1//===- llvm/Support/Unix/Program.inc ----------------------------*- 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
15//===          code that is guaranteed to work on *all* UNIX variants.
16//===----------------------------------------------------------------------===//
17
18#include "llvm/Support/Program.h"
19
20#include "Unix.h"
21#include "llvm/ADT/StringExtras.h"
22#include "llvm/Config/config.h"
23#include "llvm/Support/Compiler.h"
24#include "llvm/Support/Errc.h"
25#include "llvm/Support/FileSystem.h"
26#include "llvm/Support/Path.h"
27#include "llvm/Support/StringSaver.h"
28#include "llvm/Support/raw_ostream.h"
29#if HAVE_SYS_STAT_H
30#include <sys/stat.h>
31#endif
32#if HAVE_SYS_RESOURCE_H
33#include <sys/resource.h>
34#endif
35#if HAVE_SIGNAL_H
36#include <signal.h>
37#endif
38#if HAVE_FCNTL_H
39#include <fcntl.h>
40#endif
41#if HAVE_UNISTD_H
42#include <unistd.h>
43#endif
44#ifdef HAVE_POSIX_SPAWN
45#include <spawn.h>
46
47#if defined(__APPLE__)
48#include <TargetConditionals.h>
49#endif
50
51#if defined(__APPLE__) && !(defined(TARGET_OS_IPHONE) && TARGET_OS_IPHONE)
52#define USE_NSGETENVIRON 1
53#else
54#define USE_NSGETENVIRON 0
55#endif
56
57#if !USE_NSGETENVIRON
58extern char **environ;
59#else
60#include <crt_externs.h> // _NSGetEnviron
61#endif
62#endif
63
64using namespace llvm;
65using namespace sys;
66
67ProcessInfo::ProcessInfo() : Pid(0), ReturnCode(0) {}
68
69ErrorOr<std::string> sys::findProgramByName(StringRef Name,
70                                            ArrayRef<StringRef> Paths) {
71  assert(!Name.empty() && "Must have a name!");
72  // Use the given path verbatim if it contains any slashes; this matches
73  // the behavior of sh(1) and friends.
74  if (Name.contains('/'))
75    return std::string(Name);
76
77  SmallVector<StringRef, 16> EnvironmentPaths;
78  if (Paths.empty())
79    if (const char *PathEnv = std::getenv("PATH")) {
80      SplitString(PathEnv, EnvironmentPaths, ":");
81      Paths = EnvironmentPaths;
82    }
83
84  for (auto Path : Paths) {
85    if (Path.empty())
86      continue;
87
88    // Check to see if this first directory contains the executable...
89    SmallString<128> FilePath(Path);
90    sys::path::append(FilePath, Name);
91    if (sys::fs::can_execute(FilePath.c_str()))
92      return std::string(FilePath.str()); // Found the executable!
93  }
94  return errc::no_such_file_or_directory;
95}
96
97static bool RedirectIO(std::optional<StringRef> Path, int FD, std::string *ErrMsg) {
98  if (!Path) // Noop
99    return false;
100  std::string File;
101  if (Path->empty())
102    // Redirect empty paths to /dev/null
103    File = "/dev/null";
104  else
105    File = std::string(*Path);
106
107  // Open the file
108  int InFD = open(File.c_str(), FD == 0 ? O_RDONLY : O_WRONLY | O_CREAT, 0666);
109  if (InFD == -1) {
110    MakeErrMsg(ErrMsg, "Cannot open file '" + File + "' for " +
111                           (FD == 0 ? "input" : "output"));
112    return true;
113  }
114
115  // Install it as the requested FD
116  if (dup2(InFD, FD) == -1) {
117    MakeErrMsg(ErrMsg, "Cannot dup2");
118    close(InFD);
119    return true;
120  }
121  close(InFD); // Close the original FD
122  return false;
123}
124
125#ifdef HAVE_POSIX_SPAWN
126static bool RedirectIO_PS(const std::string *Path, int FD, std::string *ErrMsg,
127                          posix_spawn_file_actions_t *FileActions) {
128  if (!Path) // Noop
129    return false;
130  const char *File;
131  if (Path->empty())
132    // Redirect empty paths to /dev/null
133    File = "/dev/null";
134  else
135    File = Path->c_str();
136
