1 //===- VirtualFileSystem.cpp - Virtual File System Layer ------------------===//
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 VirtualFileSystem interface.
10 //
11 //===----------------------------------------------------------------------===//
12
13 #include "llvm/Support/VirtualFileSystem.h"
14 #include "llvm/ADT/ArrayRef.h"
15 #include "llvm/ADT/DenseMap.h"
16 #include "llvm/ADT/IntrusiveRefCntPtr.h"
17 #include "llvm/ADT/None.h"
18 #include "llvm/ADT/Optional.h"
19 #include "llvm/ADT/STLExtras.h"
20 #include "llvm/ADT/SmallString.h"
21 #include "llvm/ADT/SmallVector.h"
22 #include "llvm/ADT/StringRef.h"
23 #include "llvm/ADT/StringSet.h"
24 #include "llvm/ADT/Twine.h"
25 #include "llvm/ADT/iterator_range.h"
26 #include "llvm/Config/llvm-config.h"
27 #include "llvm/Support/Casting.h"
28 #include "llvm/Support/Chrono.h"
29 #include "llvm/Support/Compiler.h"
30 #include "llvm/Support/Debug.h"
31 #include "llvm/Support/Errc.h"
32 #include "llvm/Support/ErrorHandling.h"
33 #include "llvm/Support/ErrorOr.h"
34 #include "llvm/Support/FileSystem.h"
35 #include "llvm/Support/MemoryBuffer.h"
36 #include "llvm/Support/Path.h"
37 #include "llvm/Support/Process.h"
38 #include "llvm/Support/SMLoc.h"
39 #include "llvm/Support/SourceMgr.h"
40 #include "llvm/Support/YAMLParser.h"
41 #include "llvm/Support/raw_ostream.h"
42 #include <algorithm>
43 #include <atomic>
44 #include <cassert>
45 #include <cstdint>
46 #include <iterator>
47 #include <limits>
48 #include <map>
49 #include <memory>
50 #include <mutex>
51 #include <string>
52 #include <system_error>
53 #include <utility>
54 #include <vector>
55
56 using namespace llvm;
57 using namespace llvm::vfs;
58
59 using llvm::sys::fs::file_t;
60 using llvm::sys::fs::file_status;
61 using llvm::sys::fs::file_type;
62 using llvm::sys::fs::kInvalidFile;
63 using llvm::sys::fs::perms;
64 using llvm::sys::fs::UniqueID;
65
Status(const file_status & Status)66 Status::Status(const file_status &Status)
67 : UID(Status.getUniqueID()), MTime(Status.getLastModificationTime()),
68 User(Status.getUser()), Group(Status.getGroup()), Size(Status.getSize()),
69 Type(Status.type()), Perms(Status.permissions()) {}
70
Status(const Twine & Name,UniqueID UID,sys::TimePoint<> MTime,uint32_t User,uint32_t Group,uint64_t Size,file_type Type,perms Perms)71 Status::Status(const Twine &Name, UniqueID UID, sys::TimePoint<> MTime,
72 uint32_t User, uint32_t Group, uint64_t Size, file_type Type,
73 perms Perms)
74 : Name(Name.str()), UID(UID), MTime(MTime), User(User), Group(Group),
75 Size(Size), Type(Type), Perms(Perms) {}
76
copyWithNewName(const Status & In,const Twine & NewName)77 Status Status::copyWithNewName(const Status &In, const Twine &NewName) {
78 return Status(NewName, In.getUniqueID(), In.getLastModificationTime(),
79 In.getUser(), In.getGroup(), In.getSize(), In.getType(),
80 In.getPermissions());
81 }
82
copyWithNewName(const file_status & In,const Twine & NewName)83 Status Status::copyWithNewName(const file_status &In, const Twine &NewName) {
84 return Status(NewName, In.getUniqueID(), In.getLastModificationTime(),
85 In.getUser(), In.getGroup(), In.getSize(), In.type(),
86 In.permissions());
87 }
88
equivalent(const Status & Other) const89 bool Status::equivalent(const Status &Other) const {
90 assert(isStatusKnown() && Other.isStatusKnown());
91 return getUniqueID() == Other.getUniqueID();
92 }
93
isDirectory() const94 bool Status::isDirectory() const { return Type == file_type::directory_file; }
95
isRegularFile() const96 bool Status::isRegularFile() const { return Type == file_type::regular_file; }
97
isOther() const98 bool Status::isOther() const {
99 return exists() && !isRegularFile() && !isDirectory() && !isSymlink();
100 }
101
isSymlink() const102 bool Status::isSymlink() const { return Type == file_type::symlink_file; }
103
isStatusKnown() const104 bool Status::isStatusKnown() const { return Type != file_type::status_error; }
105
exists() const106 bool Status::exists() const {
107 return isStatusKnown() && Type != file_type::file_not_found;
108 }
109
110 File::~File() = default;
111
112 FileSystem::~FileSystem() = default;
113
114 ErrorOr<std::unique_ptr<MemoryBuffer>>
getBufferForFile(const llvm::Twine & Name,int64_t FileSize,bool RequiresNullTerminator,bool IsVolatile)115 FileSystem::getBufferForFile(const llvm::Twine &Name, int64_t FileSize,
116 bool RequiresNullTerminator, bool IsVolatile) {
117 auto F = openFileForRead(Name);
118 if (!F)
119 return F.getError();
120
121 return (*F)->getBuffer(Name, FileSize, RequiresNullTerminator, IsVolatile);
122 }
123
makeAbsolute(SmallVectorImpl<char> & Path) const124 std::error_code FileSystem::makeAbsolute(SmallVectorImpl<char> &Path) const {
125 if (llvm::sys::path::is_absolute(Path))
126 return {};
127
128 auto WorkingDir = getCurrentWorkingDirectory();
129 if (!WorkingDir)
130 return WorkingDir.getError();
131
132 llvm::sys::fs::make_absolute(WorkingDir.get(), Path);
133 return {};
134 }
135
getRealPath(const Twine & Path,SmallVectorImpl<char> & Output) const136 std::error_code FileSystem::getRealPath(const Twine &Path,
137 SmallVectorImpl<char> &Output) const {
138 return errc::operation_not_permitted;
139 }
140
isLocal(const Twine & Path,bool & Result)141 std::error_code FileSystem::isLocal(const Twine &Path, bool &Result) {
142 return errc::operation_not_permitted;
143 }
144
exists(const Twine & Path)145 bool FileSystem::exists(const Twine &Path) {
146 auto Status = status(Path);
147 return Status && Status->exists();
148 }
149
150 #ifndef NDEBUG
isTraversalComponent(StringRef Component)151 static bool isTraversalComponent(StringRef Component) {
152 return Component.equals("..") || Component.equals(".");
153 }
154
pathHasTraversal(StringRef Path)155 static bool pathHasTraversal(StringRef Path) {
156 using namespace llvm::sys;
157
158 for (StringRef Comp : llvm::make_range(path::begin(Path), path::end(Path)))
159 if (isTraversalComponent(Comp))
160 return true;
161 return false;
162 }
163 #endif
164
165 //===-----------------------------------------------------------------------===/
166 // RealFileSystem implementation
167 //===-----------------------------------------------------------------------===/
168
169 namespace {
170
171 /// Wrapper around a raw file descriptor.
172 class RealFile : public File {
173 friend class RealFileSystem;
174
175 file_t FD;
176 Status S;
177 std::string RealName;
178
RealFile(file_t RawFD,StringRef NewName,StringRef NewRealPathName)179 RealFile(file_t RawFD, StringRef NewName, StringRef NewRealPathName)
180 : FD(RawFD), S(NewName, {}, {}, {}, {}, {},
181 llvm::sys::fs::file_type::status_error, {}),
182 RealName(NewRealPathName.str()) {
183 assert(FD != kInvalidFile && "Invalid or inactive file descriptor");
184 }
185
186 public:
187 ~RealFile() override;
188
189 ErrorOr<Status> status() override;
190 ErrorOr<std::string> getName() override;
191 ErrorOr<std::unique_ptr<MemoryBuffer>> getBuffer(const Twine &Name,
192 int64_t FileSize,
193 bool RequiresNullTerminator,
194 bool IsVolatile) override;
195 std::error_code close() override;
196 };
197
198 } // namespace
199
~RealFile()200 RealFile::~RealFile() { close(); }
201
status()202 ErrorOr<Status> RealFile::status() {
203 assert(FD != kInvalidFile && "cannot stat closed file");
204 if (!S.isStatusKnown()) {
205 file_status RealStatus;
206 if (std::error_code EC = sys::fs::status(FD, RealStatus))
207 return EC;
208 S = Status::copyWithNewName(RealStatus, S.getName());
209 }
210 return S;
211 }
212
getName()213 ErrorOr<std::string> RealFile::getName() {
214 return RealName.empty() ? S.getName().str() : RealName;
215 }
216
217 ErrorOr<std::unique_ptr<MemoryBuffer>>
getBuffer(const Twine & Name,int64_t FileSize,bool RequiresNullTerminator,bool IsVolatile)218 RealFile::getBuffer(const Twine &Name, int64_t FileSize,
219 bool RequiresNullTerminator, bool IsVolatile) {
220 assert(FD != kInvalidFile && "cannot get buffer for closed file");
221 return MemoryBuffer::getOpenFile(FD, Name, FileSize, RequiresNullTerminator,
222 IsVolatile);
223 }
224
close()225 std::error_code RealFile::close() {
226 std::error_code EC = sys::fs::closeFile(FD);
227 FD = kInvalidFile;
228 return EC;
229 }
230
231 namespace {
232
233 /// A file system according to your operating system.
234 /// This may be linked to the process's working directory, or maintain its own.
235 ///
236 /// Currently, its own working directory is emulated by storing the path and
237 /// sending absolute paths to llvm::sys::fs:: functions.
238 /// A more principled approach would be to push this down a level, modelling
239 /// the working dir as an llvm::sys::fs::WorkingDir or similar.
240 /// This would enable the use of openat()-style functions on some platforms.
241 class RealFileSystem : public FileSystem {
242 public:
RealFileSystem(bool LinkCWDToProcess)243 explicit RealFileSystem(bool LinkCWDToProcess) {
244 if (!LinkCWDToProcess) {
245 SmallString<128> PWD, RealPWD;
246 if (llvm::sys::fs::current_path(PWD))
247 return; // Awful, but nothing to do here.
248 if (llvm::sys::fs::real_path(PWD, RealPWD))
249 WD = {PWD, PWD};
250 else
251 WD = {PWD, RealPWD};
252 }
253 }
254
255 ErrorOr<Status> status(const Twine &Path) override;
256 ErrorOr<std::unique_ptr<File>> openFileForRead(const Twine &Path) override;
257 directory_iterator dir_begin(const Twine &Dir, std::error_code &EC) override;
258
259 llvm::ErrorOr<std::string> getCurrentWorkingDirectory() const override;
260 std::error_code setCurrentWorkingDirectory(const Twine &Path) override;
261 std::error_code isLocal(const Twine &Path, bool &Result) override;
262 std::error_code getRealPath(const Twine &Path,
263 SmallVectorImpl<char> &Output) const override;
264
265 private:
266 // If this FS has its own working dir, use it to make Path absolute.
267 // The returned twine is safe to use as long as both Storage and Path live.
adjustPath(const Twine & Path,SmallVectorImpl<char> & Storage) const268 Twine adjustPath(const Twine &Path, SmallVectorImpl<char> &Storage) const {
269 if (!WD)
270 return Path;
271 Path.toVector(Storage);
272 sys::fs::make_absolute(WD->Resolved, Storage);
273 return Storage;
274 }
275
276 struct WorkingDirectory {
277 // The current working directory, without symlinks resolved. (echo $PWD).
278 SmallString<128> Specified;
279 // The current working directory, with links resolved. (readlink .).
