_DELETE
	       && (!pc.isremote () || pc.fs_is_ncfsd ()))
	{
	  HANDLE fh2;

	  debug_printf ("Cannot delete %S, try delete-on-close",
			pc.get_nt_native_path ());
	  /* Re-open from handle so we open the correct file no matter if it
	     has been moved to the bin or not. */
	  status = NtOpenFile (&fh2, DELETE,
			       pc.init_reopen_attr (attr, fh), &io,
			       bin_stat == move_to_bin ? FILE_SHARE_VALID_FLAGS
						       : FILE_SHARE_DELETE,
			       flags | FILE_DELETE_ON_CLOSE);
	  if (!NT_SUCCESS (status))
	    {
	      debug_printf ("Setting delete-on-close on %S failed, status = %y",
			    pc.get_nt_native_path (), status);
	      /* This is really the last chance.  If it hasn't been moved
		 to the bin already, try it now.  If moving to the bin
		 succeeds, we got rid of the file in some way, even if
		 unlinking didn't work. */
	      if (bin_stat == dont_move)
		bin_stat = try_to_bin (pc, fh, access, flags);
	      if (bin_stat >= has_been_moved)
		status = bin_stat == has_been_moved
				     ? STATUS_SUCCESS
				     : STATUS_DIRECTORY_NOT_EMPTY;
	    }
	  else
	    NtClose (fh2);
	}
    }
  if (fh)
    {
      if (access & FILE_WRITE_ATTRIBUTES)
	{
	  /* Restore R/O attribute if setting the delete disposition failed. */
	  if (!NT_SUCCESS (status))
	    NtSetAttributesFile (fh, pc.file_attributes ());
	  /* If we succeeded, restore R/O attribute to accommodate hardlinks.
	     Only ever try to do this for our own winsymlinks, because there's
	     a problem with setting the delete disposition:
	     http://msdn.microsoft.com/en-us/library/ff545765%28VS.85%29.aspx
	     "Subsequently, the only legal operation by such a caller is
	     to close the open file handle."

	     FIXME?  We could use FILE_HARD_LINK_INFORMATION to find all
	     hardlinks and use one of them to restore the R/O bit, after the
	     NtClose, but before we stop the transaction.  This avoids the
	     aforementioned problem entirely . */
	  else if (pc.is_lnk_symlink () && num_links > 1)
	    NtSetAttributesFile (fh, pc.file_attributes ());
	}

      NtClose (fh);

    }
out:
  /* Stop transaction if we started one. */
  if (trans)
    stop_transaction (status, old_trans, trans);
  syscall_printf ("%S, return status = %y", pc.get_nt_native_path (), status);
  return status;
}

extern "C" int
unlink (const char *ourname)
{
  int res = -1;
  dev_t devn;
  NTSTATUS status;

  path_conv win32_name (ourname, PC_SYM_NOFOLLOW, stat_suffixes);

  if (win32_name.error)
    {
      set_errno (win32_name.error);
      goto done;
    }

  devn = win32_name.get_device ();
  if (isproc_dev (devn))
    {
      set_errno (EROFS);
      goto done;
    }

  if (!win32_name.exists ())
    {
      debug_printf ("unlinking a nonexistent file");
      set_errno (ENOENT);
      goto done;
    }
  else if (win32_name.isdir ())
    {
      debug_printf ("unlinking a directory");
      set_errno (EPERM);
      goto done;
    }

  status = unlink_nt (win32_name);
  if (NT_SUCCESS (status))
    res = 0;
  else
    __seterrno_from_nt_status (status);

 done:
  syscall_printf ("%R = unlink(%s)", res, ourname);
  return res;
}

extern "C" int
_remove_r (struct _reent *, const char *ourname)
{
  path_conv win32_name (ourname, PC_SYM_NOFOLLOW);

  if (win32_name.error)
    {
      set_errno (win32_name.error);
      syscall_printf ("%R = remove(%s)",-1, ourname);
      return -1;
    }

  return win32_name.isdir () ? rmdir (ourname) : unlink (ourname);
}

extern "C" int
remove (const char *ourname)
{
  path_conv win32_name (ourname, PC_SYM_NOFOLLOW);

  if (win32_name.error)
    {
      set_errno (win32_name.error);
      syscall_printf ("-1 = remove (%s)", ourname);
      return -1;
    }

  int res = win32_name.isdir () ? rmdir (ourname) : unlink (ourname);
  syscall_printf ("%R = remove(%s)", res, ourname);
  return res;
}

extern "C" pid_t
getpid ()
{
  syscall_printf ("%d = getpid()", myself->pid);
  return myself->pid;
}

extern "C" pid_t
_getpid_r (struct _reent *)
{
  return getpid ();
}

/* getppid: POSIX 4.1.1.1 */
extern "C" pid_t
getppid ()
{
  syscall_printf ("%d = getppid()", myself->ppid);
  return myself->ppid;
}

/* setsid: POSIX 4.3.2.1 */
extern "C" pid_t
setsid (void)
{
  if (myself->pgid == myself->pid)
    syscall_printf ("hmm.  pgid %d pid %d", myself->pgid, myself->pid);
  else
    {
      myself->ctty = -2;
      myself->sid = myself->pid;
      myself->pgid = myself->pid;
      if (cygheap->ctty)
	cygheap->close_ctty ();
      syscall_printf ("sid %d, pgid %d, %s", myself->sid, myself->pgid, myctty ());
      return myself->sid;
    }

  set_errno (EPERM);
  return -1;
}

extern "C" pid_t
getsid (pid_t pid)
{
  pid_t res;
  if (!pid)
    res = myself->sid;
  else
    {
      pinfo p (pid);
      if (p)
	res = p->sid;
      else
	{
	  set_errno (ESRCH);
	  res = -1;
	}
    }
  syscall_printf ("%R = getsid(%d)", pid);
  return res;
}

extern "C" ssize_t
read (int fd, void *ptr, size_t len)
{
  size_t res = (size_t) -1;

  pthread_testcancel ();

  __try
    {
      cygheap_fdget cfd (fd);
      if (cfd < 0)
	__leave;

      if ((cfd->get_flags () & O_ACCMODE) == O_WRONLY)
	{
	  set_errno (EBADF);
	  __leave;
	}

      /* Could block, so let user know we at least got here.  */
      syscall_printf ("read(%d, %p, %d) %sblocking",
		      fd, ptr, len, cfd->is_nonblocking () ? "non" : "");

      cfd->read (ptr, len);
      res = len;
    }
  __except (EFAULT) {}
  __endtry
  syscall_printf ("%lR = read(%d, %p, %d)", res, fd, ptr, len);
  return (ssize_t) res;
}

EXPORT_ALIAS (read, _read)

extern "C" ssize_t
readv (int fd, const struct iovec *const iov, const int iovcnt)
{
  ssize_t res = -1;

  pthread_testcancel ();

  __try
    {
      const ssize_t tot = check_iovec_for_read (iov, iovcnt);

      cygheap_fdget cfd (fd);
      if (cfd < 0)
	__leave;

      if (tot <= 0)
	{
	  res = tot;
	  __leave;
	}

      if ((cfd->get_flags () & O_ACCMODE) == O_WRONLY)
	{
	  set_errno (EBADF);
	  __leave;
	}

      /* Could block, so let user know we at least got here.  */
      syscall_printf ("readv(%d, %p, %d) %sblocking",
		      fd, iov, iovcnt, cfd->is_nonblocking () ? "non" : "");

      res = cfd->readv (iov, iovcnt, tot);
    }
  __except (EFAULT) {}
  __endtry
  syscall_printf ("%lR = readv(%d, %p, %d)", res, fd, iov, iovcnt);
  return res;
}

extern "C" ssize_t
pread (int fd, void *ptr, size_t len, off_t off)
{
  ssize_t res;

  pthread_testcancel ();

  cygheap_fdget cfd (fd);
  if (cfd < 0)
    res = -1;
  else
    res = cfd->pread (ptr, len, off);

  syscall_printf ("%lR = pread(%d, %p, %d, %d)", res, fd, ptr, len, off);
  return res;
}

extern "C" ssize_t
write (int fd, const void *ptr, size_t len)
{
  ssize_t res = -1;

  pthread_testcancel ();

  __try
    {
      cygheap_fdget cfd (fd);
      if (cfd < 0)
	__leave;

      if ((cfd->get_flags () & O_ACCMODE) == O_RDONLY)
	{
	  set_errno (EBADF);
	  __leave;
	}

      /* Could block, so let user know we at least got here.  */
      if (fd == 1 || fd == 2)
	paranoid_printf ("write(%d, %p, %d)", fd, ptr, len);
      else
	syscall_printf  ("write(%d, %p, %d)", fd, ptr, len);

      res = cfd->write (ptr, len);
    }
  __except (EFAULT) {}
  __endtry
  syscall_printf ("%lR = write(%d, %p, %d)", res, fd, ptr, len);
  return res;
}

EXPORT_ALIAS (write, _write)

extern "C" ssize_t
writev (const int fd, const struct iovec *const iov, const int iovcnt)
{
  ssize_t res = -1;

  pthread_testcancel ();

  __try
    {
      const ssize_t tot = check_iovec_for_write (iov, iovcnt);

      cygheap_fdget cfd (fd);
      if (cfd < 0)
	__leave;

      if (tot <= 0)
	{
	  res = tot;
	  __leave;
	}

      if ((cfd->get_flags () & O_ACCMODE) == O_RDONLY)
	{
	  set_errno (EBADF);
	  __leave;
	}

      /* Could block, so let user know we at least got here.  */
      if (fd == 1 || fd == 2)
	paranoid_printf ("writev(%d, %p, %d)", fd, iov, iovcnt);
      else
	syscall_printf  ("writev(%d, %p, %d)", fd, iov, iovcnt);

      res = cfd->writev (iov, iovcnt, tot);
    }
  __except (EFAULT) {}
  __endtry
  if (fd == 1 || fd == 2)
    paranoid_printf ("%lR = writev(%d, %p, %d)", res, fd, iov, iovcnt);
  else
    syscall_printf ("%lR = writev(%d, %p, %d)", res, fd, iov, iovcnt);
  return res;
}

extern "C" ssize_t
pwrite (int fd, void *ptr, size_t len, off_t off)
{
  pthread_testcancel ();

  ssize_t res;
  cygheap_fdget cfd (fd);
  if (cfd < 0)
    res = -1;
  else
    res = cfd->pwrite (ptr, len, off);

  syscall_printf ("%lR = pwrite(%d, %p, %d, %d)", res, fd, ptr, len, off);
  return res;
}

/* _open */
/* newlib's fcntl.h defines _open as taking variable args so we must
   correspond.  The third arg if it exists is: mode_t mode. */
extern "C" int
open (const char *unix_path, int flags, ...)
{
  int res = -1;
  va_list ap;
  mode_t mode = 0;
  fhandler_base *fh = NULL;

  pthread_testcancel ();

  __try
    {
      syscall_printf ("open(%s, %y)", unix_path, flags);
      if (!*unix_path)
	{
	  set_errno (ENOENT);
	  __leave;
	}

      /* check for optional mode argument */
      va_start (ap, flags);
      mode = va_arg (ap, mode_t);
      va_end (ap);

      cygheap_fdnew fd;

      if (fd < 0)
	__leave;		/* errno already set */

      /* This is a temporary kludge until all utilities can catch up
	 with a change in behavior that implements linux functionality:
	 opening a tty should not automatically cause it to become the
	 controlling tty for the process.  */
      int opt = PC_OPEN | ((flags & (O_NOFOLLOW | O_EXCL))
			   ?  PC_SYM_NOFOLLOW : PC_SYM_FOLLOW);
      if (!(flags & O_NOCTTY) && fd > 2 && myself->ctty != -2)
	{
	  flags |= O_NOCTTY;
	  /* flag that, if opened, this fhandler could later be capable
	     of being a controlling terminal if /dev/tty is opened. */
	  opt |= PC_CTTY;
	}

      if (!(fh = build_fh_name (unix_path, opt, stat_suffixes)))
	__leave;		/* errno already set */
      if ((flags & O_NOFOLLOW) && fh->issymlink ())
	{
	  set_errno (ELOOP);
	  __leave;
	}
      if ((flags & O_DIRECTORY) && fh->exists () && !fh->pc.isdir ())
	{
	  set_errno (ENOTDIR);
	  __leave;
	}
      if (((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) && fh->exists ())
	{
	  set_errno (EEXIST);
	  __leave;
	}
      if (flags & O_TMPFILE)
	{
	  if ((flags & O_ACCMODE) != O_WRONLY && (flags & O_ACCMODE) != O_RDWR)
	    {
	      set_errno (EINVAL);
	      __leave;
	    }
	  if (!fh->pc.isdir ())
	    {
	      set_errno (fh->exists () ? ENOTDIR : ENOENT);
	      __leave;
	    }
	  /* Unfortunately Windows does not allow to create a nameless file.
	     So create unique filename instead.  It starts with ".cyg_tmp_",
	     followed by an 8 byte unique hex number, followed by an 8 byte
	     random hex number. */
	  int64_t rnd;
	  fhandler_base *fh_file;
	  char *new_path;

	  new_path = (char *) malloc (strlen (fh->get_name ())
				      + 1  /* slash */
				      + 10 /* prefix */
				      + 16 /* 64 bit unique id as hex*/
				      + 16 /* 64 bit random number as hex */
				      + 1  /* trailing NUL */);
	  if (!new_path)
	    __leave;
	  fh->set_unique_id ();
	  RtlGenRandom (&rnd, sizeof rnd);
	  __small_sprintf (new_path, "%s/%s%016X%016X",
			   fh->get_name (), ".cyg_tmp_",
			   fh->get_unique_id (), rnd);

	  if (!(fh_file = build_fh_name (new_path, opt, NULL)))
	    {
	      free (new_path);
	      __leave;		/* errno already set */
	    }
	  delete fh;
	  fh = fh_file;
	}

      if ((fh->is_fs_special () && fh->device_access_denied (flags))
	  || !fh->open_with_arch (flags, mode & 07777))
	__leave;		/* errno already set */

      fd = fh;
      if (fd <= 2)
	set_std_handle (fd);
      res = fd;
    }
  __except (EFAULT) {}
  __endtry
  if (res < 0 && fh)
    delete fh;
  syscall_printf ("%R = open(%s, %y)", res, unix_path, flags);
  return res;
}

EXPORT_ALIAS (open, _open )
EXPORT_ALIAS (open, _open64 )

extern "C" off_t
lseek64 (int fd, off_t pos, int dir)
{
  off_t res;

  if (dir != SEEK_SET && dir != SEEK_CUR && dir != SEEK_END)
    {
      set_errno (EINVAL);
      res = -1;
    }
  else
    {
      cygheap_fdget cfd (fd);
      if (cfd >= 0)
	res = cfd->lseek (pos, dir);
      else
	res = -1;
    }
  /* Can't use %R/%lR here since res is always 8 bytes */
  syscall_printf (res == -1 ? "%D = lseek(%d, %D, %d), errno %d"
			    : "%D = lseek(%d, %D, %d)",
		  res, fd, pos, dir, get_errno ());

  return res;
}

EXPORT_ALIAS (lseek64, _lseek64)

#ifdef __i386__
extern "C" _off_t
lseek (int fd, _off_t pos, int dir)
{
  return lseek64 (fd, (off_t) pos, dir);
}
EXPORT_ALIAS (lseek, _lseek)
#else
EXPORT_ALIAS (lseek64, lseek)
EXPORT_ALIAS (lseek64, _lseek)
#endif


extern "C" int
close (int fd)
{
  int res;

  syscall_printf ("close(%d)", fd);

  pthread_testcancel ();

  cygheap_fdget cfd (fd, true);
  if (cfd < 0)
    res = -1;
  else
    {
      cfd->isclosed (true);
      res = cfd->close_with_arch ();
      cfd.release ();
    }

  syscall_printf ("%R = close(%d)", res, fd);
  return res;
}

EXPORT_ALIAS (close, _close)

extern "C" int
isatty (int fd)
{
  int res;

  cygheap_fdget cfd (fd);
  if (cfd < 0)
    res = 0;
  else
    res = cfd->is_tty ();
  syscall_printf ("%R = isatty(%d)", res, fd);
  return res;
}
EXPORT_ALIAS (isatty, _isatty)

extern "C" int
link (const char *oldpath, const char *newpath)
{
  int res = -1;
  fhandler_base *fh;

  if (!(fh = build_fh_name (oldpath, PC_SYM_NOFOLLOW | PC_KEEP_HANDLE,
			    stat_suffixes)))
    goto error;

  if (fh->error ())
    {
      debug_printf ("got %d error from build_fh_name", fh->error ());
      set_errno (fh->error ());
    }
  else if (fh->pc.isdir ())
    set_errno (EPERM); /* We do not permit linking directories.  */
  else if (!fh->pc.exists ())
    set_errno (ENOENT);
  else
    res = fh->link (newpath);

  delete fh;
 error:
  syscall_printf ("%R = link(%s, %s)", res, oldpath, newpath);
  return res;
}

/* chown: POSIX 5.6.5.1 */
/*
 * chown () is only implemented for Windows NT.  Under other operating
 * systems, it is only a stub that always returns zero.
 */
static int
chown_worker (const char *name, unsigned fmode, uid_t uid, gid_t gid)
{
  int res = -1;
  fhandler_base *fh;

  if (!(fh = build_fh_name (name, fmode, stat_suffixes)))
    goto error;

  if (fh->error ())
    {
      debug_printf ("got %d error from build_fh_name", fh->error ());
      set_errno (fh->error ());
    }
  else
    res = fh->fchown (uid, gid);

  delete fh;
 error:
  syscall_printf ("%R = %schown(%s,...)",
		  res, (fmode & PC_SYM_NOFOLLOW) ? "l" : "", name);
  return res;
}

extern "C" int
chown32 (const char * name, uid_t uid, gid_t gid)
{
  return chown_worker (name, PC_SYM_FOLLOW, uid, gid);
}

#ifdef __i386__
extern "C" int
chown (const char * name, __uid16_t uid, __gid16_t gid)
{
  return chown_worker (name, PC_SYM_FOLLOW,
		       uid16touid32 (uid), gid16togid32 (gid));
}
#else
EXPORT_ALIAS (chown32, chown)
#endif

extern "C" int
lchown32 (const char * name, uid_t uid, gid_t gid)
{
  return chown_worker (name, PC_SYM_NOFOLLOW, uid, gid);
}

#ifdef __i386__
extern "C" int
lchown (const char * name, __uid16_t uid, __gid16_t gid)
{
  return chown_worker (name, PC_SYM_NOFOLLOW,
		       uid16touid32 (uid), gid16togid32 (gid));
}
#else
EXPORT_ALIAS (lchown32, lchown)
#endif

extern "C" int
fchown32 (int fd, uid_t uid, gid_t gid)
{
  cygheap_fdget cfd (fd);
  if (cfd < 0)
    {
      syscall_printf ("-1 = fchown (%d,...)", fd);
      return -1;
    }

  int res = cfd->fchown (uid, gid);

  syscall_printf ("%R = fchown(%s,...)", res, cfd->get_name ());
  return res;
}

#ifdef __i386__
extern "C" int
fchown (int fd, __uid16_t uid, __gid16_t gid)
{
  return fchown32 (fd, uid16touid32 (uid), gid16togid32 (gid));
}
#else
EXPORT_ALIAS (fchown32, fchown)
#endif

/* umask: POSIX 5.3.3.1 */
extern "C" mode_t
umask (mode_t mask)
{
  mode_t oldmask;

  oldmask = cygheap->umask;
  cygheap->umask = mask & 0777;
  return oldmask;
}

int
chmod_device (path_conv& pc, mode_t mode)
{
  return mknod_worker (pc.get_win32 (), pc.dev.mode () & S_IFMT, mode, pc.dev.get_major (), pc.dev.get_minor ());
}

#define FILTERED_MODE(m)	((m) & (S_ISUID | S_ISGID | S_ISVTX \
					| S_IRWXU | S_IRWXG | S_IRWXO))

/* chmod: POSIX 5.6.4.1 */
extern "C" int
chmod (const char *path, mode_t mode)
{
  int res = -1;
  fhandler_base *fh;
  if (!(fh = build_fh_name (path, PC_SYM_FOLLOW, stat_suffixes)))
    goto error;

  if (fh->error ())
    {
      debug_printf ("got %d error from build_fh_name", fh->error ());
      set_errno (fh->error ());
    }
  else
    res = fh->fchmod (FILTERED_MODE (mode));

  delete fh;
 error:
  syscall_printf ("%R = chmod(%s, 0%o)", res, path, mode);
  return res;
}

/* fchmod: P96 5.6.4.1 */

extern "C" int
fchmod (int fd, mode_t mode)
{
  cygheap_fdget cfd (fd);
  if (cfd < 0)
    {
      syscall_printf ("-1 = fchmod (%d, 0%o)", fd, mode);
      return -1;
    }

  return cfd->fchmod (FILTERED_MODE (mode));
}

#ifdef __i386__
static void
stat64_to_stat32 (struct stat *src, struct __stat32 *dst)
{
  dst->st_dev = ((src->st_dev >> 8) & 0xff00) | (src->st_dev & 0xff);
  dst->st_ino = ((unsigned) (src->st_ino >> 32)) | (unsigned) src->st_ino;
  dst->st_mode = src->st_mode;
  dst->st_nlink = src->st_nlink;
  dst->st_uid = src->st_uid;
  dst->st_gid = src->st_gid;
  dst->st_rdev = ((src->st_rdev >> 8) & 0xff00) | (src->st_rdev & 0xff);
  dst->st_size = src->st_size;
  dst->st_atim = src->st_atim;
  dst->st_mtim = src->st_mtim;
  dst->st_ctim = src->st_ctim;
  dst->st_blksize = src->st_blksize;
  dst->st_blocks = src->st_blocks;
}
#endif

static struct stat dev_st;
static bool dev_st_inited;

void
fhandler_base::stat_fixup (struct stat *buf)
{
  /* For devices, set inode number to device number.  This gives us a valid,
     unique inode number without having to call hash_path_name.  /dev/tty needs
     a bit of persuasion to get the same st_ino value in stat and fstat. */
  if (!buf->st_ino)
    {
      if (get_major () == DEV_VIRTFS_MAJOR)
	buf->st_ino = get_ino ();
      else if (dev () == FH_TTY ||
	       ((get_major () == DEV_PTYS_MAJOR
		 || get_major () == DEV_CONS_MAJOR)
		&& !strcmp (get_name (), "/dev/tty")))
	buf->st_ino = FH_TTY;
      else
	buf->st_ino = get_device ();
      	
    }
  /* For /dev-based devices, st_dev must be set to the device number of /dev,
     not it's own device major/minor numbers.  What we do here to speed up
     the process is to fetch the device number of /dev only once, liberally
     assuming that /dev doesn't change over the lifetime of a process. */
  if (!buf->st_dev)
    {
      if (dev ().is_dev_resident ())
	{
	  if (!dev_st_inited)
	    {
	      stat64 ("/dev", &dev_st);
	      dev_st_inited = true;
	    }
	  buf->st_dev = dev_st.st_dev;
	}
      else
	buf->st_dev = get_device ();
    }
  /* Only set st_rdev if it's a device. */
  if (!buf->st_rdev && get_major () != DEV_VIRTFS_MAJOR)
    {
      buf->st_rdev = get_device ();
      /* consX, console, conin, and conout point to the same device.
	 Make sure the link count is correct. */
      if (buf->st_rdev == (dev_t) myself->ctty && iscons_dev (myself->ctty))
	buf->st_nlink = 4;
      /* CD-ROM drives have two links, /dev/srX and /dev/scdX. */
      else if (gnu_dev_major (buf->st_rdev) == DEV_CDROM_MAJOR)
	buf->st_nlink = 2;
    }
}

extern "C" int
fstat64 (int fd, struct stat *buf)
{
  int res;

  cygheap_fdget cfd (fd);
  if (cfd < 0)
    res = -1;
  else
    {
      memset (buf, 0, sizeof (struct stat));
      res = cfd->fstat (buf);
      if (!res)
	cfd->stat_fixup (buf);
    }

  syscall_printf ("%R = fstat(%d, %p)", res, fd, buf);
  return res;
}

extern "C" int
_fstat64_r (struct _reent *ptr, int fd, struct stat *buf)
{
  int ret;

  if ((ret = fstat64 (fd, buf)) == -1)
    ptr->_errno = get_errno ();
  return ret;
}

#ifdef __i386__
extern "C" int
fstat (int fd, struct stat *buf)
{
  struct stat buf64;
  int ret = fstat64 (fd, &buf64);
  if (!ret)
    stat64_to_stat32 (&buf64, (struct __stat32 *) buf);
  return ret;
}

extern "C" int
_fstat_r (struct _reent *ptr, int fd, struct stat *buf)
{
  int ret;

  if ((ret = fstat (fd, buf)) == -1)
    ptr->_errno = get_errno ();
  return ret;
}
#else
EXPORT_ALIAS (fstat64, fstat)
EXPORT_ALIAS (_fstat64_r, _fstat_r)
#endif

/* fsync: P96 6.6.1.1 */
extern "C" int
fsync (int fd)
{
  pthread_testcancel ();
  cygheap_fdget cfd (fd);
  if (cfd < 0)
    {
      syscall_printf ("-1 = fsync (%d)", fd);
      return -1;
    }
  return cfd->fsync ();
}

EXPORT_ALIAS (fsync, fdatasync)

static void
sync_worker (HANDLE dir, USHORT len, LPCWSTR vol)
{
  NTSTATUS status;
  HANDLE fh;
  IO_STATUS_BLOCK io;
  OBJECT_ATTRIBUTES attr;
  UNICODE_STRING uvol = { len, len, (WCHAR *) vol };

  InitializeObjectAttributes (&attr, &uvol, OBJ_CASE_INSENSITIVE, dir, NULL);
  status = NtOpenFile (&fh, GENERIC_WRITE, &attr, &io,
		       FILE_SHARE_VALID_FLAGS, 0);
  if (!NT_SUCCESS (status))
    debug_printf ("NtOpenFile (%S), status %y", &uvol, status);
  else
    {
      status = NtFlushBuffersFile (fh, &io);
      if (!NT_SUCCESS (status))
	debug_printf ("NtFlushBuffersFile (%S), status %y", &uvol, status);
      NtClose (fh);
    }
}

/* sync: SUSv3 */
extern "C" void
sync ()
{
  OBJECT_ATTRIBUTES attr;
  NTSTATUS status;
  HANDLE devhdl;
  UNICODE_STRING device;

  /* Open \Device object directory. */
  RtlInitUnicodeString (&device, L"\\Device");
  InitializeObjectAttributes (&attr, &device, OBJ_CASE_INSENSITIVE, NULL, NULL);
  status = NtOpenDirectoryObject (&devhdl, DIRECTORY_QUERY, &attr);
  if (!NT_SUCCESS (status))
    {
      debug_printf ("NtOpenDirectoryObject, status %y", status);
      return;
    }
  /* Traverse \Device directory ... */
  PDIRECTORY_BASIC_INFORMATION dbi = (PDIRECTORY_BASIC_INFORMATION)
				     alloca (640);
  BOOLEAN restart = TRUE;
  ULONG context = 0;
  while (NT_SUCCESS (NtQueryDirectoryObject (devhdl, dbi, 640, TRUE, restart,
					     &context, NULL)))
    {
      restart = FALSE;
      /* ... and call sync_worker for each HarddiskVolumeX entry. */
      if (dbi->ObjectName.Length >= 15 * sizeof (WCHAR)
	  && !wcsncasecmp (dbi->ObjectName.Buffer, L"HarddiskVolume", 14)
	  && iswdigit (dbi->ObjectName.Buffer[14]))
	sync_worker (devhdl, dbi->ObjectName.Length, dbi->ObjectName.Buffer);
    }
  NtClose (devhdl);
}

/* Cygwin internal */
int __reg2
stat_worker (path_conv &pc, struct stat *buf)
{
  int res = -1;

  __try
    {
      if (pc.error)
	{
	  debug_printf ("got %d error from path_conv", pc.error);
	  set_errno (pc.error);
	}
      else if (pc.exists ())
	{
	  fhandler_base *fh;

	  if (!(fh = build_fh_pc (pc)))
	    __leave;

	  debug_printf ("(%S, %p, %p), file_attributes %d",
			pc.get_nt_native_path (), buf, fh, (DWORD) *fh);
	  memset (buf, 0, sizeof (*buf));
	  res = fh->fstat (buf);
	  if (!res)
	    fh->stat_fixup (buf);
	  delete fh;
	}
      else
	set_errno (ENOENT);
    }
  __except (EFAULT) {}
  __endtry
  syscall_printf ("%d = (%S,%p)", res, pc.get_nt_native_path (), buf);
  return res;
}

extern "C" int
stat64 (const char *__restrict name, struct stat *__restrict buf)
{
  syscall_printf ("entering");
  path_conv pc (name, PC_SYM_FOLLOW | PC_POSIX | PC_KEEP_HANDLE,
		stat_suffixes);
  return stat_worker (pc, buf);
}

extern "C" int
_stat64_r (struct _reent *__restrict ptr, const char *__restrict name,
	   struct stat *buf)
{
  int ret;

  if ((ret = stat64 (name, buf)) == -1)
    ptr->_errno = get_errno ();
  return ret;
}

#ifdef __i386__
extern "C" int
stat (const char *__restrict name, struct stat *__restrict buf)
{
  struct stat buf64;
  int ret = stat64 (name, &buf64);
  if (!ret)
    stat64_to_stat32 (&buf64, (struct __stat32 *) buf);
  return ret;
}

extern "C" int
_stat_r (struct _reent *__restrict ptr, const char *__restrict name,
	 struct stat *__restrict buf)
{
  int ret;

  if ((ret = stat (name, buf)) == -1)
    ptr->_errno = get_errno ();
  return ret;
}
#else
EXPORT_ALIAS (stat64, stat)
EXPORT_ALIAS (_stat64_r, _stat_r)
#endif

/* lstat: Provided by SVR4 and 4.3+BSD, POSIX? */
extern "C" int
lstat64 (const char *__restrict name, struct stat *__restrict buf)
{
  syscall_printf ("entering");
  path_conv pc (name, PC_SYM_NOFOLLOW | PC_POSIX | PC_KEEP_HANDLE,
		stat_suffixes);
  return stat_worker (pc, buf);
}

#ifdef __i386__
/* lstat: Provided by SVR4 and 4.3+BSD, POSIX? */
extern "C" int
lstat (const char *__restrict name, struct stat *__restrict buf)
{
  struct stat buf64;
  int ret = lstat64 (name, &buf64);
  if (!ret)
    stat64_to_stat32 (&buf64, (struct __stat32 *) buf);
  return ret;
}
#else
EXPORT_ALIAS (lstat64, lstat)
#endif

extern "C" int
access (const char *fn, int flags)
{
  // flags were incorrectly specified
  int res = -1;
  if (flags & ~(F_OK|R_OK|W_OK|X_OK))
    set_errno (EINVAL);
  else
    {
      fhandler_base *fh = build_fh_name (fn, PC_SYM_FOLLOW | PC_KEEP_HANDLE,
					 stat_suffixes);
      if (fh)
	{
	  res =  fh->fhaccess (flags, false);
	  delete fh;
	}
    }
  debug_printf ("returning %d", res);
  return res;
}

/* Linux provides this extension; it is basically a wrapper around the
   POSIX:2008 faccessat (AT_FDCWD, fn, flags, AT_EACCESS).  We also
   provide eaccess as an alias for this, in cygwin.din.  */
extern "C" int
euidaccess (const char *fn, int flags)
{
  // flags were incorrectly specified
  int res = -1;
  if (flags & ~(F_OK|R_OK|W_OK|X_OK))
    set_errno (EINVAL);
  else
    {
      fhandler_base *fh = build_fh_name (fn, PC_SYM_FOLLOW | PC_KEEP_HANDLE,
					 stat_suffixes);
      if (fh)
	{
	  res =  fh->fhaccess (flags, true);
	  delete fh;
	}
    }
  debug_printf ("returning %d", res);
  return res;
}

static void
rename_append_suffix (path_conv &pc, const char *path, size_t len,
		      const char *suffix)
{
  char buf[len + 5];

  if (ascii_strcasematch (path + len - 4, ".lnk")
      || ascii_strcasematch (path + len - 4, ".exe"))
    len -= 4;
  stpcpy (stpncpy (buf, path, len), suffix);
  pc.check (buf, PC_SYM_NOFOLLOW);
}

/* This function tests if a filename has one of the "approved" executable
   suffix.  This list is probably not complete... */
static inline bool
nt_path_has_executable_suffix (PUNICODE_STRING upath)
{
  static const PUNICODE_STRING blessed_executable_suffixes[] =
  {
    &ro_u_com,
    &ro_u_dll,	/* Messy, messy.  Per MSDN, the GetBinaryType function is
		   supposed to return with ERROR_BAD_EXE_FORMAT. if the file
		   is a DLL.  On 64-bit Windows, this works as expected for
		   32-bit and 64-bit DLLs.  On 32-bit Windows this only works
		   for 32-bit DLLs.  For 64-bit DLLs, 32-bit Windows returns
		   true with the type set to SCS_64BIT_BINARY. */
    &ro_u_exe,
    &ro_u_scr,
    &ro_u_sys,
    NULL
  };

  USHORT pos = upath->Length / sizeof (WCHAR);
  PWCHAR path;
  UNICODE_STRING usuf;
  const PUNICODE_STRING *suf;

  /* Too short for a native path? */
  if (pos < 8)
    return false;
  /* Assumption: All executable suffixes have a length of three. */
  path = upath->Buffer + pos - 4;
  if (*path != L'.')
    return false;
  RtlInitCountedUnicodeString (&usuf, path, 4 * sizeof (WCHAR));
  for (suf = blessed_executable_suffixes; *suf; ++suf)
    if (RtlEqualUnicodeString (&usuf, *suf, TRUE))
      return true;
  return false;
}

/* If newpath names an existing file and the RENAME_NOREPLACE flag is
   specified, fail with EEXIST.  Exception: Don't fail if the purpose
   of the rename is just to change the case of oldpath on a
   case-insensitive file system. */
static int
rename2 (const char *oldpath, const char *newpath, unsigned int flags)
{
  tmp_pathbuf tp;
  int res = -1;
  path_conv oldpc, newpc, new2pc, *dstpc, *removepc = NULL;
  bool old_dir_requested = false, new_dir_requested = false;
  bool old_explicit_suffix = false, new_explicit_suffix = false;
  bool noreplace = flags & RENAME_NOREPLACE;
  size_t olen, nlen;
  bool equal_path;
  NTSTATUS status = STATUS_SUCCESS;
  HANDLE fh = NULL, nfh;
  HANDLE old_trans = NULL, trans = NULL;
  OBJECT_ATTRIBUTES attr;
  IO_STATUS_BLOCK io;
  FILE_STANDARD_INFORMATION ofsi;
  PFILE_RENAME_INFORMATION pfri;

  __try
    {
      if (flags & ~RENAME_NOREPLACE)
	/* RENAME_NOREPLACE is the only flag currently supported. */
	{
	  set_errno (EINVAL);
	  __leave;
	}
      if (!*oldpath || !*newpath)
	{
	  /* Reject rename("","x"), rename("x","").  */
	  set_errno (ENOENT);
	  __leave;
	}
      if (has_dot_last_component (oldpath, true))
	{
	  /* Reject rename("dir/.","x").  */
	  oldpc.check (oldpath, PC_SYM_NOFOLLOW, stat_suffixes);
	  set_errno (oldpc.isdir () ? EINVAL : ENOTDIR);
	  __leave;
	}
      if (has_dot_last_component (newpath, true))
	{
	  /* Reject rename("dir","x/.").  */
	  newpc.check (newpath, PC_SYM_NOFOLLOW, stat_suffixes);
	  set_errno (!newpc.exists () ? ENOENT
				      : newpc.isdir () ? EINVAL : ENOTDIR);
	  __leave;
	}

      /* A trailing slash requires that the pathname points to an existing
	 directory.  If it's not, it's a ENOTDIR condition.  The same goes
	 for newpath a bit further down this function. */
      olen = strlen (oldpath);
      if (isdirsep (oldpath[olen - 1]))
	{
	  char *buf;
	  char *p = stpcpy (buf = tp.c_get (), oldpath) - 1;
	  oldpath = buf;
	  while (p >= oldpath && isdirsep (*p))
	    *p-- = '\0';
	  olen = p + 1 - oldpath;
	  if (!olen)
	    {
	      /* The root directory cannot be renamed.  This also rejects
		 the corner case of rename("/","/"), even though it is the
		 same file.  */
	      set_errno (EINVAL);
	      __leave;
	    }
	  old_dir_requested = true;
	}
      oldpc.check (oldpath, PC_SYM_NOFOLLOW, stat_suffixes);
      if (oldpc.error)
	{
	  set_errno (oldpc.error);
	  __leave;
	}
      if (!oldpc.exists ())
	{
	  set_errno (ENOENT);
	  __leave;
	}
      if (oldpc.isspecial () && !oldpc.issocket () && !oldpc.is_fs_special ())
	{
	  /* No renames from virtual FS */
	  set_errno (EROFS);
	  __leave;
	}
      if (oldpc.has_attribute (FILE_ATTRIBUTE_REPARSE_POINT)
	  && !oldpc.issymlink ())
	{
	  /* Volume mount point.  If we try to rename a volume mount point, NT
	     returns STATUS_NOT_SAME_DEVICE ==> Win32 ERROR_NOT_SAME_DEVICE ==>
	     errno EXDEV.  That's bad since mv(1) will now perform a
	     cross-device move.  So what we do here is to treat the volume
	     mount point just like Linux treats a mount point. */
	  set_errno (EBUSY);
	  __leave;
	}
      if (old_dir_requested && !oldpc.isdir ())
	{
	  /* Reject rename("file/","x").  */
	  set_errno (ENOTDIR);
	  __leave;
	}
      if (oldpc.known_suffix ()
	   && (ascii_strcasematch (oldpath + olen - 4, ".lnk")
	       || ascii_strcasematch (oldpath + olen - 4, ".exe")))
	old_explicit_suffix = true;

      nlen = strlen (newpath);
      if (isdirsep (newpath[nlen - 1]))
	{
	  char *buf;
	  char *p = stpcpy (buf = tp.c_get (), newpath) - 1;
	  newpath = buf;
	  while (p >= newpath && isdirsep (*p))
	    *p-- = '\0';
	  nlen = p + 1 - newpath;
	  if (!nlen) /* The root directory is never empty.  */
	    {
	      set_errno (ENOTEMPTY);
	      __leave;
	    }
	  new_dir_requested = true;
	}
      newpc.check (newpath, PC_SYM_NOFOLLOW, stat_suffixes);
      if (newpc.error)
	{
	  set_errno (newpc.error);
	  __leave;
	}
      if (newpc.isspecial () && !newpc.issocket ())
	{
	  /* No renames to virtual FSes */
	  set_errno (EROFS);
	  __leave;
	}
      if (new_dir_requested && !(newpc.exists ()
				 ? newpc.isdir () : oldpc.isdir ()))
	{
	  /* Reject rename("file1","file2/"), but allow rename("dir","d/").  */
	  set_errno (newpc.exists () ? ENOTDIR : ENOENT);
	  __leave;
	}
      if (newpc.exists ()
	  && (oldpc.isdir () ? !newpc.isdir () : newpc.isdir ()))
	{
	  /* Reject rename("file","dir") and rename("dir","file").  */
	  set_errno (newpc.isdir () ? EISDIR : ENOTDIR);
	  __leave;
	}
      if (newpc.known_suffix ()
	  && (ascii_strcasematch (newpath + nlen - 4, ".lnk")
	      || ascii_strcasematch (newpath + nlen - 4, ".exe")))
	new_explicit_suffix = true;

      /* This test is necessary in almost every case, so do it once here. */
      equal_path = RtlEqualUnicodeString (oldpc.get_nt_native_path (),
					  newpc.get_nt_native_path (),
					  oldpc.objcaseinsensitive ());

      /* First check if oldpath and newpath only differ by case.  If so, it's
	 just a request to change the case of the filename.  By simply setting
	 the file attributes to INVALID_FILE_ATTRIBUTES (which translates to
	 "file doesn't exist"), all later tests are skipped. */
      if (oldpc.objcaseinsensitive () && newpc.exists () && equal_path
	  && old_explicit_suffix == new_explicit_suffix)
	{
	  if (RtlEqualUnicodeString (oldpc.get_nt_native_path (),
				     newpc.get_nt_native_path (),
				     FALSE))
	    {
	      res = 0;
	      __leave;
	    }
	  newpc.file_attributes (INVALID_FILE_ATTRIBUTES);
	}
      else if (oldpc.isdir ())
	{
	  /* Check for newpath being identical or a subdir of oldpath. */
	  if (RtlPrefixUnicodeString (oldpc.get_nt_native_path (),
				      newpc.get_nt_native_path (),
				      oldpc.objcaseinsensitive ()))
	    {
	      if (newpc.get_nt_native_path ()->Length
		  == oldpc.get_nt_native_path ()->Length)
		{
		  res = 0;
		  __leave;
		}
	      if (*(PWCHAR) ((PBYTE) newpc.get_nt_native_path ()->Buffer
			     + oldpc.get_nt_native_path ()->Length) == L'\\')
		{
		  set_errno (EINVAL);
		  __leave;
		}
	    }
	}
      else if (!newpc.exists ())
	{
	  if (equal_path && old_explicit_suffix != new_explicit_suffix)
	    {
	      newpc.check (newpath, PC_SYM_NOFOLLOW);
	      if (RtlEqualUnicodeString (oldpc.get_nt_native_path (),
					 newpc.get_nt_native_path (),
					 oldpc.objcaseinsensitive ()))
		{
		  res = 0;
		  __leave;
		}
	    }
	  else if (oldpc.is_lnk_special ()
		   && !RtlEqualUnicodePathSuffix (newpc.get_nt_native_path (),
						  &ro_u_lnk, TRUE))
	    rename_append_suffix (newpc, newpath, nlen, ".lnk");
	  else if (oldpc.is_binary () && !old_explicit_suffix
		   && oldpc.known_suffix ()
		   && !nt_path_has_executable_suffix
		   				(newpc.get_nt_native_path ()))
	    /* Never append .exe suffix if oldpath had .exe suffix given
	       explicitely, or if oldpath wasn't already a .exe file, or
	       if the destination filename has one of the blessed executable
	       suffixes.
	       Note: To rename an executable foo.exe to bar-without-suffix,
	       the .exe suffix must be given explicitly in oldpath. */
	    rename_append_suffix (newpc, newpath, nlen, ".exe");
	}
      else
	{
	  if (equal_path && old_explicit_suffix != new_explicit_suffix)
	    {
	      newpc.check (newpath, PC_SYM_NOFOLLOW);
	      if (RtlEqualUnicodeString (oldpc.get_nt_native_path (),
					 newpc.get_nt_native_path (),
					 oldpc.objcaseinsensitive ()))
		{
		  res = 0;
		  __leave;
		}
	    }
	  else if (oldpc.is_lnk_special ())
	    {
	      if (!newpc.is_lnk_special ()
		  && !RtlEqualUnicodePathSuffix (newpc.get_nt_native_path (),
						 &ro_u_lnk, TRUE))
		{
		  rename_append_suffix (new2pc, newpath, nlen, ".lnk");
		  removepc = &newpc;
		}
	    }
	  else if (oldpc.is_binary ())
	    {
	      /* Never append .exe suffix if oldpath had .exe suffix given
		 explicitely, or if newfile is a binary (in which case the given
		 name probably makes sense as it is), or if the destination
		 filename has one of the blessed executable suffixes. */
	      if (!old_explicit_suffix && oldpc.known_suffix ()
		  && !newpc.is_binary ()
		  && !nt_path_has_executable_suffix
		  				(newpc.get_nt_native_path ()))
		{
		  rename_append_suffix (new2pc, newpath, nlen, ".exe");
		  removepc = &newpc;
		}
	    }
	  else
	    {
	      /* If the new path is an existing .lnk symlink or a .exe file,
		 but the new path has not been specified with explicit suffix,
		 rename to the new name without suffix, as expected, but also
		 remove the clashing symlink or executable.  Did I ever mention
		 how I hate the file suffix idea? */
	      if ((newpc.is_lnk_special ()
		   || RtlEqualUnicodePathSuffix (newpc.get_nt_native_path (),
						 &ro_u_exe, TRUE))
		  && !new_explicit_suffix)
		{
		  new2pc.check (newpath, PC_SYM_NOFOLLOW, stat_suffixes);
		  newpc.get_nt_native_path ()->Length -= 4 * sizeof (WCHAR);
		  if (new2pc.is_binary () || new2pc.is_lnk_special ())
		    removepc = &new2pc;
		}
	    }
	}
      dstpc = (removepc == &newpc) ? &new2pc : &newpc;