137  if (int Err = posix_spawn_file_actions_addopen(
138          FileActions, FD, File, FD == 0 ? O_RDONLY : O_WRONLY | O_CREAT, 0666))
139    return MakeErrMsg(ErrMsg, "Cannot posix_spawn_file_actions_addopen", Err);
140  return false;
141}
142#endif
143
144static void TimeOutHandler(int Sig) {}
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
164static std::vector<const char *>
165toNullTerminatedCStringArray(ArrayRef<StringRef> Strings, StringSaver &Saver) {
166  std::vector<const char *> Result;
167  for (StringRef S : Strings)
168    Result.push_back(Saver.save(S).data());
169  Result.push_back(nullptr);
170  return Result;
171}
172
173static bool Execute(ProcessInfo &PI, StringRef Program,
174                    ArrayRef<StringRef> Args, std::optional<ArrayRef<StringRef>> Env,
175                    ArrayRef<std::optional<StringRef>> Redirects,
176                    unsigned MemoryLimit, std::string *ErrMsg,
177                    BitVector *AffinityMask) {
178  if (!llvm::sys::fs::exists(Program)) {
179    if (ErrMsg)
180      *ErrMsg = std::string("Executable \"") + Program.str() +
181                std::string("\" doesn't exist!");
182    return false;
183  }
184
185  assert(!AffinityMask && "Starting a process with an affinity mask is "
186                          "currently not supported on Unix!");
187
188  BumpPtrAllocator Allocator;
189  StringSaver Saver(Allocator);
190  std::vector<const char *> ArgVector, EnvVector;
191  const char **Argv = nullptr;
192  const char **Envp = nullptr;
193  ArgVector = toNullTerminatedCStringArray(Args, Saver);
194  Argv = ArgVector.data();
195  if (Env) {
196    EnvVector = toNullTerminatedCStringArray(*Env, Saver);
197    Envp = EnvVector.data();
198  }
199
200  // If this OS has posix_spawn and there is no memory limit being implied, use
201  // posix_spawn.  It is more efficient than fork/exec.
202#ifdef HAVE_POSIX_SPAWN
203  if (MemoryLimit == 0) {
204    posix_spawn_file_actions_t FileActionsStore;
205    posix_spawn_file_actions_t *FileActions = nullptr;
206
207    // If we call posix_spawn_file_actions_addopen we have to make sure the
208    // c strings we pass to it stay alive until the call to posix_spawn,
209    // so we copy any StringRefs into this variable.
210    std::string RedirectsStorage[3];
211
212    if (!Redirects.empty()) {
213      assert(Redirects.size() == 3);
214      std::string *RedirectsStr[3] = {nullptr, nullptr, nullptr};
215      for (int I = 0; I < 3; ++I) {
216        if (Redirects[I]) {
217          RedirectsStorage[I] = std::string(*Redirects[I]);
218          RedirectsStr[I] = &RedirectsStorage[I];
219        }
220      }
221
222      FileActions = &FileActionsStore;
223      posix_spawn_file_actions_init(FileActions);
224
225      // Redirect stdin/stdout.
226      if (RedirectIO_PS(RedirectsStr[0], 0, ErrMsg, FileActions) ||
227          RedirectIO_PS(RedirectsStr[1], 1, ErrMsg, FileActions))
228        return false;
229      if (!Redirects[1] || !Redirects[2] || *Redirects[1] != *Redirects[2]) {
230        // Just redirect stderr
231        if (RedirectIO_PS(RedirectsStr[2], 2, ErrMsg, FileActions))
232          return false;
233      } else {
234        // If stdout and stderr should go to the same place, redirect stderr
235        // to the FD already open for stdout.
236        if (int Err = posix_spawn_file_actions_adddup2(FileActions, 1, 2))
237          return !MakeErrMsg(ErrMsg, "Can't redirect stderr to stdout", Err);
238      }
239    }
240
241    if (!Envp)
242#if !USE_NSGETENVIRON
243      Envp = const_cast<const char **>(environ);
244#else
245      // environ is missing in dylibs.
