1 //===- Job.cpp - Command to Execute ---------------------------------------===//
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 #include "clang/Driver/Job.h"
10 #include "clang/Basic/LLVM.h"
11 #include "clang/Driver/Driver.h"
12 #include "clang/Driver/DriverDiagnostic.h"
13 #include "clang/Driver/InputInfo.h"
14 #include "clang/Driver/Tool.h"
15 #include "clang/Driver/ToolChain.h"
16 #include "llvm/ADT/ArrayRef.h"
17 #include "llvm/ADT/SmallString.h"
18 #include "llvm/ADT/SmallVector.h"
19 #include "llvm/ADT/StringRef.h"
20 #include "llvm/ADT/StringSet.h"
21 #include "llvm/ADT/StringSwitch.h"
22 #include "llvm/Support/CrashRecoveryContext.h"
23 #include "llvm/Support/FileSystem.h"
24 #include "llvm/Support/Path.h"
25 #include "llvm/Support/PrettyStackTrace.h"
26 #include "llvm/Support/Program.h"
27 #include "llvm/Support/raw_ostream.h"
28 #include <algorithm>
29 #include <cassert>
30 #include <cstddef>
31 #include <string>
32 #include <system_error>
33 #include <utility>
34 
35 using namespace clang;
36 using namespace driver;
37 
38 Command::Command(const Action &Source, const Tool &Creator,
39                  ResponseFileSupport ResponseSupport, const char *Executable,
40                  const llvm::opt::ArgStringList &Arguments,
41                  ArrayRef<InputInfo> Inputs, ArrayRef<InputInfo> Outputs)
42     : Source(Source), Creator(Creator), ResponseSupport(ResponseSupport),
43       Executable(Executable), Arguments(Arguments) {
44   for (const auto &II : Inputs)
45     if (II.isFilename())
46       InputInfoList.push_back(II);
47   for (const auto &II : Outputs)
48     if (II.isFilename())
49       OutputFilenames.push_back(II.getFilename());
50 }
51 
52 /// Check if the compiler flag in question should be skipped when
53 /// emitting a reproducer. Also track how many arguments it has and if the
54 /// option is some kind of include path.
55 static bool skipArgs(const char *Flag, bool HaveCrashVFS, int &SkipNum,
56                      bool &IsInclude) {
57   SkipNum = 2;
58   // These flags are all of the form -Flag <Arg> and are treated as two
59   // arguments.  Therefore, we need to skip the flag and the next argument.
60   bool ShouldSkip = llvm::StringSwitch<bool>(Flag)
61     .Cases("-MF", "-MT", "-MQ", "-serialize-diagnostic-file", true)
62     .Cases("-o", "-dependency-file", true)
63     .Cases("-fdebug-compilation-dir", "-diagnostic-log-file", true)
64     .Cases("-dwarf-debug-flags", "-ivfsoverlay", true)
65     .Default(false);
66   if (ShouldSkip)
67     return true;
68 
69   // Some include flags shouldn't be skipped if we have a crash VFS
70   IsInclude = llvm::StringSwitch<bool>(Flag)
71     .Cases("-include", "-header-include-file", true)
72     .Cases("-idirafter", "-internal-isystem", "-iwithprefix", true)
73     .Cases("-internal-externc-isystem", "-iprefix", true)
74     .Cases("-iwithprefixbefore", "-isystem", "-iquote", true)
75     .Cases("-isysroot", "-I", "-F", "-resource-dir", true)
76     .Cases("-iframework", "-include-pch", true)
77     .Default(false);
78   if (IsInclude)
79     return !HaveCrashVFS;
80 
81   // The remaining flags are treated as a single argument.
82 
83   // These flags are all of the form -Flag and have no second argument.
84   ShouldSkip = llvm::StringSwitch<bool>(Flag)
85     .Cases("-M", "-MM", "-MG", "-MP", "-MD", true)
86     .Case("-MMD", true)
87     .Default(false);
88 
89   // Match found.
90   SkipNum = 1;
91   if (ShouldSkip)
92     return true;
93 
94   // These flags are treated as a single argument (e.g., -F<Dir>).
95   StringRef FlagRef(Flag);
96   IsInclude = FlagRef.startswith("-F") || FlagRef.startswith("-I");
97   if (IsInclude)
98     return !HaveCrashVFS;
99   if (FlagRef.startswith("-fmodules-cache-path="))
100     return true;
101 
102   SkipNum = 0;
103   return false;
104 }
105 
106 void Command::writeResponseFile(raw_ostream &OS) const {
107   // In a file list, we only write the set of inputs to the response file
108   if (ResponseSupport.ResponseKind == ResponseFileSupport::RF_FileList) {
109     for (const auto *Arg : InputFileList) {
110       OS << Arg << '\n';
111     }
112     return;
113   }
114 
115   // In regular response files, we send all arguments to the response file.
116   // Wrapping all arguments in double quotes ensures that both Unix tools and
117   // Windows tools understand the response file.
118   for (const auto *Arg : Arguments) {
119     OS << '"';
120 
121     for (; *Arg != '\0'; Arg++) {
122       if (*Arg == '\"' || *Arg == '\\') {
123         OS << '\\';
124       }
125       OS << *Arg;
126     }
127 
128     OS << "\" ";
129   }
130 }
131 
132 void Command::buildArgvForResponseFile(
133     llvm::SmallVectorImpl<const char *> &Out) const {
134   // When not a file list, all arguments are sent to the response file.
135   // This leaves us to set the argv to a single parameter, requesting the tool
136   // to read the response file.
137   if (ResponseSupport.ResponseKind != ResponseFileSupport::RF_FileList) {
138     Out.push_back(Executable);
139     Out.push_back(ResponseFileFlag.c_str());
140     return;
141   }
142 
143   llvm::StringSet<> Inputs;
144   for (const auto *InputName : InputFileList)
145     Inputs.insert(InputName);
146   Out.push_back(Executable);
147   // In a file list, build args vector ignoring parameters that will go in the
148   // response file (elements of the InputFileList vector)
149   bool FirstInput = true;
150   for (const auto *Arg : Arguments) {
151     if (Inputs.count(Arg) == 0) {
152       Out.push_back(Arg);
153     } else if (FirstInput) {
154       FirstInput = false;
155       Out.push_back(ResponseSupport.ResponseFlag);
156       Out.push_back(ResponseFile);
157     }
158   }
159 }
160 
161 /// Rewrite relative include-like flag paths to absolute ones.
162 static void
163 rewriteIncludes(const llvm::ArrayRef<const char *> &Args, size_t Idx,
164                 size_t NumArgs,
165                 llvm::SmallVectorImpl<llvm::SmallString<128>> &IncFlags) {
166   using namespace llvm;
167   using namespace sys;
168 
169   auto getAbsPath = [](StringRef InInc, SmallVectorImpl<char> &OutInc) -> bool {
170     if (path::is_absolute(InInc)) // Nothing to do here...
171       return false;
172     std::error_code EC = fs::current_path(OutInc);
173     if (EC)
174       return false;
175     path::append(OutInc, InInc);
176     return true;
177   };
178 
179   SmallString<128> NewInc;
180   if (NumArgs == 1) {
181     StringRef FlagRef(Args[Idx + NumArgs - 1]);
182     assert((FlagRef.startswith("-F") || FlagRef.startswith("-I")) &&
183             "Expecting -I or -F");
184     StringRef Inc = FlagRef.slice(2, StringRef::npos);
185     if (getAbsPath(Inc, NewInc)) {
186       SmallString<128> NewArg(FlagRef.slice(0, 2));
187       NewArg += NewInc;
188       IncFlags.push_back(std::move(NewArg));
189     }
190     return;
191   }
192 
193   assert(NumArgs == 2 && "Not expecting more than two arguments");
194   StringRef Inc(Args[Idx + NumArgs - 1]);
195   if (!getAbsPath(Inc, NewInc))
196     return;
197   IncFlags.push_back(SmallString<128>(Args[Idx]));
198   IncFlags.push_back(std::move(NewInc));
199 }
200 
201 void Command::Print(raw_ostream &OS, const char *Terminator, bool Quote,
202                     CrashReportInfo *CrashInfo) const {
203   // Always quote the exe.
204   OS << ' ';
205   llvm::sys::printArg(OS, Executable, /*Quote=*/true);
206 
207   ArrayRef<const char *> Args = Arguments;
208   SmallVector<const char *, 128> ArgsRespFile;
209   if (ResponseFile != nullptr) {
210     buildArgvForResponseFile(ArgsRespFile);
211     Args = ArrayRef<const char *>(ArgsRespFile).slice(1); // no executable name
212   }
213 
214   bool HaveCrashVFS = CrashInfo && !CrashInfo->VFSPath.empty();
215   for (size_t i = 0, e = Args.size(); i < e; ++i) {
216     const char *const Arg = Args[i];
217 
218     if (CrashInfo) {
219       int NumArgs = 0;
220       bool IsInclude = false;
221       if (skipArgs(Arg, HaveCrashVFS, NumArgs, IsInclude)) {
222         i += NumArgs - 1;
223         continue;
224       }
225 
226       // Relative includes need to be expanded to absolute paths.
227       if (HaveCrashVFS && IsInclude) {
228         SmallVector<SmallString<128>, 2> NewIncFlags;
229         rewriteIncludes(Args, i, NumArgs, NewIncFlags);
230         if (!NewIncFlags.empty()) {
231           for (auto &F : NewIncFlags) {
232             OS << ' ';
233             llvm::sys::printArg(OS, F.c_str(), Quote);
234           }
235           i += NumArgs - 1;
236           continue;
237         }
238       }
239 
240       auto Found = llvm::find_if(InputInfoList, [&Arg](const InputInfo &II) {
241         return II.getFilename() == Arg;
242       });
243       if (Found != InputInfoList.end() &&
244           (i == 0 || StringRef(Args[i - 1]) != "-main-file-name")) {
245         // Replace the input file name with the crashinfo's file name.
246         OS << ' ';
247         StringRef ShortName = llvm::sys::path::filename(CrashInfo->Filename);
248         llvm::sys::printArg(OS, ShortName.str(), Quote);
249         continue;
250       }
251     }
252 
253     OS << ' ';
254     llvm::sys::printArg(OS, Arg, Quote);
255   }
256 
257   if (CrashInfo && HaveCrashVFS) {
258     OS << ' ';
259     llvm::sys::printArg(OS, "-ivfsoverlay", Quote);
260     OS << ' ';
261     llvm::sys::printArg(OS, CrashInfo->VFSPath.str(), Quote);
262 
263     // The leftover modules from the crash are stored in
264     //  <name>.cache/vfs/modules
265     // Leave it untouched for pcm inspection and provide a clean/empty dir
266     // path to contain the future generated module cache:
267     //  <name>.cache/vfs/repro-modules
268     SmallString<128> RelModCacheDir = llvm::sys::path::parent_path(
269         llvm::sys::path::parent_path(CrashInfo->VFSPath));
270     llvm::sys::path::append(RelModCacheDir, "repro-modules");
271 
272     std::string ModCachePath = "-fmodules-cache-path=";
273     ModCachePath.append(RelModCacheDir.c_str());
274 
275     OS << ' ';
276     llvm::sys::printArg(OS, ModCachePath, Quote);
277   }
278 
279   if (ResponseFile != nullptr) {
280     OS << "\n Arguments passed via response file:\n";
281     writeResponseFile(OS);
282     // Avoiding duplicated newline terminator, since FileLists are
283     // newline-separated.
