1 //===- VirtualFileSystem.h - Virtual File System Layer ----------*- C++ -*-===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 /// \file 10 /// Defines the virtual file system interface vfs::FileSystem. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #ifndef LLVM_SUPPORT_VIRTUALFILESYSTEM_H 15 #define LLVM_SUPPORT_VIRTUALFILESYSTEM_H 16 17 #include "llvm/ADT/IntrusiveRefCntPtr.h" 18 #include "llvm/ADT/None.h" 19 #include "llvm/ADT/Optional.h" 20 #include "llvm/ADT/SmallVector.h" 21 #include "llvm/ADT/StringRef.h" 22 #include "llvm/Support/Chrono.h" 23 #include "llvm/Support/ErrorOr.h" 24 #include "llvm/Support/FileSystem.h" 25 #include "llvm/Support/Path.h" 26 #include "llvm/Support/SourceMgr.h" 27 #include <cassert> 28 #include <cstdint> 29 #include <ctime> 30 #include <memory> 31 #include <stack> 32 #include <string> 33 #include <system_error> 34 #include <utility> 35 #include <vector> 36 37 namespace llvm { 38 39 class MemoryBuffer; 40 class MemoryBufferRef; 41 class Twine; 42 43 namespace vfs { 44 45 /// The result of a \p status operation. 46 class Status { 47 std::string Name; 48 llvm::sys::fs::UniqueID UID; 49 llvm::sys::TimePoint<> MTime; 50 uint32_t User; 51 uint32_t Group; 52 uint64_t Size; 53 llvm::sys::fs::file_type Type = llvm::sys::fs::file_type::status_error; 54 llvm::sys::fs::perms Perms; 55 56 public: 57 // FIXME: remove when files support multiple names 58 bool IsVFSMapped = false; 59 60 Status() = default; 61 Status(const llvm::sys::fs::file_status &Status); 62 Status(const Twine &Name, llvm::sys::fs::UniqueID UID, 63 llvm::sys::TimePoint<> MTime, uint32_t User, uint32_t Group, 64 uint64_t Size, llvm::sys::fs::file_type Type, 65 llvm::sys::fs::perms Perms); 66 67 /// Get a copy of a Status with a different name. 68 static Status copyWithNewName(const Status &In, const Twine &NewName); 69 static Status copyWithNewName(const llvm::sys::fs::file_status &In, 70 const Twine &NewName); 71 72 /// Returns the name that should be used for this file or directory. getName()73 StringRef getName() const { return Name; } 74 75 /// @name Status interface from llvm::sys::fs 76 /// @{ getType()77 llvm::sys::fs::file_type getType() const { return Type; } getPermissions()78 llvm::sys::fs::perms getPermissions() const { return Perms; } getLastModificationTime()79 llvm::sys::TimePoint<> getLastModificationTime() const { return MTime; } getUniqueID()80 llvm::sys::fs::UniqueID getUniqueID() const { return UID; } getUser()81 uint32_t getUser() const { return User; } getGroup()82 uint32_t getGroup() const { return Group; } getSize()83 uint64_t getSize() const { return Size; } 84 /// @} 85 /// @name Status queries 86 /// These are static queries in llvm::sys::fs. 87 /// @{ 88 bool equivalent(const Status &Other) const; 89 bool isDirectory() const; 90 bool isRegularFile() const; 91 bool isOther() const; 92 bool isSymlink() const; 93 bool isStatusKnown() const; 94 bool exists() const; 95 /// @} 96 }; 97 98 /// Represents an open file. 99 class File { 100 public: 101 /// Destroy the file after closing it (if open). 102 /// Sub-classes should generally call close() inside their destructors. We 103 /// cannot do that from the base class, since close is virtual. 104 virtual ~File(); 105 106 /// Get the status of the file. 107 virtual llvm::ErrorOr<Status> status() = 0; 108 109 /// Get the name of the file getName()110 virtual llvm::ErrorOr<std::string> getName() { 111 if (auto Status = status()) 112 return Status->getName().str(); 113 else 114 return Status.getError(); 115 } 116 117 /// Get the contents of the file as a \p MemoryBuffer. 118 virtual llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> 119 getBuffer(const Twine &Name, int64_t FileSize = -1, 120 bool RequiresNullTerminator = true, bool IsVolatile = false) = 0; 121 122 /// Closes the file. 123 virtual std::error_code close() = 0; 124 }; 125 126 /// A member of a directory, yielded by a directory_iterator. 127 /// Only information available on most platforms is included. 