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 FD,StringRef NewName,StringRef NewRealPathName)179 RealFile(file_t FD, StringRef NewName, StringRef NewRealPathName)
180 : FD(FD), 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 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 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 llvm::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(llvm::sys::path::filename(FileName)) {}
539 virtual ~InMemoryNode() = default;
540
541 /// Get the filename of this node (the name without the directory part).
getFileName() const542 StringRef getFileName() const { return FileName; }
getKind() const543 InMemoryNodeKind getKind() const { return Kind; }
544 virtual std::string toString(unsigned Indent) const = 0;
545 };
546
547 class InMemoryFile : public InMemoryNode {
548 Status Stat;
549 std::unique_ptr<llvm::MemoryBuffer> Buffer;
550
551 public:
InMemoryFile(Status Stat,std::unique_ptr<llvm::MemoryBuffer> Buffer)552 InMemoryFile(Status Stat, std::unique_ptr<llvm::MemoryBuffer> Buffer)
553 : InMemoryNode(Stat.getName(), IME_File), Stat(std::move(Stat)),
554 Buffer(std::move(Buffer)) {}
555
556 /// Return the \p Status for this node. \p RequestedName should be the name
557 /// through which the caller referred to this node. It will override
558 /// \p Status::Name in the return value, to mimic the behavior of \p RealFile.
getStatus(const Twine & RequestedName) const559 Status getStatus(const Twine &RequestedName) const {
560 return Status::copyWithNewName(Stat, RequestedName);
561 }
getBuffer() const562 llvm::MemoryBuffer *getBuffer() const { return Buffer.get(); }
563
toString(unsigned Indent) const564 std::string toString(unsigned Indent) const override {
565 return (std::string(Indent, ' ') + Stat.getName() + "\n").str();
566 }
567
classof(const InMemoryNode * N)568 static bool classof(const InMemoryNode *N) {
569 return N->getKind() == IME_File;
570 }
571 };
572
573 namespace {
574
575 class InMemoryHardLink : public InMemoryNode {
576 const InMemoryFile &ResolvedFile;
577
578 public:
InMemoryHardLink(StringRef Path,const InMemoryFile & ResolvedFile)579 InMemoryHardLink(StringRef Path, const InMemoryFile &ResolvedFile)
580 : InMemoryNode(Path, IME_HardLink), ResolvedFile(ResolvedFile) {}
getResolvedFile() const581 const InMemoryFile &getResolvedFile() const { return ResolvedFile; }
582
toString(unsigned Indent) const583 std::string toString(unsigned Indent) const override {
584 return std::string(Indent, ' ') + "HardLink to -> " +
585 ResolvedFile.toString(0);
586 }
587
classof(const InMemoryNode * N)588 static bool classof(const InMemoryNode *N) {
589 return N->getKind() == IME_HardLink;
590 }
591 };
592
593 /// Adapt a InMemoryFile for VFS' File interface. The goal is to make
594 /// \p InMemoryFileAdaptor mimic as much as possible the behavior of
595 /// \p RealFile.
596 class InMemoryFileAdaptor : public File {
597 const InMemoryFile &Node;
598 /// The name to use when returning a Status for this file.
599 std::string RequestedName;
600
601 public:
InMemoryFileAdaptor(const InMemoryFile & Node,std::string RequestedName)602 explicit InMemoryFileAdaptor(const InMemoryFile &Node,
603 std::string RequestedName)
604 : Node(Node), RequestedName(std::move(RequestedName)) {}
605
status()606 llvm::ErrorOr<Status> status() override {
607 return Node.getStatus(RequestedName);
608 }
609
610 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
getBuffer(const Twine & Name,int64_t FileSize,bool RequiresNullTerminator,bool IsVolatile)611 getBuffer(const Twine &Name, int64_t FileSize, bool RequiresNullTerminator,
612 bool IsVolatile) override {
613 llvm::MemoryBuffer *Buf = Node.getBuffer();
614 return llvm::MemoryBuffer::getMemBuffer(
615 Buf->getBuffer(), Buf->getBufferIdentifier(), RequiresNullTerminator);
616 }
617
close()618 std::error_code close() override { return {}; }
619 };
620 } // namespace
621
622 class InMemoryDirectory : public InMemoryNode {
623 Status Stat;
624 llvm::StringMap<std::unique_ptr<InMemoryNode>> Entries;
625
626 public:
InMemoryDirectory(Status Stat)627 InMemoryDirectory(Status Stat)
628 : InMemoryNode(Stat.getName(), IME_Directory), Stat(std::move(Stat)) {}
629
630 /// Return the \p Status for this node. \p RequestedName should be the name
631 /// through which the caller referred to this node. It will override
632 /// \p Status::Name in the return value, to mimic the behavior of \p RealFile.
getStatus(const Twine & RequestedName) const633 Status getStatus(const Twine &RequestedName) const {
634 return Status::copyWithNewName(Stat, RequestedName);
635 }
getChild(StringRef Name)636 InMemoryNode *getChild(StringRef Name) {
637 auto I = Entries.find(Name);
638 if (I != Entries.end())
639 return I->second.get();
640 return nullptr;
641 }
642
addChild(StringRef Name,std::unique_ptr<InMemoryNode> Child)643 InMemoryNode *addChild(StringRef Name, std::unique_ptr<InMemoryNode> Child) {
644 return Entries.insert(make_pair(Name, std::move(Child)))
645 .first->second.get();
646 }
647
648 using const_iterator = decltype(Entries)::const_iterator;
649
begin() const650 const_iterator begin() const { return Entries.begin(); }
end() const651 const_iterator end() const { return Entries.end(); }
652
toString(unsigned Indent) const653 std::string toString(unsigned Indent) const override {
654 std::string Result =
655 (std::string(Indent, ' ') + Stat.getName() + "\n").str();
656 for (const auto &Entry : Entries)
657 Result += Entry.second->toString(Indent + 2);
658 return Result;
659 }
660
classof(const InMemoryNode * N)661 static bool classof(const InMemoryNode *N) {
662 return N->getKind() == IME_Directory;
663 }
664 };
665
666 namespace {
getNodeStatus(const InMemoryNode * Node,const Twine & RequestedName)667 Status getNodeStatus(const InMemoryNode *Node, const Twine &RequestedName) {
668 if (auto Dir = dyn_cast<detail::InMemoryDirectory>(Node))
669 return Dir->getStatus(RequestedName);
670 if (auto File = dyn_cast<detail::InMemoryFile>(Node))
671 return File->getStatus(RequestedName);
672 if (auto Link = dyn_cast<detail::InMemoryHardLink>(Node))
673 return Link->getResolvedFile().getStatus(RequestedName);
674 llvm_unreachable("Unknown node type");
675 }
676 } // namespace
677 } // namespace detail
678
InMemoryFileSystem(bool UseNormalizedPaths)679 InMemoryFileSystem::InMemoryFileSystem(bool UseNormalizedPaths)
680 : Root(new detail::InMemoryDirectory(
681 Status("", getNextVirtualUniqueID(), llvm::sys::TimePoint<>(), 0, 0,
682 0, llvm::sys::fs::file_type::directory_file,
683 llvm::sys::fs::perms::all_all))),
684 UseNormalizedPaths(UseNormalizedPaths) {}
685
686 InMemoryFileSystem::~InMemoryFileSystem() = default;
687
toString() const688 std::string InMemoryFileSystem::toString() const {
689 return Root->toString(/*Indent=*/0);
690 }
691
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)692 bool InMemoryFileSystem::addFile(const Twine &P, time_t ModificationTime,
693 std::unique_ptr<llvm::MemoryBuffer> Buffer,
694 Optional<uint32_t> User,
695 Optional<uint32_t> Group,
696 Optional<llvm::sys::fs::file_type> Type,
697 Optional<llvm::sys::fs::perms> Perms,
698 const detail::InMemoryFile *HardLinkTarget) {
699 SmallString<128> Path;
700 P.toVector(Path);
701
702 // Fix up relative paths. This just prepends the current working directory.
703 std::error_code EC = makeAbsolute(Path);
704 assert(!EC);
705 (void)EC;
706
707 if (useNormalizedPaths())
708 llvm::sys::path::remove_dots(Path, /*remove_dot_dot=*/true);
709
710 if (Path.empty())
711 return false;
712
713 detail::InMemoryDirectory *Dir = Root.get();
714 auto I = llvm::sys::path::begin(Path), E = sys::path::end(Path);
715 const auto ResolvedUser = User.getValueOr(0);
716 const auto ResolvedGroup = Group.getValueOr(0);
717 const auto ResolvedType = Type.getValueOr(sys::fs::file_type::regular_file);
718 const auto ResolvedPerms = Perms.getValueOr(sys::fs::all_all);
719 assert(!(HardLinkTarget && Buffer) && "HardLink cannot have a buffer");
720 // Any intermediate directories we create should be accessible by
721 // the owner, even if Perms says otherwise for the final path.