284     if (ResponseSupport.ResponseKind != ResponseFileSupport::RF_FileList)
285       OS << "\n";
286     OS << " (end of response file)";
287   }
288 
289   OS << Terminator;
290 }
291 
292 void Command::setResponseFile(const char *FileName) {
293   ResponseFile = FileName;
294   ResponseFileFlag = ResponseSupport.ResponseFlag;
295   ResponseFileFlag += FileName;
296 }
297 
298 void Command::setEnvironment(llvm::ArrayRef<const char *> NewEnvironment) {
299   Environment.reserve(NewEnvironment.size() + 1);
300   Environment.assign(NewEnvironment.begin(), NewEnvironment.end());
301   Environment.push_back(nullptr);
302 }
303 
304 void Command::setRedirectFiles(
305     const std::vector<std::optional<std::string>> &Redirects) {
306   RedirectFiles = Redirects;
307 }
308 
309 void Command::PrintFileNames() const {
310   if (PrintInputFilenames) {
311     for (const auto &Arg : InputInfoList)
312       llvm::outs() << llvm::sys::path::filename(Arg.getFilename()) << "\n";
313     llvm::outs().flush();
314   }
315 }
316 
317 int Command::Execute(ArrayRef<std::optional<StringRef>> Redirects,
318                      std::string *ErrMsg, bool *ExecutionFailed) const {
319   PrintFileNames();
320 
321   SmallVector<const char *, 128> Argv;
322   if (ResponseFile == nullptr) {
323     Argv.push_back(Executable);
324     Argv.append(Arguments.begin(), Arguments.end());
325     Argv.push_back(nullptr);
326   } else {
327     // If the command is too large, we need to put arguments in a response file.
328     std::string RespContents;
329     llvm::raw_string_ostream SS(RespContents);
330 
331     // Write file contents and build the Argv vector
332     writeResponseFile(SS);
333     buildArgvForResponseFile(Argv);
334     Argv.push_back(nullptr);
335     SS.flush();
336 
337     // Save the response file in the appropriate encoding
338     if (std::error_code EC = writeFileWithEncoding(
339             ResponseFile, RespContents, ResponseSupport.ResponseEncoding)) {
340       if (ErrMsg)
341         *ErrMsg = EC.message();
342       if (ExecutionFailed)
343         *ExecutionFailed = true;
344       // Return -1 by convention (see llvm/include/llvm/Support/Program.h) to
345       // indicate the requested executable cannot be started.
346       return -1;
347     }
348   }
349 
350   std::optional<ArrayRef<StringRef>> Env;
351   std::vector<StringRef> ArgvVectorStorage;
352   if (!Environment.empty()) {
353     assert(Environment.back() == nullptr &&
354            "Environment vector should be null-terminated by now");
355     ArgvVectorStorage = llvm::toStringRefArray(Environment.data());
356     Env = ArrayRef(ArgvVectorStorage);
357   }
358 
359   auto Args = llvm::toStringRefArray(Argv.data());
360 
361   // Use Job-specific redirect files if they are present.