      /* Check cross-device before touching anything.  Otherwise we might end
	 up with an unlinked target dir even if the actual rename didn't work.*/
      if (oldpc.fs_type () != dstpc->fs_type ()
	  || oldpc.fs_serial_number () != dstpc->fs_serial_number ())
	{
	  set_errno (EXDEV);
	  __leave;
	}

      /* Should we replace an existing file? */
      if ((removepc || dstpc->exists ()) && noreplace)
	{
	  set_errno (EEXIST);
	  __leave;
	}

      /* Opening the file must be part of the transaction.  It's not sufficient
	 to call only NtSetInformationFile under the transaction.  Therefore we
	 have to start the transaction here, if necessary. */
      if ((dstpc->fs_flags () & FILE_SUPPORTS_TRANSACTIONS)
	  && (dstpc->isdir ()
	      || (!removepc && dstpc->has_attribute (FILE_ATTRIBUTE_READONLY))))
	start_transaction (old_trans, trans);

      int retry_count;
      retry_count = 0;
    retry:
      /* Talking about inconsistent behaviour...
	 - DELETE is required to rename a file.  So far, so good.
	 - At least one cifs FS (Tru64) needs FILE_READ_ATTRIBUTE, otherwise the
	   FileRenameInformation call fails with STATUS_ACCESS_DENIED.  However,
	   on NFS we get a STATUS_ACCESS_DENIED if FILE_READ_ATTRIBUTE is used
	   and the file we try to rename is a symlink.  Urgh.
	 - Samba (only some versions?) doesn't like the FILE_SHARE_DELETE
	   mode if the file has the R/O attribute set and returns
	   STATUS_ACCESS_DENIED in that case. */
      {
	ULONG access = DELETE
		       | (oldpc.fs_is_cifs () ? FILE_READ_ATTRIBUTES : 0);
	ULONG sharing = FILE_SHARE_READ | FILE_SHARE_WRITE
			| (oldpc.fs_is_samba () ? 0 : FILE_SHARE_DELETE);
	ULONG flags = FILE_OPEN_FOR_BACKUP_INTENT
		      | (oldpc.is_rep_symlink () ? FILE_OPEN_REPARSE_POINT : 0);
	status = NtOpenFile (&fh, access,
			     oldpc.get_object_attr (attr, sec_none_nih),
			     &io, sharing, flags);
      }
      if (!NT_SUCCESS (status))
	{
	  debug_printf ("status %y", status);
	  if (status == STATUS_SHARING_VIOLATION
	      && cygwait (10L) != WAIT_SIGNALED)
	    {
	      /* Typical BLODA problem.  Some virus scanners check newly
		 generated files and while doing that disallow DELETE access.
		 That's really bad because it breaks applications which copy
		 files by creating a temporary filename and then rename the
		 temp filename to the target filename.  This renaming fails due
		 to the jealous virus scanner and the application fails to
		 create the target file.

		 This kludge tries to work around that by yielding until the
		 sharing violation goes away, or a signal arrived, or after
		 about a second, give or take. */
	      if (++retry_count < 40)
		{
		  yield ();
		  goto retry;
		}
	    }
	  else if (NT_TRANSACTIONAL_ERROR (status) && trans)
	    {
	      /* If NtOpenFile fails due to transactional problems, stop
		 transaction and go ahead without. */
	      stop_transaction (status, old_trans, trans);
	      debug_printf ("Transaction failure.  Retry open.");
	      goto retry;
	    }
	  __seterrno_from_nt_status (status);
	  __leave;
	}

      /* Renaming a dir to another, existing dir fails always, even if
	 ReplaceIfExists is set to TRUE and the existing dir is empty.  So
	 we have to remove the destination dir first.  This also covers the
	 case that the destination directory is not empty.  In that case,
	 unlink_nt returns with STATUS_DIRECTORY_NOT_EMPTY. */
      if (dstpc->isdir ())
	{
	  status = unlink_nt (*dstpc);
	  if (!NT_SUCCESS (status))
	    {
	      __seterrno_from_nt_status (status);
	      __leave;
	    }
	}
      /* You can't copy a file if the destination exists and has the R/O
	 attribute set.  Remove the R/O attribute first.  But first check
	 if a removepc exists.  If so, dstpc points to a non-existing file
	 due to a mangled suffix. */
      else if (!removepc && dstpc->has_attribute (FILE_ATTRIBUTE_READONLY))
	{
	  status = NtOpenFile (&nfh, FILE_WRITE_ATTRIBUTES,
			       dstpc->get_object_attr (attr, sec_none_nih),
			       &io, FILE_SHARE_VALID_FLAGS,
			       FILE_OPEN_FOR_BACKUP_INTENT
			       | (dstpc->is_rep_symlink ()
				  ? FILE_OPEN_REPARSE_POINT : 0));
	  if (!NT_SUCCESS (status))
	    {
	      __seterrno_from_nt_status (status);
	      __leave;
	    }
	  status = NtSetAttributesFile (nfh, dstpc->file_attributes ()
					     & ~FILE_ATTRIBUTE_READONLY);
	  NtClose (nfh);
	  if (!NT_SUCCESS (status))
	    {
	      __seterrno_from_nt_status (status);
	      __leave;
	    }
	}

      /* SUSv3: If the old argument and the new argument resolve to the same
	 existing file, rename() shall return successfully and perform no
	 other action.
	 The test tries to be as quick as possible.  Due to the above cross
	 device check we already know both files are on the same device.  So
	 it just tests if oldpath has more than 1 hardlink, then it opens
	 newpath and tests for identical file ids.  If so, oldpath and newpath
	 refer to the same file. */
      if ((removepc || dstpc->exists ())
	  && !oldpc.isdir ()
	  && NT_SUCCESS (NtQueryInformationFile (fh, &io, &ofsi, sizeof ofsi,
						 FileStandardInformation))
	  && ofsi.NumberOfLinks > 1
	  && NT_SUCCESS (NtOpenFile (&nfh, READ_CONTROL,
		     (removepc ?: dstpc)->get_object_attr (attr, sec_none_nih),
		     &io, FILE_SHARE_VALID_FLAGS,
		     FILE_OPEN_FOR_BACKUP_INTENT
		     | ((removepc ?: dstpc)->is_rep_symlink ()
			? FILE_OPEN_REPARSE_POINT : 0))))
	{
	  FILE_INTERNAL_INFORMATION ofii, nfii;

	  if (NT_SUCCESS (NtQueryInformationFile (fh, &io, &ofii, sizeof ofii,
						  FileInternalInformation))
	      && NT_SUCCESS (NtQueryInformationFile (nfh, &io, &nfii,
						     sizeof nfii,
						     FileInternalInformation))
	      && ofii.IndexNumber.QuadPart == nfii.IndexNumber.QuadPart)
	    {
	      debug_printf ("%s and %s are the same file", oldpath, newpath);
	      NtClose (nfh);
	      res = 0;
	      __leave;
	    }
	  NtClose (nfh);
	}
      /* Create FILE_RENAME_INFORMATION struct.  Using a tmp_pathbuf area
	 allows for paths of up to 32757 chars.  This test is just for
	 paranoia's sake. */
      if (dstpc->get_nt_native_path ()->Length
	  > NT_MAX_PATH * sizeof (WCHAR) - sizeof (FILE_RENAME_INFORMATION))
	{
	  debug_printf ("target filename too long");
	  set_errno (EINVAL);
	  __leave;
	}
      pfri = (PFILE_RENAME_INFORMATION) tp.w_get ();
      pfri->ReplaceIfExists = !noreplace;
      pfri->RootDirectory = NULL;
      pfri->FileNameLength = dstpc->get_nt_native_path ()->Length;
      memcpy (&pfri->FileName,  dstpc->get_nt_native_path ()->Buffer,
	      pfri->FileNameLength);
      /* If dstpc points to an existing file and RENAME_NOREPLACE has
	 been specified, then we should get NT error
	 STATUS_OBJECT_NAME_COLLISION ==> Win32 error
	 ERROR_ALREADY_EXISTS ==> Cygwin error EEXIST. */
      status = NtSetInformationFile (fh, &io, pfri,
				     sizeof *pfri + pfri->FileNameLength,
				     FileRenameInformation);
      /* This happens if the access rights don't allow deleting the destination.
	 Even if the handle to the original file is opened with BACKUP
	 and/or RECOVERY, these flags don't apply to the destination of the
	 rename operation.  So, a privileged user can't rename a file to an
	 existing file, if the permissions of the existing file aren't right.
	 Like directories, we have to handle this separately by removing the
	 destination before renaming. */
      if (status == STATUS_ACCESS_DENIED && dstpc->exists ()
	  && !dstpc->isdir ())
	{
	  bool need_open = false;

	  if ((dstpc->fs_flags () & FILE_SUPPORTS_TRANSACTIONS) && !trans)
	    {
	      /* As mentioned earlier, opening the file must be part of the
		 transaction.  Therefore we have to reopen the file here if the
		 transaction hasn't been started already.  Unfortunately we
		 can't use the NT "reopen file from existing handle" feature.
		 In that case NtOpenFile returns STATUS_TRANSACTIONAL_CONFLICT.
		 We *have* to close the handle to the file first, *then* we can
		 re-open it.  Fortunately nothing has happened yet, so the
		 atomicity of the rename functionality is not spoiled. */
	      NtClose (fh);
	      start_transaction (old_trans, trans);
	      need_open = true;
	    }
	  while (true)
	    {
	      status = STATUS_SUCCESS;
	      if (need_open)
		status = NtOpenFile (&fh, DELETE,
				     oldpc.get_object_attr (attr, sec_none_nih),
				     &io, FILE_SHARE_VALID_FLAGS,
				     FILE_OPEN_FOR_BACKUP_INTENT
				     | (oldpc.is_rep_symlink ()
					? FILE_OPEN_REPARSE_POINT : 0));
	      if (NT_SUCCESS (status))
		{
		  status = unlink_nt (*dstpc);
		  if (NT_SUCCESS (status))
		    break;
		}
	      if (!NT_TRANSACTIONAL_ERROR (status) || !trans)
		break;
	      /* If NtOpenFile or unlink_nt fail due to transactional problems,
		 stop transaction and retry without. */
	      NtClose (fh);
	      stop_transaction (status, old_trans, trans);
	      debug_printf ("Transaction failure %y.  Retry open.", status);
	    }
	  if (NT_SUCCESS (status))
	    status = NtSetInformationFile (fh, &io, pfri,
					   sizeof *pfri + pfri->FileNameLength,
					   FileRenameInformation);
	}
      if (NT_SUCCESS (status))
	{
	  if (removepc)
	    unlink_nt (*removepc);
	  res = 0;
	}
      else
	__seterrno_from_nt_status (status);
    }
  __except (EFAULT)
    {
      res = -1;
    }
  __endtry
  if (fh)
    NtClose (fh);
  /* Stop transaction if we started one. */
  if (trans)
    stop_transaction (status, old_trans, trans);
  if (get_errno () != EFAULT)
    syscall_printf ("%R = rename(%s, %s)", res, oldpath, newpath);
  return res;
}

extern "C" int
rename (const char *oldpath, const char *newpath)
{
  return rename2 (oldpath, newpath, 0);
}

extern "C" int
system (const char *cmdstring)
{
  pthread_testcancel ();

  if (cmdstring == NULL)
    return 1;

  int res = -1;
  const char* command[4];

  __try
    {
      command[0] = "sh";
      command[1] = "-c";
      command[2] = cmdstring;
      command[3] = (const char *) NULL;

      if ((res = spawnvp (_P_SYSTEM, "/bin/sh", command)) == -1)
	{
	  // when exec fails, return value should be as if shell
	  // executed exit (127)
	  res = 127;
	}
    }
  __except (EFAULT) {}
  __endtry
  return res;
}

extern "C" int
setdtablesize (int size)
{
  if (size < 0)
    {
      set_errno (EINVAL);
      return -1;
    }

  if (size <= (int) cygheap->fdtab.size
      || cygheap->fdtab.extend (size - cygheap->fdtab.size, OPEN_MAX_MAX))
    return 0;

  return -1;
}

extern "C" int
getdtablesize ()
{
  return cygheap->fdtab.size;
}

extern "C" int
getpagesize ()
{
  return (size_t) wincap.allocation_granularity ();
}

/* FIXME: not all values are correct... */
extern "C" long int
fpathconf (int fd, int v)
{
  cygheap_fdget cfd (fd);
  if (cfd < 0)
    return -1;
  return cfd->fpathconf (v);
}

extern "C" long int
pathconf (const char *file, int v)
{
  fhandler_base *fh = NULL;
  long ret = -1;

  __try
    {
      if (!*file)
	{
	  set_errno (ENOENT);
	  return -1;
	}
      if (!(fh = build_fh_name (file, PC_SYM_FOLLOW, stat_suffixes)))
	return -1;
      if (!fh->exists ())
	set_errno (ENOENT);
      else
	ret = fh->fpathconf (v);
    }
  __except (EFAULT) {}
  __endtry
  delete fh;
  return ret;
}

extern "C" int
ttyname_r (int fd, char *buf, size_t buflen)
{
  int ret = 0;

  __try
    {
      cygheap_fdget cfd (fd, true);
      if (cfd < 0)
	ret = EBADF;
      else if (!cfd->is_tty ())
	ret = ENOTTY;
      else if (buflen < strlen (cfd->ttyname ()) + 1)
	ret = ERANGE;
      else
	strcpy (buf, cfd->ttyname ());
      debug_printf ("returning %d tty: %s", ret, ret ? "NULL" : buf);
    }
  __except (NO_ERROR)
    {
      ret = EFAULT;
    }
  __endtry
  return ret;
}

extern "C" char *
ttyname (int fd)
{
  static char name[TTY_NAME_MAX];
  int ret = ttyname_r (fd, name, TTY_NAME_MAX);
  if (ret)
    {
      set_errno (ret);
      return NULL;
    }
  return name;
}

extern "C" char *
ctermid (char *str)
{
  if (str == NULL)
    str = _my_tls.locals.ttybuf;
  if (myself->ctty < 0)
    strcpy (str, "no tty");
  else
    {
      device d;
      d.parse (myself->ctty);
      strcpy (str, d.name ());
    }
  return str;
}

/* Tells stdio if it should do the cr/lf conversion for this file */
extern "C" int
_cygwin_istext_for_stdio (int fd)
{
  if (CYGWIN_VERSION_OLD_STDIO_CRLF_HANDLING)
    {
      syscall_printf ("fd %d: old API", fd);
      return 0; /* we do it for old apps, due to getc/putc macros */
    }

  cygheap_fdget cfd (fd, false, false);
  if (cfd < 0)
    {
      syscall_printf ("fd %d: not open", fd);
      return 0;
    }

#if 0
  if (cfd->get_device () != FH_FS)
    {
      syscall_printf ("fd not disk file.  Defaulting to binary.");
      return 0;
    }
#endif

  if (cfd->wbinary () || cfd->rbinary ())
    {
      syscall_printf ("fd %d: opened as binary", fd);
      return 0;
    }

  syscall_printf ("fd %d: defaulting to text", fd);
  return 1;
}

/* internal newlib function */
extern "C" int _fwalk (struct _reent *ptr, int (*function) (FILE *));

static int
setmode_helper (FILE *f)
{
  if (fileno (f) != _my_tls.locals.setmode_file)
    {
      syscall_printf ("improbable, but %d != %d", fileno (f), _my_tls.locals.setmode_file);
      return 0;
    }
  syscall_printf ("file was %s now %s", f->_flags & __SCLE ? "text" : "binary",
		  _my_tls.locals.setmode_mode & O_TEXT ? "text" : "binary");
  if (_my_tls.locals.setmode_mode & O_TEXT)
    f->_flags |= __SCLE;
  else
    f->_flags &= ~__SCLE;
  return 0;
}

extern "C" int
getmode (int fd)
{
  cygheap_fdget cfd (fd);
  if (cfd < 0)
    return -1;

  return cfd->get_flags () & (O_BINARY | O_TEXT);
}

/* Set a file descriptor into text or binary mode, returning the
   previous mode.  */

extern "C" int
_setmode (int fd, int mode)
{
  cygheap_fdget cfd (fd);
  if (cfd < 0)
    return -1;
  if (mode != O_BINARY  && mode != O_TEXT && mode != 0)
    {
      set_errno (EINVAL);
      return -1;
    }

  /* Note that we have no way to indicate the case that writes are
     binary but not reads, or vice-versa.  These cases can arise when
     using the tty or console interface.  People using those
     interfaces should not use setmode.  */

  int res;
  if (cfd->wbinary () && cfd->rbinary ())
    res = O_BINARY;
  else if (cfd->wbinset () && cfd->rbinset ())
    res = O_TEXT;	/* Specifically set O_TEXT */
  else
    res = 0;

  if (!mode)
    cfd->reset_to_open_binmode ();
  else
    cfd->set_flags ((cfd->get_flags () & ~(O_TEXT | O_BINARY)) | mode);

  syscall_printf ("(%d<%S>, %p) returning %s", fd,
		  cfd->pc.get_nt_native_path (), mode,
		  res & O_TEXT ? "text" : "binary");
  return res;
}

extern "C" int
cygwin_setmode (int fd, int mode)
{
  int res = _setmode (fd, mode);
  if (res != -1)
    {
      _my_tls.locals.setmode_file = fd;
      if (_cygwin_istext_for_stdio (fd))
	_my_tls.locals.setmode_mode = O_TEXT;
      else
	_my_tls.locals.setmode_mode = O_BINARY;
      _fwalk (_GLOBAL_REENT, setmode_helper);
    }
  return res;
}

extern "C" int
posix_fadvise (int fd, off_t offset, off_t len, int advice)
{
  int res = -1;
  cygheap_fdget cfd (fd);
  if (cfd >= 0)
    res = cfd->fadvise (offset, len, advice);
  else
    res = EBADF;
  syscall_printf ("%R = posix_fadvice(%d, %D, %D, %d)",
		  res, fd, offset, len, advice);
  return res;
}

extern "C" int
posix_fallocate (int fd, off_t offset, off_t len)
{
  int res = 0;
  if (offset < 0 || len == 0)
    res = EINVAL;
  else
    {
      cygheap_fdget cfd (fd);
      if (cfd >= 0)
	res = cfd->ftruncate (offset + len, false);
      else
	res = EBADF;
    }
  syscall_printf ("%R = posix_fallocate(%d, %D, %D)", res, fd, offset, len);
  return res;
}

extern "C" int
ftruncate64 (int fd, off_t length)
{
  int res = -1;
  cygheap_fdget cfd (fd);
  if (cfd >= 0)
    {
      res = cfd->ftruncate (length, true);
      if (res)
	{
	  set_errno (res);
	  res = -1;
	}
    }
  else
    set_errno (EBADF);
  syscall_printf ("%R = ftruncate(%d, %D)", res, fd, length);
  return res;
}

#ifdef __i386__
/* ftruncate: P96 5.6.7.1 */
extern "C" int
ftruncate (int fd, _off_t length)
{
  return ftruncate64 (fd, (off_t)length);
}
#else
EXPORT_ALIAS (ftruncate64, ftruncate)
#endif

/* truncate: Provided by SVR4 and 4.3+BSD.  Not part of POSIX.1 or XPG3 */
extern "C" int
truncate64 (const char *pathname, off_t length)
{
  int fd;
  int res = -1;

  fd = open (pathname, O_RDWR);

  if (fd != -1)
    {
      res = ftruncate64 (fd, length);
      close (fd);
    }
  syscall_printf ("%R = truncate(%s, %D)", res, pathname, length);

  return res;
}

#ifdef __i386__
/* truncate: Provided by SVR4 and 4.3+BSD.  Not part of POSIX.1 or XPG3 */
extern "C" int
truncate (const char *pathname, _off_t length)
{
  return truncate64 (pathname, (off_t)length);
}
#else
EXPORT_ALIAS (truncate64, truncate)
#endif

extern "C" long
_get_osfhandle (int fd)
{
  long res;

  cygheap_fdget cfd (fd);
  if (cfd >= 0)
    res = (long) cfd->get_handle ();
  else
    res = -1;

  syscall_printf ("%R = get_osfhandle(%d)", res, fd);
  return res;
}

extern "C" int
fstatvfs (int fd, struct statvfs *sfs)
{
  __try
    {
      cygheap_fdget cfd (fd);
      if (cfd < 0)
	__leave;
      return cfd->fstatvfs (sfs);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
statvfs (const char *name, struct statvfs *sfs)
{
  int res = -1;
  fhandler_base *fh = NULL;

  __try
    {
      if (!(fh = build_fh_name (name, PC_SYM_FOLLOW, stat_suffixes)))
	__leave;

      if (fh->error ())
	{
	  debug_printf ("got %d error from build_fh_name", fh->error ());
	  set_errno (fh->error ());
	}
      else if (fh->exists ())
	{
	  debug_printf ("(%s, %p), file_attributes %d", name, sfs, (DWORD) *fh);
	  res = fh->fstatvfs (sfs);
	}
      else
	set_errno (ENOENT);

    }
  __except (EFAULT) {}
  __endtry
  delete fh;
  if (get_errno () != EFAULT)
    syscall_printf ("%R = statvfs(%s,%p)", res, name, sfs);
  return res;
}

extern "C" int
fstatfs (int fd, struct statfs *sfs)
{
  struct statvfs vfs;
  int ret = fstatvfs (fd, &vfs);
  if (!ret)
    {
      sfs->f_type = vfs.f_flag;
      sfs->f_bsize = vfs.f_bsize;
      sfs->f_blocks = vfs.f_blocks;
      sfs->f_bavail = vfs.f_bavail;
      sfs->f_bfree = vfs.f_bfree;
      sfs->f_files = -1;
      sfs->f_ffree = -1;
      sfs->f_fsid = vfs.f_fsid;
      sfs->f_namelen = vfs.f_namemax;
    }
  return ret;
}

extern "C" int
statfs (const char *fname, struct statfs *sfs)
{
  struct statvfs vfs;
  int ret = statvfs (fname, &vfs);
  if (!ret)
    {
      sfs->f_type = vfs.f_flag;
      sfs->f_bsize = vfs.f_bsize;
      sfs->f_blocks = vfs.f_blocks;
      sfs->f_bavail = vfs.f_bavail;
      sfs->f_bfree = vfs.f_bfree;
      sfs->f_files = -1;
      sfs->f_ffree = -1;
      sfs->f_fsid = vfs.f_fsid;
      sfs->f_namelen = vfs.f_namemax;
    }
  return ret;
}

/* setpgid: POSIX 4.3.3.1 */
extern "C" int
setpgid (pid_t pid, pid_t pgid)
{
  int res = -1;
  if (pid == 0)
    pid = getpid ();
  if (pgid == 0)
    pgid = pid;

  if (pgid < 0)
    set_errno (EINVAL);
  else
    {
      pinfo p (pid, PID_MAP_RW);
      if (!p)
	set_errno (ESRCH);
      else if (p->pgid == pgid)
	res = 0;
      /* A process may only change the process group of itself and its children */
      else if (p != myself && p->ppid != myself->pid)
	set_errno (EPERM);
      else
	{
	  p->pgid = pgid;
	  if (p->pid != p->pgid)
	    p->set_has_pgid_children (0);
	  res = 0;
	}
    }

  syscall_printf ("pid %d, pgid %d, res %d", pid, pgid, res);
  return res;
}

extern "C" pid_t
getpgid (pid_t pid)
{
  if (pid == 0)
    pid = getpid ();

  pinfo p (pid);
  if (!p)
    {
      set_errno (ESRCH);
      return -1;
    }
  return p->pgid;
}

extern "C" int
setpgrp (void)
{
  return setpgid (0, 0);
}

extern "C" pid_t
getpgrp (void)
{
  return getpgid (0);
}

extern "C" char *
ptsname (int fd)
{
  static char buf[TTY_NAME_MAX];
  return ptsname_r (fd, buf, sizeof (buf)) == 0 ? buf : NULL;
}

extern "C" int
ptsname_r (int fd, char *buf, size_t buflen)
{
  if (!buf)
    {
      set_errno (EINVAL);
      return EINVAL;
    }

  cygheap_fdget cfd (fd);
  if (cfd < 0)
    return 0;
  return cfd->ptsname_r (buf, buflen);
}

static int __stdcall
mknod_worker (const char *path, mode_t type, mode_t mode, _major_t major,
	      _minor_t minor)
{
  char buf[sizeof (":\\00000000:00000000:00000000") + PATH_MAX];
  sprintf (buf, ":\\%x:%x:%x", major, minor,
	   type | (mode & (S_IRWXU | S_IRWXG | S_IRWXO)));
  return symlink_worker (buf, path, true);
}

extern "C" int
mknod32 (const char *path, mode_t mode, dev_t dev)
{
  __try
    {
      if (!*path)
	{
	  set_errno (ENOENT);
	  __leave;
	}

      if (strlen (path) >= PATH_MAX)
	__leave;

      path_conv w32path (path, PC_SYM_NOFOLLOW);
      if (w32path.exists ())
	{
	  set_errno (EEXIST);
	  __leave;
	}

      mode_t type = mode & S_IFMT;
      _major_t major = _major (dev);
      _minor_t minor = _minor (dev);
      switch (type)
	{
	case S_IFCHR:
	case S_IFBLK:
	  break;

	case S_IFIFO:
	  major = _major (FH_FIFO);
	  minor = _minor (FH_FIFO);
	  break;

	case 0:
	case S_IFREG:
	  {
	    int fd = open (path, O_CREAT, mode);
	    if (fd < 0)
	      __leave;
	    close (fd);
	    return 0;
	  }

	default:
	  set_errno (EINVAL);
	  __leave;
	}

      return mknod_worker (w32path.get_win32 (), type, mode, major, minor);
    }
  __except (EFAULT)
  __endtry
  return -1;
}

extern "C" int
mknod (const char *_path, mode_t mode, __dev16_t dev)
{
  return mknod32 (_path, mode, (dev_t) dev);
}

extern "C" int
mkfifo (const char *path, mode_t mode)
{
  return mknod32 (path, (mode & ~S_IFMT) | S_IFIFO, 0);
}

/* seteuid: standards? */
extern "C" int
seteuid32 (uid_t uid)
{
  debug_printf ("uid: %u myself->uid: %u myself->gid: %u",
		uid, myself->uid, myself->gid);

  /* Same uid as we're just running under is usually a no-op.

     Except we have an external token which is a restricted token.  Or,
     the external token is NULL, but the current impersonation token is
     a restricted token.  This allows to restrict user rights temporarily
     like this:

       cygwin_internal(CW_SET_EXTERNAL_TOKEN, restricted_token,
		       CW_TOKEN_RESTRICTED);
       setuid (getuid ());
       [...do stuff with restricted rights...]
       cygwin_internal(CW_SET_EXTERNAL_TOKEN, INVALID_HANDLE_VALUE,
		       CW_TOKEN_RESTRICTED);
       setuid (getuid ());

    Note that using the current uid is a requirement!  We have restricted
    tokens galore (UAC), so this is really just a special case to restrict
    your own processes to lesser rights. */
  bool request_restricted_uid_switch = (uid == myself->uid
      && cygheap->user.ext_token_is_restricted);
  if (uid == myself->uid && !cygheap->user.groups.ischanged
      && !request_restricted_uid_switch)
    {
      debug_printf ("Nothing happens");
      return 0;
    }

  cygsid usersid;
  user_groups &groups = cygheap->user.groups;
  HANDLE new_token = INVALID_HANDLE_VALUE;
  struct passwd * pw_new;
  bool token_is_internal, issamesid = false;

  pw_new = internal_getpwuid (uid);
  if (!usersid.getfrompw (pw_new))
    {
      set_errno (EINVAL);
      return -1;
    }

  cygheap->user.deimpersonate ();

  /* Verify if the process token is suitable. */
  /* First of all, skip all checks if a switch to a restricted token has been
     requested, or if trying to switch back from it. */
  if (request_restricted_uid_switch)
    {
      if (cygheap->user.external_token != NO_IMPERSONATION)
	{
	  debug_printf ("Switch to restricted token");
	  new_token = cygheap->user.external_token;
	}
      else
	{
	  debug_printf ("Switch back from restricted token");
	  new_token = hProcToken;
	  cygheap->user.ext_token_is_restricted = false;
	}
    }
  /* TODO, CV 2008-11-25: The check against saved_sid is a kludge and a
     shortcut.  We must check if it's really feasible in the long run.
     The reason to add this shortcut is this:  sshd switches back to the
     privileged user running sshd at least twice in the process of
     authentication.  It calls seteuid first, then setegid.  Due to this
     order, the setgroups group list is still active when calling seteuid
     and verify_token treats the original token of the privileged user as
     insufficient.  This in turn results in creating a new user token for
     the privileged user instead of using the orignal token.  This can have
     unfortunate side effects.  The created token has different group
     memberships, different user rights, and misses possible network
     credentials.
     Therefore we try this shortcut now.  When switching back to the
     privileged user, we probably always want a correct (aka original)
     user token for this privileged user, not only in sshd. */
  else if ((uid == cygheap->user.saved_uid
	   && usersid == cygheap->user.saved_sid ())
	   || verify_token (hProcToken, usersid, groups))
    new_token = hProcToken;
  /* Verify if the external token is suitable */
  else if (cygheap->user.external_token != NO_IMPERSONATION
	   && verify_token (cygheap->user.external_token, usersid, groups))
    new_token = cygheap->user.external_token;
  /* Verify if the current token (internal or former external) is suitable */
  else if (cygheap->user.curr_primary_token != NO_IMPERSONATION
	   && cygheap->user.curr_primary_token != cygheap->user.external_token
	   && verify_token (cygheap->user.curr_primary_token, usersid, groups,
			    &token_is_internal))
    new_token = cygheap->user.curr_primary_token;
  /* Verify if the internal token is suitable */
  else if (cygheap->user.internal_token != NO_IMPERSONATION
	   && cygheap->user.internal_token != cygheap->user.curr_primary_token
	   && verify_token (cygheap->user.internal_token, usersid, groups,
			    &token_is_internal))
    new_token = cygheap->user.internal_token;

  debug_printf ("Found token %p", new_token);

  /* If no impersonation token is available, try to authenticate using
     LSA private data stored password, LSA authentication using our own
     LSA module, or, as last chance, NtCreateToken. */
  if (new_token == INVALID_HANDLE_VALUE)
    {
      new_token = lsaprivkeyauth (pw_new);
      if (new_token)
	{
	  /* We have to verify this token since settings in /etc/group
	     might render it unusable im terms of group membership. */
	  if (!verify_token (new_token, usersid, groups))
	    {
	      CloseHandle (new_token);
	      new_token = NULL;
	    }
	}
      if (!new_token)
	{
	  debug_printf ("lsaprivkeyauth failed, try lsaauth.");
	  if (!(new_token = lsaauth (usersid, groups)))
	    {
	      debug_printf ("lsaauth failed, try create_token.");
	      new_token = create_token (usersid, groups);
	      if (new_token == INVALID_HANDLE_VALUE)
		{
		  debug_printf ("create_token failed, bail out of here");
		  cygheap->user.reimpersonate ();
		  return -1;
		}
	    }
	}

      /* Keep at most one internal token */
      if (cygheap->user.internal_token != NO_IMPERSONATION)
	CloseHandle (cygheap->user.internal_token);
      cygheap->user.internal_token = new_token;
    }

  if (new_token != hProcToken)
    {
      NTSTATUS status;

      if (!request_restricted_uid_switch)
	load_user_profile (new_token, pw_new, usersid);

      /* Try setting owner to same value as user. */
      status = NtSetInformationToken (new_token, TokenOwner,
				      &usersid, sizeof usersid);
      if (!NT_SUCCESS (status))
	debug_printf ("NtSetInformationToken (user.token, TokenOwner), %y",
		      status);
      /* Try setting primary group in token to current group */
      status = NtSetInformationToken (new_token, TokenPrimaryGroup,
				      &groups.pgsid, sizeof (cygsid));
      if (!NT_SUCCESS (status))
	debug_printf ("NtSetInformationToken (user.token, TokenPrimaryGroup),"
		      "%y", status);
      /* Try setting default DACL */
      PACL dacl_buf = (PACL) alloca (MAX_DACL_LEN (5));
      if (sec_acl (dacl_buf, true, true, usersid))
	{
	  TOKEN_DEFAULT_DACL tdacl = { dacl_buf };
	  status = NtSetInformationToken (new_token, TokenDefaultDacl,
					  &tdacl, sizeof (tdacl));
	  if (!NT_SUCCESS (status))
	    debug_printf ("NtSetInformationToken (TokenDefaultDacl), %y",
			  status);
	}
    }

  issamesid = (usersid == cygheap->user.sid ());
  cygheap->user.set_sid (usersid);
  cygheap->user.curr_primary_token = new_token == hProcToken ? NO_IMPERSONATION
							: new_token;
  cygheap->user.curr_token_is_restricted = false;
  cygheap->user.setuid_to_restricted = false;
  if (cygheap->user.curr_imp_token != NO_IMPERSONATION)
    {
      CloseHandle (cygheap->user.curr_imp_token);
      cygheap->user.curr_imp_token = NO_IMPERSONATION;
    }
  if (cygheap->user.curr_primary_token != NO_IMPERSONATION)
    {
      /* HANDLE_FLAG_INHERIT may be missing in external token.  */
      if (!SetHandleInformation (cygheap->user.curr_primary_token,
				 HANDLE_FLAG_INHERIT, HANDLE_FLAG_INHERIT)
	  || !DuplicateTokenEx (cygheap->user.curr_primary_token,
				MAXIMUM_ALLOWED, &sec_none,
				SecurityImpersonation, TokenImpersonation,
				&cygheap->user.curr_imp_token))
	{
	  __seterrno ();
	  cygheap->user.curr_primary_token = NO_IMPERSONATION;
	  return -1;
	}
      cygheap->user.curr_token_is_restricted = request_restricted_uid_switch;
      set_cygwin_privileges (cygheap->user.curr_primary_token);
      set_cygwin_privileges (cygheap->user.curr_imp_token);
    }
  if (!cygheap->user.reimpersonate ())
    {
      __seterrno ();
      return -1;
    }

  cygheap->user.set_name (pw_new->pw_name);
  myself->uid = uid;
  groups.ischanged = FALSE;
  if (!issamesid)
    /* Recreate and fill out the user shared region for a new user. */
    user_info::create (true);
  return 0;
}

#ifdef __i386__
extern "C" int
seteuid (__uid16_t uid)
{
  return seteuid32 (uid16touid32 (uid));
}
#else
EXPORT_ALIAS (seteuid32, seteuid)
#endif

/* setuid: POSIX 4.2.2.1 */
extern "C" int
setuid32 (uid_t uid)
{
  int ret = seteuid32 (uid);
  if (!ret)
    {
      cygheap->user.real_uid = myself->uid;
      /* If restricted token, forget original privileges on exec ().  */
      cygheap->user.setuid_to_restricted = cygheap->user.curr_token_is_restricted;
    }
  debug_printf ("real: %d, effective: %d", cygheap->user.real_uid, myself->uid);
  return ret;
}

#ifdef __i386__
extern "C" int
setuid (__uid16_t uid)
{
  return setuid32 (uid16touid32 (uid));
}
#else
EXPORT_ALIAS (setuid32, setuid)
#endif

extern "C" int
setreuid32 (uid_t ruid, uid_t euid)
{
  int ret = 0;
  bool tried = false;
  uid_t old_euid = myself->uid;

  if (ruid != ILLEGAL_UID && cygheap->user.real_uid != ruid && euid != ruid)
    tried = !(ret = seteuid32 (ruid));
  if (!ret && euid != ILLEGAL_UID)
    ret = seteuid32 (euid);
  if (tried && (ret || euid == ILLEGAL_UID) && seteuid32 (old_euid))
    system_printf ("Cannot restore original euid %u", old_euid);
  if (!ret && ruid != ILLEGAL_UID)
    cygheap->user.real_uid = ruid;
  debug_printf ("real: %u, effective: %u", cygheap->user.real_uid, myself->uid);
  return ret;
}

#ifdef __i386__
extern "C" int
setreuid (__uid16_t ruid, __uid16_t euid)
{
  return setreuid32 (uid16touid32 (ruid), uid16touid32 (euid));
}
#else
EXPORT_ALIAS (setreuid32, setreuid)
#endif

/* setegid: from System V.  */
extern "C" int
setegid32 (gid_t gid)
{
  debug_printf ("new egid: %u current: %u", gid, myself->gid);

  if (gid == myself->gid)
    {
      myself->gid = gid;
      return 0;
    }

  NTSTATUS status;
  user_groups * groups = &cygheap->user.groups;
  cygsid gsid;
  struct group * gr = internal_getgrgid (gid);

  if (!gsid.getfromgr (gr))
    {
      set_errno (EINVAL);
      return -1;
    }
  myself->gid = gid;

  groups->update_pgrp (gsid);
  if (cygheap->user.issetuid ())
    {
      /* If impersonated, update impersonation token... */
      status = NtSetInformationToken (cygheap->user.primary_token (),
				      TokenPrimaryGroup, &gsid, sizeof gsid);
      if (!NT_SUCCESS (status))
	debug_printf ("NtSetInformationToken (primary_token, "
		      "TokenPrimaryGroup), %y", status);
      status = NtSetInformationToken (cygheap->user.imp_token (),
				      TokenPrimaryGroup, &gsid, sizeof gsid);
      if (!NT_SUCCESS (status))
	debug_printf ("NtSetInformationToken (token, TokenPrimaryGroup), %y",
		      status);
    }
  cygheap->user.deimpersonate ();
  status = NtSetInformationToken (hProcToken, TokenPrimaryGroup,
				  &gsid, sizeof gsid);
  if (!NT_SUCCESS (status))
    debug_printf ("NtSetInformationToken (hProcToken, TokenPrimaryGroup), %y",
		  status);
  clear_procimptoken ();
  cygheap->user.reimpersonate ();
  return 0;
}

#ifdef __i386__
extern "C" int
setegid (__gid16_t gid)
{
  return setegid32 (gid16togid32 (gid));
}
#else
EXPORT_ALIAS (setegid32, setegid)
#endif

/* setgid: POSIX 4.2.2.1 */
extern "C" int
setgid32 (gid_t gid)
{
  int ret = setegid32 (gid);
  if (!ret)
    cygheap->user.real_gid = myself->gid;
  return ret;
}

#ifdef __i386__
extern "C" int
setgid (__gid16_t gid)
{
  int ret = setegid32 (gid16togid32 (gid));
  if (!ret)
    cygheap->user.real_gid = myself->gid;
  return ret;
}
#else
EXPORT_ALIAS (setgid32, setgid)
#endif

extern "C" int
setregid32 (gid_t rgid, gid_t egid)
{
  int ret = 0;
  bool tried = false;
  gid_t old_egid = myself->gid;

  if (rgid != ILLEGAL_GID && cygheap->user.real_gid != rgid && egid != rgid)
    tried = !(ret = setegid32 (rgid));
  if (!ret && egid != ILLEGAL_GID)
    ret = setegid32 (egid);
  if (tried && (ret || egid == ILLEGAL_GID) && setegid32 (old_egid))
    system_printf ("Cannot restore original egid %u", old_egid);
  if (!ret && rgid != ILLEGAL_GID)
    cygheap->user.real_gid = rgid;
  debug_printf ("real: %u, effective: %u", cygheap->user.real_gid, myself->gid);
  return ret;
}

#ifdef __i386__
extern "C" int
setregid (__gid16_t rgid, __gid16_t egid)
{
  return setregid32 (gid16togid32 (rgid), gid16togid32 (egid));
}
#else
EXPORT_ALIAS (setregid32, setregid)
#endif

/* chroot: privileged Unix system call.  */
/* FIXME: Not privileged here. How should this be done? */
extern "C" int
chroot (const char *newroot)
{
  path_conv path (newroot, PC_SYM_FOLLOW | PC_POSIX);

  int ret = -1;
  if (path.error)
    set_errno (path.error);
  else if (!path.exists ())
    set_errno (ENOENT);
  else if (!path.isdir ())
    set_errno (ENOTDIR);
  else if (path.isspecial ())
    set_errno (EPERM);
  else
    {
      getwinenv("PATH="); /* Save the native PATH */
      cygheap->root.set (path.get_posix (), path.get_win32 (),
			 !!path.objcaseinsensitive ());
      ret = 0;
    }

  syscall_printf ("%R = chroot(%s)", ret, newroot ?: "NULL");
  return ret;
}

extern "C" int
creat (const char *path, mode_t mode)
{
  return open (path, O_WRONLY | O_CREAT | O_TRUNC, mode);
}

extern "C" void
__assertfail ()
{
  exit (99);
}

extern "C" int
vhangup ()
{
  set_errno (ENOSYS);
  return -1;
}

extern "C" int
setpriority (int which, id_t who, int value)
{
  DWORD prio = nice_to_winprio (value);
  int error = 0;

  switch (which)
    {
    case PRIO_PROCESS:
      if (!who)
	who = myself->pid;
      if ((pid_t) who == myself->pid)
	{
	  if (!SetPriorityClass (GetCurrentProcess (), prio))
	    {
	      set_errno (EACCES);
	      return -1;
	    }
	  myself->nice = value;
	  debug_printf ("Set nice to %d", myself->nice);
	  return 0;
	}
      break;
    case PRIO_PGRP:
      if (!who)
	who = myself->pgid;
      break;
    case PRIO_USER:
      if (!who)
	who = myself->uid;
      break;
    default:
      set_errno (EINVAL);
      return -1;
    }
  winpids pids ((DWORD) PID_MAP_RW);
  for (DWORD i = 0; i < pids.npids; ++i)
    {
      _pinfo *p = pids[i];
      if (p)
	{
	  switch (which)
	    {
	    case PRIO_PROCESS:
	      if ((pid_t) who != p->pid)
		continue;
	      break;
	    case PRIO_PGRP:
	      if ((pid_t) who != p->pgid)
		continue;
	      break;
	    case PRIO_USER:
		if ((uid_t) who != p->uid)
		continue;
	      break;
	    }
	  HANDLE proc_h = OpenProcess (PROCESS_SET_INFORMATION, FALSE,
				       p->dwProcessId);
	  if (!proc_h)
	    error = EPERM;
	  else
	    {
	      if (!SetPriorityClass (proc_h, prio))
		error = EACCES;
	      else
		p->nice = value;
	      CloseHandle (proc_h);
	    }
	}
    }
  pids.reset ();
  if (error)
    {
      set_errno (error);
      return -1;
    }
  return 0;
}

extern "C" int
getpriority (int which, id_t who)
{
  int nice = NZERO * 2; /* Illegal value */

  switch (which)
    {
    case PRIO_PROCESS:
      if (!who)
	who = myself->pid;
      if ((pid_t) who == myself->pid)
	return myself->nice;
      break;
    case PRIO_PGRP:
      if (!who)
	who = myself->pgid;
      break;
    case PRIO_USER:
      if (!who)
	who = myself->uid;
      break;
    default:
      set_errno (EINVAL);
      return -1;
    }
  winpids pids ((DWORD) 0);
  for (DWORD i = 0; i < pids.npids; ++i)
    {
      _pinfo *p = pids[i];
      if (p)
	switch (which)
	  {
	  case PRIO_PROCESS:
	    if ((pid_t) who == p->pid)
	      {
		nice = p->nice;
		goto out;
	      }
	    break;
	  case PRIO_PGRP:
	    if ((pid_t) who == p->pgid && p->nice < nice)
	      nice = p->nice;
	    break;
	  case PRIO_USER:
	    if ((uid_t) who == p->uid && p->nice < nice)
	      nice = p->nice;
	    break;
	  }
    }
out:
  pids.reset ();
  if (nice == NZERO * 2)
    {
      set_errno (ESRCH);
      return -1;
    }
  return nice;
}

extern "C" int
nice (int incr)
{
  return setpriority (PRIO_PROCESS, myself->pid, myself->nice + incr);
}

static void
locked_append (int fd, const void * buf, size_t size)
{
  struct flock lock_buffer = {F_WRLCK, SEEK_SET, 0, 0, 0};
  int count = 0;

  do
    if ((lock_buffer.l_start = lseek64 (fd, 0, SEEK_END)) != (off_t) -1
	&& fcntl64 (fd, F_SETLKW, &lock_buffer) != -1)
      {
	if (lseek64 (fd, 0, SEEK_END) != (off_t) -1)
	  write (fd, buf, size);
	lock_buffer.l_type = F_UNLCK;
	fcntl64 (fd, F_SETLK, &lock_buffer);
	break;
      }
  while (count++ < 1000
	 && (errno == EACCES || errno == EAGAIN)
	 && !usleep (1000));
}

extern "C" void
updwtmp (const char *wtmp_file, const struct utmp *ut)
{
  int fd;

  if ((fd = open (wtmp_file, O_WRONLY | O_BINARY, 0)) >= 0)
    {
      locked_append (fd, ut, sizeof *ut);
      close (fd);
    }
}

static int utmp_fd = -1;
static bool utmp_readonly = false;
static char *utmp_file = (char *) _PATH_UTMP;

static void
internal_setutent (bool force_readwrite)
{
  if (force_readwrite && utmp_readonly)
    endutent ();
  if (utmp_fd < 0)
    {
      utmp_fd = open (utmp_file, O_RDWR | O_BINARY);
      /* If open fails, we assume an unprivileged process (who?).  In this
	 case we try again for reading only unless the process calls
	 pututline() (==force_readwrite) in which case opening just fails. */
      if (utmp_fd < 0 && !force_readwrite)
	{
	  utmp_fd = open (utmp_file, O_RDONLY | O_BINARY);
	  if (utmp_fd >= 0)
	    utmp_readonly = true;
	}
    }
  else
    lseek (utmp_fd, 0, SEEK_SET);
}

extern "C" void
setutent ()
{
  internal_setutent (false);
}

extern "C" void
endutent ()
{
  if (utmp_fd >= 0)
    {
      close (utmp_fd);
      utmp_fd = -1;
      utmp_readonly = false;
    }
}

extern "C" int
utmpname (const char *file)
{
  __try
    {
      if (*file)
	{
	  endutent ();
	  utmp_file = strdup (file);
	  if (utmp_file)
	    {
	      debug_printf ("New UTMP file: %s", utmp_file);
	      return 0;
	    }
	}
    }
  __except (EFAULT) {}
  __endtry
  debug_printf ("Setting UTMP file failed");
  return -1;
}

EXPORT_ALIAS (utmpname, utmpxname)

/* Note: do not make NO_COPY */
static struct utmp utmp_data_buf[16];
static unsigned utix = 0;
#define nutdbuf (sizeof (utmp_data_buf) / sizeof (utmp_data_buf[0]))
#define utmp_data ({ \
  if (utix >= nutdbuf) \
    utix = 0; \
  utmp_data_buf + utix++; \
})

static struct utmpx *
copy_ut_to_utx (struct utmp *ut, struct utmpx *utx)
{
  if (!ut)
    return NULL;
  memcpy (utx, ut, sizeof *ut);
  utx->ut_tv.tv_sec = ut->ut_time;
  utx->ut_tv.tv_usec = 0;
  return utx;
}

extern "C" struct utmp *
getutent ()
{
  if (utmp_fd < 0)
    {
      internal_setutent (false);
      if (utmp_fd < 0)
	return NULL;
    }

  utmp *ut = utmp_data;
  if (read (utmp_fd, ut, sizeof *ut) != sizeof *ut)
    return NULL;
  return ut;
}

extern "C" struct utmp *
getutid (const struct utmp *id)
{
  __try
    {
      if (utmp_fd < 0)
	{
	  internal_setutent (false);
	  if (utmp_fd < 0)
	    __leave;
	}
      utmp *ut = utmp_data;
      while (read (utmp_fd, ut, sizeof *ut) == sizeof *ut)
	{
	  switch (id->ut_type)
	    {
	    case RUN_LVL:
	    case BOOT_TIME:
	    case OLD_TIME:
	    case NEW_TIME:
	      if (id->ut_type == ut->ut_type)
		return ut;
	      break;
	    case INIT_PROCESS:
	    case LOGIN_PROCESS:
	    case USER_PROCESS:
	    case DEAD_PROCESS:
	       if (strncmp (id->ut_id, ut->ut_id, UT_IDLEN) == 0)
		return ut;
	      break;
	    default:
	      break;
	    }
	}
    }
  __except (EFAULT) {}
  __endtry
  return NULL;
}

extern "C" struct utmp *
getutline (const struct utmp *line)
{
  __try
    {
      if (utmp_fd < 0)
	{
	  internal_setutent (false);
	  if (utmp_fd < 0)
	    __leave;
	}

      utmp *ut = utmp_data;
      while (read (utmp_fd, ut, sizeof *ut) == sizeof *ut)
	if ((ut->ut_type == LOGIN_PROCESS ||
	     ut->ut_type == USER_PROCESS) &&
	    !strncmp (ut->ut_line, line->ut_line, sizeof (ut->ut_line)))
	  return ut;
    }
  __except (EFAULT) {}
  __endtry
  return NULL;
}

extern "C" struct utmp *
pututline (const struct utmp *ut)
{
  __try
    {
      internal_setutent (true);
      if (utmp_fd < 0)
	{
	  debug_printf ("error: utmp_fd %d", utmp_fd);
	  __leave;
	}
      debug_printf ("ut->ut_type %d, ut->ut_pid %d, ut->ut_line '%s', ut->ut_id '%s'\n",
		    ut->ut_type, ut->ut_pid, ut->ut_line, ut->ut_id);
      debug_printf ("ut->ut_user '%s', ut->ut_host '%s'\n",
		    ut->ut_user, ut->ut_host);

      struct utmp *u;
      if ((u = getutid (ut)))
	{
	  lseek (utmp_fd, -sizeof *ut, SEEK_CUR);
	  write (utmp_fd, ut, sizeof *ut);
	}
      else
	locked_append (utmp_fd, ut, sizeof *ut);
      /* The documentation says to return a pointer to this which implies that
	 this has to be cast from a const.  That doesn't seem right but the
	 documentation seems pretty clear on this.  */
      return (struct utmp *) ut;
    }
  __except (EFAULT) {}
  __endtry
  return NULL;
}

extern "C" void
setutxent ()
{
  internal_setutent (false);
}

extern "C" void
endutxent ()
{
  endutent ();
}

extern "C" struct utmpx *
getutxent ()
{
  /* POSIX: Not required to be thread safe. */
  static struct utmpx utx;
  return copy_ut_to_utx (getutent (), &utx);
}

extern "C" struct utmpx *
getutxid (const struct utmpx *id)
{
  /* POSIX: Not required to be thread safe. */
  static struct utmpx utx;

  __try
    {
      ((struct utmpx *)id)->ut_time = id->ut_tv.tv_sec;
      return copy_ut_to_utx (getutid ((struct utmp *) id), &utx);
    }
  __except (EFAULT) {}
  __endtry
  return NULL;
}

extern "C" struct utmpx *
getutxline (const struct utmpx *line)
{
  /* POSIX: Not required to be thread safe. */
  static struct utmpx utx;

  __try
    {
      ((struct utmpx *)line)->ut_time = line->ut_tv.tv_sec;
      return copy_ut_to_utx (getutline ((struct utmp *) line), &utx);
    }
  __except (EFAULT) {}
  __endtry
  return NULL;
}

extern "C" struct utmpx *
pututxline (const struct utmpx *utmpx)
{
  /* POSIX: Not required to be thread safe. */
  static struct utmpx utx;

  __try
    {
      ((struct utmpx *)utmpx)->ut_time = utmpx->ut_tv.tv_sec;
      return copy_ut_to_utx (pututline ((struct utmp *) utmpx), &utx);
    }
  __except (EFAULT) {}
  __endtry
  return NULL;
}

extern "C" void
updwtmpx (const char *wtmpx_file, const struct utmpx *utmpx)
{
  ((struct utmpx *)utmpx)->ut_time = utmpx->ut_tv.tv_sec;
  updwtmp (wtmpx_file, (const struct utmp *) utmpx);
}

extern "C" long
gethostid (void)
{
  /* Fetch the globally unique MachineGuid value from
     HKLM/Software/Microsoft/Cryptography and hash it. */

  /* Caution: sizeof long might become > 4 when we go 64 bit, but gethostid
     is supposed to return a 32 bit value, despite the return type long.
     That's why hostid is *not* long here. */
  int32_t hostid = 0x40291372; /* Choose a nice start value */
  WCHAR wguid[38];

  reg_key key (HKEY_LOCAL_MACHINE,
	       KEY_READ | (wincap.is_wow64() ? KEY_WOW64_64KEY : 0),
	       L"SOFTWARE", L"Microsoft", L"Cryptography", NULL);
  key.get_string (L"MachineGuid", wguid, 38,
		  L"00000000-0000-0000-0000-000000000000");
  /* SDBM hash */
  for (PWCHAR wp = wguid; *wp; ++wp)
    hostid = *wp + (hostid << 6) + (hostid << 16) - hostid;
  debug_printf ("hostid %08y from MachineGuid %W", hostid, wguid);
  return (int32_t) hostid; /* Avoid sign extension. */
}

#define ETC_SHELLS "/etc/shells"
static int shell_index;
static struct __sFILE64 *shell_fp;

extern "C" char *
getusershell ()
{
  /* List of default shells if no /etc/shells exists, defined as on Linux.
     FIXME: SunOS has a far longer list, containing all shells which
     might be shipped with the OS.  Should we do the same for the Cygwin
     distro, adding bash, tcsh, ksh, pdksh and zsh?  */
  static const char *def_shells[] = {
    "/bin/sh",
    "/bin/csh",
    "/usr/bin/sh",
    "/usr/bin/csh",
    NULL
  };
  static char buf[PATH_MAX];
  int ch, buf_idx;

  if (!shell_fp && !(shell_fp = fopen64 (ETC_SHELLS, "rt")))
    {
      if (def_shells[shell_index])
	return strcpy (buf, def_shells[shell_index++]);
      return NULL;
    }
  /* Skip white space characters. */
  while ((ch = getc (shell_fp)) != EOF && isspace (ch))
    ;
  /* Get each non-whitespace character as part of the shell path as long as
     it fits in buf. */
  for (buf_idx = 0;
       ch != EOF && !isspace (ch) && buf_idx < (PATH_MAX - 1);
       buf_idx++, ch = getc (shell_fp))
    buf[buf_idx] = ch;
  /* Skip any trailing non-whitespace character not fitting in buf.  If the
     path is longer than PATH_MAX, it's invalid anyway. */
  while (ch != EOF && !isspace (ch))
    ch = getc (shell_fp);
  if (buf_idx)
    {
      buf[buf_idx] = '\0';
      return buf;
    }
  return NULL;
}

extern "C" void
setusershell ()
{
  if (shell_fp)
    fseek (shell_fp, 0L, SEEK_SET);
  shell_index = 0;
}

extern "C" void
endusershell ()
{
  if (shell_fp)
    {
      fclose (shell_fp);
      shell_fp = NULL;
    }
  shell_index = 0;
}

extern "C" void
flockfile (FILE *file)
{
  _flockfile (file);
}

extern "C" int
ftrylockfile (FILE *file)
{
  return _ftrylockfile (file);
}

extern "C" void
funlockfile (FILE *file)
{
  _funlockfile (file);
}

extern "C" FILE *
popen (const char *command, const char *in_type)
{
  const char *type = in_type;
  char fdopen_flags[3] = "\0\0";
  int pipe_flags = 0;

#define rw	fdopen_flags[0]
#define bintext	fdopen_flags[1]

  /* Sanity check.  GLibc allows any order and any number of repetition,
     as long as the string doesn't contradict itself.  We do the same here. */
  while (*type)
    {
      if (*type == 'r' || *type == 'w')
	{
	  if (rw && rw != *type)
	    break;
	  rw = *type++;
	}
      else if (*type == 'b' || *type == 't')
	{
	  if (bintext && bintext != *type)
	    break;
	  bintext = *type++;
	}
      else if (*type == 'e')
	{
	  pipe_flags = O_CLOEXEC;
	  ++type;
	}
      else
	break;
    }
  if ((rw != 'r' && rw != 'w') || (*type != '\0'))
    {
      set_errno (EINVAL);
      return NULL;
    }

  int fds[2];
  if (pipe2 (fds, pipe_flags) < 0)
    return NULL;

  int myix = rw == 'r' ? 0 : 1;

  lock_process now;
  FILE *fp = fdopen (fds[myix], fdopen_flags);
  if (fp)
    {
      /* If fds are in the range of stdin/stdout/stderr, move them
	 out of the way (possibly temporarily).  Otherwise, spawn_guts
	 will be confused.  We do this here rather than adding logic to
	 spawn_guts because spawn_guts is likely to be a more frequently
	 used routine and having stdin/stdout/stderr closed and reassigned
	 to pipe handles is an unlikely event. */
      int orig_fds[2] = {fds[0], fds[1]};
      for (int i = 0; i < 2; i++)
	if (fds[i] <= 2)
	  {
	    cygheap_fdnew newfd(3);
	    cygheap->fdtab.move_fd (fds[i], newfd);
	    fds[i] = newfd;
	  }

      int myfd = fds[myix];	/* myfd - convenience variable for manipulation
				   of the "parent" end of the pipe. */
      int stdchild = myix ^ 1;	/* stdchild denotes the index into fd for the
				   handle which will be redirected to
				   stdin/stdout */
      int __std[2];
      __std[myix] = -1;		/* -1 means don't pass this fd to the child
				   process */
      __std[stdchild] = fds[stdchild]; /* Do pass this as the std handle */

      const char *argv[4] =
	{
	  "/bin/sh",
	  "-c",
	  command,
	  NULL
	};

      /* With 'e' flag given, we have to revert the close-on-exec on the child
         end of the pipe.  Otherwise don't pass our end of the pipe to the
	 child process. */
      if (pipe_flags & O_CLOEXEC)
	fcntl64 (__std[stdchild], F_SETFD, 0);
      else
	fcntl64 (myfd, F_SETFD, FD_CLOEXEC);

      /* Also don't pass the file handle currently associated with stdin/stdout
	 to the child.  This function may actually fail if the stdchild fd
	 is closed.  But that's ok. */
      int stdchild_state = fcntl64 (stdchild, F_GETFD, 0);
      fcntl64 (stdchild, F_SETFD, stdchild_state | FD_CLOEXEC);

      /* Start a shell process to run the given command without forking. */
      pid_t pid = ch_spawn.worker ("/bin/sh", argv, cur_environ (), _P_NOWAIT,
				   __std[0], __std[1]);

      /* Reinstate the close-on-exec state */
      fcntl64 (stdchild, F_SETFD, stdchild_state);

      /* If pid >= 0 then spawn_guts succeeded.  */
      if (pid >= 0)
	{
	  close (fds[stdchild]);	/* Close the child end of the pipe. */
	  /* Move the fd back to its original slot if it has been moved since
	     we're always supposed to open the lowest numbered available fd
	     and, if fds[mix] != orig_fds[myix] then orig_fds[myix] is
	     presumably lower. */
	  if (fds[myix] != orig_fds[myix])
	    cygheap->fdtab.move_fd (fds[myix], myfd = orig_fds[myix]);
	  fhandler_pipe *fh = (fhandler_pipe *) cygheap->fdtab[myfd];
	  /* Flag that this handle is associated with popen. */
	  fh->set_popen_pid (pid);
	  return fp;
	}
    }

  /* If we reach here we've seen an error but the pipe handles are open.
     Close them and return NULL. */
  int save_errno = get_errno ();
  if (fp)
    {
      /* Must fclose fp to avoid memory leak. */
      fclose (fp);
      close (fds[myix ^ 1]);
    }
  else
    {
      close (fds[0]);
      close (fds[1]);
    }
  set_errno (save_errno);

#undef rw
#undef bintext

  return NULL;
}

int
pclose (FILE *fp)
{
  fhandler_pipe *fh = (fhandler_pipe *) cygheap->fdtab[fileno(fp)];

  if (fh->get_device () != FH_PIPEW && fh->get_device () != FH_PIPER)
    {
      set_errno (EBADF);
      return -1;
    }

  int pid = fh->get_popen_pid ();
  if (!pid)
    {
      set_errno (ECHILD);
      return -1;
    }

  if (fclose (fp))
    return -1;

  int status;
  while (1)
    if (waitpid (pid, &status, 0) == pid)
      break;
    else if (get_errno () == EINTR)
      continue;
    else
      return -1;

  return status;
}

/* Preliminary(?) implementation of the openat family of functions. */

static int
gen_full_path_at (char *path_ret, int dirfd, const char *pathname,
		  bool null_pathname_allowed = false)
{
  /* Set null_pathname_allowed to true to allow GLIBC compatible behaviour
     for NULL pathname.  Only used by futimesat. */
  if (!pathname && !null_pathname_allowed)
    {
      set_errno (EFAULT);
      return -1;
    }
  if (pathname)
    {
      if (!*pathname)
	{
	  set_errno (ENOENT);
	  return -1;
	}
      if (strlen (pathname) >= PATH_MAX)
	{
	  set_errno (ENAMETOOLONG);
	  return -1;
	}
    }
  if (pathname && isabspath (pathname))
    stpcpy (path_ret, pathname);
  else
    {
      char *p;

      if (dirfd == AT_FDCWD)
	{
	  cwdstuff::cwd_lock.acquire ();
	  p = stpcpy (path_ret, cygheap->cwd.get_posix ());
	  cwdstuff::cwd_lock.release ();
	}
      else
	{
	  cygheap_fdget cfd (dirfd);
	  if (cfd < 0)
	    return -1;
	  if (!cfd->pc.isdir ())
	    {
	      set_errno (ENOTDIR);
	      return -1;
	    }
	  p = stpcpy (path_ret, cfd->get_name ());
	}
      if (!p)
	{
	  set_errno (ENOTDIR);
	  return -1;
	}
      if (pathname)
	{
	  if (p[-1] != '/')
	    *p++ = '/';
	  stpcpy (p, pathname);
	}
    }
  return 0;
}

extern "C" int
openat (int dirfd, const char *pathname, int flags, ...)
{
  tmp_pathbuf tp;
  __try
    {
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;

      va_list ap;
      mode_t mode;

      va_start (ap, flags);
      mode = va_arg (ap, mode_t);
      va_end (ap);
      return open (path, flags, mode);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
faccessat (int dirfd, const char *pathname, int mode, int flags)
{
  tmp_pathbuf tp;
  int res = -1;

  __try
    {
      char *path = tp.c_get ();
      if (!gen_full_path_at (path, dirfd, pathname))
	{
	  if ((mode & ~(F_OK|R_OK|W_OK|X_OK))
	      || (flags & ~(AT_SYMLINK_NOFOLLOW|AT_EACCESS)))
	    set_errno (EINVAL);
	  else
	    {
	      fhandler_base *fh = build_fh_name (path,
						 (flags & AT_SYMLINK_NOFOLLOW
						  ? PC_SYM_NOFOLLOW
						  : PC_SYM_FOLLOW)
						 | PC_KEEP_HANDLE,
						 stat_suffixes);
	      if (fh)
		{
		  res = fh->fhaccess (mode, !!(flags & AT_EACCESS));
		  delete fh;
		}
	    }
	}
    }
  __except (EFAULT) {}
  __endtry
  debug_printf ("returning %d", res);
  return res;
}

extern "C" int
fchmodat (int dirfd, const char *pathname, mode_t mode, int flags)
{
  tmp_pathbuf tp;
  __try
    {
      if (flags)
	{
	  /* BSD has lchmod, but Linux does not.  POSIX says
	     AT_SYMLINK_NOFOLLOW is allowed to fail on symlinks; but Linux
	     blindly fails even for non-symlinks.  */
	  set_errno ((flags & ~AT_SYMLINK_NOFOLLOW) ? EINVAL : EOPNOTSUPP);
	  __leave;
	}
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;
      return chmod (path, mode);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
fchownat (int dirfd, const char *pathname, uid_t uid, gid_t gid, int flags)
{
  tmp_pathbuf tp;
  __try
    {
      if (flags & ~AT_SYMLINK_NOFOLLOW)
	{
	  set_errno (EINVAL);
	  __leave;
	}
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;
      return chown_worker (path, (flags & AT_SYMLINK_NOFOLLOW)
				 ? PC_SYM_NOFOLLOW : PC_SYM_FOLLOW, uid, gid);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
fstatat (int dirfd, const char *__restrict pathname, struct stat *__restrict st,
	 int flags)
{
  tmp_pathbuf tp;
  __try
    {
      if (flags & ~AT_SYMLINK_NOFOLLOW)
	{
	  set_errno (EINVAL);
	  __leave;
	}
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;
      path_conv pc (path, ((flags & AT_SYMLINK_NOFOLLOW)
			   ? PC_SYM_NOFOLLOW : PC_SYM_FOLLOW)
			  | PC_POSIX | PC_KEEP_HANDLE, stat_suffixes);
      return stat_worker (pc, st);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern int utimens_worker (path_conv &, const struct timespec *);

extern "C" int
utimensat (int dirfd, const char *pathname, const struct timespec *times,
	   int flags)
{
  tmp_pathbuf tp;
  __try
    {
      char *path = tp.c_get ();
      if (flags & ~AT_SYMLINK_NOFOLLOW)
	{
	  set_errno (EINVAL);
	  __leave;
	}
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;
      path_conv win32 (path, PC_POSIX | ((flags & AT_SYMLINK_NOFOLLOW)
					 ? PC_SYM_NOFOLLOW : PC_SYM_FOLLOW),
		       stat_suffixes);
      return utimens_worker (win32, times);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
futimesat (int dirfd, const char *pathname, const struct timeval *times)
{
  tmp_pathbuf tp;
  __try
    {
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname, true))
	__leave;
      return utimes (path, times);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
linkat (int olddirfd, const char *oldpathname,
	int newdirfd, const char *newpathname,
	int flags)
{
  tmp_pathbuf tp;
  __try
    {
      if (flags & ~AT_SYMLINK_FOLLOW)
	{
	  set_errno (EINVAL);
	  __leave;
	}
      char *oldpath = tp.c_get ();
      if (gen_full_path_at (oldpath, olddirfd, oldpathname))
	__leave;
      char *newpath = tp.c_get ();
      if (gen_full_path_at (newpath, newdirfd, newpathname))
	__leave;
      if (flags & AT_SYMLINK_FOLLOW)
	{
	  path_conv old_name (oldpath, PC_SYM_FOLLOW | PC_POSIX, stat_suffixes);
	  if (old_name.error)
	    {
	      set_errno (old_name.error);
	      __leave;
	    }
	  strcpy (oldpath, old_name.get_posix ());
	}
      return link (oldpath, newpath);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
mkdirat (int dirfd, const char *pathname, mode_t mode)
{
  tmp_pathbuf tp;
  __try
    {
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;
      return mkdir (path, mode);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
mkfifoat (int dirfd, const char *pathname, mode_t mode)
{
  tmp_pathbuf tp;
  __try
    {
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;
      return mkfifo (path, mode);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
mknodat (int dirfd, const char *pathname, mode_t mode, dev_t dev)
{
  tmp_pathbuf tp;
  __try
    {
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;
      return mknod32 (path, mode, dev);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" ssize_t
readlinkat (int dirfd, const char *__restrict pathname, char *__restrict buf,
	    size_t bufsize)
{
  tmp_pathbuf tp;
  __try
    {
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;
      return readlink (path, buf, bufsize);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
renameat2 (int olddirfd, const char *oldpathname,
	   int newdirfd, const char *newpathname, unsigned int flags)
{
  tmp_pathbuf tp;
  __try
    {
      char *oldpath = tp.c_get ();
      if (gen_full_path_at (oldpath, olddirfd, oldpathname))
	__leave;
      char *newpath = tp.c_get ();
      if (gen_full_path_at (newpath, newdirfd, newpathname))
	__leave;
      return rename2 (oldpath, newpath, flags);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
renameat (int olddirfd, const char *oldpathname,
	  int newdirfd, const char *newpathname)
{
  return renameat2 (olddirfd, oldpathname, newdirfd, newpathname, 0);
}

extern "C" int
scandirat (int dirfd, const char *pathname, struct dirent ***namelist,
	   int (*select) (const struct dirent *),
	   int (*compar) (const struct dirent **, const struct dirent **))
{
  tmp_pathbuf tp;
  __try
    {
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;
      return scandir (path, namelist, select, compar);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
symlinkat (const char *oldpath, int newdirfd, const char *newpathname)
{
  tmp_pathbuf tp;
  __try
    {
      char *newpath = tp.c_get ();
      if (gen_full_path_at (newpath, newdirfd, newpathname))
	__leave;
      return symlink (oldpath, newpath);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}

extern "C" int
unlinkat (int dirfd, const char *pathname, int flags)
{
  tmp_pathbuf tp;
  __try
    {
      if (flags & ~AT_REMOVEDIR)
	{
	  set_errno (EINVAL);
	  __leave;
	}
      char *path = tp.c_get ();
      if (gen_full_path_at (path, dirfd, pathname))
	__leave;
      return (flags & AT_REMOVEDIR) ? rmdir (path) : unlink (path);
    }
  __except (EFAULT) {}
  __endtry
  return -1;
}