1//===- llvm/Support/Unix/Program.cpp -----------------------------*- 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 implements the Unix specific portion of the Program class.
10//
11//===----------------------------------------------------------------------===//
12
13//===----------------------------------------------------------------------===//
14//=== WARNING: Implementation here must contain only generic UNIX code that
15//===          is guaranteed to work on *all* UNIX variants.
16//===----------------------------------------------------------------------===//
17
18#include "Unix.h"
19#include "llvm/ADT/StringExtras.h"
20#include "llvm/Config/config.h"
21#include "llvm/Support/Compiler.h"
22#include "llvm/Support/Errc.h"
23#include "llvm/Support/FileSystem.h"
24#include "llvm/Support/Path.h"
25#include "llvm/Support/StringSaver.h"
26#include "llvm/Support/raw_ostream.h"
27#if HAVE_SYS_STAT_H
28#include <sys/stat.h>
29#endif
30#if HAVE_SYS_RESOURCE_H
31#include <sys/resource.h>
32#endif
33#if HAVE_SIGNAL_H
34#include <signal.h>
35#endif
36#if HAVE_FCNTL_H
37#include <fcntl.h>
38#endif
39#if HAVE_UNISTD_H
40#include <unistd.h>
41#endif
42#ifdef HAVE_POSIX_SPAWN
43#include <spawn.h>
44
45#if defined(__APPLE__)
46#include <TargetConditionals.h>
47#endif
48
49#if defined(__APPLE__) && !(defined(TARGET_OS_IPHONE) && TARGET_OS_IPHONE)
50#define USE_NSGETENVIRON 1
51#else
52#define USE_NSGETENVIRON 0
53#endif
54
55#if !USE_NSGETENVIRON
56  extern char **environ;
57#else
58#include <crt_externs.h> // _NSGetEnviron
59#endif
60#endif
61
62namespace llvm {
63
64using namespace sys;
65
66ProcessInfo::ProcessInfo() : Pid(0), ReturnCode(0) {}
67
68ErrorOr<std::string> sys::findProgramByName(StringRef Name,
69                                            ArrayRef<StringRef> Paths) {
70  assert(!Name.empty() && "Must have a name!");
71  // Use the given path verbatim if it contains any slashes; this matches
72  // the behavior of sh(1) and friends.
73  if (Name.find('/') != StringRef::npos) return std::string(Name);
74
75  SmallVector<StringRef, 16> EnvironmentPaths;
76  if (Paths.empty())
77    if (const char *PathEnv = std::getenv("PATH")) {
78      SplitString(PathEnv, EnvironmentPaths, ":");
79      Paths = EnvironmentPaths;
80    }
81
82  for (auto Path : Paths) {
83    if (Path.empty())
84      continue;
85
86    // Check to see if this first directory contains the executable...
87    SmallString<128> FilePath(Path);
88    sys::path::append(FilePath, Name);
89    if (sys::fs::can_execute(FilePath.c_str()))
90      return std::string(FilePath.str());  // Found the executable!
91  }
92  return errc::no_such_file_or_directory;
93}
94
95static bool RedirectIO(Optional<StringRef> Path, int FD, std::string* ErrMsg) {
96  if (!Path) // Noop
97    return false;
98  std::string File;
99  if (Path->empty())
100    // Redirect empty paths to /dev/null
101    File = "/dev/null";
102  else
103    File = std::string(*Path);
104
105  // Open the file
106  int InFD = open(File.c_str(), FD == 0 ? O_RDONLY : O_WRONLY|O_CREAT, 0666);
107  if (InFD == -1) {
108    MakeErrMsg(ErrMsg, "Cannot open file '" + File + "' for "
109              + (FD == 0 ? "input" : "output"));
110    return true;
111  }
112
113  // Install it as the requested FD
114  if (dup2(InFD, FD) == -1) {
115    MakeErrMsg(ErrMsg, "Cannot dup2");
116    close(InFD);
117    return true;
118  }
119  close(InFD);      // Close the original FD
120  return false;
121}
122
123#ifdef HAVE_POSIX_SPAWN
124static bool RedirectIO_PS(const std::string *Path, int FD, std::string *ErrMsg,
125                          posix_spawn_file_actions_t *FileActions) {
126  if (!Path) // Noop
127    return false;
128  const char *File;
129  if (Path->empty())
130    // Redirect empty paths to /dev/null
131    File = "/dev/null";
132  else
133    File = Path->c_str();
134
135  if (int Err = posix_spawn_file_actions_addopen(
136          FileActions, FD, File,
137          FD == 0 ? O_RDONLY : O_WRONLY | O_CREAT, 0666))
138    return MakeErrMsg(ErrMsg, "Cannot posix_spawn_file_actions_addopen", Err);
139  return false;
140}
141#endif
142
143static void TimeOutHandler(int Sig) {
144}
145
146static void SetMemoryLimits(unsigned size) {
147#if HAVE_SYS_RESOURCE_H && HAVE_GETRLIMIT && HAVE_SETRLIMIT
148  struct rlimit r;
149  __typeof__ (r.rlim_cur) limit = (__typeof__ (r.rlim_cur)) (size) * 1048576;
150
151  // Heap size
152  getrlimit (RLIMIT_DATA, &r);
153  r.rlim_cur = limit;
154  setrlimit (RLIMIT_DATA, &r);
155#ifdef RLIMIT_RSS
156  // Resident set size.