362   if (!RedirectFiles.empty()) {
363     std::vector<std::optional<StringRef>> RedirectFilesOptional;
364     for (const auto &Ele : RedirectFiles)
365       if (Ele)
366         RedirectFilesOptional.push_back(std::optional<StringRef>(*Ele));
367       else
368         RedirectFilesOptional.push_back(std::nullopt);
369 
370     return llvm::sys::ExecuteAndWait(Executable, Args, Env,
371                                      ArrayRef(RedirectFilesOptional),
372                                      /*secondsToWait=*/0, /*memoryLimit=*/0,
373                                      ErrMsg, ExecutionFailed, &ProcStat);
374   }
375 
376   return llvm::sys::ExecuteAndWait(Executable, Args, Env, Redirects,
377                                    /*secondsToWait*/ 0, /*memoryLimit*/ 0,
378                                    ErrMsg, ExecutionFailed, &ProcStat);
379 }
380 
381 CC1Command::CC1Command(const Action &Source, const Tool &Creator,
382                        ResponseFileSupport ResponseSupport,
383                        const char *Executable,
384                        const llvm::opt::ArgStringList &Arguments,
385                        ArrayRef<InputInfo> Inputs, ArrayRef<InputInfo> Outputs)
386     : Command(Source, Creator, ResponseSupport, Executable, Arguments, Inputs,
387               Outputs) {
388   InProcess = true;
389 }
390 
391 void CC1Command::Print(raw_ostream &OS, const char *Terminator, bool Quote,
392                        CrashReportInfo *CrashInfo) const {
393   if (InProcess)
394     OS << " (in-process)\n";
395   Command::Print(OS, Terminator, Quote, CrashInfo);
396 }
397 
398 int CC1Command::Execute(ArrayRef<std::optional<StringRef>> Redirects,
399                         std::string *ErrMsg, bool *ExecutionFailed) const {
400   // FIXME: Currently, if there're more than one job, we disable
401   // -fintegrate-cc1. If we're no longer a integrated-cc1 job, fallback to
402   // out-of-process execution. See discussion in https://reviews.llvm.org/D74447
403   if (!InProcess)
404     return Command::Execute(Redirects, ErrMsg, ExecutionFailed);
405 
406   PrintFileNames();
407 
408   SmallVector<const char *, 128> Argv;
409   Argv.push_back(getExecutable());
410   Argv.append(getArguments().begin(), getArguments().end());
411   Argv.push_back(nullptr);
412   Argv.pop_back(); // The terminating null element shall not be part of the
413                    // slice (main() behavior).
414 
415   // This flag simply indicates that the program couldn't start, which isn't
416   // applicable here.
417   if (ExecutionFailed)
418     *ExecutionFailed = false;
419 
420   llvm::CrashRecoveryContext CRC;
421   CRC.DumpStackAndCleanupOnFailure = true;
422 
423   const void *PrettyState = llvm::SavePrettyStackState();
424   const Driver &D = getCreator().getToolChain().getDriver();
425 
426   int R = 0;
427   // Enter ExecuteCC1Tool() instead of starting up a new process
428   if (!CRC.RunSafely([&]() { R = D.CC1Main(Argv); })) {
429     llvm::RestorePrettyStackState(PrettyState);
430     return CRC.RetCode;
431   }
432   return R;
433 }
434 
435 void CC1Command::setEnvironment(llvm::ArrayRef<const char *> NewEnvironment) {
436   // We don't support set a new environment when calling into ExecuteCC1Tool()
437   llvm_unreachable(
438       "The CC1Command doesn't support changing the environment vars!");
439 }
440 
441 ForceSuccessCommand::ForceSuccessCommand(
442     const Action &Source_, const Tool &Creator_,
443     ResponseFileSupport ResponseSupport, const char *Executable_,
444     const llvm::opt::ArgStringList &Arguments_, ArrayRef<InputInfo> Inputs,
445     ArrayRef<InputInfo> Outputs)
446     : Command(Source_, Creator_, ResponseSupport, Executable_, Arguments_,
447               Inputs, Outputs) {}
448 
449 void ForceSuccessCommand::Print(raw_ostream &OS, const char *Terminator,
450                             bool Quote, CrashReportInfo *CrashInfo) const {
451   Command::Print(OS, "", Quote, CrashInfo);
452   OS << " || (exit 0)" << Terminator;
453 }
454 
455 int ForceSuccessCommand::Execute(ArrayRef<std::optional<StringRef>> Redirects,
456                                  std::string *ErrMsg,
457                                  bool *ExecutionFailed) const {
458   int Status = Command::Execute(Redirects, ErrMsg, ExecutionFailed);
459   (void)Status;
460   if (ExecutionFailed)
461     *ExecutionFailed = false;
462   return 0;
463 }
464 
465 void JobList::Print(raw_ostream &OS, const char *Terminator, bool Quote,
466                     CrashReportInfo *CrashInfo) const {
467   for (const auto &Job : *this)
468     Job.Print(OS, Terminator, Quote, CrashInfo);
469 }
470 
471 void JobList::clear() { Jobs.clear(); }
472