722 const auto NewDirectoryPerms = ResolvedPerms | sys::fs::owner_all;
723 while (true) {
724 StringRef Name = *I;
725 detail::InMemoryNode *Node = Dir->getChild(Name);
726 ++I;
727 if (!Node) {
728 if (I == E) {
729 // End of the path.
730 std::unique_ptr<detail::InMemoryNode> Child;
731 if (HardLinkTarget)
732 Child.reset(new detail::InMemoryHardLink(P.str(), *HardLinkTarget));
733 else {
734 // Create a new file or directory.
735 Status Stat(P.str(), getNextVirtualUniqueID(),
736 llvm::sys::toTimePoint(ModificationTime), ResolvedUser,
737 ResolvedGroup, Buffer->getBufferSize(), ResolvedType,
738 ResolvedPerms);
739 if (ResolvedType == sys::fs::file_type::directory_file) {
740 Child.reset(new detail::InMemoryDirectory(std::move(Stat)));
741 } else {
742 Child.reset(
743 new detail::InMemoryFile(std::move(Stat), std::move(Buffer)));
744 }
745 }
746 Dir->addChild(Name, std::move(Child));
747 return true;
748 }
749
750 // Create a new directory. Use the path up to here.
751 Status Stat(
752 StringRef(Path.str().begin(), Name.end() - Path.str().begin()),
753 getNextVirtualUniqueID(), llvm::sys::toTimePoint(ModificationTime),
754 ResolvedUser, ResolvedGroup, 0, sys::fs::file_type::directory_file,
755 NewDirectoryPerms);
756 Dir = cast<detail::InMemoryDirectory>(Dir->addChild(
757 Name, llvm::make_unique<detail::InMemoryDirectory>(std::move(Stat))));
758 continue;
759 }
760
761 if (auto *NewDir = dyn_cast<detail::InMemoryDirectory>(Node)) {
762 Dir = NewDir;
763 } else {
764 assert((isa<detail::InMemoryFile>(Node) ||
765 isa<detail::InMemoryHardLink>(Node)) &&
766 "Must be either file, hardlink or directory!");
767
768 // Trying to insert a directory in place of a file.
769 if (I != E)
770 return false;
771
772 // Return false only if the new file is different from the existing one.
773 if (auto Link = dyn_cast<detail::InMemoryHardLink>(Node)) {
774 return Link->getResolvedFile().getBuffer()->getBuffer() ==
775 Buffer->getBuffer();
776 }
777 return cast<detail::InMemoryFile>(Node)->getBuffer()->getBuffer() ==
778 Buffer->getBuffer();
779 }
780 }
781 }
782
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)783 bool InMemoryFileSystem::addFile(const Twine &P, time_t ModificationTime,
784 std::unique_ptr<llvm::MemoryBuffer> Buffer,
785 Optional<uint32_t> User,
786 Optional<uint32_t> Group,
787 Optional<llvm::sys::fs::file_type> Type,
788 Optional<llvm::sys::fs::perms> Perms) {
789 return addFile(P, ModificationTime, std::move(Buffer), User, Group, Type,
790 Perms, /*HardLinkTarget=*/nullptr);
791 }
792
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)793 bool InMemoryFileSystem::addFileNoOwn(const Twine &P, time_t ModificationTime,
794 llvm::MemoryBuffer *Buffer,
795 Optional<uint32_t> User,
796 Optional<uint32_t> Group,
797 Optional<llvm::sys::fs::file_type> Type,
798 Optional<llvm::sys::fs::perms> Perms) {
799 return addFile(P, ModificationTime,
800 llvm::MemoryBuffer::getMemBuffer(
801 Buffer->getBuffer(), Buffer->getBufferIdentifier()),
802 std::move(User), std::move(Group), std::move(Type),
803 std::move(Perms));
804 }
805
806 static ErrorOr<const detail::InMemoryNode *>
lookupInMemoryNode(const InMemoryFileSystem & FS,detail::InMemoryDirectory * Dir,const Twine & P)807 lookupInMemoryNode(const InMemoryFileSystem &FS, detail::InMemoryDirectory *Dir,
808 const Twine &P) {
809 SmallString<128> Path;
810 P.toVector(Path);
811
812 // Fix up relative paths. This just prepends the current working directory.
813 std::error_code EC = FS.makeAbsolute(Path);
814 assert(!EC);
815 (void)EC;
816
817 if (FS.useNormalizedPaths())
818 llvm::sys::path::remove_dots(Path, /*remove_dot_dot=*/true);
819
820 if (Path.empty())
821 return Dir;
822
823 auto I = llvm::sys::path::begin(Path), E = llvm::sys::path::end(Path);
824 while (true) {
825 detail::InMemoryNode *Node = Dir->getChild(*I);
826 ++I;
827 if (!Node)
828 return errc::no_such_file_or_directory;
829
830 // Return the file if it's at the end of the path.
831 if (auto File = dyn_cast<detail::InMemoryFile>(Node)) {
832 if (I == E)
833 return File;
834 return errc::no_such_file_or_directory;
835 }
836
837 // If Node is HardLink then return the resolved file.
838 if (auto File = dyn_cast<detail::InMemoryHardLink>(Node)) {
839 if (I == E)
840 return &File->getResolvedFile();
841 return errc::no_such_file_or_directory;
842 }
843 // Traverse directories.
844 Dir = cast<detail::InMemoryDirectory>(Node);
845 if (I == E)
846 return Dir;
847 }
848 }
849
addHardLink(const Twine & FromPath,const Twine & ToPath)850 bool InMemoryFileSystem::addHardLink(const Twine &FromPath,
851 const Twine &ToPath) {
852 auto FromNode = lookupInMemoryNode(*this, Root.get(), FromPath);
853 auto ToNode = lookupInMemoryNode(*this, Root.get(), ToPath);
854 // FromPath must not have been added before. ToPath must have been added
855 // before. Resolved ToPath must be a File.
856 if (!ToNode || FromNode || !isa<detail::InMemoryFile>(*ToNode))
857 return false;
858 return this->addFile(FromPath, 0, nullptr, None, None, None, None,
859 cast<detail::InMemoryFile>(*ToNode));
860 }
861
status(const Twine & Path)862 llvm::ErrorOr<Status> InMemoryFileSystem::status(const Twine &Path) {
863 auto Node = lookupInMemoryNode(*this, Root.get(), Path);
864 if (Node)
865 return detail::getNodeStatus(*Node, Path);
866 return Node.getError();
867 }
868
869 llvm::ErrorOr<std::unique_ptr<File>>
openFileForRead(const Twine & Path)870 InMemoryFileSystem::openFileForRead(const Twine &Path) {
871 auto Node = lookupInMemoryNode(*this, Root.get(), Path);
872 if (!Node)
873 return Node.getError();
874
875 // When we have a file provide a heap-allocated wrapper for the memory buffer
876 // to match the ownership semantics for File.
877 if (auto *F = dyn_cast<detail::InMemoryFile>(*Node))
878 return std::unique_ptr<File>(
879 new detail::InMemoryFileAdaptor(*F, Path.str()));
880
881 // FIXME: errc::not_a_file?
882 return make_error_code(llvm::errc::invalid_argument);
883 }
884
885 namespace {
886
887 /// Adaptor from InMemoryDir::iterator to directory_iterator.
888 class InMemoryDirIterator : public llvm::vfs::detail::DirIterImpl {
889 detail::InMemoryDirectory::const_iterator I;
890 detail::InMemoryDirectory::const_iterator E;
891 std::string RequestedDirName;
892
setCurrentEntry()893 void setCurrentEntry() {
894 if (I != E) {
895 SmallString<256> Path(RequestedDirName);
896 llvm::sys::path::append(Path, I->second->getFileName());
897 sys::fs::file_type Type;
898 switch (I->second->getKind()) {
899 case detail::IME_File:
900 case detail::IME_HardLink:
901 Type = sys::fs::file_type::regular_file;
902 break;
903 case detail::IME_Directory:
904 Type = sys::fs::file_type::directory_file;
905 break;
906 }
907 CurrentEntry = directory_entry(Path.str(), Type);
908 } else {
909 // When we're at the end, make CurrentEntry invalid and DirIterImpl will
910 // do the rest.
911 CurrentEntry = directory_entry();
912 }
913 }
914
915 public:
916 InMemoryDirIterator() = default;
917
InMemoryDirIterator(const detail::InMemoryDirectory & Dir,std::string RequestedDirName)918 explicit InMemoryDirIterator(const detail::InMemoryDirectory &Dir,
919 std::string RequestedDirName)
920 : I(Dir.begin()), E(Dir.end()),
921 RequestedDirName(std::move(RequestedDirName)) {
922 setCurrentEntry();
923 }
924
increment()925 std::error_code increment() override {
926 ++I;
927 setCurrentEntry();
928 return {};
929 }
930 };
931
932 } // namespace
933
dir_begin(const Twine & Dir,std::error_code & EC)934 directory_iterator InMemoryFileSystem::dir_begin(const Twine &Dir,
935 std::error_code &EC) {
936 auto Node = lookupInMemoryNode(*this, Root.get(), Dir);
937 if (!Node) {
938 EC = Node.getError();
939 return directory_iterator(std::make_shared<InMemoryDirIterator>());
940 }
941
942 if (auto *DirNode = dyn_cast<detail::InMemoryDirectory>(*Node))
943 return directory_iterator(
944 std::make_shared<InMemoryDirIterator>(*DirNode, Dir.str()));
945
946 EC = make_error_code(llvm::errc::not_a_directory);
947 return directory_iterator(std::make_shared<InMemoryDirIterator>());
948 }
949
setCurrentWorkingDirectory(const Twine & P)950 std::error_code InMemoryFileSystem::setCurrentWorkingDirectory(const Twine &P) {
951 SmallString<128> Path;
952 P.toVector(Path);
953
954 // Fix up relative paths. This just prepends the current working directory.