157  getrlimit (RLIMIT_RSS, &r);
158  r.rlim_cur = limit;
159  setrlimit (RLIMIT_RSS, &r);
160#endif
161#endif
162}
163
164}
165
166static std::vector<const char *>
167toNullTerminatedCStringArray(ArrayRef<StringRef> Strings, StringSaver &Saver) {
168  std::vector<const char *> Result;
169  for (StringRef S : Strings)
170    Result.push_back(Saver.save(S).data());
171  Result.push_back(nullptr);
172  return Result;
173}
174
175static bool Execute(ProcessInfo &PI, StringRef Program,
176                    ArrayRef<StringRef> Args, Optional<ArrayRef<StringRef>> Env,
177                    ArrayRef<Optional<StringRef>> Redirects,
178                    unsigned MemoryLimit, std::string *ErrMsg) {
179  if (!llvm::sys::fs::exists(Program)) {
180    if (ErrMsg)
181      *ErrMsg = std::string("Executable \"") + Program.str() +
182                std::string("\" doesn't exist!");
183    return false;
184  }
185
186  BumpPtrAllocator Allocator;
187  StringSaver Saver(Allocator);
188  std::vector<const char *> ArgVector, EnvVector;
189  const char **Argv = nullptr;
190  const char **Envp = nullptr;
191  ArgVector = toNullTerminatedCStringArray(Args, Saver);
192  Argv = ArgVector.data();
193  if (Env) {
194    EnvVector = toNullTerminatedCStringArray(*Env, Saver);
195    Envp = EnvVector.data();
196  }
197
198  // If this OS has posix_spawn and there is no memory limit being implied, use
199  // posix_spawn.  It is more efficient than fork/exec.
200#ifdef HAVE_POSIX_SPAWN
201  if (MemoryLimit == 0) {
202    posix_spawn_file_actions_t FileActionsStore;
203    posix_spawn_file_actions_t *FileActions = nullptr;
204
205    // If we call posix_spawn_file_actions_addopen we have to make sure the
206    // c strings we pass to it stay alive until the call to posix_spawn,
207    // so we copy any StringRefs into this variable.
208    std::string RedirectsStorage[3];
209
210    if (!Redirects.empty()) {
211      assert(Redirects.size() == 3);
212      std::string *RedirectsStr[3] = {nullptr, nullptr, nullptr};
213      for (int I = 0; I < 3; ++I) {
214        if (Redirects[I]) {
215          RedirectsStorage[I] = std::string(*Redirects[I]);
216          RedirectsStr[I] = &RedirectsStorage[I];
217        }
218      }
219
220      FileActions = &FileActionsStore;
221      posix_spawn_file_actions_init(FileActions);
222
223      // Redirect stdin/stdout.
224      if (RedirectIO_PS(RedirectsStr[0], 0, ErrMsg, FileActions) ||
225          RedirectIO_PS(RedirectsStr[1], 1, ErrMsg, FileActions))
226        return false;
227      if (!Redirects[1] || !Redirects[2] || *Redirects[1] != *Redirects[2]) {
228        // Just redirect stderr
229        if (RedirectIO_PS(RedirectsStr[2], 2, ErrMsg, FileActions))
230          return false;
231      } else {
232        // If stdout and stderr should go to the same place, redirect stderr
233        // to the FD already open for stdout.
234        if (int Err = posix_spawn_file_actions_adddup2(FileActions, 1, 2))
235          return !MakeErrMsg(ErrMsg, "Can't redirect stderr to stdout", Err);
236      }
237    }
238
239    if (!Envp)
240#if !USE_NSGETENVIRON
241      Envp = const_cast<const char **>(environ);
242#else
243      // environ is missing in dylibs.