128 class directory_entry { 129 std::string Path; 130 llvm::sys::fs::file_type Type = llvm::sys::fs::file_type::type_unknown; 131 132 public: 133 directory_entry() = default; directory_entry(std::string Path,llvm::sys::fs::file_type Type)134 directory_entry(std::string Path, llvm::sys::fs::file_type Type) 135 : Path(std::move(Path)), Type(Type) {} 136 path()137 llvm::StringRef path() const { return Path; } type()138 llvm::sys::fs::file_type type() const { return Type; } 139 }; 140 141 namespace detail { 142 143 /// An interface for virtual file systems to provide an iterator over the 144 /// (non-recursive) contents of a directory. 145 struct DirIterImpl { 146 virtual ~DirIterImpl(); 147 148 /// Sets \c CurrentEntry to the next entry in the directory on success, 149 /// to directory_entry() at end, or returns a system-defined \c error_code. 150 virtual std::error_code increment() = 0; 151 152 directory_entry CurrentEntry; 153 }; 154 155 } // namespace detail 156 157 /// An input iterator over the entries in a virtual path, similar to 158 /// llvm::sys::fs::directory_iterator. 159 class directory_iterator { 160 std::shared_ptr<detail::DirIterImpl> Impl; // Input iterator semantics on copy 161 162 public: directory_iterator(std::shared_ptr<detail::DirIterImpl> I)163 directory_iterator(std::shared_ptr<detail::DirIterImpl> I) 164 : Impl(std::move(I)) { 165 assert(Impl.get() != nullptr && "requires non-null implementation"); 166 if (Impl->CurrentEntry.path().empty()) 167 Impl.reset(); // Normalize the end iterator to Impl == nullptr. 168 } 169 170 /// Construct an 'end' iterator. 171 directory_iterator() = default; 172 173 /// Equivalent to operator++, with an error code. increment(std::error_code & EC)174 directory_iterator &increment(std::error_code &EC) { 175 assert(Impl && "attempting to increment past end"); 176 EC = Impl->increment(); 177 if (Impl->CurrentEntry.path().empty()) 178 Impl.reset(); // Normalize the end iterator to Impl == nullptr. 179 return *this; 180 } 181 182 const directory_entry &operator*() const { return Impl->CurrentEntry; } 183 const directory_entry *operator->() const { return &Impl->CurrentEntry; } 184 185 bool operator==(const directory_iterator &RHS) const { 186 if (Impl && RHS.Impl) 187 return Impl->CurrentEntry.path() == RHS.Impl->CurrentEntry.path(); 188 return !Impl && !RHS.Impl; 189 } 190 bool operator!=(const directory_iterator &RHS) const { 191 return !(*this == RHS); 192 } 193 }; 194 195 class FileSystem; 196 197 namespace detail { 198 199 /// Keeps state for the recursive_directory_iterator. 200 struct RecDirIterState { 201 std::stack<directory_iterator, std::vector<directory_iterator>> Stack; 202 bool HasNoPushRequest = false; 203 }; 204 205 } // end namespace detail 206 207 /// An input iterator over the recursive contents of a virtual path, 208 /// similar to llvm::sys::fs::recursive_directory_iterator. 209 class recursive_directory_iterator { 210 FileSystem *FS; 211 std::shared_ptr<detail::RecDirIterState> 212 State; // Input iterator semantics on copy. 213 214 public: 215 recursive_directory_iterator(FileSystem &FS, const Twine &Path, 216 std::error_code &EC); 217 218 /// Construct an 'end' iterator. 219 recursive_directory_iterator() = default; 220 221 /// Equivalent to operator++, with an error code. 222 recursive_directory_iterator &increment(std::error_code &EC); 223 224 const directory_entry &operator*() const { return *State->Stack.top(); } 225 const directory_entry *operator->() const { return &*State->Stack.top(); } 226 227 bool operator==(const recursive_directory_iterator &Other) const { 228 return State == Other.State; // identity 229 } 230 bool operator!=(const recursive_directory_iterator &RHS) const { 231 return !(*this == RHS); 232 } 233 234 /// Gets the current level. Starting path is at level 0. level()235 int level() const { 236 assert(!State->Stack.empty() && 237 "Cannot get level without any iteration state"); 238 return State->Stack.size() - 1; 239 } 240 no_push()241 void no_push() { State->HasNoPushRequest = true; } 242 }; 243 244 /// The virtual file system interface. 