955 std::error_code EC = makeAbsolute(Path);
956 assert(!EC);
957 (void)EC;
958
959 if (useNormalizedPaths())
960 llvm::sys::path::remove_dots(Path, /*remove_dot_dot=*/true);
961
962 if (!Path.empty())
963 WorkingDirectory = Path.str();
964 return {};
965 }
966
967 std::error_code
getRealPath(const Twine & Path,SmallVectorImpl<char> & Output) const968 InMemoryFileSystem::getRealPath(const Twine &Path,
969 SmallVectorImpl<char> &Output) const {
970 auto CWD = getCurrentWorkingDirectory();
971 if (!CWD || CWD->empty())
972 return errc::operation_not_permitted;
973 Path.toVector(Output);
974 if (auto EC = makeAbsolute(Output))
975 return EC;
976 llvm::sys::path::remove_dots(Output, /*remove_dot_dot=*/true);
977 return {};
978 }
979
isLocal(const Twine & Path,bool & Result)980 std::error_code InMemoryFileSystem::isLocal(const Twine &Path, bool &Result) {
981 Result = false;
982 return {};
983 }
984
985 } // namespace vfs
986 } // namespace llvm
987
988 //===-----------------------------------------------------------------------===/
989 // RedirectingFileSystem implementation
990 //===-----------------------------------------------------------------------===/
991
992 // FIXME: reuse implementation common with OverlayFSDirIterImpl as these
993 // iterators are conceptually similar.
994 class llvm::vfs::VFSFromYamlDirIterImpl
995 : public llvm::vfs::detail::DirIterImpl {
996 std::string Dir;
997 RedirectingFileSystem::RedirectingDirectoryEntry::iterator Current, End;
998
999 // To handle 'fallthrough' mode we need to iterate at first through
1000 // RedirectingDirectoryEntry and then through ExternalFS. These operations are
1001 // done sequentially, we just need to keep a track of what kind of iteration
1002 // we are currently performing.
1003
1004 /// Flag telling if we should iterate through ExternalFS or stop at the last
1005 /// RedirectingDirectoryEntry::iterator.
1006 bool IterateExternalFS;
1007 /// Flag telling if we have switched to iterating through ExternalFS.
1008 bool IsExternalFSCurrent = false;
1009 FileSystem &ExternalFS;
1010 directory_iterator ExternalDirIter;
1011 llvm::StringSet<> SeenNames;
1012
1013 /// To combine multiple iterations, different methods are responsible for
1014 /// different iteration steps.
1015 /// @{
1016
1017 /// Responsible for dispatching between RedirectingDirectoryEntry iteration
1018 /// and ExternalFS iteration.
1019 std::error_code incrementImpl(bool IsFirstTime);
1020 /// Responsible for RedirectingDirectoryEntry iteration.
1021 std::error_code incrementContent(bool IsFirstTime);
1022 /// Responsible for ExternalFS iteration.
1023 std::error_code incrementExternal();
1024 /// @}
1025
1026 public:
1027 VFSFromYamlDirIterImpl(
1028 const Twine &Path,
1029 RedirectingFileSystem::RedirectingDirectoryEntry::iterator Begin,
1030 RedirectingFileSystem::RedirectingDirectoryEntry::iterator End,
1031 bool IterateExternalFS, FileSystem &ExternalFS, std::error_code &EC);
1032
1033 std::error_code increment() override;
1034 };
1035
1036 llvm::ErrorOr<std::string>
getCurrentWorkingDirectory() const1037 RedirectingFileSystem::getCurrentWorkingDirectory() const {
1038 return ExternalFS->getCurrentWorkingDirectory();
1039 }
1040
1041 std::error_code
setCurrentWorkingDirectory(const Twine & Path)1042 RedirectingFileSystem::setCurrentWorkingDirectory(const Twine &Path) {
1043 return ExternalFS->setCurrentWorkingDirectory(Path);
1044 }
1045
isLocal(const Twine & Path,bool & Result)1046 std::error_code RedirectingFileSystem::isLocal(const Twine &Path,
1047 bool &Result) {
1048 return ExternalFS->isLocal(Path, Result);
1049 }
1050
dir_begin(const Twine & Dir,std::error_code & EC)1051 directory_iterator RedirectingFileSystem::dir_begin(const Twine &Dir,
1052 std::error_code &EC) {
1053 ErrorOr<RedirectingFileSystem::Entry *> E = lookupPath(Dir);
1054 if (!E) {
1055 EC = E.getError();
1056 if (IsFallthrough && EC == errc::no_such_file_or_directory)
1057 return ExternalFS->dir_begin(Dir, EC);
1058 return {};
1059 }
1060 ErrorOr<Status> S = status(Dir, *E);
1061 if (!S) {
1062 EC = S.getError();
1063 return {};
1064 }
1065 if (!S->isDirectory()) {
1066 EC = std::error_code(static_cast<int>(errc::not_a_directory),
1067 std::system_category());
1068 return {};
1069 }
1070
1071 auto *D = cast<RedirectingFileSystem::RedirectingDirectoryEntry>(*E);
1072 return directory_iterator(std::make_shared<VFSFromYamlDirIterImpl>(
1073 Dir, D->contents_begin(), D->contents_end(),
1074 /*IterateExternalFS=*/IsFallthrough, *ExternalFS, EC));
1075 }
1076
setExternalContentsPrefixDir(StringRef PrefixDir)1077 void RedirectingFileSystem::setExternalContentsPrefixDir(StringRef PrefixDir) {
1078 ExternalContentsPrefixDir = PrefixDir.str();
1079 }
1080
getExternalContentsPrefixDir() const1081 StringRef RedirectingFileSystem::getExternalContentsPrefixDir() const {
1082 return ExternalContentsPrefixDir;
1083 }
1084
1085 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
dump() const1086 LLVM_DUMP_METHOD void RedirectingFileSystem::dump() const {
1087 for (const auto &Root : Roots)
1088 dumpEntry(Root.get());
1089 }
1090
1091 LLVM_DUMP_METHOD void
dumpEntry(RedirectingFileSystem::Entry * E,int NumSpaces) const1092 RedirectingFileSystem::dumpEntry(RedirectingFileSystem::Entry *E,
1093 int NumSpaces) const {
1094 StringRef Name = E->getName();
1095 for (int i = 0, e = NumSpaces; i < e; ++i)
1096 dbgs() << " ";
1097 dbgs() << "'" << Name.str().c_str() << "'"
1098 << "\n";
1099
1100 if (E->getKind() == RedirectingFileSystem::EK_Directory) {
1101 auto *DE = dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(E);
1102 assert(DE && "Should be a directory");
1103
1104 for (std::unique_ptr<Entry> &SubEntry :
1105 llvm::make_range(DE->contents_begin(), DE->contents_end()))
1106 dumpEntry(SubEntry.get(), NumSpaces + 2);
1107 }
1108 }
1109 #endif
1110
1111 /// A helper class to hold the common YAML parsing state.