244      Envp = const_cast<const char **>(*_NSGetEnviron());
245#endif
246
247    constexpr int maxRetries = 8;
248    int retries = 0;
249    pid_t PID;
250    int Err;
251    do {
252      PID = 0; // Make Valgrind happy.
253      Err = posix_spawn(&PID, Program.str().c_str(), FileActions,
254                        /*attrp*/ nullptr, const_cast<char **>(Argv),
255                        const_cast<char **>(Envp));
256    } while (Err == EINTR && ++retries < maxRetries);
257
258    if (FileActions)
259      posix_spawn_file_actions_destroy(FileActions);
260
261    if (Err)
262     return !MakeErrMsg(ErrMsg, "posix_spawn failed", Err);
263
264    PI.Pid = PID;
265    PI.Process = PID;
266
267    return true;
268  }
269#endif
270
271  // Create a child process.
272  int child = fork();
273  switch (child) {
274    // An error occurred:  Return to the caller.
275    case -1:
276      MakeErrMsg(ErrMsg, "Couldn't fork");
277      return false;
278
279    // Child process: Execute the program.
280    case 0: {
281      // Redirect file descriptors...
282      if (!Redirects.empty()) {
283        // Redirect stdin
284        if (RedirectIO(Redirects[0], 0, ErrMsg)) { return false; }
285        // Redirect stdout
286        if (RedirectIO(Redirects[1], 1, ErrMsg)) { return false; }
287        if (Redirects[1] && Redirects[2] && *Redirects[1] == *Redirects[2]) {
288          // If stdout and stderr should go to the same place, redirect stderr
289          // to the FD already open for stdout.
290          if (-1 == dup2(1,2)) {
291            MakeErrMsg(ErrMsg, "Can't redirect stderr to stdout");
292            return false;
293          }
294        } else {
295          // Just redirect stderr
296          if (RedirectIO(Redirects[2], 2, ErrMsg)) { return false; }
297        }
298      }
299
300      // Set memory limits
301      if (MemoryLimit!=0) {
302        SetMemoryLimits(MemoryLimit);
303      }
304
305      // Execute!
306      std::string PathStr = std::string(Program);
307      if (Envp != nullptr)
308        execve(PathStr.c_str(), const_cast<char **>(Argv),
309               const_cast<char **>(Envp));
310      else
311        execv(PathStr.c_str(), const_cast<char **>(Argv));
312      // If the execve() failed, we should exit. Follow Unix protocol and
313      // return 127 if the executable was not found, and 126 otherwise.
314      // Use _exit rather than exit so that atexit functions and static
315      // object destructors cloned from the parent process aren't
316      // redundantly run, and so that any data buffered in stdio buffers
317      // cloned from the parent aren't redundantly written out.
318      _exit(errno == ENOENT ? 127 : 126);
319    }
320
321    // Parent process: Break out of the switch to do our processing.
322    default:
323      break;
324  }
325
326  PI.Pid = child;
327  PI.Process = child;
328
329  return true;
330}
331
332namespace llvm {
333
334ProcessInfo sys::Wait(const ProcessInfo &PI, unsigned SecondsToWait,
335                      bool WaitUntilTerminates, std::string *ErrMsg,
336                      Optional<ProcessStatistics> *ProcStat) {
337  struct sigaction Act, Old;
338  assert(PI.Pid && "invalid pid to wait on, process not started?");
339
340  int WaitPidOptions = 0;
341  pid_t ChildPid = PI.Pid;
342  if (WaitUntilTerminates) {
343    SecondsToWait = 0;
344  } else if (SecondsToWait) {
345    // Install a timeout handler.  The handler itself does nothing, but the
346    // simple fact of having a handler at all causes the wait below to return
347    // with EINTR, unlike if we used SIG_IGN.
348    memset(&Act, 0, sizeof(Act));
349    Act.sa_handler = TimeOutHandler;
350    sigemptyset(&Act.sa_mask);
351    sigaction(SIGALRM, &Act, &Old);
352    alarm(SecondsToWait);
353  } else if (SecondsToWait == 0)
354    WaitPidOptions = WNOHANG;
355
356  // Parent process: Wait for the child process to terminate.
357  int status;
358  ProcessInfo WaitResult;
359  rusage Info;
360  if (ProcStat)
361    ProcStat->reset();
362
363  do {
364    WaitResult.Pid = wait4(ChildPid, &status, WaitPidOptions, &Info);
365  } while (WaitUntilTerminates && WaitResult.Pid == -1 && errno == EINTR);
366
367  if (WaitResult.Pid != PI.Pid) {
368    if (WaitResult.Pid == 0) {
369      // Non-blocking wait.
370      return WaitResult;
371    } else {
372      if (SecondsToWait && errno == EINTR) {
373        // Kill the child.