245 class FileSystem : public llvm::ThreadSafeRefCountedBase<FileSystem> { 246 public: 247 virtual ~FileSystem(); 248 249 /// Get the status of the entry at \p Path, if one exists. 250 virtual llvm::ErrorOr<Status> status(const Twine &Path) = 0; 251 252 /// Get a \p File object for the file at \p Path, if one exists. 253 virtual llvm::ErrorOr<std::unique_ptr<File>> 254 openFileForRead(const Twine &Path) = 0; 255 256 /// This is a convenience method that opens a file, gets its content and then 257 /// closes the file. 258 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> 259 getBufferForFile(const Twine &Name, int64_t FileSize = -1, 260 bool RequiresNullTerminator = true, bool IsVolatile = false); 261 262 /// Get a directory_iterator for \p Dir. 263 /// \note The 'end' iterator is directory_iterator(). 264 virtual directory_iterator dir_begin(const Twine &Dir, 265 std::error_code &EC) = 0; 266 267 /// Set the working directory. This will affect all following operations on 268 /// this file system and may propagate down for nested file systems. 269 virtual std::error_code setCurrentWorkingDirectory(const Twine &Path) = 0; 270 271 /// Get the working directory of this file system. 272 virtual llvm::ErrorOr<std::string> getCurrentWorkingDirectory() const = 0; 273 274 /// Gets real path of \p Path e.g. collapse all . and .. patterns, resolve 275 /// symlinks. For real file system, this uses `llvm::sys::fs::real_path`. 276 /// This returns errc::operation_not_permitted if not implemented by subclass. 277 virtual std::error_code getRealPath(const Twine &Path, 278 SmallVectorImpl<char> &Output) const; 279 280 /// Check whether a file exists. Provided for convenience. 281 bool exists(const Twine &Path); 282 283 /// Is the file mounted on a local filesystem? 284 virtual std::error_code isLocal(const Twine &Path, bool &Result); 285 286 /// Make \a Path an absolute path. 287 /// 288 /// Makes \a Path absolute using the current directory if it is not already. 289 /// An empty \a Path will result in the current directory. 290 /// 291 /// /absolute/path => /absolute/path 292 /// relative/../path => <current-directory>/relative/../path 293 /// 294 /// \param Path A path that is modified to be an absolute path. 295 /// \returns success if \a path has been made absolute, otherwise a 296 /// platform-specific error_code. 297 virtual std::error_code makeAbsolute(SmallVectorImpl<char> &Path) const; 298 }; 299 300 /// Gets an \p vfs::FileSystem for the 'real' file system, as seen by 301 /// the operating system. 302 /// The working directory is linked to the process's working directory. 303 /// (This is usually thread-hostile). 304 IntrusiveRefCntPtr<FileSystem> getRealFileSystem(); 305 306 /// Create an \p vfs::FileSystem for the 'real' file system, as seen by 307 /// the operating system. 308 /// It has its own working directory, independent of (but initially equal to) 309 /// that of the process. 310 std::unique_ptr<FileSystem> createPhysicalFileSystem(); 311 312 /// A file system that allows overlaying one \p AbstractFileSystem on top 313 /// of another. 314 /// 315 /// Consists of a stack of >=1 \p FileSystem objects, which are treated as being 316 /// one merged file system. When there is a directory that exists in more than 317 /// one file system, the \p OverlayFileSystem contains a directory containing 318 /// the union of their contents. The attributes (permissions, etc.) of the 319 /// top-most (most recently added) directory are used. When there is a file 320 /// that exists in more than one file system, the file in the top-most file 321 /// system overrides the other(s). 322 class OverlayFileSystem : public FileSystem { 323 using FileSystemList = SmallVector<IntrusiveRefCntPtr<FileSystem>, 1>; 324 325 /// The stack of file systems, implemented as a list in order of 326 /// their addition. 327 FileSystemList FSList; 328 329 public: 330 OverlayFileSystem(IntrusiveRefCntPtr<FileSystem> Base); 331 332 /// Pushes a file system on top of the stack. 333 void pushOverlay(IntrusiveRefCntPtr<FileSystem> FS); 334 335 llvm::ErrorOr<Status> status(const Twine &Path) override; 336 llvm::ErrorOr<std::unique_ptr<File>> 337 openFileForRead(const Twine &Path) override; 338 directory_iterator dir_begin(const Twine &Dir, std::error_code &EC) override; 339 llvm::ErrorOr<std::string> getCurrentWorkingDirectory() const override; 340 std::error_code setCurrentWorkingDirectory(const Twine &Path) override; 341 std::error_code isLocal(const Twine &Path, bool &Result) override; 342 std::error_code getRealPath(const Twine &Path, 343 SmallVectorImpl<char> &Output) const override; 344 345 using iterator = FileSystemList::reverse_iterator; 346 using const_iterator = FileSystemList::const_reverse_iterator; 347 using reverse_iterator = FileSystemList::iterator; 348 using const_reverse_iterator = FileSystemList::const_iterator; 349 350 /// Get an iterator pointing to the most recently added file system. overlays_begin()351 iterator overlays_begin() { return FSList.rbegin(); } overlays_begin()352 const_iterator overlays_begin() const { return FSList.rbegin(); } 353 354 /// Get an iterator pointing one-past the least recently added file system. overlays_end()355 iterator overlays_end() { return FSList.rend(); } overlays_end()356 const_iterator overlays_end() const { return FSList.rend(); } 357 358 /// Get an iterator pointing to the least recently added file system. overlays_rbegin()359 reverse_iterator overlays_rbegin() { return FSList.begin(); } overlays_rbegin()360 const_reverse_iterator overlays_rbegin() const { return FSList.begin(); } 361 362 /// Get an iterator pointing one-past the most recently added file system. overlays_rend()363 reverse_iterator overlays_rend() { return FSList.end(); } overlays_rend()364 const_reverse_iterator overlays_rend() const { return FSList.end(); } 365 }; 366 367 /// By default, this delegates all calls to the underlying file system. This 368 /// is useful when derived file systems want to override some calls and still 369 /// proxy other calls. 370 class ProxyFileSystem : public FileSystem { 371 public: ProxyFileSystem(IntrusiveRefCntPtr<FileSystem> FS)372 explicit ProxyFileSystem(IntrusiveRefCntPtr<FileSystem> FS) 373 : FS(std::move(FS)) {} 374 status(const Twine & Path)375 llvm::ErrorOr<Status> status(const Twine &Path) override { 376 return FS->status(Path); 377 } 378 llvm::ErrorOr<std::unique_ptr<File>> openFileForRead(const Twine & Path)379 openFileForRead(const Twine &Path) override { 380 return FS->openFileForRead(Path); 381 } dir_begin(const Twine & Dir,std::error_code & EC)382 directory_iterator dir_begin(const Twine &Dir, std::error_code &EC) override { 383 return FS->dir_begin(Dir, EC); 384 } getCurrentWorkingDirectory()385 llvm::ErrorOr<std::string> getCurrentWorkingDirectory() const override { 386 return FS->getCurrentWorkingDirectory(); 387 } setCurrentWorkingDirectory(const Twine & Path)388 std::error_code setCurrentWorkingDirectory(const Twine &Path) override { 389 return FS->setCurrentWorkingDirectory(Path); 390 } getRealPath(const Twine & Path,SmallVectorImpl<char> & Output)391 std::error_code getRealPath(const Twine &Path, 392 SmallVectorImpl<char> &Output) const override { 393 return FS->getRealPath(Path, Output); 394 } isLocal(const Twine & Path,bool & Result)395 std::error_code isLocal(const Twine &Path, bool &Result) override { 396 return FS->isLocal(Path, Result); 397 } 398 399 protected: getUnderlyingFS()400 FileSystem &getUnderlyingFS() { return *FS; } 401 402 private: 403 IntrusiveRefCntPtr<FileSystem> FS; 404 405 virtual void anchor(); 406 }; 407 408 namespace detail { 409 410 class InMemoryDirectory; 411 class InMemoryFile; 412 413 } // namespace detail 414 415 /// An in-memory file system. 416 class InMemoryFileSystem : public FileSystem { 417 std::unique_ptr<detail::InMemoryDirectory> Root; 418 std::string WorkingDirectory; 419 bool UseNormalizedPaths = true; 420 421 /// If HardLinkTarget is non-null, a hardlink is created to the To path which 422 /// must be a file. If it is null then it adds the file as the public addFile. 423 bool addFile(const Twine &Path, time_t ModificationTime, 424 std::unique_ptr<llvm::MemoryBuffer> Buffer, 425 Optional<uint32_t> User, Optional<uint32_t> Group, 426 Optional<llvm::sys::fs::file_type> Type, 427 Optional<llvm::sys::fs::perms> Perms, 428 const detail::InMemoryFile *HardLinkTarget); 429 430 public: 431 explicit InMemoryFileSystem(bool UseNormalizedPaths = true); 432 ~InMemoryFileSystem() override; 433 434 /// Add a file containing a buffer or a directory to the VFS with a 435 /// path. The VFS owns the buffer. If present, User, Group, Type 436 /// and Perms apply to the newly-created file or directory. 437 /// \return true if the file or directory was successfully added, 438 /// false if the file or directory already exists in the file system with 439 /// different contents. 440 bool addFile(const Twine &Path, time_t ModificationTime, 441 std::unique_ptr<llvm::MemoryBuffer> Buffer, 442 Optional<uint32_t> User = None, Optional<uint32_t> Group = None, 443 Optional<llvm::sys::fs::file_type> Type = None, 444 Optional<llvm::sys::fs::perms> Perms = None); 445 446 /// Add a hard link to a file. 447 /// Here hard links are not intended to be fully equivalent to the classical 448 /// filesystem. Both the hard link and the file share the same buffer and 449 /// status (and thus have the same UniqueID). Because of this there is no way 450 /// to distinguish between the link and the file after the link has been 451 /// added. 452 /// 453 /// The To path must be an existing file or a hardlink. The From file must not 454 /// have been added before. The To Path must not be a directory. The From Node 455 /// is added as a hard link which points to the resolved file of To Node. 456 /// \return true if the above condition is satisfied and hardlink was 457 /// successfully created, false otherwise. 458 bool addHardLink(const Twine &From, const Twine &To); 459 460 /// Add a buffer to the VFS with a path. The VFS does not own the buffer. 461 /// If present, User, Group, Type and Perms apply to the newly-created file 462 /// or directory. 463 /// \return true if the file or directory was successfully added, 464 /// false if the file or directory already exists in the file system with 465 /// different contents. 466 bool addFileNoOwn(const Twine &Path, time_t ModificationTime, 467 const llvm::MemoryBufferRef &Buffer, 468 Optional<uint32_t> User = None, 469 Optional<uint32_t> Group = None, 470 Optional<llvm::sys::fs::file_type> Type = None, 471 Optional<llvm::sys::fs::perms> Perms = None); 472 473 std::string toString() const; 474 475 /// Return true if this file system normalizes . and .. in paths. useNormalizedPaths()476 bool useNormalizedPaths() const { return UseNormalizedPaths; } 477 478 llvm::ErrorOr<Status> status(const Twine &Path) override; 479 llvm::ErrorOr<std::unique_ptr<File>> 480 openFileForRead(const Twine &Path) override; 481 directory_iterator dir_begin(const Twine &Dir, std::error_code &EC) override; 482 getCurrentWorkingDirectory()483 llvm::ErrorOr<std::string> getCurrentWorkingDirectory() const override { 484 return WorkingDirectory; 485 } 486 /// Canonicalizes \p Path by combining with the current working 487 /// directory and normalizing the path (e.g. remove dots). If the current 488 /// working directory is not set, this returns errc::operation_not_permitted. 489 /// 490 /// This doesn't resolve symlinks as they are not supported in in-memory file 491 /// system. 492 std::error_code getRealPath(const Twine &Path, 493 SmallVectorImpl<char> &Output) const override; 494 std::error_code isLocal(const Twine &Path, bool &Result) override; 495 std::error_code setCurrentWorkingDirectory(const Twine &Path) override; 496 }; 497 498 /// Get a globally unique ID for a virtual file or directory. 499 llvm::sys::fs::UniqueID getNextVirtualUniqueID(); 500 501 /// Gets a \p FileSystem for a virtual file system described in YAML 502 /// format. 503 std::unique_ptr<FileSystem> 504 getVFSFromYAML(std::unique_ptr<llvm::MemoryBuffer> Buffer, 505 llvm::SourceMgr::DiagHandlerTy DiagHandler, 506 StringRef YAMLFilePath, void *DiagContext = nullptr, 507 IntrusiveRefCntPtr<FileSystem> ExternalFS = getRealFileSystem()); 508 509 struct YAMLVFSEntry { 510 template <typename T1, typename T2> 511 YAMLVFSEntry(T1 &&VPath, T2 &&RPath, bool IsDirectory = false) VPathYAMLVFSEntry512 : VPath(std::forward<T1>(VPath)), RPath(std::forward<T2>(RPath)), 513 IsDirectory(IsDirectory) {} 514 std::string VPath; 515 std::string RPath; 516 bool IsDirectory = false; 517 }; 518 519 class RedirectingFSDirIterImpl; 520 class RedirectingFileSystemParser; 521 522 /// A virtual file system parsed from a YAML file. 523 /// 524 /// Currently, this class allows creating virtual files and directories. Virtual 525 /// files map to existing external files in \c ExternalFS, and virtual 526 /// directories may either map to existing directories in \c ExternalFS or list 527 /// their contents in the form of other virtual directories and/or files. 528 /// 529 /// The basic structure of the parsed file is: 530 /// \verbatim 531 /// { 532 /// 'version': <version number>, 533 /// <optional configuration> 534 /// 'roots': [ 535 /// <directory entries> 536 /// ] 537 /// } 538 /// \endverbatim 539 /// 540 /// All configuration options are optional. 541 /// 'case-sensitive': <boolean, default=(true for Posix, false for Windows)> 542 /// 'use-external-names': <boolean, default=true> 543 /// 'overlay-relative': <boolean, default=false> 544 /// 'fallthrough': <boolean, default=true> 545 /// 546 /// Virtual directories that list their contents are represented as 547 /// \verbatim 548 /// { 549 /// 'type': 'directory', 550 /// 'name': <string>, 551 /// 'contents': [ <file or directory entries> ] 552 /// } 553 /// \endverbatim 554 /// 555 /// The default attributes for such virtual directories are: 556 /// \verbatim 557 /// MTime = now() when created 558 /// Perms = 0777 559 /// User = Group = 0 560 /// Size = 0 561 /// UniqueID = unspecified unique value 562 /// \endverbatim 563 /// 564 /// When a path prefix matches such a directory, the next component in the path 565 /// is matched against the entries in the 'contents' array. 566 /// 567 /// Re-mapped directories, on the other hand, are represented as 568 /// /// \verbatim 569 /// { 570 /// 'type': 'directory-remap', 571 /// 'name': <string>, 572 /// 'use-external-name': <boolean>, # Optional 573 /// 'external-contents': <path to external directory> 574 /// } 575 /// \endverbatim 576 /// 577 /// and inherit their attributes from the external directory. When a path 578 /// prefix matches such an entry, the unmatched components are appended to the 579 /// 'external-contents' path, and the resulting path is looked up in the 580 /// external file system instead. 581 /// 582 /// Re-mapped files are represented as 583 /// \verbatim 584 /// { 585 /// 'type': 'file', 586 /// 'name': <string>, 587 /// 'use-external-name': <boolean>, # Optional 588 /// 'external-contents': <path to external file> 589 /// } 590 /// \endverbatim 591 /// 592 /// Their attributes and file contents are determined by looking up the file at 593 /// their 'external-contents' path in the external file system. 594 /// 595 /// For 'file', 'directory' and 'directory-remap' entries the 'name' field may 596 /// contain multiple path components (e.g. /path/to/file). However, any 597 /// directory in such a path that contains more than one child must be uniquely 598 /// represented by a 'directory' entry. 599 class RedirectingFileSystem : public vfs::FileSystem { 600 public: 601 enum EntryKind { EK_Directory, EK_DirectoryRemap, EK_File }; 602 enum NameKind { NK_NotSet, NK_External, NK_Virtual }; 603 604 /// A single file or directory in the VFS. 605 class Entry { 606 EntryKind Kind; 607 std::string Name; 608 609 public: Entry(EntryKind K,StringRef Name)610 Entry(EntryKind K, StringRef Name) : Kind(K), Name(Name) {} 611 virtual ~Entry() = default; 612 getName()613 StringRef getName() const { return Name; } getKind()614 EntryKind getKind() const { return Kind; } 615 }; 616 617 /// A directory in the vfs with explicitly specified contents. 618 class DirectoryEntry : public Entry { 619 std::vector<std::unique_ptr<Entry>> Contents; 620 Status S; 621 622 public: 623 /// Constructs a directory entry with explicitly specified contents. DirectoryEntry(StringRef Name,std::vector<std::unique_ptr<Entry>> Contents,Status S)624 DirectoryEntry(StringRef Name, std::vector<std::unique_ptr<Entry>> Contents, 625 Status S) 626 : Entry(EK_Directory, Name), Contents(std::move(Contents)), 627 S(std::move(S)) {} 628 629 /// Constructs an empty directory entry. DirectoryEntry(StringRef Name,Status S)630 DirectoryEntry(StringRef Name, Status S) 631 : Entry(EK_Directory, Name), S(std::move(S)) {} 632 getStatus()633 Status getStatus() { return S; } 634 addContent(std::unique_ptr<Entry> Content)635 void addContent(std::unique_ptr<Entry> Content) { 636 Contents.push_back(std::move(Content)); 637 } 638 getLastContent()639 Entry *getLastContent() const { return Contents.back().get(); } 640 641 using iterator = decltype(Contents)::iterator; 642 contents_begin()643 iterator contents_begin() { return Contents.begin(); } contents_end()644 iterator contents_end() { return Contents.end(); } 645 classof(const Entry * E)646 static bool classof(const Entry *E) { return E->getKind() == EK_Directory; } 647 }; 648 649 /// A file or directory in the vfs that is mapped to a file or directory in 650 /// the external filesystem. 651 class RemapEntry : public Entry { 652 std::string ExternalContentsPath; 653 NameKind UseName; 654 655 protected: RemapEntry(EntryKind K,StringRef Name,StringRef ExternalContentsPath,NameKind UseName)656 RemapEntry(EntryKind K, StringRef Name, StringRef ExternalContentsPath, 657 NameKind UseName) 658 : Entry(K, Name), ExternalContentsPath(ExternalContentsPath), 659 UseName(UseName) {} 660 661 public: getExternalContentsPath()662 StringRef getExternalContentsPath() const { return ExternalContentsPath; } 663 664 /// Whether to use the external path as the name for this file or directory. useExternalName(bool GlobalUseExternalName)665 bool useExternalName(bool GlobalUseExternalName) const { 666 return UseName == NK_NotSet ? GlobalUseExternalName 667 : (UseName == NK_External); 668 } 669 getUseName()670 NameKind getUseName() const { return UseName; } 671 classof(const Entry * E)672 static bool classof(const Entry *E) { 673 switch (E->getKind()) { 674 case EK_DirectoryRemap: 675 LLVM_FALLTHROUGH; 676 case EK_File: 677 return true; 678 case EK_Directory: 679 return false; 680 } 681 llvm_unreachable("invalid entry kind"); 682 } 683 }; 684 685 /// A directory in the vfs that maps to a directory in the external file 686 /// system. 687 class DirectoryRemapEntry : public RemapEntry { 688 public: DirectoryRemapEntry(StringRef Name,StringRef ExternalContentsPath,NameKind UseName)689 DirectoryRemapEntry(StringRef Name, StringRef ExternalContentsPath, 690 NameKind UseName) 691 : RemapEntry(EK_DirectoryRemap, Name, ExternalContentsPath, UseName) {} 692 classof(const Entry * E)693 static bool classof(const Entry *E) { 694 return E->getKind() == EK_DirectoryRemap; 695 } 696 }; 697 698 /// A file in the vfs that maps to a file in the external file system. 699 class FileEntry : public RemapEntry { 700 public: FileEntry(StringRef Name,StringRef ExternalContentsPath,NameKind UseName)701 FileEntry(StringRef Name, StringRef ExternalContentsPath, NameKind UseName) 702 : RemapEntry(EK_File, Name, ExternalContentsPath, UseName) {} 703 classof(const Entry * E)704 static bool classof(const Entry *E) { return E->getKind() == EK_File; } 705 }; 706 707 /// Represents the result of a path lookup into the RedirectingFileSystem. 708 struct LookupResult { 709 /// The entry the looked-up path corresponds to. 710 Entry *E; 711 712 private: 713 /// When the found Entry is a DirectoryRemapEntry, stores the path in the 714 /// external file system that the looked-up path in the virtual file system 715 // corresponds to. 716 Optional<std::string> ExternalRedirect; 717 718 public: 719 LookupResult(Entry *E, sys::path::const_iterator Start, 720 sys::path::const_iterator End); 721 722 /// If the found Entry maps the the input path to a path in the external 723 /// file system (i.e. it is a FileEntry or DirectoryRemapEntry), returns 724 /// that path. getExternalRedirectLookupResult725 Optional<StringRef> getExternalRedirect() const { 726 if (isa<DirectoryRemapEntry>(E)) 727 return StringRef(*ExternalRedirect); 728 if (auto *FE = dyn_cast<FileEntry>(E)) 729 return FE->getExternalContentsPath(); 730 return None; 731 } 732 }; 733 734 private: 735 friend class RedirectingFSDirIterImpl; 736 friend class RedirectingFileSystemParser; 737 shouldUseExternalFS()738 bool shouldUseExternalFS() const { return IsFallthrough; } 739 740 /// Canonicalize path by removing ".", "..", "./", components. This is 741 /// a VFS request, do not bother about symlinks in the path components 742 /// but canonicalize in order to perform the correct entry search. 743 std::error_code makeCanonical(SmallVectorImpl<char> &Path) const; 744 745 /// Whether to fall back to the external file system when an operation fails 746 /// with the given error code on a path associated with the provided Entry. 747 bool shouldFallBackToExternalFS(std::error_code EC, Entry *E = nullptr) const; 748 749 // In a RedirectingFileSystem, keys can be specified in Posix or Windows 750 // style (or even a mixture of both), so this comparison helper allows 751 // slashes (representing a root) to match backslashes (and vice versa). Note 752 // that, other than the root, path components should not contain slashes or 753 // backslashes. pathComponentMatches(llvm::StringRef lhs,llvm::StringRef rhs)754 bool pathComponentMatches(llvm::StringRef lhs, llvm::StringRef rhs) const { 755 if ((CaseSensitive ? lhs.equals(rhs) : lhs.equals_insensitive(rhs))) 756 return true; 757 return (lhs == "/" && rhs == "\\") || (lhs == "\\" && rhs == "/"); 758 } 759 760 /// The root(s) of the virtual file system. 761 std::vector<std::unique_ptr<Entry>> Roots; 762 763 /// The current working directory of the file system. 764 std::string WorkingDirectory; 765 766 /// The file system to use for external references. 767 IntrusiveRefCntPtr<FileSystem> ExternalFS; 768 769 /// If IsRelativeOverlay is set, this represents the directory 770 /// path that should be prefixed to each 'external-contents' entry 771 /// when reading from YAML files. 772 std::string ExternalContentsPrefixDir; 773 774 /// @name Configuration 775 /// @{ 776 777 /// Whether to perform case-sensitive comparisons. 778 /// 779 /// Currently, case-insensitive matching only works correctly with ASCII. 780 bool CaseSensitive = 781 #ifdef _WIN32 782 false; 783 #else 784 true; 785 #endif 786 787 /// IsRelativeOverlay marks whether a ExternalContentsPrefixDir path must 788 /// be prefixed in every 'external-contents' when reading from YAML files. 789 bool IsRelativeOverlay = false; 790 791 /// Whether to use to use the value of 'external-contents' for the 792 /// names of files. This global value is overridable on a per-file basis. 793 bool UseExternalNames = true; 794 795 /// Whether to attempt a file lookup in external file system after it wasn't 796 /// found in VFS. 797 bool IsFallthrough = true; 798 /// @} 799 800 RedirectingFileSystem(IntrusiveRefCntPtr<FileSystem> ExternalFS); 801 802 /// Looks up the path <tt>[Start, End)</tt> in \p From, possibly recursing 803 /// into the contents of \p From if it is a directory. Returns a LookupResult 804 /// giving the matched entry and, if that entry is a FileEntry or 805 /// DirectoryRemapEntry, the path it redirects to in the external file system. 806 ErrorOr<LookupResult> lookupPathImpl(llvm::sys::path::const_iterator Start, 807 llvm::sys::path::const_iterator End, 808 Entry *From) const; 809 810 /// Get the status for a path with the provided \c LookupResult. 811 ErrorOr<Status> status(const Twine &Path, const LookupResult &Result); 812 813 public: 814 /// Looks up \p Path in \c Roots and returns a LookupResult giving the 815 /// matched entry and, if the entry was a FileEntry or DirectoryRemapEntry, 816 /// the path it redirects to in the external file system. 817 ErrorOr<LookupResult> lookupPath(StringRef Path) const; 818 819 /// