1//===- Win32/Program.cpp - Win32 Program Implementation ------- -*- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file provides the Win32 specific implementation of the Program class.
10//
11//===----------------------------------------------------------------------===//
12
13#include "llvm/ADT/BitVector.h"
14#include "llvm/ADT/StringExtras.h"
15#include "llvm/Support/ConvertUTF.h"
16#include "llvm/Support/Errc.h"
17#include "llvm/Support/FileSystem.h"
18#include "llvm/Support/Path.h"
19#include "llvm/Support/Windows/WindowsSupport.h"
20#include "llvm/Support/WindowsError.h"
21#include "llvm/Support/raw_ostream.h"
22#include <cstdio>
23#include <fcntl.h>
24#include <io.h>
25#include <malloc.h>
26#include <numeric>
27#include <psapi.h>
28
29//===----------------------------------------------------------------------===//
30//=== WARNING: Implementation here must contain only Win32 specific code
31//===          and must not be UNIX code
32//===----------------------------------------------------------------------===//
33
34namespace llvm {
35
36ProcessInfo::ProcessInfo() : Pid(0), Process(0), ReturnCode(0) {}
37
38ErrorOr<std::string> sys::findProgramByName(StringRef Name,
39                                            ArrayRef<StringRef> Paths) {
40  assert(!Name.empty() && "Must have a name!");
41
42  if (Name.find_first_of("/\\") != StringRef::npos)
43    return std::string(Name);
44
45  const wchar_t *Path = nullptr;
46  std::wstring PathStorage;
47  if (!Paths.empty()) {
48    PathStorage.reserve(Paths.size() * MAX_PATH);
49    for (unsigned i = 0; i < Paths.size(); ++i) {
50      if (i)
51        PathStorage.push_back(L';');
52      StringRef P = Paths[i];
53      SmallVector<wchar_t, MAX_PATH> TmpPath;
54      if (std::error_code EC = windows::UTF8ToUTF16(P, TmpPath))
55        return EC;
56      PathStorage.append(TmpPath.begin(), TmpPath.end());
57    }
58    Path = PathStorage.c_str();
59  }
60
61  SmallVector<wchar_t, MAX_PATH> U16Name;
62  if (std::error_code EC = windows::UTF8ToUTF16(Name, U16Name))
63    return EC;
64
65  SmallVector<StringRef, 12> PathExts;
66  PathExts.push_back("");
67  PathExts.push_back(".exe"); // FIXME: This must be in %PATHEXT%.
68  if (const char *PathExtEnv = std::getenv("PATHEXT"))
69    SplitString(PathExtEnv, PathExts, ";");
70
71  SmallVector<char, MAX_PATH> U8Result;
72  for (StringRef Ext : PathExts) {
73    SmallVector<wchar_t, MAX_PATH> U16Result;
74    DWORD Len = MAX_PATH;
75    do {
76      U16Result.resize_for_overwrite(Len);
77      // Lets attach the extension manually. That is needed for files
78      // with a point in name like aaa.bbb. SearchPathW will not add extension
79      // from its argument to such files because it thinks they already had one.
80      SmallVector<wchar_t, MAX_PATH> U16NameExt;
81      if (std::error_code EC =
82              windows::UTF8ToUTF16(Twine(Name + Ext).str(), U16NameExt))
83        return EC;
84
85      Len = ::SearchPathW(Path, c_str(U16NameExt), nullptr, U16Result.size(),
86                          U16Result.data(), nullptr);
87    } while (Len > U16Result.size());
88
89    if (Len == 0)
90      continue;
91
92    U16Result.truncate(Len);
93
94    if (std::error_code EC =
95            windows::UTF16ToUTF8(U16Result.data(), U16Result.size(), U8Result))
96      return EC;
97
98    if (sys::fs::can_execute(U8Result))
99      break; // Found it.