1112 class llvm::vfs::RedirectingFileSystemParser {
1113 yaml::Stream &Stream;
1114
error(yaml::Node * N,const Twine & Msg)1115 void error(yaml::Node *N, const Twine &Msg) { Stream.printError(N, Msg); }
1116
1117 // false on error
parseScalarString(yaml::Node * N,StringRef & Result,SmallVectorImpl<char> & Storage)1118 bool parseScalarString(yaml::Node *N, StringRef &Result,
1119 SmallVectorImpl<char> &Storage) {
1120 const auto *S = dyn_cast<yaml::ScalarNode>(N);
1121
1122 if (!S) {
1123 error(N, "expected string");
1124 return false;
1125 }
1126 Result = S->getValue(Storage);
1127 return true;
1128 }
1129
1130 // false on error
parseScalarBool(yaml::Node * N,bool & Result)1131 bool parseScalarBool(yaml::Node *N, bool &Result) {
1132 SmallString<5> Storage;
1133 StringRef Value;
1134 if (!parseScalarString(N, Value, Storage))
1135 return false;
1136
1137 if (Value.equals_lower("true") || Value.equals_lower("on") ||
1138 Value.equals_lower("yes") || Value == "1") {
1139 Result = true;
1140 return true;
1141 } else if (Value.equals_lower("false") || Value.equals_lower("off") ||
1142 Value.equals_lower("no") || Value == "0") {
1143 Result = false;
1144 return true;
1145 }
1146
1147 error(N, "expected boolean value");
1148 return false;
1149 }
1150
1151 struct KeyStatus {
1152 bool Required;
1153 bool Seen = false;
1154
KeyStatusllvm::vfs::RedirectingFileSystemParser::KeyStatus1155 KeyStatus(bool Required = false) : Required(Required) {}
1156 };
1157
1158 using KeyStatusPair = std::pair<StringRef, KeyStatus>;
1159
1160 // false on error
checkDuplicateOrUnknownKey(yaml::Node * KeyNode,StringRef Key,DenseMap<StringRef,KeyStatus> & Keys)1161 bool checkDuplicateOrUnknownKey(yaml::Node *KeyNode, StringRef Key,
1162 DenseMap<StringRef, KeyStatus> &Keys) {
1163 if (!Keys.count(Key)) {
1164 error(KeyNode, "unknown key");
1165 return false;
1166 }
1167 KeyStatus &S = Keys[Key];
1168 if (S.Seen) {
1169 error(KeyNode, Twine("duplicate key '") + Key + "'");
1170 return false;
1171 }
1172 S.Seen = true;
1173 return true;
1174 }
1175
1176 // false on error
checkMissingKeys(yaml::Node * Obj,DenseMap<StringRef,KeyStatus> & Keys)1177 bool checkMissingKeys(yaml::Node *Obj, DenseMap<StringRef, KeyStatus> &Keys) {
1178 for (const auto &I : Keys) {
1179 if (I.second.Required && !I.second.Seen) {
1180 error(Obj, Twine("missing key '") + I.first + "'");
1181 return false;
1182 }
1183 }
1184 return true;
1185 }
1186
1187 RedirectingFileSystem::Entry *
lookupOrCreateEntry(RedirectingFileSystem * FS,StringRef Name,RedirectingFileSystem::Entry * ParentEntry=nullptr)1188 lookupOrCreateEntry(RedirectingFileSystem *FS, StringRef Name,
1189 RedirectingFileSystem::Entry *ParentEntry = nullptr) {
1190 if (!ParentEntry) { // Look for a existent root
1191 for (const auto &Root : FS->Roots) {
1192 if (Name.equals(Root->getName())) {
1193 ParentEntry = Root.get();
1194 return ParentEntry;
1195 }
1196 }
1197 } else { // Advance to the next component
1198 auto *DE = dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(
1199 ParentEntry);
1200 for (std::unique_ptr<RedirectingFileSystem::Entry> &Content :
1201 llvm::make_range(DE->contents_begin(), DE->contents_end())) {
1202 auto *DirContent =
1203 dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(
1204 Content.get());
1205 if (DirContent && Name.equals(Content->getName()))
1206 return DirContent;
1207 }
1208 }
1209
1210 // ... or create a new one
1211 std::unique_ptr<RedirectingFileSystem::Entry> E =
1212 llvm::make_unique<RedirectingFileSystem::RedirectingDirectoryEntry>(
1213 Name, Status("", getNextVirtualUniqueID(),
1214 std::chrono::system_clock::now(), 0, 0, 0,
1215 file_type::directory_file, sys::fs::all_all));
1216
1217 if (!ParentEntry) { // Add a new root to the overlay
1218 FS->Roots.push_back(std::move(E));
1219 ParentEntry = FS->Roots.back().get();
1220 return ParentEntry;
1221 }
1222
1223 auto *DE =
1224 dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(ParentEntry);
1225 DE->addContent(std::move(E));
1226 return DE->getLastContent();
1227 }
1228
uniqueOverlayTree(RedirectingFileSystem * FS,RedirectingFileSystem::Entry * SrcE,RedirectingFileSystem::Entry * NewParentE=nullptr)1229 void uniqueOverlayTree(RedirectingFileSystem *FS,
1230 RedirectingFileSystem::Entry *SrcE,
1231 RedirectingFileSystem::Entry *NewParentE = nullptr) {
1232 StringRef Name = SrcE->getName();
1233 switch (SrcE->getKind()) {
1234 case RedirectingFileSystem::EK_Directory: {
1235 auto *DE =
1236 dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(SrcE);
1237 assert(DE && "Must be a directory");
1238 // Empty directories could be present in the YAML as a way to
1239 // describe a file for a current directory after some of its subdir
1240 // is parsed. This only leads to redundant walks, ignore it.
1241 if (!Name.empty())
1242 NewParentE = lookupOrCreateEntry(FS, Name, NewParentE);
1243 for (std::unique_ptr<RedirectingFileSystem::Entry> &SubEntry :
1244 llvm::make_range(DE->contents_begin(), DE->contents_end()))
1245 uniqueOverlayTree(FS, SubEntry.get(), NewParentE);
1246 break;
1247 }
1248 case RedirectingFileSystem::EK_File: {
1249 auto *FE = dyn_cast<RedirectingFileSystem::RedirectingFileEntry>(SrcE);
1250 assert(FE && "Must be a file");
1251 assert(NewParentE && "Parent entry must exist");
1252 auto *DE = dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(
1253 NewParentE);
1254 DE->addContent(
1255 llvm::make_unique<RedirectingFileSystem::RedirectingFileEntry>(
1256 Name, FE->getExternalContentsPath(), FE->getUseName()));
1257 break;
1258 }
1259 }
1260 }
1261
1262 std::unique_ptr<RedirectingFileSystem::Entry>
parseEntry(yaml::Node * N,RedirectingFileSystem * FS,bool IsRootEntry)1263 parseEntry(yaml::Node *N, RedirectingFileSystem *FS, bool IsRootEntry) {
1264 auto *M = dyn_cast<yaml::MappingNode>(N);
1265 if (!M) {
1266 error(N, "expected mapping node for file or directory entry");
1267 return nullptr;
1268 }
1269
1270 KeyStatusPair Fields[] = {
1271 KeyStatusPair("name", true),
1272 KeyStatusPair("type", true),
1273 KeyStatusPair("contents", false),
1274 KeyStatusPair("external-contents", false),
1275 KeyStatusPair("use-external-name", false),
1276 };
1277
1278 DenseMap<StringRef, KeyStatus> Keys(std::begin(Fields), std::end(Fields));
1279
1280 bool HasContents = false; // external or otherwise
1281 std::vector<std::unique_ptr<RedirectingFileSystem::Entry>>
1282 EntryArrayContents;
1283 std::string ExternalContentsPath;
1284 std::string Name;
1285 yaml::Node *NameValueNode = nullptr;
1286 auto UseExternalName =
1287 RedirectingFileSystem::RedirectingFileEntry::NK_NotSet;
1288 RedirectingFileSystem::EntryKind Kind;
1289
1290 for (auto &I : *M) {
1291 StringRef Key;
1292 // Reuse the buffer for key and value, since we don't look at key after
1293 // parsing value.
1294 SmallString<256> Buffer;
1295 if (!parseScalarString(I.getKey(), Key, Buffer))
1296 return nullptr;
1297
1298 if (!checkDuplicateOrUnknownKey(I.getKey(), Key, Keys))
1299 return nullptr;
1300
1301 StringRef Value;
1302 if (Key == "name") {
1303 if (!parseScalarString(I.getValue(), Value, Buffer))
1304 return nullptr;
1305
1306 NameValueNode = I.getValue();
1307 if (FS->UseCanonicalizedPaths) {
1308 SmallString<256> Path(Value);
1309 // Guarantee that old YAML files containing paths with ".." and "."
1310 // are properly canonicalized before read into the VFS.
1311 Path = sys::path::remove_leading_dotslash(Path);
1312 sys::path::remove_dots(Path, /*remove_dot_dot=*/true);
1313 Name = Path.str();
1314 } else {
1315 Name = Value;
1316 }
1317 } else if (Key == "type") {
1318 if (!parseScalarString(I.getValue(), Value, Buffer))
1319 return nullptr;
1320 if (Value == "file")
1321 Kind = RedirectingFileSystem::EK_File;
1322 else if (Value == "directory")
1323 Kind = RedirectingFileSystem::EK_Directory;
1324 else {
1325 error(I.getValue(), "unknown value for 'type'");
1326 return nullptr;
1327 }
1328 } else if (Key == "contents") {
1329 if (HasContents) {
1330 error(I.getKey(),
1331 "entry already has 'contents' or 'external-contents'");
1332 return nullptr;
1333 }
1334 HasContents = true;
1335 auto *Contents = dyn_cast<yaml::SequenceNode>(I.getValue());
1336 if (!Contents) {
1337 // FIXME: this is only for directories, what about files?