280 SmallString<128> Resolved;
281 };
282 Optional<WorkingDirectory> WD;
283 };
284
285 } // namespace
286
status(const Twine & Path)287 ErrorOr<Status> RealFileSystem::status(const Twine &Path) {
288 SmallString<256> Storage;
289 sys::fs::file_status RealStatus;
290 if (std::error_code EC =
291 sys::fs::status(adjustPath(Path, Storage), RealStatus))
292 return EC;
293 return Status::copyWithNewName(RealStatus, Path);
294 }
295
296 ErrorOr<std::unique_ptr<File>>
openFileForRead(const Twine & Name)297 RealFileSystem::openFileForRead(const Twine &Name) {
298 SmallString<256> RealName, Storage;
299 Expected<file_t> FDOrErr = sys::fs::openNativeFileForRead(
300 adjustPath(Name, Storage), sys::fs::OF_None, &RealName);
301 if (!FDOrErr)
302 return errorToErrorCode(FDOrErr.takeError());
303 return std::unique_ptr<File>(
304 new RealFile(*FDOrErr, Name.str(), RealName.str()));
305 }
306
getCurrentWorkingDirectory() const307 llvm::ErrorOr<std::string> RealFileSystem::getCurrentWorkingDirectory() const {
308 if (WD)
309 return std::string(WD->Specified.str());
310
311 SmallString<128> Dir;
312 if (std::error_code EC = llvm::sys::fs::current_path(Dir))
313 return EC;
314 return std::string(Dir.str());
315 }
316
setCurrentWorkingDirectory(const Twine & Path)317 std::error_code RealFileSystem::setCurrentWorkingDirectory(const Twine &Path) {
318 if (!WD)
319 return llvm::sys::fs::set_current_path(Path);
320
321 SmallString<128> Absolute, Resolved, Storage;
322 adjustPath(Path, Storage).toVector(Absolute);
323 bool IsDir;
324 if (auto Err = llvm::sys::fs::is_directory(Absolute, IsDir))
325 return Err;
326 if (!IsDir)
327 return std::make_error_code(std::errc::not_a_directory);
328 if (auto Err = llvm::sys::fs::real_path(Absolute, Resolved))
329 return Err;
330 WD = {Absolute, Resolved};
331 return std::error_code();
332 }
333
isLocal(const Twine & Path,bool & Result)334 std::error_code RealFileSystem::isLocal(const Twine &Path, bool &Result) {
335 SmallString<256> Storage;
336 return llvm::sys::fs::is_local(adjustPath(Path, Storage), Result);
337 }
338
339 std::error_code
getRealPath(const Twine & Path,SmallVectorImpl<char> & Output) const340 RealFileSystem::getRealPath(const Twine &Path,
341 SmallVectorImpl<char> &Output) const {
342 SmallString<256> Storage;
343 return llvm::sys::fs::real_path(adjustPath(Path, Storage), Output);
344 }
345
getRealFileSystem()346 IntrusiveRefCntPtr<FileSystem> vfs::getRealFileSystem() {
347 static IntrusiveRefCntPtr<FileSystem> FS(new RealFileSystem(true));
348 return FS;
349 }
350
createPhysicalFileSystem()351 std::unique_ptr<FileSystem> vfs::createPhysicalFileSystem() {
352 return std::make_unique<RealFileSystem>(false);
353 }
354
355 namespace {
356
357 class RealFSDirIter : public llvm::vfs::detail::DirIterImpl {
358 llvm::sys::fs::directory_iterator Iter;
359
360 public:
RealFSDirIter(const Twine & Path,std::error_code & EC)361 RealFSDirIter(const Twine &Path, std::error_code &EC) : Iter(Path, EC) {
362 if (Iter != llvm::sys::fs::directory_iterator())
363 CurrentEntry = directory_entry(Iter->path(), Iter->type());
364 }
365
increment()366 std::error_code increment() override {
367 std::error_code EC;
368 Iter.increment(EC);
369 CurrentEntry = (Iter == llvm::sys::fs::directory_iterator())
370 ? directory_entry()
371 : directory_entry(Iter->path(), Iter->type());
372 return EC;
373 }
374 };
375
376 } // namespace
377
dir_begin(const Twine & Dir,std::error_code & EC)378 directory_iterator RealFileSystem::dir_begin(const Twine &Dir,
379 std::error_code &EC) {
380 SmallString<128> Storage;
381 return directory_iterator(
382 std::make_shared<RealFSDirIter>(adjustPath(Dir, Storage), EC));
383 }
384
385 //===-----------------------------------------------------------------------===/
386 // OverlayFileSystem implementation
387 //===-----------------------------------------------------------------------===/
388
OverlayFileSystem(IntrusiveRefCntPtr<FileSystem> BaseFS)389 OverlayFileSystem::OverlayFileSystem(IntrusiveRefCntPtr<FileSystem> BaseFS) {
390 FSList.push_back(std::move(BaseFS));
391 }
392
pushOverlay(IntrusiveRefCntPtr<FileSystem> FS)393 void OverlayFileSystem::pushOverlay(IntrusiveRefCntPtr<FileSystem> FS) {
394 FSList.push_back(FS);
395 // Synchronize added file systems by duplicating the working directory from
396 // the first one in the list.
397 FS->setCurrentWorkingDirectory(getCurrentWorkingDirectory().get());
398 }
399
status(const Twine & Path)400 ErrorOr<Status> OverlayFileSystem::status(const Twine &Path) {
401 // FIXME: handle symlinks that cross file systems
402 for (iterator I = overlays_begin(), E = overlays_end(); I != E; ++I) {
403 ErrorOr<Status> Status = (*I)->status(Path);
404 if (Status || Status.getError() != llvm::errc::no_such_file_or_directory)
405 return Status;
406 }
407 return make_error_code(llvm::errc::no_such_file_or_directory);
408 }
409
410 ErrorOr<std::unique_ptr<File>>
openFileForRead(const llvm::Twine & Path)411 OverlayFileSystem::openFileForRead(const llvm::Twine &Path) {
412 // FIXME: handle symlinks that cross file systems
413 for (iterator I = overlays_begin(), E = overlays_end(); I != E; ++I) {
414 auto Result = (*I)->openFileForRead(Path);
415 if (Result || Result.getError() != llvm::errc::no_such_file_or_directory)
416 return Result;
417 }
418 return make_error_code(llvm::errc::no_such_file_or_directory);
419 }
420
421 llvm::ErrorOr<std::string>
getCurrentWorkingDirectory() const422 OverlayFileSystem::getCurrentWorkingDirectory() const {
423 // All file systems are synchronized, just take the first working directory.
424 return FSList.front()->getCurrentWorkingDirectory();
425 }
426
427 std::error_code
setCurrentWorkingDirectory(const Twine & Path)428 OverlayFileSystem::setCurrentWorkingDirectory(const Twine &Path) {
429 for (auto &FS : FSList)
430 if (std::error_code EC = FS->setCurrentWorkingDirectory(Path))
431 return EC;
432 return {};
433 }
434
isLocal(const Twine & Path,bool & Result)435 std::error_code OverlayFileSystem::isLocal(const Twine &Path, bool &Result) {
436 for (auto &FS : FSList)
437 if (FS->exists(Path))
438 return FS->isLocal(Path, Result);
439 return errc::no_such_file_or_directory;
440 }
441
442 std::error_code
getRealPath(const Twine & Path,SmallVectorImpl<char> & Output) const443 OverlayFileSystem::getRealPath(const Twine &Path,
444 SmallVectorImpl<char> &Output) const {
445 for (auto &FS : FSList)
446 if (FS->exists(Path))
447 return FS->getRealPath(Path, Output);
448 return errc::no_such_file_or_directory;
449 }
450
451 llvm::vfs::detail::DirIterImpl::~DirIterImpl() = default;
452
453 namespace {
454
455 class OverlayFSDirIterImpl : public llvm::vfs::detail::DirIterImpl {
456 OverlayFileSystem &Overlays;
457 std::string Path;
458 OverlayFileSystem::iterator CurrentFS;
459 directory_iterator CurrentDirIter;
460 llvm::StringSet<> SeenNames;
461
incrementFS()462 std::error_code incrementFS() {
463 assert(CurrentFS != Overlays.overlays_end() && "incrementing past end");
464 ++CurrentFS;
465 for (auto E = Overlays.overlays_end(); CurrentFS != E; ++CurrentFS) {
466 std::error_code EC;
467 CurrentDirIter = (*CurrentFS)->dir_begin(Path, EC);
468 if (EC && EC != errc::no_such_file_or_directory)
469 return EC;
470 if (CurrentDirIter != directory_iterator())
471 break; // found
472 }
473 return {};
474 }
475
incrementDirIter(bool IsFirstTime)476 std::error_code incrementDirIter(bool IsFirstTime) {
477 assert((IsFirstTime || CurrentDirIter != directory_iterator()) &&
478 "incrementing past end");
479 std::error_code EC;
480 if (!IsFirstTime)
481 CurrentDirIter.increment(EC);
482 if (!EC && CurrentDirIter == directory_iterator())
483 EC = incrementFS();
484 return EC;
485 }
486
incrementImpl(bool IsFirstTime)487 std::error_code incrementImpl(bool IsFirstTime) {
488 while (true) {
489 std::error_code EC = incrementDirIter(IsFirstTime);
490 if (EC || CurrentDirIter == directory_iterator()) {
491 CurrentEntry = directory_entry();
492 return EC;
493 }
494 CurrentEntry = *CurrentDirIter;
495 StringRef Name = llvm::sys::path::filename(CurrentEntry.path());
496 if (SeenNames.insert(Name).second)
497 return EC; // name not seen before
498 }
499 llvm_unreachable("returned above");
500 }
501
502 public:
OverlayFSDirIterImpl(const Twine & Path,OverlayFileSystem & FS,std::error_code & EC)503 OverlayFSDirIterImpl(const Twine &Path, OverlayFileSystem &FS,
504 std::error_code &EC)
505 : Overlays(FS), Path(Path.str()), CurrentFS(Overlays.overlays_begin()) {
506 CurrentDirIter = (*CurrentFS)->dir_begin(Path, EC);
507 EC = incrementImpl(true);
508 }
509
increment()510 std::error_code increment() override { return incrementImpl(false); }
511 };
512
513 } // namespace
514
dir_begin(const Twine & Dir,std::error_code & EC)515 directory_iterator OverlayFileSystem::dir_begin(const Twine &Dir,
516 std::error_code &EC) {
517 return directory_iterator(
518 std::make_shared<OverlayFSDirIterImpl>(Dir, *this, EC));
519 }
520
anchor()521 void ProxyFileSystem::anchor() {}
522
523 namespace llvm {
524 namespace vfs {
525
526 namespace detail {
527
528 enum InMemoryNodeKind { IME_File, IME_Directory, IME_HardLink };
529
530 /// The in memory file system is a tree of Nodes. Every node can either be a
531 /// file , hardlink or a directory.
532 class InMemoryNode {
533 InMemoryNodeKind Kind;
534 std::string FileName;
535
536 public:
InMemoryNode(llvm::StringRef FileName,InMemoryNodeKind Kind)537 InMemoryNode(llvm::StringRef FileName, InMemoryNodeKind Kind)
538 : Kind(Kind), FileName(std::string(llvm::sys::path::filename(FileName))) {
539 }
540 virtual ~InMemoryNode() = default;
541
542 /// Get the filename of this node (the name without the directory part).
getFileName() const543 StringRef getFileName() const { return FileName; }
getKind() const544 InMemoryNodeKind getKind() const { return Kind; }
545 virtual std::string toString(unsigned Indent) const = 0;
546 };
547
548 class InMemoryFile : public InMemoryNode {
549 Status Stat;
550 std::unique_ptr<llvm::MemoryBuffer> Buffer;
551
552 public:
InMemoryFile(Status Stat,std::unique_ptr<llvm::MemoryBuffer> Buffer)553 InMemoryFile(Status Stat, std::unique_ptr<llvm::MemoryBuffer> Buffer)
554 : InMemoryNode(Stat.getName(), IME_File), Stat(std::move(Stat)),
555 Buffer(std::move(Buffer)) {}
556
557 /// Return the \p Status for this node. \p RequestedName should be the name
558 /// through which the caller referred to this node. It will override
559 /// \p Status::Name in the return value, to mimic the behavior of \p RealFile.
getStatus(const Twine & RequestedName) const560 Status getStatus(const Twine &RequestedName) const {
561 return Status::copyWithNewName(Stat, RequestedName);
562 }
getBuffer() const563 llvm::MemoryBuffer *getBuffer() const { return Buffer.get(); }
564
toString(unsigned Indent) const565 std::string toString(unsigned Indent) const override {
566 return (std::string(Indent, ' ') + Stat.getName() + "\n").str();
567 }
568
classof(const InMemoryNode * N)569 static bool classof(const InMemoryNode *N) {
570 return N->getKind() == IME_File;
571 }
572 };
573
574 namespace {
575
576 class InMemoryHardLink : public InMemoryNode {
577 const InMemoryFile &ResolvedFile;
578
579 public:
InMemoryHardLink(StringRef Path,const InMemoryFile & ResolvedFile)580 InMemoryHardLink(StringRef Path, const InMemoryFile &ResolvedFile)
581 : InMemoryNode(Path, IME_HardLink), ResolvedFile(ResolvedFile) {}
getResolvedFile() const582 const InMemoryFile &getResolvedFile() const { return ResolvedFile; }
583
toString(unsigned Indent) const584 std::string toString(unsigned Indent) const override {
585 return std::string(Indent, ' ') + "HardLink to -> " +
586 ResolvedFile.toString(0);
587 }
588
classof(const InMemoryNode * N)589 static bool classof(const InMemoryNode *N) {
590 return N->getKind() == IME_HardLink;
591 }
592 };
593
594 /// Adapt a InMemoryFile for VFS' File interface. The goal is to make
595 /// \p InMemoryFileAdaptor mimic as much as possible the behavior of
596 /// \p RealFile.