246      Envp = const_cast<const char **>(*_NSGetEnviron());
247#endif
248
249    constexpr int maxRetries = 8;
250    int retries = 0;
251    pid_t PID;
252    int Err;
253    do {
254      PID = 0; // Make Valgrind happy.
255      Err = posix_spawn(&PID, Program.str().c_str(), FileActions,
256                        /*attrp*/ nullptr, const_cast<char **>(Argv),
257                        const_cast<char **>(Envp));
258    } while (Err == EINTR && ++retries < maxRetries);
259
260    if (FileActions)
261      posix_spawn_file_actions_destroy(FileActions);
262
263    if (Err)
264      return !MakeErrMsg(ErrMsg, "posix_spawn failed", Err);
265
266    PI.Pid = PID;
267    PI.Process = PID;
268
269    return true;
270  }
271#endif
272
273  // Create a child process.
274  int child = fork();
275  switch (child) {
276  // An error occurred:  Return to the caller.
277  case -1:
278    MakeErrMsg(ErrMsg, "Couldn't fork");
279    return false;
280
281  // Child process: Execute the program.
282  case 0: {
283    // Redirect file descriptors...
284    if (!Redirects.empty()) {
285      // Redirect stdin
286      if (RedirectIO(Redirects[0], 0, ErrMsg)) {
287        return false;
288      }
289      // Redirect stdout
290      if (RedirectIO(Redirects[1], 1, ErrMsg)) {
291        return false;
292      }
293      if (Redirects[1] && Redirects[2] && *Redirects[1] == *Redirects[2]) {
294        // If stdout and stderr should go to the same place, redirect stderr
295        // to the FD already open for stdout.
296        if (-1 == dup2(1, 2)) {
297          MakeErrMsg(ErrMsg, "Can't redirect stderr to stdout");
298          return false;
299        }
300      } else {
301        // Just redirect stderr
302        if (RedirectIO(Redirects[2], 2, ErrMsg)) {
303          return false;
304        }
305      }
306    }
307
308    // Set memory limits
309    if (MemoryLimit != 0) {
310      SetMemoryLimits(MemoryLimit);
311    }
312
313    // Execute!
314    std::string PathStr = std::string(Program);
315    if (Envp != nullptr)
316      execve(PathStr.c_str(), const_cast<char **>(Argv),
317             const_cast<char **>(Envp));
318    else
319      execv(PathStr.c_str(), const_cast<char **>(Argv));
320    // If the execve() failed, we should exit. Follow Unix protocol and
321    // return 127 if the executable was not found, and 126 otherwise.
322    // Use _exit rather than exit so that atexit functions and static
323    // object destructors cloned from the parent process aren't
324    // redundantly run, and so that any data buffered in stdio buffers
325    // cloned from the parent aren't redundantly written out.
326    _exit(errno == ENOENT ? 127 : 126);
327  }
328
329  // Parent process: Break out of the switch to do our processing.
330  default:
331    break;
332  }
333
334  PI.Pid = child;
335  PI.Process = child;
336
337  return true;
338}
339
340namespace llvm {
341namespace sys {
342
343#ifndef _AIX
344using ::wait4;
345#else
346static pid_t(wait4)(pid_t pid, int *status, int options, struct rusage *usage);
347#endif
348
349} // namespace sys
350} // namespace llvm
351
352#ifdef _AIX
353#ifndef _ALL_SOURCE
354extern "C" pid_t(wait4)(pid_t pid, int *status, int options,
355                        struct rusage *usage);
356#endif
357pid_t(llvm::sys::wait4)(pid_t pid, int *status, int options,
358                        struct rusage *usage) {
359  assert(pid > 0 && "Only expecting to handle actual PID values!");
360  assert((options & ~WNOHANG) == 0 && "Expecting WNOHANG at most!");
361  assert(usage && "Expecting usage collection!");
362
363  // AIX wait4 does not work well with WNOHANG.
364  if (!(options & WNOHANG))
365    return ::wait4(pid, status, options, usage);
366
367  // For WNOHANG, we use waitid (which supports WNOWAIT) until the child process
368  // has terminated.