1338 error(I.getValue(), "expected array");
1339 return nullptr;
1340 }
1341
1342 for (auto &I : *Contents) {
1343 if (std::unique_ptr<RedirectingFileSystem::Entry> E =
1344 parseEntry(&I, FS, /*IsRootEntry*/ false))
1345 EntryArrayContents.push_back(std::move(E));
1346 else
1347 return nullptr;
1348 }
1349 } else if (Key == "external-contents") {
1350 if (HasContents) {
1351 error(I.getKey(),
1352 "entry already has 'contents' or 'external-contents'");
1353 return nullptr;
1354 }
1355 HasContents = true;
1356 if (!parseScalarString(I.getValue(), Value, Buffer))
1357 return nullptr;
1358
1359 SmallString<256> FullPath;
1360 if (FS->IsRelativeOverlay) {
1361 FullPath = FS->getExternalContentsPrefixDir();
1362 assert(!FullPath.empty() &&
1363 "External contents prefix directory must exist");
1364 llvm::sys::path::append(FullPath, Value);
1365 } else {
1366 FullPath = Value;
1367 }
1368
1369 if (FS->UseCanonicalizedPaths) {
1370 // Guarantee that old YAML files containing paths with ".." and "."
1371 // are properly canonicalized before read into the VFS.
1372 FullPath = sys::path::remove_leading_dotslash(FullPath);
1373 sys::path::remove_dots(FullPath, /*remove_dot_dot=*/true);
1374 }
1375 ExternalContentsPath = FullPath.str();
1376 } else if (Key == "use-external-name") {
1377 bool Val;
1378 if (!parseScalarBool(I.getValue(), Val))
1379 return nullptr;
1380 UseExternalName =
1381 Val ? RedirectingFileSystem::RedirectingFileEntry::NK_External
1382 : RedirectingFileSystem::RedirectingFileEntry::NK_Virtual;
1383 } else {
1384 llvm_unreachable("key missing from Keys");
1385 }
1386 }
1387
1388 if (Stream.failed())
1389 return nullptr;
1390
1391 // check for missing keys
1392 if (!HasContents) {
1393 error(N, "missing key 'contents' or 'external-contents'");
1394 return nullptr;
1395 }
1396 if (!checkMissingKeys(N, Keys))
1397 return nullptr;
1398
1399 // check invalid configuration
1400 if (Kind == RedirectingFileSystem::EK_Directory &&
1401 UseExternalName !=
1402 RedirectingFileSystem::RedirectingFileEntry::NK_NotSet) {
1403 error(N, "'use-external-name' is not supported for directories");
1404 return nullptr;
1405 }
1406
1407 if (IsRootEntry && !sys::path::is_absolute(Name)) {
1408 assert(NameValueNode && "Name presence should be checked earlier");
1409 error(NameValueNode,
1410 "entry with relative path at the root level is not discoverable");
1411 return nullptr;
1412 }
1413
1414 // Remove trailing slash(es), being careful not to remove the root path
1415 StringRef Trimmed(Name);
1416 size_t RootPathLen = sys::path::root_path(Trimmed).size();
1417 while (Trimmed.size() > RootPathLen &&
1418 sys::path::is_separator(Trimmed.back()))
1419 Trimmed = Trimmed.slice(0, Trimmed.size() - 1);
1420 // Get the last component
1421 StringRef LastComponent = sys::path::filename(Trimmed);
1422
1423 std::unique_ptr<RedirectingFileSystem::Entry> Result;
1424 switch (Kind) {
1425 case RedirectingFileSystem::EK_File:
1426 Result = llvm::make_unique<RedirectingFileSystem::RedirectingFileEntry>(
1427 LastComponent, std::move(ExternalContentsPath), UseExternalName);
1428 break;
1429 case RedirectingFileSystem::EK_Directory:
1430 Result =
1431 llvm::make_unique<RedirectingFileSystem::RedirectingDirectoryEntry>(
1432 LastComponent, std::move(EntryArrayContents),
1433 Status("", getNextVirtualUniqueID(),
1434 std::chrono::system_clock::now(), 0, 0, 0,
1435 file_type::directory_file, sys::fs::all_all));
1436 break;
1437 }
1438
1439 StringRef Parent = sys::path::parent_path(Trimmed);
1440 if (Parent.empty())
1441 return Result;
1442
1443 // if 'name' contains multiple components, create implicit directory entries
1444 for (sys::path::reverse_iterator I = sys::path::rbegin(Parent),
1445 E = sys::path::rend(Parent);
1446 I != E; ++I) {
1447 std::vector<std::unique_ptr<RedirectingFileSystem::Entry>> Entries;
1448 Entries.push_back(std::move(Result));
1449 Result =
1450 llvm::make_unique<RedirectingFileSystem::RedirectingDirectoryEntry>(
1451 *I, std::move(Entries),
1452 Status("", getNextVirtualUniqueID(),
1453 std::chrono::system_clock::now(), 0, 0, 0,
1454 file_type::directory_file, sys::fs::all_all));
1455 }
1456 return Result;
1457 }
1458
1459 public:
RedirectingFileSystemParser(yaml::Stream & S)1460 RedirectingFileSystemParser(yaml::Stream &S) : Stream(S) {}
1461
1462 // false on error
parse(yaml::Node * Root,RedirectingFileSystem * FS)1463 bool parse(yaml::Node *Root, RedirectingFileSystem *FS) {
1464 auto *Top = dyn_cast<yaml::MappingNode>(Root);
1465 if (!Top) {
1466 error(Root, "expected mapping node");
1467 return false;
1468 }
1469
1470 KeyStatusPair Fields[] = {
1471 KeyStatusPair("version", true),
1472 KeyStatusPair("case-sensitive", false),
1473 KeyStatusPair("use-external-names", false),
1474 KeyStatusPair("overlay-relative", false),
1475 KeyStatusPair("fallthrough", false),
1476 KeyStatusPair("roots", true),
1477 };
1478
1479 DenseMap<StringRef, KeyStatus> Keys(std::begin(Fields), std::end(Fields));
1480 std::vector<std::unique_ptr<RedirectingFileSystem::Entry>> RootEntries;
1481
1482 // Parse configuration and 'roots'
1483 for (auto &I : *Top) {
1484 SmallString<10> KeyBuffer;
1485 StringRef Key;
1486 if (!parseScalarString(I.getKey(), Key, KeyBuffer))
1487 return false;
1488
1489 if (!checkDuplicateOrUnknownKey(I.getKey(), Key, Keys))
1490 return false;
1491
1492 if (Key == "roots") {
1493 auto *Roots = dyn_cast<yaml::SequenceNode>(I.getValue());
1494 if (!Roots) {
1495 error(I.getValue(), "expected array");
1496 return false;
1497 }
1498
1499 for (auto &I : *Roots) {
1500 if (std::unique_ptr<RedirectingFileSystem::Entry> E =
1501 parseEntry(&I, FS, /*IsRootEntry*/ true))
1502 RootEntries.push_back(std::move(E));
1503 else
1504 return false;
1505 }
1506 } else if (Key == "version") {
1507 StringRef VersionString;
1508 SmallString<4> Storage;
1509 if (!parseScalarString(I.getValue(), VersionString, Storage))
1510 return false;
1511 int Version;
1512 if (VersionString.getAsInteger<int>(10, Version)) {
1513 error(I.getValue(), "expected integer");
1514 return false;
1515 }
1516 if (Version < 0) {
1517 error(I.getValue(), "invalid version number");
1518 return false;
1519 }
1520 if (Version != 0) {
1521 error(I.getValue(), "version mismatch, expected 0");
1522 return false;
1523 }
1524 } else if (Key == "case-sensitive") {
1525 if (!parseScalarBool(I.getValue(), FS->CaseSensitive))
1526 return false;
1527 } else if (Key == "overlay-relative") {
1528 if (!parseScalarBool(I.getValue(), FS->IsRelativeOverlay))
1529 return false;
1530 } else if (Key == "use-external-names") {
1531 if (!parseScalarBool(I.getValue(), FS->UseExternalNames))
1532 return false;
1533 } else if (Key == "fallthrough") {
1534 if (!parseScalarBool(I.getValue(), FS->IsFallthrough))
1535 return false;
1536 } else {
1537 llvm_unreachable("key missing from Keys");
1538 }
1539 }
1540
1541 if (Stream.failed())
1542 return false;
1543
1544 if (!checkMissingKeys(Top, Keys))
1545 return false;
1546
1547 // Now that we sucessefully parsed the YAML file, canonicalize the internal
1548 // representation to a proper directory tree so that we can search faster
1549 // inside the VFS.
1550 for (auto &E : RootEntries)
1551 uniqueOverlayTree(FS, E.get());
1552
1553 return true;
1554 }
1555 };
1556
1557 RedirectingFileSystem *
create(std::unique_ptr<MemoryBuffer> Buffer,SourceMgr::DiagHandlerTy DiagHandler,StringRef YAMLFilePath,void * DiagContext,IntrusiveRefCntPtr<FileSystem> ExternalFS)1558 RedirectingFileSystem::create(std::unique_ptr<MemoryBuffer> Buffer,
1559 SourceMgr::DiagHandlerTy DiagHandler,
1560 StringRef YAMLFilePath, void *DiagContext,
1561 IntrusiveRefCntPtr<FileSystem> ExternalFS) {
1562 SourceMgr SM;
1563 yaml::Stream Stream(Buffer->getMemBufferRef(), SM);
1564
1565 SM.setDiagHandler(DiagHandler, DiagContext);
1566 yaml::document_iterator DI = Stream.begin();
1567 yaml::Node *Root = DI->getRoot();
1568 if (DI == Stream.end() || !Root) {
1569 SM.PrintMessage(SMLoc(), SourceMgr::DK_Error, "expected root node");
1570 return nullptr;
1571 }
1572
1573 RedirectingFileSystemParser P(Stream);
1574
1575 std::unique_ptr<RedirectingFileSystem> FS(
1576 new RedirectingFileSystem(std::move(ExternalFS)));
1577
1578 if (!YAMLFilePath.empty()) {
1579 // Use the YAML path from -ivfsoverlay to compute the dir to be prefixed
1580 // to each 'external-contents' path.
1581 //
1582 // Example:
1583 // -ivfsoverlay dummy.cache/vfs/vfs.yaml
1584 // yields:
1585 // FS->ExternalContentsPrefixDir => /<absolute_path_to>/dummy.cache/vfs
1586 //
1587 SmallString<256> OverlayAbsDir = sys::path::parent_path(YAMLFilePath);
1588 std::error_code EC = llvm::sys::fs::make_absolute(OverlayAbsDir);
1589 assert(!EC && "Overlay dir final path must be absolute");
1590 (void)EC;
1591 FS->setExternalContentsPrefixDir(OverlayAbsDir);
1592 }
1593
1594 if (!P.parse(Root, FS.get()))
1595 return nullptr;
1596
1597 return FS.release();
1598 }
1599
1600 ErrorOr<RedirectingFileSystem::Entry *>
lookupPath(const Twine & Path_) const1601 RedirectingFileSystem::lookupPath(const Twine &Path_) const {
1602 SmallString<256> Path;
1603 Path_.toVector(Path);
1604
1605 // Handle relative paths
1606 if (std::error_code EC = makeAbsolute(Path))
1607 return EC;
1608
1609 // Canonicalize path by removing ".", "..", "./", etc components. This is
1610 // a VFS request, do bot bother about symlinks in the path components
1611 // but canonicalize in order to perform the correct entry search.
1612 if (UseCanonicalizedPaths) {
1613 Path = sys::path::remove_leading_dotslash(Path);
1614 sys::path::remove_dots(Path, /*remove_dot_dot=*/true);
1615 }
1616
1617 if (Path.empty())
1618 return make_error_code(llvm::errc::invalid_argument);
1619
1620 sys::path::const_iterator Start = sys::path::begin(Path);
1621 sys::path::const_iterator End = sys::path::end(Path);
1622 for (const auto &Root : Roots) {
1623 ErrorOr<RedirectingFileSystem::Entry *> Result =
1624 lookupPath(Start, End, Root.get());
1625 if (Result || Result.getError() != llvm::errc::no_such_file_or_directory)
1626 return Result;
1627 }
1628 return make_error_code(llvm::errc::no_such_file_or_directory);
1629 }
1630
1631 ErrorOr<RedirectingFileSystem::Entry *>
lookupPath(sys::path::const_iterator Start,sys::path::const_iterator End,RedirectingFileSystem::Entry * From) const1632 RedirectingFileSystem::lookupPath(sys::path::const_iterator Start,
1633 sys::path::const_iterator End,
1634 RedirectingFileSystem::Entry *From) const {
1635 #ifndef _WIN32
1636 assert(!isTraversalComponent(*Start) &&
1637 !isTraversalComponent(From->getName()) &&
1638 "Paths should not contain traversal components");
1639 #else
1640 // FIXME: this is here to support windows, remove it once canonicalized
1641 // paths become globally default.
1642 if (Start->equals("."))
1643 ++Start;
1644 #endif
1645
1646 StringRef FromName = From->getName();
1647
1648 // Forward the search to the next component in case this is an empty one.
1649 if (!FromName.empty()) {
1650 if (CaseSensitive ? !Start->equals(FromName)
1651 : !Start->equals_lower(FromName))
1652 // failure to match
1653 return make_error_code(llvm::errc::no_such_file_or_directory);
1654
1655 ++Start;
1656
1657 if (Start == End) {
1658 // Match!
1659 return From;
1660 }
1661 }
1662
1663 auto *DE = dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(From);
1664 if (!DE)
1665 return make_error_code(llvm::errc::not_a_directory);
1666
1667 for (const std::unique_ptr<RedirectingFileSystem::Entry> &DirEntry :
1668 llvm::make_range(DE->contents_begin(), DE->contents_end())) {
1669 ErrorOr<RedirectingFileSystem::Entry *> Result =
1670 lookupPath(Start, End, DirEntry.get());
1671 if (Result || Result.getError() != llvm::errc::no_such_file_or_directory)
1672 return Result;
1673 }
1674 return make_error_code(llvm::errc::no_such_file_or_directory);
1675 }
1676
getRedirectedFileStatus(const Twine & Path,bool UseExternalNames,Status ExternalStatus)1677 static Status getRedirectedFileStatus(const Twine &Path, bool UseExternalNames,
1678 Status ExternalStatus) {
1679 Status S = ExternalStatus;
1680 if (!UseExternalNames)
1681 S = Status::copyWithNewName(S, Path);
1682 S.IsVFSMapped = true;
1683 return S;
1684 }
1685
status(const Twine & Path,RedirectingFileSystem::Entry * E)1686 ErrorOr<Status> RedirectingFileSystem::status(const Twine &Path,
1687 RedirectingFileSystem::Entry *E) {
1688 assert(E != nullptr);
1689 if (auto *F = dyn_cast<RedirectingFileSystem::RedirectingFileEntry>(E)) {
1690 ErrorOr<Status> S = ExternalFS->status(F->getExternalContentsPath());
1691 assert(!S || S->getName() == F->getExternalContentsPath());
1692 if (S)
1693 return getRedirectedFileStatus(Path, F->useExternalName(UseExternalNames),
1694 *S);
1695 return S;
1696 } else { // directory
1697 auto *DE = cast<RedirectingFileSystem::RedirectingDirectoryEntry>(E);
1698 return Status::copyWithNewName(DE->getStatus(), Path);
1699 }
1700 }
1701
status(const Twine & Path)1702 ErrorOr<Status> RedirectingFileSystem::status(const Twine &Path) {
1703 ErrorOr<RedirectingFileSystem::Entry *> Result = lookupPath(Path);
1704 if (!Result) {
1705 if (IsFallthrough &&
1706 Result.getError() == llvm::errc::no_such_file_or_directory) {
1707 return ExternalFS->status(Path);
1708 }
1709 return Result.getError();
1710 }
1711 return status(Path, *Result);
1712 }
1713
1714 namespace {
1715
1716 /// Provide a file wrapper with an overriden status.
1717 class FileWithFixedStatus : public File {
1718 std::unique_ptr<File> InnerFile;
1719 Status S;
1720
1721 public:
FileWithFixedStatus(std::unique_ptr<File> InnerFile,Status S)1722 FileWithFixedStatus(std::unique_ptr<File> InnerFile, Status S)
1723 : InnerFile(std::move(InnerFile)), S(std::move(S)) {}
1724
status()1725 ErrorOr<Status> status() override { return S; }
1726 ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
1727
getBuffer(const Twine & Name,int64_t FileSize,bool RequiresNullTerminator,bool IsVolatile)1728 getBuffer(const Twine &Name, int64_t FileSize, bool RequiresNullTerminator,
1729 bool IsVolatile) override {
1730 return InnerFile->getBuffer(Name, FileSize, RequiresNullTerminator,
1731 IsVolatile);
1732 }
1733
close()1734 std::error_code close() override { return InnerFile->close(); }
1735 };
1736
1737 } // namespace
1738
1739 ErrorOr<std::unique_ptr<File>>
openFileForRead(const Twine & Path)1740 RedirectingFileSystem::openFileForRead(const Twine &Path) {
1741 ErrorOr<RedirectingFileSystem::Entry *> E = lookupPath(Path);
1742 if (!E) {
1743 if (IsFallthrough &&
1744 E.getError() == llvm::errc::no_such_file_or_directory) {
1745 return ExternalFS->openFileForRead(Path);
1746 }
1747 return E.getError();
1748 }
1749
1750 auto *F = dyn_cast<RedirectingFileSystem::RedirectingFileEntry>(*E);
1751 if (!F) // FIXME: errc::not_a_file?
1752 return make_error_code(llvm::errc::invalid_argument);
1753
1754 auto Result = ExternalFS->openFileForRead(F->getExternalContentsPath());
1755 if (!Result)
1756 return Result;
1757
1758 auto ExternalStatus = (*Result)->status();
1759 if (!ExternalStatus)
1760 return ExternalStatus.getError();
1761
1762 // FIXME: Update the status with the name and VFSMapped.