597 class InMemoryFileAdaptor : public File {
598 const InMemoryFile &Node;
599 /// The name to use when returning a Status for this file.
600 std::string RequestedName;
601
602 public:
InMemoryFileAdaptor(const InMemoryFile & Node,std::string RequestedName)603 explicit InMemoryFileAdaptor(const InMemoryFile &Node,
604 std::string RequestedName)
605 : Node(Node), RequestedName(std::move(RequestedName)) {}
606
status()607 llvm::ErrorOr<Status> status() override {
608 return Node.getStatus(RequestedName);
609 }
610
611 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
getBuffer(const Twine & Name,int64_t FileSize,bool RequiresNullTerminator,bool IsVolatile)612 getBuffer(const Twine &Name, int64_t FileSize, bool RequiresNullTerminator,
613 bool IsVolatile) override {
614 llvm::MemoryBuffer *Buf = Node.getBuffer();
615 return llvm::MemoryBuffer::getMemBuffer(
616 Buf->getBuffer(), Buf->getBufferIdentifier(), RequiresNullTerminator);
617 }
618
close()619 std::error_code close() override { return {}; }
620 };
621 } // namespace
622
623 class InMemoryDirectory : public InMemoryNode {
624 Status Stat;
625 llvm::StringMap<std::unique_ptr<InMemoryNode>> Entries;
626
627 public:
InMemoryDirectory(Status Stat)628 InMemoryDirectory(Status Stat)
629 : InMemoryNode(Stat.getName(), IME_Directory), Stat(std::move(Stat)) {}
630
631 /// Return the \p Status for this node. \p RequestedName should be the name
632 /// through which the caller referred to this node. It will override
633 /// \p Status::Name in the return value, to mimic the behavior of \p RealFile.
getStatus(const Twine & RequestedName) const634 Status getStatus(const Twine &RequestedName) const {
635 return Status::copyWithNewName(Stat, RequestedName);
636 }
getChild(StringRef Name)637 InMemoryNode *getChild(StringRef Name) {
638 auto I = Entries.find(Name);
639 if (I != Entries.end())
640 return I->second.get();
641 return nullptr;
642 }
643
addChild(StringRef Name,std::unique_ptr<InMemoryNode> Child)644 InMemoryNode *addChild(StringRef Name, std::unique_ptr<InMemoryNode> Child) {
645 return Entries.insert(make_pair(Name, std::move(Child)))
646 .first->second.get();
647 }
648
649 using const_iterator = decltype(Entries)::const_iterator;
650
begin() const651 const_iterator begin() const { return Entries.begin(); }
end() const652 const_iterator end() const { return Entries.end(); }
653
toString(unsigned Indent) const654 std::string toString(unsigned Indent) const override {
655 std::string Result =
656 (std::string(Indent, ' ') + Stat.getName() + "\n").str();
657 for (const auto &Entry : Entries)
658 Result += Entry.second->toString(Indent + 2);
659 return Result;
660 }
661
classof(const InMemoryNode * N)662 static bool classof(const InMemoryNode *N) {
663 return N->getKind() == IME_Directory;
664 }
665 };
666
667 namespace {
getNodeStatus(const InMemoryNode * Node,const Twine & RequestedName)668 Status getNodeStatus(const InMemoryNode *Node, const Twine &RequestedName) {
669 if (auto Dir = dyn_cast<detail::InMemoryDirectory>(Node))
670 return Dir->getStatus(RequestedName);
671 if (auto File = dyn_cast<detail::InMemoryFile>(Node))
672 return File->getStatus(RequestedName);
673 if (auto Link = dyn_cast<detail::InMemoryHardLink>(Node))
674 return Link->getResolvedFile().getStatus(RequestedName);
675 llvm_unreachable("Unknown node type");
676 }
677 } // namespace
678 } // namespace detail
679
InMemoryFileSystem(bool UseNormalizedPaths)680 InMemoryFileSystem::InMemoryFileSystem(bool UseNormalizedPaths)
681 : Root(new detail::InMemoryDirectory(
682 Status("", getNextVirtualUniqueID(), llvm::sys::TimePoint<>(), 0, 0,
683 0, llvm::sys::fs::file_type::directory_file,
684 llvm::sys::fs::perms::all_all))),
685 UseNormalizedPaths(UseNormalizedPaths) {}
686
687 InMemoryFileSystem::~InMemoryFileSystem() = default;
688
toString() const689 std::string InMemoryFileSystem::toString() const {
690 return Root->toString(/*Indent=*/0);
691 }
692
addFile(const Twine & P,time_t ModificationTime,std::unique_ptr<llvm::MemoryBuffer> Buffer,Optional<uint32_t> User,Optional<uint32_t> Group,Optional<llvm::sys::fs::file_type> Type,Optional<llvm::sys::fs::perms> Perms,const detail::InMemoryFile * HardLinkTarget)693 bool InMemoryFileSystem::addFile(const Twine &P, time_t ModificationTime,
694 std::unique_ptr<llvm::MemoryBuffer> Buffer,
695 Optional<uint32_t> User,
696 Optional<uint32_t> Group,
697 Optional<llvm::sys::fs::file_type> Type,
698 Optional<llvm::sys::fs::perms> Perms,
699 const detail::InMemoryFile *HardLinkTarget) {
700 SmallString<128> Path;
701 P.toVector(Path);
702
703 // Fix up relative paths. This just prepends the current working directory.
704 std::error_code EC = makeAbsolute(Path);
705 assert(!EC);
706 (void)EC;
707
708 if (useNormalizedPaths())
709 llvm::sys::path::remove_dots(Path, /*remove_dot_dot=*/true);
710
711 if (Path.empty())
712 return false;
713
714 detail::InMemoryDirectory *Dir = Root.get();
715 auto I = llvm::sys::path::begin(Path), E = sys::path::end(Path);
716 const auto ResolvedUser = User.getValueOr(0);
717 const auto ResolvedGroup = Group.getValueOr(0);
718 const auto ResolvedType = Type.getValueOr(sys::fs::file_type::regular_file);
719 const auto ResolvedPerms = Perms.getValueOr(sys::fs::all_all);
720 assert(!(HardLinkTarget && Buffer) && "HardLink cannot have a buffer");
721 // Any intermediate directories we create should be accessible by
722 // the owner, even if Perms says otherwise for the final path.
723 const auto NewDirectoryPerms = ResolvedPerms | sys::fs::owner_all;
724 while (true) {
725 StringRef Name = *I;
726 detail::InMemoryNode *Node = Dir->getChild(Name);
727 ++I;
728 if (!Node) {
729 if (I == E) {
730 // End of the path.
731 std::unique_ptr<detail::InMemoryNode> Child;
732 if (HardLinkTarget)
733 Child.reset(new detail::InMemoryHardLink(P.str(), *HardLinkTarget));
734 else {
735 // Create a new file or directory.
736 Status Stat(P.str(), getNextVirtualUniqueID(),
737 llvm::sys::toTimePoint(ModificationTime), ResolvedUser,
738 ResolvedGroup, Buffer->getBufferSize(), ResolvedType,
739 ResolvedPerms);
740 if (ResolvedType == sys::fs::file_type::directory_file) {
741 Child.reset(new detail::InMemoryDirectory(std::move(Stat)));
742 } else {
743 Child.reset(
744 new detail::InMemoryFile(std::move(Stat), std::move(Buffer)));
745 }
746 }
747 Dir->addChild(Name, std::move(Child));
748 return true;
749 }
750
751 // Create a new directory. Use the path up to here.
752 Status Stat(
753 StringRef(Path.str().begin(), Name.end() - Path.str().begin()),
754 getNextVirtualUniqueID(), llvm::sys::toTimePoint(ModificationTime),
755 ResolvedUser, ResolvedGroup, 0, sys::fs::file_type::directory_file,
756 NewDirectoryPerms);
757 Dir = cast<detail::InMemoryDirectory>(Dir->addChild(
758 Name, std::make_unique<detail::InMemoryDirectory>(std::move(Stat))));
759 continue;
760 }
761
762 if (auto *NewDir = dyn_cast<detail::InMemoryDirectory>(Node)) {
763 Dir = NewDir;
764 } else {
765 assert((isa<detail::InMemoryFile>(Node) ||
766 isa<detail::InMemoryHardLink>(Node)) &&
767 "Must be either file, hardlink or directory!");
768
769 // Trying to insert a directory in place of a file.
770 if (I != E)
771 return false;
772
773 // Return false only if the new file is different from the existing one.
774 if (auto Link = dyn_cast<detail::InMemoryHardLink>(Node)) {
775 return Link->getResolvedFile().getBuffer()->getBuffer() ==
776 Buffer->getBuffer();
777 }
778 return cast<detail::InMemoryFile>(Node)->getBuffer()->getBuffer() ==
779 Buffer->getBuffer();
780 }
781 }
782 }
783
addFile(const Twine & P,time_t ModificationTime,std::unique_ptr<llvm::MemoryBuffer> Buffer,Optional<uint32_t> User,Optional<uint32_t> Group,Optional<llvm::sys::fs::file_type> Type,Optional<llvm::sys::fs::perms> Perms)784 bool InMemoryFileSystem::addFile(const Twine &P, time_t ModificationTime,
785 std::unique_ptr<llvm::MemoryBuffer> Buffer,
786 Optional<uint32_t> User,
787 Optional<uint32_t> Group,
788 Optional<llvm::sys::fs::file_type> Type,
789 Optional<llvm::sys::fs::perms> Perms) {
790 return addFile(P, ModificationTime, std::move(Buffer), User, Group, Type,
791 Perms, /*HardLinkTarget=*/nullptr);
792 }
793
addFileNoOwn(const Twine & P,time_t ModificationTime,llvm::MemoryBuffer * Buffer,Optional<uint32_t> User,Optional<uint32_t> Group,Optional<llvm::sys::fs::file_type> Type,Optional<llvm::sys::fs::perms> Perms)794 bool InMemoryFileSystem::addFileNoOwn(const Twine &P, time_t ModificationTime,
795 llvm::MemoryBuffer *Buffer,
796 Optional<uint32_t> User,
797 Optional<uint32_t> Group,
798 Optional<llvm::sys::fs::file_type> Type,
799 Optional<llvm::sys::fs::perms> Perms) {
800 return addFile(P, ModificationTime,
801 llvm::MemoryBuffer::getMemBuffer(
802 Buffer->getBuffer(), Buffer->getBufferIdentifier()),
803 std::move(User), std::move(Group), std::move(Type),
804 std::move(Perms));
805 }
806
807 static ErrorOr<const detail::InMemoryNode *>
lookupInMemoryNode(const InMemoryFileSystem & FS,detail::InMemoryDirectory * Dir,const Twine & P)808 lookupInMemoryNode(const InMemoryFileSystem &FS, detail::InMemoryDirectory *Dir,
809 const Twine &P) {
810 SmallString<128> Path;
811 P.toVector(Path);
812
813 // Fix up relative paths. This just prepends the current working directory.
814 std::error_code EC = FS.makeAbsolute(Path);
815 assert(!EC);
816 (void)EC;
817
818 if (FS.useNormalizedPaths())
819 llvm::sys::path::remove_dots(Path, /*remove_dot_dot=*/true);
820
821 if (Path.empty())
822 return Dir;
823
824 auto I = llvm::sys::path::begin(Path), E = llvm::sys::path::end(Path);
825 while (true) {
826 detail::InMemoryNode *Node = Dir->getChild(*I);
827 ++I;
828 if (!Node)
829 return errc::no_such_file_or_directory;
830
831 // Return the file if it's at the end of the path.