369  siginfo_t WaitIdInfo;
370  WaitIdInfo.si_pid = 0;
371  int WaitIdRetVal =
372      waitid(P_PID, pid, &WaitIdInfo, WNOWAIT | WEXITED | options);
373
374  if (WaitIdRetVal == -1 || WaitIdInfo.si_pid == 0)
375    return WaitIdRetVal;
376
377  assert(WaitIdInfo.si_pid == pid);
378
379  // The child has already terminated, so a blocking wait on it is okay in the
380  // absence of indiscriminate `wait` calls from the current process (which
381  // would cause the call here to fail with ECHILD).
382  return ::wait4(pid, status, options & ~WNOHANG, usage);
383}
384#endif
385
386ProcessInfo llvm::sys::Wait(const ProcessInfo &PI,
387                            std::optional<unsigned> SecondsToWait,
388                            std::string *ErrMsg,
389                            std::optional<ProcessStatistics> *ProcStat,
390                            bool Polling) {
391  struct sigaction Act, Old;
392  assert(PI.Pid && "invalid pid to wait on, process not started?");
393
394  int WaitPidOptions = 0;
395  pid_t ChildPid = PI.Pid;
396  bool WaitUntilTerminates = false;
397  if (!SecondsToWait) {
398    WaitUntilTerminates = true;
399  } else {
400    if (*SecondsToWait == 0)
401      WaitPidOptions = WNOHANG;
402
403    // Install a timeout handler.  The handler itself does nothing, but the
404    // simple fact of having a handler at all causes the wait below to return
405    // with EINTR, unlike if we used SIG_IGN.
406    memset(&Act, 0, sizeof(Act));
407    Act.sa_handler = TimeOutHandler;
408    sigemptyset(&Act.sa_mask);
409    sigaction(SIGALRM, &Act, &Old);
410    // FIXME The alarm signal may be delivered to another thread.
411    alarm(*SecondsToWait);
412  }
413
414  // Parent process: Wait for the child process to terminate.
415  int status = 0;
416  ProcessInfo WaitResult;
417  rusage Info;
418  if (ProcStat)
419    ProcStat->reset();
420
421  do {
422    WaitResult.Pid = sys::wait4(ChildPid, &status, WaitPidOptions, &Info);
423  } while (WaitUntilTerminates && WaitResult.Pid == -1 && errno == EINTR);
424
425  if (WaitResult.Pid != PI.Pid) {
426    if (WaitResult.Pid == 0) {
427      // Non-blocking wait.
428      return WaitResult;
429    } else {
430      if (SecondsToWait && errno == EINTR && !Polling) {
431        // Kill the child.
432        kill(PI.Pid, SIGKILL);
433
434        // Turn off the alarm and restore the signal handler
435        alarm(0);
436        sigaction(SIGALRM, &Old, nullptr);
437
438        // Wait for child to die
439        // FIXME This could grab some other child process out from another
440        // waiting thread and then leave a zombie anyway.
441        if (wait(&status) != ChildPid)
442          MakeErrMsg(ErrMsg, "Child timed out but wouldn't die");
443        else
444          MakeErrMsg(ErrMsg, "Child timed out", 0);
445
446        WaitResult.ReturnCode = -2; // Timeout detected
447        return WaitResult;
448      } else if (errno != EINTR) {
449        MakeErrMsg(ErrMsg, "Error waiting for child process");
450        WaitResult.ReturnCode = -1;
451        return WaitResult;
452      }
453    }
454  }
455
456  // We exited normally without timeout, so turn off the timer.
457  if (SecondsToWait && !WaitUntilTerminates) {
458    alarm(0);
459    sigaction(SIGALRM, &Old, nullptr);
460  }
461
462  if (ProcStat) {
463    std::chrono::microseconds UserT = toDuration(Info.ru_utime);
464    std::chrono::microseconds KernelT = toDuration(Info.ru_stime);
465    uint64_t PeakMemory = 0;
466#ifndef __HAIKU__
467    PeakMemory = static_cast<uint64_t>(Info.ru_maxrss);
468#endif
469    *ProcStat = ProcessStatistics{UserT + KernelT, UserT, PeakMemory};
470  }
471
472  // Return the proper exit status. Detect error conditions
473  // so we can return -1 for them and set ErrMsg informatively.