100
101    U8Result.clear();
102  }
103
104  if (U8Result.empty())
105    return mapWindowsError(::GetLastError());
106
107  llvm::sys::path::make_preferred(U8Result);
108  return std::string(U8Result.begin(), U8Result.end());
109}
110
111bool MakeErrMsg(std::string *ErrMsg, const std::string &prefix) {
112  if (!ErrMsg)
113    return true;
114  char *buffer = NULL;
115  DWORD LastError = GetLastError();
116  DWORD R = FormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER |
117                               FORMAT_MESSAGE_FROM_SYSTEM |
118                               FORMAT_MESSAGE_MAX_WIDTH_MASK,
119                           NULL, LastError, 0, (LPSTR)&buffer, 1, NULL);
120  if (R)
121    *ErrMsg = prefix + ": " + buffer;
122  else
123    *ErrMsg = prefix + ": Unknown error";
124  *ErrMsg += " (0x" + llvm::utohexstr(LastError) + ")";
125
126  LocalFree(buffer);
127  return R != 0;
128}
129
130static HANDLE RedirectIO(std::optional<StringRef> Path, int fd,
131                         std::string *ErrMsg) {
132  HANDLE h;
133  if (!Path) {
134    if (!DuplicateHandle(GetCurrentProcess(), (HANDLE)_get_osfhandle(fd),
135                         GetCurrentProcess(), &h, 0, TRUE,
136                         DUPLICATE_SAME_ACCESS))
137      return INVALID_HANDLE_VALUE;
138    return h;
139  }
140
141  std::string fname;
142  if (Path->empty())
143    fname = "NUL";
144  else
145    fname = std::string(*Path);
146
147  SECURITY_ATTRIBUTES sa;
148  sa.nLength = sizeof(sa);
149  sa.lpSecurityDescriptor = 0;
150  sa.bInheritHandle = TRUE;
151
152  SmallVector<wchar_t, 128> fnameUnicode;
153  if (Path->empty()) {
154    // Don't play long-path tricks on "NUL".
155    if (windows::UTF8ToUTF16(fname, fnameUnicode))
156      return INVALID_HANDLE_VALUE;
157  } else {
158    if (sys::windows::widenPath(fname, fnameUnicode))
159      return INVALID_HANDLE_VALUE;
160  }
161  h = CreateFileW(fnameUnicode.data(), fd ? GENERIC_WRITE : GENERIC_READ,
162                  FILE_SHARE_READ, &sa, fd == 0 ? OPEN_EXISTING : CREATE_ALWAYS,
163                  FILE_ATTRIBUTE_NORMAL, NULL);
164  if (h == INVALID_HANDLE_VALUE) {
165    MakeErrMsg(ErrMsg,
166               fname + ": Can't open file for " + (fd ? "input" : "output"));
167  }
168
169  return h;
170}
171
172} // namespace llvm
173
174static bool Execute(ProcessInfo &PI, StringRef Program,
175                    ArrayRef<StringRef> Args, std::optional<ArrayRef<StringRef>> Env,
176                    ArrayRef<std::optional<StringRef>> Redirects,
177                    unsigned MemoryLimit, std::string *ErrMsg,
178                    BitVector *AffinityMask) {
179  if (!sys::fs::can_execute(Program)) {
180    if (ErrMsg)
181      *ErrMsg = "program not executable";
182    return false;
183  }
184
185  // can_execute may succeed by looking at Program + ".exe". CreateProcessW
186  // will implicitly add the .exe if we provide a command line without an
187  // executable path, but since we use an explicit executable, we have to add
188  // ".exe" ourselves.
189  SmallString<64> ProgramStorage;
190  if (!sys::fs::exists(Program))
191    Program = Twine(Program + ".exe").toStringRef(ProgramStorage);
192
193  // Windows wants a command line, not an array of args, to pass to the new
194  // process.  We have to concatenate them all, while quoting the args that
195  // have embedded spaces (or are empty).
196  auto Result = flattenWindowsCommandLine(Args);
197  if (std::error_code ec = Result.getError()) {
198    SetLastError(ec.value());
199    MakeErrMsg(ErrMsg, std::string("Unable to convert command-line to UTF-16"));
200    return false;
201  }
202  std::wstring Command = *Result;
203
204  // The pointer to the environment block for the new process.
205  std::vector<wchar_t> EnvBlock;
206
207  if (Env) {
208    // An environment block consists of a null-terminated block of
209    // null-terminated strings. Convert the array of environment variables to
210    // an environment block by concatenating them.
211    for (StringRef E : *Env) {
212      SmallVector<wchar_t, MAX_PATH> EnvString;
213      if (std::error_code ec = windows::UTF8ToUTF16(E, EnvString)) {
214        SetLastError(ec.value());
215        MakeErrMsg(ErrMsg, "Unable to convert environment variable to UTF-16");
216        return false;
217      }
218
219      llvm::append_range(EnvBlock, EnvString);
220      EnvBlock.push_back(0);
221    }
222    EnvBlock.push_back(0);
223  }
224
225  // Create a child process.
226  STARTUPINFOW si;
227  memset(&si, 0, sizeof(si));
228  si.cb = sizeof(si);
229  si.hStdInput = INVALID_HANDLE_VALUE;
230  si.hStdOutput = INVALID_HANDLE_VALUE;
231  si.hStdError = INVALID_HANDLE_VALUE;
232
233  if (!Redirects.empty()) {
234    si.dwFlags = STARTF_USESTDHANDLES;
235
236    si.hStdInput = RedirectIO(Redirects[0], 0, ErrMsg);
237    if (si.hStdInput == INVALID_HANDLE_VALUE) {
238      MakeErrMsg(ErrMsg, "can't redirect stdin");
239      return false;
240    }
241    si.hStdOutput = RedirectIO(Redirects[1], 1, ErrMsg);
242    if (si.hStdOutput == INVALID_HANDLE_VALUE) {
243      CloseHandle(si.hStdInput);
244      MakeErrMsg(ErrMsg, "can't redirect stdout");
245      return false;
246    }
247    if (Redirects[1] && Redirects[2] && *Redirects[1] == *Redirects[2]) {
248      // If stdout and stderr should go to the same place, redirect stderr
249      // to the handle already open for stdout.
250      if (!DuplicateHandle(GetCurrentProcess(), si.hStdOutput,
251                           GetCurrentProcess(), &si.hStdError, 0, TRUE,
252                           DUPLICATE_SAME_ACCESS)) {
253        CloseHandle(si.hStdInput);
254        CloseHandle(si.hStdOutput);
255        MakeErrMsg(ErrMsg, "can't dup stderr to stdout");
256        return false;
257      }
258    } else {
259      // Just redirect stderr
260      si.hStdError = RedirectIO(Redirects[2], 2, ErrMsg);
261      if (si.hStdError == INVALID_HANDLE_VALUE) {
262        CloseHandle(si.hStdInput);
263        CloseHandle(si.hStdOutput);
264        MakeErrMsg(ErrMsg, "can't redirect stderr");
265        return false;
266      }
267    }
268  }
269
270  PROCESS_INFORMATION pi;
271  memset(&pi, 0, sizeof(pi));
272
273  fflush(stdout);
274  fflush(stderr);
275
276  SmallVector<wchar_t, MAX_PATH> ProgramUtf16;
277  if (std::error_code ec = sys::windows::widenPath(Program, ProgramUtf16)) {
278    SetLastError(ec.value());
279    MakeErrMsg(ErrMsg,
280               std::string("Unable to convert application name to UTF-16"));
281    return false;
282  }
283
284  unsigned CreateFlags = CREATE_UNICODE_ENVIRONMENT;
285  if (AffinityMask)
286    CreateFlags |= CREATE_SUSPENDED;
287
288  std::vector<wchar_t> CommandUtf16(Command.size() + 1, 0);
289  std::copy(Command.begin(), Command.end(), CommandUtf16.begin());
290  BOOL rc = CreateProcessW(ProgramUtf16.data(), CommandUtf16.data(), 0, 0, TRUE,
291                           CreateFlags, EnvBlock.empty() ? 0 : EnvBlock.data(),
292                           0, &si, &pi);
293  DWORD err = GetLastError();
294
295  // Regardless of whether the process got created or not, we are done with
296  // the handles we created for it to inherit.
297  CloseHandle(si.hStdInput);
298  CloseHandle(si.hStdOutput);
299  CloseHandle(si.hStdError);
300
301  // Now return an error if the process didn't get created.
302  if (!rc) {
303    SetLastError(err);
304    MakeErrMsg(ErrMsg,
305               std::string("Couldn't execute program '") + Program.str() + "'");
306    return false;
307  }
308
309  PI.Pid = pi.dwProcessId;
310  PI.Process = pi.hProcess;
311
312  // Make sure these get closed no matter what.
313  ScopedCommonHandle hThread(pi.hThread);
314
315  // Assign the process to a job if a memory limit is defined.
316  ScopedJobHandle hJob;
317  if (MemoryLimit != 0) {
318    hJob = CreateJobObjectW(0, 0);
319    bool success = false;
320    if (hJob) {
321      JOBOBJECT_EXTENDED_LIMIT_INFORMATION jeli;
322      memset(&jeli, 0, sizeof(jeli));
323      jeli.BasicLimitInformation.LimitFlags = JOB_OBJECT_LIMIT_PROCESS_MEMORY;
324      jeli.ProcessMemoryLimit = uintptr_t(MemoryLimit) * 1048576;
325      if (SetInformationJobObject(hJob, JobObjectExtendedLimitInformation,
326                                  &jeli, sizeof(jeli))) {
327        if (AssignProcessToJobObject(hJob, pi.hProcess))
328          success = true;
329      }
330    }
331    if (!success) {
332      SetLastError(GetLastError());
333      MakeErrMsg(ErrMsg, std::string("Unable to set memory limit"));
334      TerminateProcess(pi.hProcess, 1);
335      WaitForSingleObject(pi.hProcess, INFINITE);
336      return false;
337    }
338  }
339
340  // Set the affinity mask
341  if (AffinityMask) {
342    ::SetProcessAffinityMask(pi.hProcess,
343                             (DWORD_PTR)AffinityMask->getData().front());
344    ::ResumeThread(pi.hThread);
345  }
346
347  return true;
348}
349
350static bool argNeedsQuotes(StringRef Arg) {
351  if (Arg.empty())
352    return true;
353  return StringRef::npos != Arg.find_first_of("\t \"&\'()*<>\\`^|\n");
354}
355
356static std::string quoteSingleArg(StringRef Arg) {
357  std::string Result;
358  Result.push_back('"');
359
360  while (!Arg.empty()) {
361    size_t FirstNonBackslash = Arg.find_first_not_of('\\');
362    size_t BackslashCount = FirstNonBackslash;
363    if (FirstNonBackslash == StringRef::npos) {
364      // The entire remainder of the argument is backslashes.  Escape all of
365      // them and just early out.
366      BackslashCount = Arg.size();
367      Result.append(BackslashCount * 2, '\\');
368      break;
369    }
370
371    if (Arg[FirstNonBackslash] == '\"') {
372      // This is an embedded quote.  Escape all preceding backslashes, then
373      // add one additional backslash to escape the quote.
374      Result.append(BackslashCount * 2 + 1, '\\');
375      Result.push_back('\"');
376    } else {
377      // This is just a normal character.  Don't escape any of the preceding
378      // backslashes, just append them as they are and then append the
379      // character.
380      Result.append(BackslashCount, '\\');
381      Result.push_back(Arg[FirstNonBackslash]);
382    }
383
384    // Drop all the backslashes, plus the following character.
385    Arg = Arg.drop_front(FirstNonBackslash + 1);
386  }
387
388  Result.push_back('"');
389  return Result;
390}
391
392namespace llvm {
393ErrorOr<std::wstring> sys::flattenWindowsCommandLine(ArrayRef<StringRef> Args) {
394  std::string Command;
395  for (StringRef Arg : Args) {
396    if (argNeedsQuotes(Arg))
397      Command += quoteSingleArg(Arg);
398    else
399      Command += Arg;
400
401    Command.push_back(' ');
402  }
403
404  SmallVector<wchar_t, MAX_PATH> CommandUtf16;
405  if (std::error_code ec = windows::UTF8ToUTF16(Command, CommandUtf16))
406    return ec;
407
408  return std::wstring(CommandUtf16.begin(), CommandUtf16.end());
409}
410
411ProcessInfo sys::Wait(const ProcessInfo &PI,
412                      std::optional<unsigned> SecondsToWait,
413                      std::string *ErrMsg,
414                      std::optional<ProcessStatistics> *ProcStat,
415                      bool Polling) {
416  assert(PI.Pid && "invalid pid to wait on, process not started?");
417  assert((PI.Process && PI.Process != INVALID_HANDLE_VALUE) &&
418         "invalid process handle to wait on, process not started?");
419  DWORD milliSecondsToWait = SecondsToWait ? *SecondsToWait * 1000 : INFINITE;
420
421  ProcessInfo WaitResult = PI;
422  if (ProcStat)
423    ProcStat->reset();
424  DWORD WaitStatus = WaitForSingleObject(PI.Process, milliSecondsToWait);
425  if (WaitStatus == WAIT_TIMEOUT) {
426    if (!Polling && *SecondsToWait > 0) {
427      if (!TerminateProcess(PI.Process, 1)) {
428        if (ErrMsg)
429          MakeErrMsg(ErrMsg, "Failed to terminate timed-out program");
430
431        // -2 indicates a crash or timeout as opposed to failure to execute.
432        WaitResult.ReturnCode = -2;
433        CloseHandle(PI.Process);
434        return WaitResult;
435      }
436      WaitForSingleObject(PI.Process, INFINITE);
437      CloseHandle(PI.Process);
438    } else {
439      // Non-blocking wait.
440      return ProcessInfo();
441    }
442  }
443
444  // Get process execution statistics.
445  if (ProcStat) {
446    FILETIME CreationTime, ExitTime, KernelTime, UserTime;
447    PROCESS_MEMORY_COUNTERS MemInfo;
448    if (GetProcessTimes(PI.Process, &CreationTime, &ExitTime, &KernelTime,
449                        &UserTime) &&
450        GetProcessMemoryInfo(PI.Process, &MemInfo, sizeof(MemInfo))) {
451      auto UserT = std::chrono::duration_cast<std::chrono::microseconds>(
452          toDuration(UserTime));
453      auto KernelT = std::chrono::duration_cast<std::chrono::microseconds>(
454          toDuration(KernelTime));
455      uint64_t PeakMemory = MemInfo.PeakPagefileUsage / 1024;
456      *ProcStat = ProcessStatistics{UserT + KernelT, UserT, PeakMemory};
457    }
458  }
459
460  // Get its exit status.
461  DWORD status;
462  BOOL rc = GetExitCodeProcess(PI.Process, &status);
463  DWORD err = GetLastError();
464  if (err != ERROR_INVALID_HANDLE)
465    CloseHandle(PI.Process);
466
467  if (!rc) {
468    SetLastError(err);
469    if (ErrMsg)
470      MakeErrMsg(ErrMsg, "Failed getting status for program");
471
472    // -2 indicates a crash or timeout as opposed to failure to execute.
473    WaitResult.ReturnCode = -2;
474    return WaitResult;
475  }
476
477  if (!status)
478    return WaitResult;
479
480  // Pass 10(Warning) and 11(Error) to the callee as negative value.
481  if ((status & 0xBFFF0000U) == 0x80000000U)
482    WaitResult.ReturnCode = static_cast<int>(status);
483  else if (status & 0xFF)
484    WaitResult.ReturnCode = status & 0x7FFFFFFF;
485  else
486    WaitResult.ReturnCode = 1;
487
488  return WaitResult;
489}
490
491std::error_code llvm::sys::ChangeStdinMode(sys::fs::OpenFlags Flags) {
492  if (!(Flags & fs::OF_CRLF))
493    return ChangeStdinToBinary();
494  return std::error_code();
495}
496
497std::error_code llvm::sys::ChangeStdoutMode(sys::fs::OpenFlags Flags) {
498  if (!(Flags & fs::OF_CRLF))
499    return ChangeStdoutToBinary();
500  return std::error_code();
501}
502
503std::error_code sys::ChangeStdinToBinary() {
504  int result = _setmode(_fileno(stdin), _O_BINARY);
505  if (result == -1)
506    return std::error_code(errno, std::generic_category());
507  return std::error_code();
508}
509
510std::error_code sys::ChangeStdoutToBinary() {
511  int result = _setmode(_fileno(stdout), _O_BINARY);
512  if (result == -1)
513    return std::error_code(errno, std::generic_category());
514  return std::error_code();
515}
516
517std::error_code
518llvm::sys::writeFileWithEncoding(StringRef FileName, StringRef Contents,
519                                 WindowsEncodingMethod Encoding) {
520  std::error_code EC;
521  llvm::raw_fd_ostream OS(FileName, EC, llvm::sys::fs::OF_TextWithCRLF);
522  if (EC)
523    return EC;
524
525  if (Encoding == WEM_UTF8) {
526    OS << Contents;
527  } else if (Encoding == WEM_CurrentCodePage) {
528    SmallVector<wchar_t, 1> ArgsUTF16;
529    SmallVector<char, 1> ArgsCurCP;
530
531    if ((EC = windows::UTF8ToUTF16(Contents, ArgsUTF16)))
532      return EC;
533
534    if ((EC = windows::UTF16ToCurCP(ArgsUTF16.data(), ArgsUTF16.size(),
535                                    ArgsCurCP)))
536      return EC;
537
538    OS.write(ArgsCurCP.data(), ArgsCurCP.size());
539  } else if (Encoding == WEM_UTF16) {
540    SmallVector<wchar_t, 1> ArgsUTF16;
541
542    if ((EC = windows::UTF8ToUTF16(Contents, ArgsUTF16)))
543      return EC;
544
545    // Endianness guessing
546    char BOM[2];
547    uint16_t src = UNI_UTF16_BYTE_ORDER_MARK_NATIVE;
548    memcpy(BOM, &src, 2);
549    OS.write(BOM, 2);
550    OS.write((char *)ArgsUTF16.data(), ArgsUTF16.size() << 1);
551  } else {
552    llvm_unreachable("Unknown encoding");
553  }
554
555  if (OS.has_error())
556    return make_error_code(errc::io_error);
557
558  return EC;
559}
560
561bool llvm::sys::commandLineFitsWithinSystemLimits(StringRef Program,
562                                                  ArrayRef<StringRef> Args) {
563  // The documentation on CreateProcessW states that the size of the argument
564  // lpCommandLine must not be greater than 32767 characters, including the
565  // Unicode terminating null character. We use smaller value to reduce risk
566  // of getting invalid command line due to unaccounted factors.
567  static const size_t MaxCommandStringLength = 32000;
568  SmallVector<StringRef, 8> FullArgs;
569  FullArgs.push_back(Program);
570  FullArgs.append(Args.begin(), Args.end());
571  auto Result = flattenWindowsCommandLine(FullArgs);
572  assert(!Result.getError());
573  return (Result->size() + 1) <= MaxCommandStringLength;
574}
575} // namespace llvm
576