832 if (auto File = dyn_cast<detail::InMemoryFile>(Node)) {
833 if (I == E)
834 return File;
835 return errc::no_such_file_or_directory;
836 }
837
838 // If Node is HardLink then return the resolved file.
839 if (auto File = dyn_cast<detail::InMemoryHardLink>(Node)) {
840 if (I == E)
841 return &File->getResolvedFile();
842 return errc::no_such_file_or_directory;
843 }
844 // Traverse directories.
845 Dir = cast<detail::InMemoryDirectory>(Node);
846 if (I == E)
847 return Dir;
848 }
849 }
850
addHardLink(const Twine & FromPath,const Twine & ToPath)851 bool InMemoryFileSystem::addHardLink(const Twine &FromPath,
852 const Twine &ToPath) {
853 auto FromNode = lookupInMemoryNode(*this, Root.get(), FromPath);
854 auto ToNode = lookupInMemoryNode(*this, Root.get(), ToPath);
855 // FromPath must not have been added before. ToPath must have been added
856 // before. Resolved ToPath must be a File.
857 if (!ToNode || FromNode || !isa<detail::InMemoryFile>(*ToNode))
858 return false;
859 return this->addFile(FromPath, 0, nullptr, None, None, None, None,
860 cast<detail::InMemoryFile>(*ToNode));
861 }
862
status(const Twine & Path)863 llvm::ErrorOr<Status> InMemoryFileSystem::status(const Twine &Path) {
864 auto Node = lookupInMemoryNode(*this, Root.get(), Path);
865 if (Node)
866 return detail::getNodeStatus(*Node, Path);
867 return Node.getError();
868 }
869
870 llvm::ErrorOr<std::unique_ptr<File>>
openFileForRead(const Twine & Path)871 InMemoryFileSystem::openFileForRead(const Twine &Path) {
872 auto Node = lookupInMemoryNode(*this, Root.get(), Path);
873 if (!Node)
874 return Node.getError();
875
876 // When we have a file provide a heap-allocated wrapper for the memory buffer
877 // to match the ownership semantics for File.
878 if (auto *F = dyn_cast<detail::InMemoryFile>(*Node))
879 return std::unique_ptr<File>(
880 new detail::InMemoryFileAdaptor(*F, Path.str()));
881
882 // FIXME: errc::not_a_file?
883 return make_error_code(llvm::errc::invalid_argument);
884 }
885
886 namespace {
887
888 /// Adaptor from InMemoryDir::iterator to directory_iterator.
889 class InMemoryDirIterator : public llvm::vfs::detail::DirIterImpl {
890 detail::InMemoryDirectory::const_iterator I;
891 detail::InMemoryDirectory::const_iterator E;
892 std::string RequestedDirName;
893
setCurrentEntry()894 void setCurrentEntry() {
895 if (I != E) {
896 SmallString<256> Path(RequestedDirName);
897 llvm::sys::path::append(Path, I->second->getFileName());
898 sys::fs::file_type Type = sys::fs::file_type::type_unknown;
899 switch (I->second->getKind()) {
900 case detail::IME_File:
901 case detail::IME_HardLink:
902 Type = sys::fs::file_type::regular_file;
903 break;
904 case detail::IME_Directory:
905 Type = sys::fs::file_type::directory_file;
906 break;
907 }
908 CurrentEntry = directory_entry(std::string(Path.str()), Type);
909 } else {
910 // When we're at the end, make CurrentEntry invalid and DirIterImpl will
911 // do the rest.
912 CurrentEntry = directory_entry();
913 }
914 }
915
916 public:
917 InMemoryDirIterator() = default;
918
InMemoryDirIterator(const detail::InMemoryDirectory & Dir,std::string RequestedDirName)919 explicit InMemoryDirIterator(const detail::InMemoryDirectory &Dir,
920 std::string RequestedDirName)
921 : I(Dir.begin()), E(Dir.end()),
922 RequestedDirName(std::move(RequestedDirName)) {
923 setCurrentEntry();
924 }
925
increment()926 std::error_code increment() override {
927 ++I;
928 setCurrentEntry();
929 return {};
930 }
931 };
932
933 } // namespace
934
dir_begin(const Twine & Dir,std::error_code & EC)935 directory_iterator InMemoryFileSystem::dir_begin(const Twine &Dir,
936 std::error_code &EC) {
937 auto Node = lookupInMemoryNode(*this, Root.get(), Dir);
938 if (!Node) {
939 EC = Node.getError();
940 return directory_iterator(std::make_shared<InMemoryDirIterator>());
941 }
942
943 if (auto *DirNode = dyn_cast<detail::InMemoryDirectory>(*Node))
944 return directory_iterator(
945 std::make_shared<InMemoryDirIterator>(*DirNode, Dir.str()));
946
947 EC = make_error_code(llvm::errc::not_a_directory);
948 return directory_iterator(std::make_shared<InMemoryDirIterator>());
949 }
950
setCurrentWorkingDirectory(const Twine & P)951 std::error_code InMemoryFileSystem::setCurrentWorkingDirectory(const Twine &P) {
952 SmallString<128> Path;
953 P.toVector(Path);
954
955 // Fix up relative paths. This just prepends the current working directory.
956 std::error_code EC = makeAbsolute(Path);
957 assert(!EC);
958 (void)EC;
959
960 if (useNormalizedPaths())
961 llvm::sys::path::remove_dots(Path, /*remove_dot_dot=*/true);
962
963 if (!Path.empty())
964 WorkingDirectory = std::string(Path.str());
965 return {};
966 }
967
968 std::error_code
getRealPath(const Twine & Path,SmallVectorImpl<char> & Output) const969 InMemoryFileSystem::getRealPath(const Twine &Path,
970 SmallVectorImpl<char> &Output) const {
971 auto CWD = getCurrentWorkingDirectory();
972 if (!CWD || CWD->empty())
973 return errc::operation_not_permitted;
974 Path.toVector(Output);
975 if (auto EC = makeAbsolute(Output))
976 return EC;
977 llvm::sys::path::remove_dots(Output, /*remove_dot_dot=*/true);
978 return {};
979 }
980
isLocal(const Twine & Path,bool & Result)981 std::error_code InMemoryFileSystem::isLocal(const Twine &Path, bool &Result) {
982 Result = false;
983 return {};
984 }
985
986 } // namespace vfs
987 } // namespace llvm
988
989 //===-----------------------------------------------------------------------===/
990 // RedirectingFileSystem implementation
991 //===-----------------------------------------------------------------------===/
992
993 namespace {
994
995 /// Removes leading "./" as well as path components like ".." and ".".
canonicalize(llvm::StringRef Path)996 static llvm::SmallString<256> canonicalize(llvm::StringRef Path) {
997 // First detect the path style in use by checking the first separator.
998 llvm::sys::path::Style style = llvm::sys::path::Style::native;
999 const size_t n = Path.find_first_of("/\\");
1000 if (n != static_cast<size_t>(-1))
1001 style = (Path[n] == '/') ? llvm::sys::path::Style::posix
1002 : llvm::sys::path::Style::windows;
1003
1004 // Now remove the dots. Explicitly specifying the path style prevents the
1005 // direction of the slashes from changing.
1006 llvm::SmallString<256> result =
1007 llvm::sys::path::remove_leading_dotslash(Path, style);
1008 llvm::sys::path::remove_dots(result, /*remove_dot_dot=*/true, style);
1009 return result;
1010 }
1011
1012 } // anonymous namespace
1013
1014
RedirectingFileSystem(IntrusiveRefCntPtr<FileSystem> FS)1015 RedirectingFileSystem::RedirectingFileSystem(IntrusiveRefCntPtr<FileSystem> FS)
1016 : ExternalFS(std::move(FS)) {
1017 if (ExternalFS)
1018 if (auto ExternalWorkingDirectory =
1019 ExternalFS->getCurrentWorkingDirectory()) {
1020 WorkingDirectory = *ExternalWorkingDirectory;
1021 ExternalFSValidWD = true;
1022 }
1023 }
1024
1025 // FIXME: reuse implementation common with OverlayFSDirIterImpl as these
1026 // iterators are conceptually similar.
1027 class llvm::vfs::VFSFromYamlDirIterImpl
1028 : public llvm::vfs::detail::DirIterImpl {
1029 std::string Dir;
1030 RedirectingFileSystem::RedirectingDirectoryEntry::iterator Current, End;
1031
1032 // To handle 'fallthrough' mode we need to iterate at first through
1033 // RedirectingDirectoryEntry and then through ExternalFS. These operations are
1034 // done sequentially, we just need to keep a track of what kind of iteration
1035 // we are currently performing.
1036
1037 /// Flag telling if we should iterate through ExternalFS or stop at the last
1038 /// RedirectingDirectoryEntry::iterator.
1039 bool IterateExternalFS;
1040 /// Flag telling if we have switched to iterating through ExternalFS.
1041 bool IsExternalFSCurrent = false;
1042 FileSystem &ExternalFS;
1043 directory_iterator ExternalDirIter;
1044 llvm::StringSet<> SeenNames;
1045
1046 /// To combine multiple iterations, different methods are responsible for
1047 /// different iteration steps.
1048 /// @{
1049
1050 /// Responsible for dispatching between RedirectingDirectoryEntry iteration
1051 /// and ExternalFS iteration.
1052 std::error_code incrementImpl(bool IsFirstTime);
1053 /// Responsible for RedirectingDirectoryEntry iteration.
1054 std::error_code incrementContent(bool IsFirstTime);
1055 /// Responsible for ExternalFS iteration.
1056 std::error_code incrementExternal();
1057 /// @}
1058
1059 public:
1060 VFSFromYamlDirIterImpl(
1061 const Twine &Path,
1062 RedirectingFileSystem::RedirectingDirectoryEntry::iterator Begin,
1063 RedirectingFileSystem::RedirectingDirectoryEntry::iterator End,
1064 bool IterateExternalFS, FileSystem &ExternalFS, std::error_code &EC);
1065
1066 std::error_code increment() override;
1067 };
1068
1069 llvm::ErrorOr<std::string>
getCurrentWorkingDirectory() const1070 RedirectingFileSystem::getCurrentWorkingDirectory() const {
1071 return WorkingDirectory;
1072 }
1073
1074 std::error_code
setCurrentWorkingDirectory(const Twine & Path)1075 RedirectingFileSystem::setCurrentWorkingDirectory(const Twine &Path) {
1076 // Don't change the working directory if the path doesn't exist.
1077 if (!exists(Path))
1078 return errc::no_such_file_or_directory;
1079
1080 // Always change the external FS but ignore its result.
1081 if (ExternalFS) {
1082 auto EC = ExternalFS->setCurrentWorkingDirectory(Path);
1083 ExternalFSValidWD = !static_cast<bool>(EC);
1084 }
1085
1086 SmallString<128> AbsolutePath;
1087 Path.toVector(AbsolutePath);
1088 if (std::error_code EC = makeAbsolute(AbsolutePath))
1089 return EC;
1090 WorkingDirectory = std::string(AbsolutePath.str());
1091 return {};
1092 }
1093
isLocal(const Twine & Path,bool & Result)1094 std::error_code RedirectingFileSystem::isLocal(const Twine &Path,
1095 bool &Result) {
1096 return ExternalFS->isLocal(Path, Result);
1097 }
1098
makeAbsolute(SmallVectorImpl<char> & Path) const1099 std::error_code RedirectingFileSystem::makeAbsolute(SmallVectorImpl<char> &Path) const {
1100 if (llvm::sys::path::is_absolute(Path, llvm::sys::path::Style::posix) ||
1101 llvm::sys::path::is_absolute(Path, llvm::sys::path::Style::windows))
1102 return {};
1103
1104 auto WorkingDir = getCurrentWorkingDirectory();
1105 if (!WorkingDir)
1106 return WorkingDir.getError();
1107
1108 // We can't use sys::fs::make_absolute because that assumes the path style
1109 // is native and there is no way to override that. Since we know WorkingDir
1110 // is absolute, we can use it to determine which style we actually have and
1111 // append Path ourselves.
1112 sys::path::Style style = sys::path::Style::windows;
1113 if (sys::path::is_absolute(WorkingDir.get(), sys::path::Style::posix)) {
1114 style = sys::path::Style::posix;
1115 }
1116
1117 std::string Result = WorkingDir.get();
1118 StringRef Dir(Result);
1119 if (!Dir.endswith(sys::path::get_separator(style))) {
1120 Result += sys::path::get_separator(style);
1121 }
1122 Result.append(Path.data(), Path.size());
1123 Path.assign(Result.begin(), Result.end());
1124
1125 return {};
1126 }
1127
dir_begin(const Twine & Dir,std::error_code & EC)1128 directory_iterator RedirectingFileSystem::dir_begin(const Twine &Dir,
1129 std::error_code &EC) {
1130 ErrorOr<RedirectingFileSystem::Entry *> E = lookupPath(Dir);
1131 if (!E) {
1132 EC = E.getError();
1133 if (shouldUseExternalFS() && EC == errc::no_such_file_or_directory)
1134 return ExternalFS->dir_begin(Dir, EC);
1135 return {};
1136 }
1137 ErrorOr<Status> S = status(Dir, *E);
1138 if (!S) {
1139 EC = S.getError();
1140 return {};
1141 }
1142 if (!S->isDirectory()) {
1143 EC = std::error_code(static_cast<int>(errc::not_a_directory),
1144 std::system_category());
1145 return {};
1146 }
1147
1148 auto *D = cast<RedirectingFileSystem::RedirectingDirectoryEntry>(*E);
1149 return directory_iterator(std::make_shared<VFSFromYamlDirIterImpl>(
1150 Dir, D->contents_begin(), D->contents_end(),
1151 /*IterateExternalFS=*/shouldUseExternalFS(), *ExternalFS, EC));
1152 }
1153
setExternalContentsPrefixDir(StringRef PrefixDir)1154 void RedirectingFileSystem::setExternalContentsPrefixDir(StringRef PrefixDir) {
1155 ExternalContentsPrefixDir = PrefixDir.str();
1156 }
1157
getExternalContentsPrefixDir() const1158 StringRef RedirectingFileSystem::getExternalContentsPrefixDir() const {
1159 return ExternalContentsPrefixDir;
1160 }
1161
dump(raw_ostream & OS) const1162 void RedirectingFileSystem::dump(raw_ostream &OS) const {
1163 for (const auto &Root : Roots)
1164 dumpEntry(OS, Root.get());
1165 }
1166
dumpEntry(raw_ostream & OS,RedirectingFileSystem::Entry * E,int NumSpaces) const1167 void RedirectingFileSystem::dumpEntry(raw_ostream &OS,
1168 RedirectingFileSystem::Entry *E,
1169 int NumSpaces) const {
1170 StringRef Name = E->getName();
1171 for (int i = 0, e = NumSpaces; i < e; ++i)
1172 OS << " ";
1173 OS << "'" << Name.str().c_str() << "'"
1174 << "\n";
1175
1176 if (E->getKind() == RedirectingFileSystem::EK_Directory) {
1177 auto *DE = dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(E);
1178 assert(DE && "Should be a directory");
1179
1180 for (std::unique_ptr<Entry> &SubEntry :
1181 llvm::make_range(DE->contents_begin(), DE->contents_end()))
1182 dumpEntry(OS, SubEntry.get(), NumSpaces + 2);
1183 }
1184 }
1185
1186 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
dump() const1187 LLVM_DUMP_METHOD void RedirectingFileSystem::dump() const { dump(dbgs()); }
1188 #endif
1189
1190 /// A helper class to hold the common YAML parsing state.
1191 class llvm::vfs::RedirectingFileSystemParser {
1192 yaml::Stream &Stream;
1193
error(yaml::Node * N,const Twine & Msg)1194 void error(yaml::Node *N, const Twine &Msg) { Stream.printError(N, Msg); }
1195
1196 // false on error
parseScalarString(yaml::Node * N,StringRef & Result,SmallVectorImpl<char> & Storage)1197 bool parseScalarString(yaml::Node *N, StringRef &Result,
1198 SmallVectorImpl<char> &Storage) {
1199 const auto *S = dyn_cast<yaml::ScalarNode>(N);
1200
1201 if (!S) {
1202 error(N, "expected string");
1203 return false;
1204 }
1205 Result = S->getValue(Storage);
1206 return true;
1207 }
1208
1209 // false on error
parseScalarBool(yaml::Node * N,bool & Result)1210 bool parseScalarBool(yaml::Node *N, bool &Result) {
1211 SmallString<5> Storage;
1212 StringRef Value;
1213 if (!parseScalarString(N, Value, Storage))
1214 return false;
1215
1216 if (Value.equals_lower("true") || Value.equals_lower("on") ||
1217 Value.equals_lower("yes") || Value == "1") {
1218 Result = true;
1219 return true;
1220 } else if (Value.equals_lower("false") || Value.equals_lower("off") ||
1221 Value.equals_lower("no") || Value == "0") {
1222 Result = false;
1223 return true;
1224 }
1225
1226 error(N, "expected boolean value");
1227 return false;
1228 }
1229
1230 struct KeyStatus {
1231 bool Required;
1232 bool Seen = false;
1233
KeyStatusllvm::vfs::RedirectingFileSystemParser::KeyStatus1234 KeyStatus(bool Required = false) : Required(Required) {}
1235 };
1236
1237 using KeyStatusPair = std::pair<StringRef, KeyStatus>;
1238
1239 // false on error
checkDuplicateOrUnknownKey(yaml::Node * KeyNode,StringRef Key,DenseMap<StringRef,KeyStatus> & Keys)1240 bool checkDuplicateOrUnknownKey(yaml::Node *KeyNode, StringRef Key,
1241 DenseMap<StringRef, KeyStatus> &Keys) {
1242 if (!Keys.count(Key)) {
1243 error(KeyNode, "unknown key");
1244 return false;
1245 }
1246 KeyStatus &S = Keys[Key];
1247 if (S.Seen) {
1248 error(KeyNode, Twine("duplicate key '") + Key + "'");
1249 return false;
1250 }
1251 S.Seen = true;
1252 return true;
1253 }
1254
1255 // false on error
checkMissingKeys(yaml::Node * Obj,DenseMap<StringRef,KeyStatus> & Keys)1256 bool checkMissingKeys(yaml::Node *Obj, DenseMap<StringRef, KeyStatus> &Keys) {
1257 for (const auto &I : Keys) {
1258 if (I.second.Required && !I.second.Seen) {
1259 error(Obj, Twine("missing key '") + I.first + "'");
1260 return false;
1261 }
1262 }
1263 return true;
1264 }
1265
1266 RedirectingFileSystem::Entry *
lookupOrCreateEntry(RedirectingFileSystem * FS,StringRef Name,RedirectingFileSystem::Entry * ParentEntry=nullptr)1267 lookupOrCreateEntry(RedirectingFileSystem *FS, StringRef Name,
1268 RedirectingFileSystem::Entry *ParentEntry = nullptr) {
1269 if (!ParentEntry) { // Look for a existent root
1270 for (const auto &Root : FS->Roots) {
1271 if (Name.equals(Root->getName())) {
1272 ParentEntry = Root.get();
1273 return ParentEntry;
1274 }
1275 }
1276 } else { // Advance to the next component
1277 auto *DE = dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(
1278 ParentEntry);
1279 for (std::unique_ptr<RedirectingFileSystem::Entry> &Content :
1280 llvm::make_range(DE->contents_begin(), DE->contents_end())) {
1281 auto *DirContent =
1282 dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(
1283 Content.get());
1284 if (DirContent && Name.equals(Content->getName()))
1285 return DirContent;
1286 }
1287 }
1288
1289 // ... or create a new one
1290 std::unique_ptr<RedirectingFileSystem::Entry> E =
1291 std::make_unique<RedirectingFileSystem::RedirectingDirectoryEntry>(
1292 Name, Status("", getNextVirtualUniqueID(),
1293 std::chrono::system_clock::now(), 0, 0, 0,
1294 file_type::directory_file, sys::fs::all_all));
1295
1296 if (!ParentEntry) { // Add a new root to the overlay
1297 FS->Roots.push_back(std::move(E));
1298 ParentEntry = FS->Roots.back().get();
1299 return ParentEntry;
1300 }
1301
1302 auto *DE =
1303 cast<RedirectingFileSystem::RedirectingDirectoryEntry>(ParentEntry);
1304 DE->addContent(std::move(E));
1305 return DE->getLastContent();
1306 }
1307
uniqueOverlayTree(RedirectingFileSystem * FS,RedirectingFileSystem::Entry * SrcE,RedirectingFileSystem::Entry * NewParentE=nullptr)1308 void uniqueOverlayTree(RedirectingFileSystem *FS,
1309 RedirectingFileSystem::Entry *SrcE,
1310 RedirectingFileSystem::Entry *NewParentE = nullptr) {
1311 StringRef Name = SrcE->getName();
1312 switch (SrcE->getKind()) {
1313 case RedirectingFileSystem::EK_Directory: {
1314 auto *DE = cast<RedirectingFileSystem::RedirectingDirectoryEntry>(SrcE);
1315 // Empty directories could be present in the YAML as a way to
1316 // describe a file for a current directory after some of its subdir
1317 // is parsed. This only leads to redundant walks, ignore it.
1318 if (!Name.empty())
1319 NewParentE = lookupOrCreateEntry(FS, Name, NewParentE);
1320 for (std::unique_ptr<RedirectingFileSystem::Entry> &SubEntry :
1321 llvm::make_range(DE->contents_begin(), DE->contents_end()))
1322 uniqueOverlayTree(FS, SubEntry.get(), NewParentE);
1323 break;
1324 }
1325 case RedirectingFileSystem::EK_File: {
1326 assert(NewParentE && "Parent entry must exist");
1327 auto *FE = cast<RedirectingFileSystem::RedirectingFileEntry>(SrcE);
1328 auto *DE =
1329 cast<RedirectingFileSystem::RedirectingDirectoryEntry>(NewParentE);
1330 DE->addContent(
1331 std::make_unique<RedirectingFileSystem::RedirectingFileEntry>(
1332 Name, FE->getExternalContentsPath(), FE->getUseName()));
1333 break;
1334 }
1335 }
1336 }
1337
1338 std::unique_ptr<RedirectingFileSystem::Entry>
parseEntry(yaml::Node * N,RedirectingFileSystem * FS,bool IsRootEntry)1339 parseEntry(yaml::Node *N, RedirectingFileSystem *FS, bool IsRootEntry) {
1340 auto *M = dyn_cast<yaml::MappingNode>(N);
1341 if (!M) {
1342 error(N, "expected mapping node for file or directory entry");
1343 return nullptr;
1344 }
1345
1346 KeyStatusPair Fields[] = {
1347 KeyStatusPair("name", true),
1348 KeyStatusPair("type", true),
1349 KeyStatusPair("contents", false),
1350 KeyStatusPair("external-contents", false),
1351 KeyStatusPair("use-external-name", false),
1352 };
1353
1354 DenseMap<StringRef, KeyStatus> Keys(std::begin(Fields), std::end(Fields));
1355
1356 bool HasContents = false; // external or otherwise
1357 std::vector<std::unique_ptr<RedirectingFileSystem::Entry>>
1358 EntryArrayContents;
1359 SmallString<256> ExternalContentsPath;
1360 SmallString<256> Name;
1361 yaml::Node *NameValueNode = nullptr;
1362 auto UseExternalName =
1363 RedirectingFileSystem::RedirectingFileEntry::NK_NotSet;
1364 RedirectingFileSystem::EntryKind Kind;
1365
1366 for (auto &I : *M) {
1367 StringRef Key;
1368 // Reuse the buffer for key and value, since we don't look at key after
1369 // parsing value.
1370 SmallString<256> Buffer;
1371 if (!parseScalarString(I.getKey(), Key, Buffer))
1372 return nullptr;
1373
1374 if (!checkDuplicateOrUnknownKey(I.getKey(), Key, Keys))
1375 return nullptr;
1376
1377 StringRef Value;
1378 if (Key == "name") {
1379 if (!parseScalarString(I.getValue(), Value, Buffer))
1380 return nullptr;
1381
1382 NameValueNode = I.getValue();
1383 // Guarantee that old YAML files containing paths with ".." and "."
1384 // are properly canonicalized before read into the VFS.
1385 Name = canonicalize(Value).str();
1386 } else if (Key == "type") {
1387 if (!parseScalarString(I.getValue(), Value, Buffer))
1388 return nullptr;
1389 if (Value == "file")
1390 Kind = RedirectingFileSystem::EK_File;
1391 else if (Value == "directory")
1392 Kind = RedirectingFileSystem::EK_Directory;
1393 else {
1394 error(I.getValue(), "unknown value for 'type'");
1395 return nullptr;
1396 }
1397 } else if (Key == "contents") {
1398 if (HasContents) {
1399 error(I.getKey(),
1400 "entry already has 'contents' or 'external-contents'");
1401 return nullptr;
1402 }
1403 HasContents = true;
1404 auto *Contents = dyn_cast<yaml::SequenceNode>(I.getValue());
1405 if (!Contents) {
1406 // FIXME: this is only for directories, what about files?
1407 error(I.getValue(), "expected array");
1408 return nullptr;
1409 }
1410
1411 for (auto &I : *Contents) {
1412 if (std::unique_ptr<RedirectingFileSystem::Entry> E =
1413 parseEntry(&I, FS, /*IsRootEntry*/ false))
1414 EntryArrayContents.push_back(std::move(E));
1415 else
1416 return nullptr;
1417 }
1418 } else if (Key == "external-contents") {
1419 if (HasContents) {
1420 error(I.getKey(),
1421 "entry already has 'contents' or 'external-contents'");
1422 return nullptr;
1423 }
1424 HasContents = true;
1425 if (!parseScalarString(I.getValue(), Value, Buffer))
1426 return nullptr;
1427
1428 SmallString<256> FullPath;
1429 if (FS->IsRelativeOverlay) {
1430 FullPath = FS->getExternalContentsPrefixDir();
1431 assert(!FullPath.empty() &&
1432 "External contents prefix directory must exist");
1433 llvm::sys::path::append(FullPath, Value);
1434 } else {
1435 FullPath = Value;
1436 }
1437
1438 // Guarantee that old YAML files containing paths with ".." and "."
1439 // are properly canonicalized before read into the VFS.
1440 FullPath = canonicalize(FullPath);
1441 ExternalContentsPath = FullPath.str();
1442 } else if (Key == "use-external-name") {
1443 bool Val;
1444 if (!parseScalarBool(I.getValue(), Val))
1445 return nullptr;
1446 UseExternalName =
1447 Val ? RedirectingFileSystem::RedirectingFileEntry::NK_External
1448 : RedirectingFileSystem::RedirectingFileEntry::NK_Virtual;
1449 } else {
1450 llvm_unreachable("key missing from Keys");
1451 }
1452 }
1453
1454 if (Stream.failed())
1455 return nullptr;
1456
1457 // check for missing keys
1458 if (!HasContents) {
1459 error(N, "missing key 'contents' or 'external-contents'");
1460 return nullptr;
1461 }
1462 if (!checkMissingKeys(N, Keys))
1463 return nullptr;
1464
1465 // check invalid configuration
1466 if (Kind == RedirectingFileSystem::EK_Directory &&
1467 UseExternalName !=
1468 RedirectingFileSystem::RedirectingFileEntry::NK_NotSet) {
1469 error(N, "'use-external-name' is not supported for directories");
1470 return nullptr;
1471 }
1472
1473 sys::path::Style path_style = sys::path::Style::native;
1474 if (IsRootEntry) {
1475 // VFS root entries may be in either Posix or Windows style. Figure out
1476 // which style we have, and use it consistently.
1477 if (sys::path::is_absolute(Name, sys::path::Style::posix)) {
1478 path_style = sys::path::Style::posix;
1479 } else if (sys::path::is_absolute(Name, sys::path::Style::windows)) {
1480 path_style = sys::path::Style::windows;
1481 } else {
1482 assert(NameValueNode && "Name presence should be checked earlier");
1483 error(NameValueNode,
1484 "entry with relative path at the root level is not discoverable");
1485 return nullptr;
1486 }
1487 }
1488
1489 // Remove trailing slash(es), being careful not to remove the root path
1490 StringRef Trimmed(Name);
1491 size_t RootPathLen = sys::path::root_path(Trimmed, path_style).size();
1492 while (Trimmed.size() > RootPathLen &&
1493 sys::path::is_separator(Trimmed.back(), path_style))
1494 Trimmed = Trimmed.slice(0, Trimmed.size() - 1);
1495
1496 // Get the last component
1497 StringRef LastComponent = sys::path::filename(Trimmed, path_style);
1498
1499 std::unique_ptr<RedirectingFileSystem::Entry> Result;
1500 switch (Kind) {
1501 case RedirectingFileSystem::EK_File:
1502 Result = std::make_unique<RedirectingFileSystem::RedirectingFileEntry>(
1503 LastComponent, std::move(ExternalContentsPath), UseExternalName);
1504 break;
1505 case RedirectingFileSystem::EK_Directory:
1506 Result =
1507 std::make_unique<RedirectingFileSystem::RedirectingDirectoryEntry>(
1508 LastComponent, std::move(EntryArrayContents),
1509 Status("", getNextVirtualUniqueID(),
1510 std::chrono::system_clock::now(), 0, 0, 0,
1511 file_type::directory_file, sys::fs::all_all));
1512 break;
1513 }
1514
1515 StringRef Parent = sys::path::parent_path(Trimmed, path_style);
1516 if (Parent.empty())
1517 return Result;
1518
1519 // if 'name' contains multiple components, create implicit directory entries
1520 for (sys::path::reverse_iterator I = sys::path::rbegin(Parent, path_style),
1521 E = sys::path::rend(Parent);
1522 I != E; ++I) {
1523 std::vector<std::unique_ptr<RedirectingFileSystem::Entry>> Entries;
1524 Entries.push_back(std::move(Result));
1525 Result =
1526 std::make_unique<RedirectingFileSystem::RedirectingDirectoryEntry>(
1527 *I, std::move(Entries),
1528 Status("", getNextVirtualUniqueID(),
1529 std::chrono::system_clock::now(), 0, 0, 0,
1530 file_type::directory_file, sys::fs::all_all));
1531 }
1532 return Result;
1533 }
1534
1535 public:
RedirectingFileSystemParser(yaml::Stream & S)1536 RedirectingFileSystemParser(yaml::Stream &S) : Stream(S) {}
1537
1538 // false on error
parse(yaml::Node * Root,RedirectingFileSystem * FS)1539 bool parse(yaml::Node *Root, RedirectingFileSystem *FS) {
1540 auto *Top = dyn_cast<yaml::MappingNode>(Root);
1541 if (!Top) {
1542 error(Root, "expected mapping node");
1543 return false;
1544 }
1545
1546 KeyStatusPair Fields[] = {
1547 KeyStatusPair("version", true),
1548 KeyStatusPair("case-sensitive", false),
1549 KeyStatusPair("use-external-names", false),
1550 KeyStatusPair("overlay-relative", false),
1551 KeyStatusPair("fallthrough", false),
1552 KeyStatusPair("roots", true),
1553 };
1554
1555 DenseMap<StringRef, KeyStatus> Keys(std::begin(Fields), std::end(Fields));
1556 std::vector<std::unique_ptr<RedirectingFileSystem::Entry>> RootEntries;
1557
1558 // Parse configuration and 'roots'
1559 for (auto &I : *Top) {
1560 SmallString<10> KeyBuffer;
1561 StringRef Key;
1562 if (!parseScalarString(I.getKey(), Key, KeyBuffer))
1563 return false;
1564
1565 if (!checkDuplicateOrUnknownKey(I.getKey(), Key, Keys))
1566 return false;
1567
1568 if (Key == "roots") {
1569 auto *Roots = dyn_cast<yaml::SequenceNode>(I.getValue());
1570 if (!Roots) {
1571 error(I.getValue(), "expected array");
1572 return false;
1573 }
1574
1575 for (auto &I : *Roots) {
1576 if (std::unique_ptr<RedirectingFileSystem::Entry> E =
1577 parseEntry(&I, FS, /*IsRootEntry*/ true))
1578 RootEntries.push_back(std::move(E));
1579 else
1580 return false;
1581 }
1582 } else if (Key == "version") {
1583 StringRef VersionString;
1584 SmallString<4> Storage;
1585 if (!parseScalarString(I.getValue(), VersionString, Storage))
1586 return false;
1587 int Version;
1588 if (VersionString.getAsInteger<int>(10, Version)) {
1589 error(I.getValue(), "expected integer");
1590 return false;
1591 }
1592 if (Version < 0) {
1593 error(I.getValue(), "invalid version number");
1594 return false;
1595 }
1596 if (Version != 0) {
1597 error(I.getValue(), "version mismatch, expected 0");
1598 return false;
1599 }
1600 } else if (Key == "case-sensitive") {
1601 if (!parseScalarBool(I.getValue(), FS->CaseSensitive))
1602 return false;
1603 } else if (Key == "overlay-relative") {
1604 if (!parseScalarBool(I.getValue(), FS->IsRelativeOverlay))
1605 return false;
1606 } else if (Key == "use-external-names") {
1607 if (!parseScalarBool(I.getValue(), FS->UseExternalNames))
1608 return false;
1609 } else if (Key == "fallthrough") {
1610 if (!parseScalarBool(I.getValue(), FS->IsFallthrough))
1611 return false;
1612 } else {
1613 llvm_unreachable("key missing from Keys");
1614 }
1615 }
1616
1617 if (Stream.failed())
1618 return false;
1619
1620 if (!checkMissingKeys(Top, Keys))
1621 return false;
1622
1623 // Now that we sucessefully parsed the YAML file, canonicalize the internal
1624 // representation to a proper directory tree so that we can search faster
1625 // inside the VFS.
1626 for (auto &E : RootEntries)
1627 uniqueOverlayTree(FS, E.get());
1628
1629 return true;
1630 }
1631 };
1632
1633 RedirectingFileSystem *
create(std::unique_ptr<MemoryBuffer> Buffer,SourceMgr::DiagHandlerTy DiagHandler,StringRef YAMLFilePath,void * DiagContext,IntrusiveRefCntPtr<FileSystem> ExternalFS)1634 RedirectingFileSystem::create(std::unique_ptr<MemoryBuffer> Buffer,
1635 SourceMgr::DiagHandlerTy DiagHandler,
1636 StringRef YAMLFilePath, void *DiagContext,
1637 IntrusiveRefCntPtr<FileSystem> ExternalFS) {
1638 SourceMgr SM;
1639 yaml::Stream Stream(Buffer->getMemBufferRef(), SM);
1640
1641 SM.setDiagHandler(DiagHandler, DiagContext);
1642 yaml::document_iterator DI = Stream.begin();
1643 yaml::Node *Root = DI->getRoot();
1644 if (DI == Stream.end() || !Root) {
1645 SM.PrintMessage(SMLoc(), SourceMgr::DK_Error, "expected root node");
1646 return nullptr;
1647 }
1648
1649 RedirectingFileSystemParser P(Stream);
1650
1651 std::unique_ptr<RedirectingFileSystem> FS(
1652 new RedirectingFileSystem(ExternalFS));
1653
1654 if (!YAMLFilePath.empty()) {
1655 // Use the YAML path from -ivfsoverlay to compute the dir to be prefixed
1656 // to each 'external-contents' path.
1657 //
1658 // Example:
1659 // -ivfsoverlay dummy.cache/vfs/vfs.yaml
1660 // yields:
1661 // FS->ExternalContentsPrefixDir => /<absolute_path_to>/dummy.cache/vfs
1662 //
1663 SmallString<256> OverlayAbsDir = sys::path::parent_path(YAMLFilePath);
1664 std::error_code EC = llvm::sys::fs::make_absolute(OverlayAbsDir);
1665 assert(!EC && "Overlay dir final path must be absolute");
1666 (void)EC;
1667 FS->setExternalContentsPrefixDir(OverlayAbsDir);
1668 }
1669
1670 if (!P.parse(Root, FS.get()))
1671 return nullptr;
1672
1673 return FS.release();
1674 }
1675
1676 ErrorOr<RedirectingFileSystem::Entry *>
lookupPath(const Twine & Path_) const1677 RedirectingFileSystem::lookupPath(const Twine &Path_) const {
1678 SmallString<256> Path;
1679 Path_.toVector(Path);
1680
1681 // Handle relative paths
1682 if (std::error_code EC = makeAbsolute(Path))
1683 return EC;
1684
1685 // Canonicalize path by removing ".", "..", "./", components. This is
1686 // a VFS request, do not bother about symlinks in the path components
1687 // but canonicalize in order to perform the correct entry search.
1688 Path = canonicalize(Path);
1689 if (Path.empty())
1690 return make_error_code(llvm::errc::invalid_argument);
1691
1692 sys::path::const_iterator Start = sys::path::begin(Path);
1693 sys::path::const_iterator End = sys::path::end(Path);
1694 for (const auto &Root : Roots) {
1695 ErrorOr<RedirectingFileSystem::Entry *> Result =
1696 lookupPath(Start, End, Root.get());
1697 if (Result || Result.getError() != llvm::errc::no_such_file_or_directory)
1698 return Result;
1699 }
1700 return make_error_code(llvm::errc::no_such_file_or_directory);
1701 }
1702
1703 ErrorOr<RedirectingFileSystem::Entry *>
lookupPath(sys::path::const_iterator Start,sys::path::const_iterator End,RedirectingFileSystem::Entry * From) const1704 RedirectingFileSystem::lookupPath(sys::path::const_iterator Start,
1705 sys::path::const_iterator End,
1706 RedirectingFileSystem::Entry *From) const {
1707 assert(!isTraversalComponent(*Start) &&
1708 !isTraversalComponent(From->getName()) &&
1709 "Paths should not contain traversal components");
1710
1711 StringRef FromName = From->getName();
1712
1713 // Forward the search to the next component in case this is an empty one.
1714 if (!FromName.empty()) {
1715 if (!pathComponentMatches(*Start, FromName))
1716 return make_error_code(llvm::errc::no_such_file_or_directory);
1717
1718 ++Start;
1719
1720 if (Start == End) {
1721 // Match!
1722 return From;
1723 }
1724 }
1725
1726 auto *DE = dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(From);
1727 if (!DE)
1728 return make_error_code(llvm::errc::not_a_directory);
1729
1730 for (const std::unique_ptr<RedirectingFileSystem::Entry> &DirEntry :
1731 llvm::make_range(DE->contents_begin(), DE->contents_end())) {
1732 ErrorOr<RedirectingFileSystem::Entry *> Result =
1733 lookupPath(Start, End, DirEntry.get());
1734 if (Result || Result.getError() != llvm::errc::no_such_file_or_directory)
1735 return Result;
1736 }
1737
1738 return make_error_code(llvm::errc::no_such_file_or_directory);
1739 }
1740
getRedirectedFileStatus(const Twine & Path,bool UseExternalNames,Status ExternalStatus)1741 static Status getRedirectedFileStatus(const Twine &Path, bool UseExternalNames,
1742 Status ExternalStatus) {
1743 Status S = ExternalStatus;
1744 if (!UseExternalNames)
1745 S = Status::copyWithNewName(S, Path);
1746 S.IsVFSMapped = true;
1747 return S;
1748 }
1749
status(const Twine & Path,RedirectingFileSystem::Entry * E)1750 ErrorOr<Status> RedirectingFileSystem::status(const Twine &Path,
1751 RedirectingFileSystem::Entry *E) {
1752 assert(E != nullptr);
1753 if (auto *F = dyn_cast<RedirectingFileSystem::RedirectingFileEntry>(E)) {
1754 ErrorOr<Status> S = ExternalFS->status(F->getExternalContentsPath());
1755 assert(!S || S->getName() == F->getExternalContentsPath());
1756 if (S)
1757 return getRedirectedFileStatus(Path, F->useExternalName(UseExternalNames),
1758 *S);
1759 return S;
1760 } else { // directory
1761 auto *DE = cast<RedirectingFileSystem::RedirectingDirectoryEntry>(E);
1762 return Status::copyWithNewName(DE->getStatus(), Path);
1763 }
1764 }
1765
status(const Twine & Path)1766 ErrorOr<Status> RedirectingFileSystem::status(const Twine &Path) {
1767 ErrorOr<RedirectingFileSystem::Entry *> Result = lookupPath(Path);
1768 if (!Result) {
1769 if (shouldUseExternalFS() &&
1770 Result.getError() == llvm::errc::no_such_file_or_directory) {
1771 return ExternalFS->status(Path);
1772 }
1773 return Result.getError();
1774 }
1775 return status(Path, *Result);
1776 }
1777
1778 namespace {
1779
1780 /// Provide a file wrapper with an overriden status.
1781 class FileWithFixedStatus : public File {
1782 std::unique_ptr<File> InnerFile;
1783 Status S;
1784
1785 public:
FileWithFixedStatus(std::unique_ptr<File> InnerFile,Status S)1786 FileWithFixedStatus(std::unique_ptr<File> InnerFile, Status S)
1787 : InnerFile(std::move(InnerFile)), S(std::move(S)) {}
1788
status()1789 ErrorOr<Status> status() override { return S; }
1790 ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
1791
getBuffer(const Twine & Name,int64_t FileSize,bool RequiresNullTerminator,bool IsVolatile)1792 getBuffer(const Twine &Name, int64_t FileSize, bool RequiresNullTerminator,
1793 bool IsVolatile) override {
1794 return InnerFile->getBuffer(Name, FileSize, RequiresNullTerminator,
1795 IsVolatile);
1796 }
1797
close()1798 std::error_code close() override { return InnerFile->close(); }
1799 };
1800
1801 } // namespace
1802
1803 ErrorOr<std::unique_ptr<File>>
openFileForRead(const Twine & Path)1804 RedirectingFileSystem::openFileForRead(const Twine &Path) {
1805 ErrorOr<RedirectingFileSystem::Entry *> E = lookupPath(Path);
1806 if (!E) {
1807 if (shouldUseExternalFS() &&
1808 E.getError() == llvm::errc::no_such_file_or_directory) {
1809 return ExternalFS->openFileForRead(Path);
1810 }
1811 return E.getError();
1812 }
1813
1814 auto *F = dyn_cast<RedirectingFileSystem::RedirectingFileEntry>(*E);
1815 if (!F) // FIXME: errc::not_a_file?
1816 return make_error_code(llvm::errc::invalid_argument);
1817
1818 auto Result = ExternalFS->openFileForRead(F->getExternalContentsPath());
1819 if (!Result)
1820 return Result;
1821
1822 auto ExternalStatus = (*Result)->status();
1823 if (!ExternalStatus)
1824 return ExternalStatus.getError();
1825
1826 // FIXME: Update the status with the name and VFSMapped.
1827 Status S = getRedirectedFileStatus(Path, F->useExternalName(UseExternalNames),
1828 *ExternalStatus);
1829 return std::unique_ptr<File>(
1830 std::make_unique<FileWithFixedStatus>(std::move(*Result), S));
1831 }
1832
1833 std::error_code
getRealPath(const Twine & Path,SmallVectorImpl<char> & Output) const1834 RedirectingFileSystem::getRealPath(const Twine &Path,
1835 SmallVectorImpl<char> &Output) const {
1836 ErrorOr<RedirectingFileSystem::Entry *> Result = lookupPath(Path);
1837 if (!Result) {
1838 if (shouldUseExternalFS() &&
1839 Result.getError() == llvm::errc::no_such_file_or_directory) {
1840 return ExternalFS->getRealPath(Path, Output);
1841 }
1842 return Result.getError();
1843 }
1844
1845 if (auto *F =
1846 dyn_cast<RedirectingFileSystem::RedirectingFileEntry>(*Result)) {
1847 return ExternalFS->getRealPath(F->getExternalContentsPath(), Output);
1848 }
1849 // Even if there is a directory entry, fall back to ExternalFS if allowed,
1850 // because directories don't have a single external contents path.
1851 return shouldUseExternalFS() ? ExternalFS->getRealPath(Path, Output)
1852 : llvm::errc::invalid_argument;
1853 }
1854
1855 IntrusiveRefCntPtr<FileSystem>
getVFSFromYAML(std::unique_ptr<MemoryBuffer> Buffer,SourceMgr::DiagHandlerTy DiagHandler,StringRef YAMLFilePath,void * DiagContext,IntrusiveRefCntPtr<FileSystem> ExternalFS)1856 vfs::getVFSFromYAML(std::unique_ptr<MemoryBuffer> Buffer,
1857 SourceMgr::DiagHandlerTy DiagHandler,
1858 StringRef YAMLFilePath, void *DiagContext,
1859 IntrusiveRefCntPtr<FileSystem> ExternalFS) {
1860 return RedirectingFileSystem::create(std::move(Buffer), DiagHandler,
1861 YAMLFilePath, DiagContext,
1862 std::move(ExternalFS));
1863 }
1864
getVFSEntries(RedirectingFileSystem::Entry * SrcE,SmallVectorImpl<StringRef> & Path,SmallVectorImpl<YAMLVFSEntry> & Entries)1865 static void getVFSEntries(RedirectingFileSystem::Entry *SrcE,
1866 SmallVectorImpl<StringRef> &Path,
1867 SmallVectorImpl<YAMLVFSEntry> &Entries) {
1868 auto Kind = SrcE->getKind();
1869 if (Kind == RedirectingFileSystem::EK_Directory) {
1870 auto *DE = dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(SrcE);
1871 assert(DE && "Must be a directory");
1872 for (std::unique_ptr<RedirectingFileSystem::Entry> &SubEntry :
1873 llvm::make_range(DE->contents_begin(), DE->contents_end())) {
1874 Path.push_back(SubEntry->getName());
1875 getVFSEntries(SubEntry.get(), Path, Entries);
1876 Path.pop_back();
1877 }
1878 return;
1879 }
1880
1881 assert(Kind == RedirectingFileSystem::EK_File && "Must be a EK_File");
1882 auto *FE = dyn_cast<RedirectingFileSystem::RedirectingFileEntry>(SrcE);
1883 assert(FE && "Must be a file");
1884 SmallString<128> VPath;
1885 for (auto &Comp : Path)
1886 llvm::sys::path::append(VPath, Comp);
1887 Entries.push_back(YAMLVFSEntry(VPath.c_str(), FE->getExternalContentsPath()));
1888 }
1889
collectVFSFromYAML(std::unique_ptr<MemoryBuffer> Buffer,SourceMgr::DiagHandlerTy DiagHandler,StringRef YAMLFilePath,SmallVectorImpl<YAMLVFSEntry> & CollectedEntries,void * DiagContext,IntrusiveRefCntPtr<FileSystem> ExternalFS)1890 void vfs::collectVFSFromYAML(std::unique_ptr<MemoryBuffer> Buffer,
1891 SourceMgr::DiagHandlerTy DiagHandler,
1892 StringRef YAMLFilePath,
1893 SmallVectorImpl<YAMLVFSEntry> &CollectedEntries,
1894 void *DiagContext,
1895 IntrusiveRefCntPtr<FileSystem> ExternalFS) {
1896 RedirectingFileSystem *VFS = RedirectingFileSystem::create(
1897 std::move(Buffer), DiagHandler, YAMLFilePath, DiagContext,
1898 std::move(ExternalFS));
1899 ErrorOr<RedirectingFileSystem::Entry *> RootE = VFS->lookupPath("/");
1900 if (!RootE)
1901 return;
1902 SmallVector<StringRef, 8> Components;
1903 Components.push_back("/");
1904 getVFSEntries(*RootE, Components, CollectedEntries);
1905 }
1906
getNextVirtualUniqueID()1907 UniqueID vfs::getNextVirtualUniqueID() {
1908 static std::atomic<unsigned> UID;
1909 unsigned ID = ++UID;
1910 // The following assumes that uint64_t max will never collide with a real
1911 // dev_t value from the OS.
1912 return UniqueID(std::numeric_limits<uint64_t>::max(), ID);
1913 }
1914
addEntry(StringRef VirtualPath,StringRef RealPath,bool IsDirectory)1915 void YAMLVFSWriter::addEntry(StringRef VirtualPath, StringRef RealPath,
1916 bool IsDirectory) {
1917 assert(sys::path::is_absolute(VirtualPath) && "virtual path not absolute");
1918 assert(sys::path::is_absolute(RealPath) && "real path not absolute");
1919 assert(!pathHasTraversal(VirtualPath) && "path traversal is not supported");
1920 Mappings.emplace_back(VirtualPath, RealPath, IsDirectory);
1921 }
1922
addFileMapping(StringRef VirtualPath,StringRef RealPath)1923 void YAMLVFSWriter::addFileMapping(StringRef VirtualPath, StringRef RealPath) {
1924 addEntry(VirtualPath, RealPath, /*IsDirectory=*/false);
1925 }
1926
addDirectoryMapping(StringRef VirtualPath,StringRef RealPath)1927 void YAMLVFSWriter::addDirectoryMapping(StringRef VirtualPath,
1928 StringRef RealPath) {
1929 addEntry(VirtualPath, RealPath, /*IsDirectory=*/true);
1930 }
1931
1932 namespace {
1933
1934 class JSONWriter {
1935 llvm::raw_ostream &OS;
1936 SmallVector<StringRef, 16> DirStack;
1937
getDirIndent()1938 unsigned getDirIndent() { return 4 * DirStack.size(); }
getFileIndent()1939 unsigned getFileIndent() { return 4 * (DirStack.size() + 1); }
1940 bool containedIn(StringRef Parent, StringRef Path);
1941 StringRef containedPart(StringRef Parent, StringRef Path);
1942 void startDirectory(StringRef Path);
1943 void endDirectory();
1944 void writeEntry(StringRef VPath, StringRef RPath);
1945
1946 public:
JSONWriter(llvm::raw_ostream & OS)1947 JSONWriter(llvm::raw_ostream &OS) : OS(OS) {}
1948
1949 void write(ArrayRef<YAMLVFSEntry> Entries, Optional<bool> UseExternalNames,
1950 Optional<bool> IsCaseSensitive, Optional<bool> IsOverlayRelative,
1951 StringRef OverlayDir);
1952 };
1953
1954 } // namespace
1955
containedIn(StringRef Parent,StringRef Path)1956 bool JSONWriter::containedIn(StringRef Parent, StringRef Path) {
1957 using namespace llvm::sys;
1958
1959 // Compare each path component.
1960 auto IParent = path::begin(Parent), EParent = path::end(Parent);
1961 for (auto IChild = path::begin(Path), EChild = path::end(Path);
1962 IParent != EParent && IChild != EChild; ++IParent, ++IChild) {
1963 if (*IParent != *IChild)
1964 return false;
1965 }
1966 // Have we exhausted the parent path?
1967 return IParent == EParent;
1968 }
1969
containedPart(StringRef Parent,StringRef Path)1970 StringRef JSONWriter::containedPart(StringRef Parent, StringRef Path) {
1971 assert(!Parent.empty());
1972 assert(containedIn(Parent, Path));
1973 return Path.slice(Parent.size() + 1, StringRef::npos);
1974 }
1975
startDirectory(StringRef Path)1976 void JSONWriter::startDirectory(StringRef Path) {
1977 StringRef Name =
1978 DirStack.empty() ? Path : containedPart(DirStack.back(), Path);
1979 DirStack.push_back(Path);
1980 unsigned Indent = getDirIndent();
1981 OS.indent(Indent) << "{\n";
1982 OS.indent(Indent + 2) << "'type': 'directory',\n";
1983 OS.indent(Indent + 2) << "'name': \"" << llvm::yaml::escape(Name) << "\",\n";
1984 OS.indent(Indent + 2) << "'contents': [\n";
1985 }
1986
endDirectory()1987 void JSONWriter::endDirectory() {
1988 unsigned Indent = getDirIndent();
1989 OS.indent(Indent + 2) << "]\n";
1990 OS.indent(Indent) << "}";
1991
1992 DirStack.pop_back();
1993 }
1994
writeEntry(StringRef VPath,StringRef RPath)1995 void JSONWriter::writeEntry(StringRef VPath, StringRef RPath) {
1996 unsigned Indent = getFileIndent();
1997 OS.indent(Indent) << "{\n";
1998 OS.indent(Indent + 2) << "'type': 'file',\n";
1999 OS.indent(Indent + 2) << "'name': \"" << llvm::yaml::escape(VPath) << "\",\n";
2000 OS.indent(Indent + 2) << "'external-contents': \""
2001 << llvm::yaml::escape(RPath) << "\"\n";
2002 OS.indent(Indent) << "}";
2003 }
2004
write(ArrayRef<YAMLVFSEntry> Entries,Optional<bool> UseExternalNames,Optional<bool> IsCaseSensitive,Optional<bool> IsOverlayRelative,StringRef OverlayDir)2005 void JSONWriter::write(ArrayRef<YAMLVFSEntry> Entries,
2006 Optional<bool> UseExternalNames,
2007 Optional<bool> IsCaseSensitive,
2008 Optional<bool> IsOverlayRelative,
2009 StringRef OverlayDir) {
2010 using namespace llvm::sys;
2011
2012 OS << "{\n"
2013 " 'version': 0,\n";
2014 if (IsCaseSensitive.hasValue())
2015 OS << " 'case-sensitive': '"
2016 << (IsCaseSensitive.getValue() ? "true" : "false") << "',\n";
2017 if (UseExternalNames.hasValue())
2018 OS << " 'use-external-names': '"
2019 << (UseExternalNames.getValue() ? "true" : "false") << "',\n";
2020 bool UseOverlayRelative = false;
2021 if (IsOverlayRelative.hasValue()) {
2022 UseOverlayRelative = IsOverlayRelative.getValue();
2023 OS << " 'overlay-relative': '" << (UseOverlayRelative ? "true" : "false")
2024 << "',\n";
2025 }
2026 OS << " 'roots': [\n";
2027
2028 if (!Entries.empty()) {
2029 const YAMLVFSEntry &Entry = Entries.front();
2030
2031 startDirectory(
2032 Entry.IsDirectory ? Entry.VPath : path::parent_path(Entry.VPath)
2033 );
2034
2035 StringRef RPath = Entry.RPath;
2036 if (UseOverlayRelative) {
2037 unsigned OverlayDirLen = OverlayDir.size();
2038 assert(RPath.substr(0, OverlayDirLen) == OverlayDir &&
2039 "Overlay dir must be contained in RPath");
2040 RPath = RPath.slice(OverlayDirLen, RPath.size());
2041 }
2042
2043 bool IsCurrentDirEmpty = true;
2044 if (!Entry.IsDirectory) {
2045 writeEntry(path::filename(Entry.VPath), RPath);
2046 IsCurrentDirEmpty = false;
2047 }
2048
2049 for (const auto &Entry : Entries.slice(1)) {
2050 StringRef Dir =
2051 Entry.IsDirectory ? Entry.VPath : path::parent_path(Entry.VPath);
2052 if (Dir == DirStack.back()) {
2053 if (!IsCurrentDirEmpty) {
2054 OS << ",\n";
2055 }
2056 } else {
2057 bool IsDirPoppedFromStack = false;
2058 while (!DirStack.empty() && !containedIn(DirStack.back(), Dir)) {
2059 OS << "\n";
2060 endDirectory();
2061 IsDirPoppedFromStack = true;
2062 }
2063 if (IsDirPoppedFromStack || !IsCurrentDirEmpty) {
2064 OS << ",\n";
2065 }
2066 startDirectory(Dir);
2067 IsCurrentDirEmpty = true;
2068 }
2069 StringRef RPath = Entry.RPath;
2070 if (UseOverlayRelative) {
2071 unsigned OverlayDirLen = OverlayDir.size();
2072 assert(RPath.substr(0, OverlayDirLen) == OverlayDir &&
2073 "Overlay dir must be contained in RPath");
2074 RPath = RPath.slice(OverlayDirLen, RPath.size());
2075 }
2076 if (!Entry.IsDirectory) {
2077 writeEntry(path::filename(Entry.VPath), RPath);
2078 IsCurrentDirEmpty = false;
2079 }
2080 }
2081
2082 while (!DirStack.empty()) {
2083 OS << "\n";
2084 endDirectory();
2085 }
2086 OS << "\n";
2087 }
2088
2089 OS << " ]\n"
2090 << "}\n";
2091 }
2092
write(llvm::raw_ostream & OS)2093 void YAMLVFSWriter::write(llvm::raw_ostream &OS) {
2094 llvm::sort(Mappings, [](const YAMLVFSEntry &LHS, const YAMLVFSEntry &RHS) {
2095 return LHS.VPath < RHS.VPath;
2096 });
2097
2098 JSONWriter(OS).write(Mappings, UseExternalNames, IsCaseSensitive,
2099 IsOverlayRelative, OverlayDir);
2100 }
2101
VFSFromYamlDirIterImpl(const Twine & _Path,RedirectingFileSystem::RedirectingDirectoryEntry::iterator Begin,RedirectingFileSystem::RedirectingDirectoryEntry::iterator End,bool IterateExternalFS,FileSystem & ExternalFS,std::error_code & EC)2102 VFSFromYamlDirIterImpl::VFSFromYamlDirIterImpl(
2103 const Twine &_Path,
2104 RedirectingFileSystem::RedirectingDirectoryEntry::iterator Begin,
2105 RedirectingFileSystem::RedirectingDirectoryEntry::iterator End,
2106 bool IterateExternalFS, FileSystem &ExternalFS, std::error_code &EC)
2107 : Dir(_Path.str()), Current(Begin), End(End),
2108 IterateExternalFS(IterateExternalFS), ExternalFS(ExternalFS) {
2109 EC = incrementImpl(/*IsFirstTime=*/true);
2110 }
2111
increment()2112 std::error_code VFSFromYamlDirIterImpl::increment() {
2113 return incrementImpl(/*IsFirstTime=*/false);
2114 }
2115
incrementExternal()2116 std::error_code VFSFromYamlDirIterImpl::incrementExternal() {
2117 assert(!(IsExternalFSCurrent && ExternalDirIter == directory_iterator()) &&
2118 "incrementing past end");
2119 std::error_code EC;
2120 if (IsExternalFSCurrent) {
2121 ExternalDirIter.increment(EC);
2122 } else if (IterateExternalFS) {
2123 ExternalDirIter = ExternalFS.dir_begin(Dir, EC);
2124 IsExternalFSCurrent = true;
2125 if (EC && EC != errc::no_such_file_or_directory)
2126 return EC;
2127 EC = {};
2128 }
2129 if (EC || ExternalDirIter == directory_iterator()) {
2130 CurrentEntry = directory_entry();
2131 } else {
2132 CurrentEntry = *ExternalDirIter;
2133 }
2134 return EC;
2135 }
2136
incrementContent(bool IsFirstTime)2137 std::error_code VFSFromYamlDirIterImpl::incrementContent(bool IsFirstTime) {
2138 assert((IsFirstTime || Current != End) && "cannot iterate past end");
2139 if (!IsFirstTime)
2140 ++Current;
2141 while (Current != End) {
2142 SmallString<128> PathStr(Dir);
2143 llvm::sys::path::append(PathStr, (*Current)->getName());
2144 sys::fs::file_type Type = sys::fs::file_type::type_unknown;
2145 switch ((*Current)->getKind()) {
2146 case RedirectingFileSystem::EK_Directory:
2147 Type = sys::fs::file_type::directory_file;
2148 break;
2149 case RedirectingFileSystem::EK_File:
2150 Type = sys::fs::file_type::regular_file;
2151 break;
2152 }
2153 CurrentEntry = directory_entry(std::string(PathStr.str()), Type);
2154 return {};
2155 }
2156 return incrementExternal();
2157 }
2158
incrementImpl(bool IsFirstTime)2159 std::error_code VFSFromYamlDirIterImpl::incrementImpl(bool IsFirstTime) {
2160 while (true) {
2161 std::error_code EC = IsExternalFSCurrent ? incrementExternal()
2162 : incrementContent(IsFirstTime);
2163 if (EC || CurrentEntry.path().empty())
2164 return EC;
2165 StringRef Name = llvm::sys::path::filename(CurrentEntry.path());
2166 if (SeenNames.insert(Name).second)
2167 return EC; // name not seen before
2168 }
2169 llvm_unreachable("returned above");
2170 }
2171
recursive_directory_iterator(FileSystem & FS_,const Twine & Path,std::error_code & EC)2172 vfs::recursive_directory_iterator::recursive_directory_iterator(
2173 FileSystem &FS_, const Twine &Path, std::error_code &EC)
2174 : FS(&FS_) {
2175 directory_iterator I = FS->dir_begin(Path, EC);
2176 if (I != directory_iterator()) {
2177 State = std::make_shared<detail::RecDirIterState>();
2178 State->Stack.push(I);
2179 }
2180 }
2181
2182 vfs::recursive_directory_iterator &
increment(std::error_code & EC)2183 recursive_directory_iterator::increment(std::error_code &EC) {
2184 assert(FS && State && !State->Stack.empty() && "incrementing past end");
2185 assert(!State->Stack.top()->path().empty() && "non-canonical end iterator");
2186 vfs::directory_iterator End;
2187
2188 if (State->HasNoPushRequest)
2189 State->HasNoPushRequest = false;
2190 else {
2191 if (State->Stack.top()->type() == sys::fs::file_type::directory_file) {
2192 vfs::directory_iterator I = FS->dir_begin(State->Stack.top()->path(), EC);
2193 if (I != End) {
2194 State->Stack.push(I);
2195 return *this;
2196 }
2197 }
2198 }
2199
2200 while (!State->Stack.empty() && State->Stack.top().increment(EC) == End)
2201 State->Stack.pop();
2202
2203 if (State->Stack.empty())
2204 State.reset(); // end iterator
2205
2206 return *this;
2207 }
2208