474  int result = 0;
475  if (WIFEXITED(status)) {
476    result = WEXITSTATUS(status);
477    WaitResult.ReturnCode = result;
478
479    if (result == 127) {
480      if (ErrMsg)
481        *ErrMsg = llvm::sys::StrError(ENOENT);
482      WaitResult.ReturnCode = -1;
483      return WaitResult;
484    }
485    if (result == 126) {
486      if (ErrMsg)
487        *ErrMsg = "Program could not be executed";
488      WaitResult.ReturnCode = -1;
489      return WaitResult;
490    }
491  } else if (WIFSIGNALED(status)) {
492    if (ErrMsg) {
493      *ErrMsg = strsignal(WTERMSIG(status));
494#ifdef WCOREDUMP
495      if (WCOREDUMP(status))
496        *ErrMsg += " (core dumped)";
497#endif
498    }
499    // Return a special value to indicate that the process received an unhandled
500    // signal during execution as opposed to failing to execute.
501    WaitResult.ReturnCode = -2;
502  }
503  return WaitResult;
504}
505
506std::error_code llvm::sys::ChangeStdinMode(fs::OpenFlags Flags) {
507  if (!(Flags & fs::OF_Text))
508    return ChangeStdinToBinary();
509  return std::error_code();
510}
511
512std::error_code llvm::sys::ChangeStdoutMode(fs::OpenFlags Flags) {
513  if (!(Flags & fs::OF_Text))
514    return ChangeStdoutToBinary();
515  return std::error_code();
516}
517
518std::error_code llvm::sys::ChangeStdinToBinary() {
519  // Do nothing, as Unix doesn't differentiate between text and binary.
520  return std::error_code();
521}
522
523std::error_code llvm::sys::ChangeStdoutToBinary() {
524  // Do nothing, as Unix doesn't differentiate between text and binary.
525  return std::error_code();
526}
527
528std::error_code
529llvm::sys::writeFileWithEncoding(StringRef FileName, StringRef Contents,
530                                 WindowsEncodingMethod Encoding /*unused*/) {
531  std::error_code EC;
532  llvm::raw_fd_ostream OS(FileName, EC,
533                          llvm::sys::fs::OpenFlags::OF_TextWithCRLF);
534
535  if (EC)
536    return EC;
537
538  OS << Contents;
539
540  if (OS.has_error())
541    return make_error_code(errc::io_error);
542
543  return EC;
544}
545
546bool llvm::sys::commandLineFitsWithinSystemLimits(StringRef Program,
547                                                  ArrayRef<StringRef> Args) {
548  static long ArgMax = sysconf(_SC_ARG_MAX);
549  // POSIX requires that _POSIX_ARG_MAX is 4096, which is the lowest possible
550  // value for ARG_MAX on a POSIX compliant system.
551  static long ArgMin = _POSIX_ARG_MAX;
552
553  // This the same baseline used by xargs.
554  long EffectiveArgMax = 128 * 1024;
555
556  if (EffectiveArgMax > ArgMax)
557    EffectiveArgMax = ArgMax;
558  else if (EffectiveArgMax < ArgMin)
559    EffectiveArgMax = ArgMin;
560
561  // System says no practical limit.
562  if (ArgMax == -1)
563    return true;
564
565  // Conservatively account for space required by environment variables.
566  long HalfArgMax = EffectiveArgMax / 2;
567
568  size_t ArgLength = Program.size() + 1;
569  for (StringRef Arg : Args) {
570    // Ensure that we do not exceed the MAX_ARG_STRLEN constant on Linux, which
571    // does not have a constant unlike what the man pages would have you
572    // believe. Since this limit is pretty high, perform the check
573    // unconditionally rather than trying to be aggressive and limiting it to
574    // Linux only.
575    if (Arg.size() >= (32 * 4096))
576      return false;
577
578    ArgLength += Arg.size() + 1;
579    if (ArgLength > size_t(HalfArgMax)) {
580      return false;
581    }
582  }
583
584  return true;
585}
586