1763 Status S = getRedirectedFileStatus(Path, F->useExternalName(UseExternalNames),
1764 *ExternalStatus);
1765 return std::unique_ptr<File>(
1766 llvm::make_unique<FileWithFixedStatus>(std::move(*Result), S));
1767 }
1768
1769 std::error_code
getRealPath(const Twine & Path,SmallVectorImpl<char> & Output) const1770 RedirectingFileSystem::getRealPath(const Twine &Path,
1771 SmallVectorImpl<char> &Output) const {
1772 ErrorOr<RedirectingFileSystem::Entry *> Result = lookupPath(Path);
1773 if (!Result) {
1774 if (IsFallthrough &&
1775 Result.getError() == llvm::errc::no_such_file_or_directory) {
1776 return ExternalFS->getRealPath(Path, Output);
1777 }
1778 return Result.getError();
1779 }
1780
1781 if (auto *F =
1782 dyn_cast<RedirectingFileSystem::RedirectingFileEntry>(*Result)) {
1783 return ExternalFS->getRealPath(F->getExternalContentsPath(), Output);
1784 }
1785 // Even if there is a directory entry, fall back to ExternalFS if allowed,
1786 // because directories don't have a single external contents path.
1787 return IsFallthrough ? ExternalFS->getRealPath(Path, Output)
1788 : llvm::errc::invalid_argument;
1789 }
1790
1791 IntrusiveRefCntPtr<FileSystem>
getVFSFromYAML(std::unique_ptr<MemoryBuffer> Buffer,SourceMgr::DiagHandlerTy DiagHandler,StringRef YAMLFilePath,void * DiagContext,IntrusiveRefCntPtr<FileSystem> ExternalFS)1792 vfs::getVFSFromYAML(std::unique_ptr<MemoryBuffer> Buffer,
1793 SourceMgr::DiagHandlerTy DiagHandler,
1794 StringRef YAMLFilePath, void *DiagContext,
1795 IntrusiveRefCntPtr<FileSystem> ExternalFS) {
1796 return RedirectingFileSystem::create(std::move(Buffer), DiagHandler,
1797 YAMLFilePath, DiagContext,
1798 std::move(ExternalFS));
1799 }
1800
getVFSEntries(RedirectingFileSystem::Entry * SrcE,SmallVectorImpl<StringRef> & Path,SmallVectorImpl<YAMLVFSEntry> & Entries)1801 static void getVFSEntries(RedirectingFileSystem::Entry *SrcE,
1802 SmallVectorImpl<StringRef> &Path,
1803 SmallVectorImpl<YAMLVFSEntry> &Entries) {
1804 auto Kind = SrcE->getKind();
1805 if (Kind == RedirectingFileSystem::EK_Directory) {
1806 auto *DE = dyn_cast<RedirectingFileSystem::RedirectingDirectoryEntry>(SrcE);
1807 assert(DE && "Must be a directory");
1808 for (std::unique_ptr<RedirectingFileSystem::Entry> &SubEntry :
1809 llvm::make_range(DE->contents_begin(), DE->contents_end())) {
1810 Path.push_back(SubEntry->getName());
1811 getVFSEntries(SubEntry.get(), Path, Entries);
1812 Path.pop_back();
1813 }
1814 return;
1815 }
1816
1817 assert(Kind == RedirectingFileSystem::EK_File && "Must be a EK_File");
1818 auto *FE = dyn_cast<RedirectingFileSystem::RedirectingFileEntry>(SrcE);
1819 assert(FE && "Must be a file");
1820 SmallString<128> VPath;
1821 for (auto &Comp : Path)
1822 llvm::sys::path::append(VPath, Comp);
1823 Entries.push_back(YAMLVFSEntry(VPath.c_str(), FE->getExternalContentsPath()));
1824 }
1825
collectVFSFromYAML(std::unique_ptr<MemoryBuffer> Buffer,SourceMgr::DiagHandlerTy DiagHandler,StringRef YAMLFilePath,SmallVectorImpl<YAMLVFSEntry> & CollectedEntries,void * DiagContext,IntrusiveRefCntPtr<FileSystem> ExternalFS)1826 void vfs::collectVFSFromYAML(std::unique_ptr<MemoryBuffer> Buffer,
1827 SourceMgr::DiagHandlerTy DiagHandler,
1828 StringRef YAMLFilePath,
1829 SmallVectorImpl<YAMLVFSEntry> &CollectedEntries,
1830 void *DiagContext,
1831 IntrusiveRefCntPtr<FileSystem> ExternalFS) {
1832 RedirectingFileSystem *VFS = RedirectingFileSystem::create(
1833 std::move(Buffer), DiagHandler, YAMLFilePath, DiagContext,
1834 std::move(ExternalFS));
1835 ErrorOr<RedirectingFileSystem::Entry *> RootE = VFS->lookupPath("/");
1836 if (!RootE)
1837 return;
1838 SmallVector<StringRef, 8> Components;
1839 Components.push_back("/");
1840 getVFSEntries(*RootE, Components, CollectedEntries);
1841 }
1842
getNextVirtualUniqueID()1843 UniqueID vfs::getNextVirtualUniqueID() {
1844 static std::atomic<unsigned> UID;
1845 unsigned ID = ++UID;
1846 // The following assumes that uint64_t max will never collide with a real
1847 // dev_t value from the OS.
1848 return UniqueID(std::numeric_limits<uint64_t>::max(), ID);
1849 }
1850
addFileMapping(StringRef VirtualPath,StringRef RealPath)1851 void YAMLVFSWriter::addFileMapping(StringRef VirtualPath, StringRef RealPath) {
1852 assert(sys::path::is_absolute(VirtualPath) && "virtual path not absolute");
1853 assert(sys::path::is_absolute(RealPath) && "real path not absolute");
1854 assert(!pathHasTraversal(VirtualPath) && "path traversal is not supported");
1855 Mappings.emplace_back(VirtualPath, RealPath);
1856 }
1857
1858 namespace {
1859
1860 class JSONWriter {
1861 llvm::raw_ostream &OS;
1862 SmallVector<StringRef, 16> DirStack;
1863
getDirIndent()1864 unsigned getDirIndent() { return 4 * DirStack.size(); }
getFileIndent()1865 unsigned getFileIndent() { return 4 * (DirStack.size() + 1); }
1866 bool containedIn(StringRef Parent, StringRef Path);
1867 StringRef containedPart(StringRef Parent, StringRef Path);
1868 void startDirectory(StringRef Path);
1869 void endDirectory();
1870 void writeEntry(StringRef VPath, StringRef RPath);
1871
1872 public:
JSONWriter(llvm::raw_ostream & OS)1873 JSONWriter(llvm::raw_ostream &OS) : OS(OS) {}
1874
1875 void write(ArrayRef<YAMLVFSEntry> Entries, Optional<bool> UseExternalNames,
1876 Optional<bool> IsCaseSensitive, Optional<bool> IsOverlayRelative,
1877 StringRef OverlayDir);
1878 };
1879
1880 } // namespace
1881
containedIn(StringRef Parent,StringRef Path)1882 bool JSONWriter::containedIn(StringRef Parent, StringRef Path) {
1883 using namespace llvm::sys;
1884
1885 // Compare each path component.
1886 auto IParent = path::begin(Parent), EParent = path::end(Parent);
1887 for (auto IChild = path::begin(Path), EChild = path::end(Path);
1888 IParent != EParent && IChild != EChild; ++IParent, ++IChild) {
1889 if (*IParent != *IChild)
1890 return false;
1891 }
1892 // Have we exhausted the parent path?
1893 return IParent == EParent;
1894 }
1895
containedPart(StringRef Parent,StringRef Path)1896 StringRef JSONWriter::containedPart(StringRef Parent, StringRef Path) {
1897 assert(!Parent.empty());
1898 assert(containedIn(Parent, Path));
1899 return Path.slice(Parent.size() + 1, StringRef::npos);
1900 }
1901
startDirectory(StringRef Path)1902 void JSONWriter::startDirectory(StringRef Path) {
1903 StringRef Name =
1904 DirStack.empty() ? Path : containedPart(DirStack.back(), Path);
1905 DirStack.push_back(Path);
1906 unsigned Indent = getDirIndent();
1907 OS.indent(Indent) << "{\n";
1908 OS.indent(Indent + 2) << "'type': 'directory',\n";
1909 OS.indent(Indent + 2) << "'name': \"" << llvm::yaml::escape(Name) << "\",\n";
1910 OS.indent(Indent + 2) << "'contents': [\n";
1911 }
1912
endDirectory()1913 void JSONWriter::endDirectory() {
1914 unsigned Indent = getDirIndent();
1915 OS.indent(Indent + 2) << "]\n";
1916 OS.indent(Indent) << "}";
1917
1918 DirStack.pop_back();
1919 }
1920
writeEntry(StringRef VPath,StringRef RPath)1921 void JSONWriter::writeEntry(StringRef VPath, StringRef RPath) {
1922 unsigned Indent = getFileIndent();
1923 OS.indent(Indent) << "{\n";
1924 OS.indent(Indent + 2) << "'type': 'file',\n";
1925 OS.indent(Indent + 2) << "'name': \"" << llvm::yaml::escape(VPath) << "\",\n";
1926 OS.indent(Indent + 2) << "'external-contents': \""
1927 << llvm::yaml::escape(RPath) << "\"\n";
1928 OS.indent(Indent) << "}";
1929 }
1930
write(ArrayRef<YAMLVFSEntry> Entries,Optional<bool> UseExternalNames,Optional<bool> IsCaseSensitive,Optional<bool> IsOverlayRelative,StringRef OverlayDir)1931 void JSONWriter::write(ArrayRef<YAMLVFSEntry> Entries,
1932 Optional<bool> UseExternalNames,
1933 Optional<bool> IsCaseSensitive,
1934 Optional<bool> IsOverlayRelative,
1935 StringRef OverlayDir) {
1936 using namespace llvm::sys;
1937
1938 OS << "{\n"
1939 " 'version': 0,\n";
1940 if (IsCaseSensitive.hasValue())
1941 OS << " 'case-sensitive': '"
1942 << (IsCaseSensitive.getValue() ? "true" : "false") << "',\n";
1943 if (UseExternalNames.hasValue())
1944 OS << " 'use-external-names': '"
1945 << (UseExternalNames.getValue() ? "true" : "false") << "',\n";
1946 bool UseOverlayRelative = false;
1947 if (IsOverlayRelative.hasValue()) {
1948 UseOverlayRelative = IsOverlayRelative.getValue();
1949 OS << " 'overlay-relative': '" << (UseOverlayRelative ? "true" : "false")
1950 << "',\n";
1951 }
1952 OS << " 'roots': [\n";
1953
1954 if (!Entries.empty()) {
1955 const YAMLVFSEntry &Entry = Entries.front();
1956 startDirectory(path::parent_path(Entry.VPath));
1957
1958 StringRef RPath = Entry.RPath;
1959 if (UseOverlayRelative) {
1960 unsigned OverlayDirLen = OverlayDir.size();
1961 assert(RPath.substr(0, OverlayDirLen) == OverlayDir &&
1962 "Overlay dir must be contained in RPath");
1963 RPath = RPath.slice(OverlayDirLen, RPath.size());
1964 }
1965
1966 writeEntry(path::filename(Entry.VPath), RPath);
1967
1968 for (const auto &Entry : Entries.slice(1)) {
1969 StringRef Dir = path::parent_path(Entry.VPath);
1970 if (Dir == DirStack.back())
1971 OS << ",\n";
1972 else {
1973 while (!DirStack.empty() && !containedIn(DirStack.back(), Dir)) {
1974 OS << "\n";
1975 endDirectory();
1976 }
1977 OS << ",\n";
1978 startDirectory(Dir);
1979 }
1980 StringRef RPath = Entry.RPath;
1981 if (UseOverlayRelative) {
1982 unsigned OverlayDirLen = OverlayDir.size();
1983 assert(RPath.substr(0, OverlayDirLen) == OverlayDir &&
1984 "Overlay dir must be contained in RPath");
1985 RPath = RPath.slice(OverlayDirLen, RPath.size());
1986 }
1987 writeEntry(path::filename(Entry.VPath), RPath);
1988 }
1989
1990 while (!DirStack.empty()) {
1991 OS << "\n";
1992 endDirectory();
1993 }
1994 OS << "\n";
1995 }
1996
1997 OS << " ]\n"
1998 << "}\n";
1999 }
2000
write(llvm::raw_ostream & OS)2001 void YAMLVFSWriter::write(llvm::raw_ostream &OS) {
2002 llvm::sort(Mappings, [](const YAMLVFSEntry &LHS, const YAMLVFSEntry &RHS) {
2003 return LHS.VPath < RHS.VPath;
2004 });
2005
2006 JSONWriter(OS).write(Mappings, UseExternalNames, IsCaseSensitive,
2007 IsOverlayRelative, OverlayDir);
2008 }
2009
VFSFromYamlDirIterImpl(const Twine & _Path,RedirectingFileSystem::RedirectingDirectoryEntry::iterator Begin,RedirectingFileSystem::RedirectingDirectoryEntry::iterator End,bool IterateExternalFS,FileSystem & ExternalFS,std::error_code & EC)2010 VFSFromYamlDirIterImpl::VFSFromYamlDirIterImpl(
2011 const Twine &_Path,
2012 RedirectingFileSystem::RedirectingDirectoryEntry::iterator Begin,
2013 RedirectingFileSystem::RedirectingDirectoryEntry::iterator End,
2014 bool IterateExternalFS, FileSystem &ExternalFS, std::error_code &EC)
2015 : Dir(_Path.str()), Current(Begin), End(End),
2016 IterateExternalFS(IterateExternalFS), ExternalFS(ExternalFS) {
2017 EC = incrementImpl(/*IsFirstTime=*/true);
2018 }
2019
increment()2020 std::error_code VFSFromYamlDirIterImpl::increment() {
2021 return incrementImpl(/*IsFirstTime=*/false);
2022 }
2023
incrementExternal()2024 std::error_code VFSFromYamlDirIterImpl::incrementExternal() {
2025 assert(!(IsExternalFSCurrent && ExternalDirIter == directory_iterator()) &&
2026 "incrementing past end");
2027 std::error_code EC;
2028 if (IsExternalFSCurrent) {
2029 ExternalDirIter.increment(EC);
2030 } else if (IterateExternalFS) {
2031 ExternalDirIter = ExternalFS.dir_begin(Dir, EC);
2032 IsExternalFSCurrent = true;
2033 if (EC && EC != errc::no_such_file_or_directory)
2034 return EC;
2035 EC = {};
2036 }
2037 if (EC || ExternalDirIter == directory_iterator()) {
2038 CurrentEntry = directory_entry();
2039 } else {
2040 CurrentEntry = *ExternalDirIter;
2041 }
2042 return EC;
2043 }
2044
incrementContent(bool IsFirstTime)2045 std::error_code VFSFromYamlDirIterImpl::incrementContent(bool IsFirstTime) {
2046 assert((IsFirstTime || Current != End) && "cannot iterate past end");
2047 if (!IsFirstTime)
2048 ++Current;
2049 while (Current != End) {
2050 SmallString<128> PathStr(Dir);
2051 llvm::sys::path::append(PathStr, (*Current)->getName());
2052 sys::fs::file_type Type;
2053 switch ((*Current)->getKind()) {
2054 case RedirectingFileSystem::EK_Directory:
2055 Type = sys::fs::file_type::directory_file;
2056 break;
2057 case RedirectingFileSystem::EK_File:
2058 Type = sys::fs::file_type::regular_file;
2059 break;
2060 }
2061 CurrentEntry = directory_entry(PathStr.str(), Type);
2062 return {};
2063 }
2064 return incrementExternal();
2065 }
2066
incrementImpl(bool IsFirstTime)2067 std::error_code VFSFromYamlDirIterImpl::incrementImpl(bool IsFirstTime) {
2068 while (true) {
2069 std::error_code EC = IsExternalFSCurrent ? incrementExternal()
2070 : incrementContent(IsFirstTime);
2071 if (EC || CurrentEntry.path().empty())
2072 return EC;
2073 StringRef Name = llvm::sys::path::filename(CurrentEntry.path());
2074 if (SeenNames.insert(Name).second)
2075 return EC; // name not seen before
2076 }
2077 llvm_unreachable("returned above");
2078 }
2079
recursive_directory_iterator(FileSystem & FS_,const Twine & Path,std::error_code & EC)2080 vfs::recursive_directory_iterator::recursive_directory_iterator(
2081 FileSystem &FS_, const Twine &Path, std::error_code &EC)
2082 : FS(&FS_) {
2083 directory_iterator I = FS->dir_begin(Path, EC);
2084 if (I != directory_iterator()) {
2085 State = std::make_shared<detail::RecDirIterState>();
2086 State->Stack.push(I);
2087 }
2088 }
2089
2090 vfs::recursive_directory_iterator &
increment(std::error_code & EC)2091 recursive_directory_iterator::increment(std::error_code &EC) {
2092 assert(FS && State && !State->Stack.empty() && "incrementing past end");
2093 assert(!State->Stack.top()->path().empty() && "non-canonical end iterator");
2094 vfs::directory_iterator End;
2095
2096 if (State->HasNoPushRequest)
2097 State->HasNoPushRequest = false;
2098 else {
2099 if (State->Stack.top()->type() == sys::fs::file_type::directory_file) {
2100 vfs::directory_iterator I = FS->dir_begin(State->Stack.top()->path(), EC);
2101 if (I != End) {
2102 State->Stack.push(I);
2103 return *this;
2104 }
2105 }
2106 }
2107
2108 while (!State->Stack.empty() && State->Stack.top().increment(EC) == End)
2109 State->Stack.pop();
2110
2111 if (State->Stack.empty())
2112 State.reset(); // end iterator
2113
2114 return *this;
2115 }
2116