374        kill(PI.Pid, SIGKILL);
375
376        // Turn off the alarm and restore the signal handler
377        alarm(0);
378        sigaction(SIGALRM, &Old, nullptr);
379
380        // Wait for child to die
381        if (wait(&status) != ChildPid)
382          MakeErrMsg(ErrMsg, "Child timed out but wouldn't die");
383        else
384          MakeErrMsg(ErrMsg, "Child timed out", 0);
385
386        WaitResult.ReturnCode = -2; // Timeout detected
387        return WaitResult;
388      } else if (errno != EINTR) {
389        MakeErrMsg(ErrMsg, "Error waiting for child process");
390        WaitResult.ReturnCode = -1;
391        return WaitResult;
392      }
393    }
394  }
395
396  // We exited normally without timeout, so turn off the timer.
397  if (SecondsToWait && !WaitUntilTerminates) {
398    alarm(0);
399    sigaction(SIGALRM, &Old, nullptr);
400  }
401
402  if (ProcStat) {
403    std::chrono::microseconds UserT = toDuration(Info.ru_utime);
404    std::chrono::microseconds KernelT = toDuration(Info.ru_stime);
405    uint64_t PeakMemory = static_cast<uint64_t>(Info.ru_maxrss);
406    *ProcStat = ProcessStatistics{UserT + KernelT, UserT, PeakMemory};
407  }
408
409  // Return the proper exit status. Detect error conditions
410  // so we can return -1 for them and set ErrMsg informatively.
411  int result = 0;
412  if (WIFEXITED(status)) {
413    result = WEXITSTATUS(status);
414    WaitResult.ReturnCode = result;
415
416    if (result == 127) {
417      if (ErrMsg)
418        *ErrMsg = llvm::sys::StrError(ENOENT);
419      WaitResult.ReturnCode = -1;
420      return WaitResult;
421    }
422    if (result == 126) {
423      if (ErrMsg)
424        *ErrMsg = "Program could not be executed";
425      WaitResult.ReturnCode = -1;
426      return WaitResult;
427    }
428  } else if (WIFSIGNALED(status)) {
429    if (ErrMsg) {
430      *ErrMsg = strsignal(WTERMSIG(status));
431#ifdef WCOREDUMP
432      if (WCOREDUMP(status))
433        *ErrMsg += " (core dumped)";
434#endif
435    }
436    // Return a special value to indicate that the process received an unhandled
437    // signal during execution as opposed to failing to execute.
438    WaitResult.ReturnCode = -2;
439  }
440  return WaitResult;
441}
442
443std::error_code sys::ChangeStdinToBinary() {
444  // Do nothing, as Unix doesn't differentiate between text and binary.
445  return std::error_code();
446}
447
448std::error_code sys::ChangeStdoutToBinary() {
449  // Do nothing, as Unix doesn't differentiate between text and binary.
450  return std::error_code();
451}
452
453std::error_code
454llvm::sys::writeFileWithEncoding(StringRef FileName, StringRef Contents,
455                                 WindowsEncodingMethod Encoding /*unused*/) {
456  std::error_code EC;
457  llvm::raw_fd_ostream OS(FileName, EC, llvm::sys::fs::OpenFlags::OF_Text);
458
459  if (EC)
460    return EC;
461
462  OS << Contents;
463
464  if (OS.has_error())
465    return make_error_code(errc::io_error);
466
467  return EC;
468}
469
470bool llvm::sys::commandLineFitsWithinSystemLimits(StringRef Program,
471                                                  ArrayRef<StringRef> Args) {
472  static long ArgMax = sysconf(_SC_ARG_MAX);
473  // POSIX requires that _POSIX_ARG_MAX is 4096, which is the lowest possible
474  // value for ARG_MAX on a POSIX compliant system.
475  static long ArgMin = _POSIX_ARG_MAX;
476
477  // This the same baseline used by xargs.
478  long EffectiveArgMax = 128 * 1024;
479
480  if (EffectiveArgMax > ArgMax)
481    EffectiveArgMax = ArgMax;
482  else if (EffectiveArgMax < ArgMin)
483    EffectiveArgMax = ArgMin;
484
485  // System says no practical limit.
486  if (ArgMax == -1)
487    return true;
488
489  // Conservatively account for space required by environment variables.
490  long HalfArgMax = EffectiveArgMax / 2;
491
492  size_t ArgLength = Program.size() + 1;
493  for (StringRef Arg : Args) {
494    // Ensure that we do not exceed the MAX_ARG_STRLEN constant on Linux, which
495    // does not have a constant unlike what the man pages would have you
496    // believe. Since this limit is pretty high, perform the check
497    // unconditionally rather than trying to be aggressive and limiting it to
498    // Linux only.
499    if (Arg.size() >= (32 * 4096))
500      return false;
501
502    ArgLength += Arg.size() + 1;
503    if (ArgLength > size_t(HalfArgMax)) {
504      return false;
505    }
506  }
507
508  return true;
509}
510}
511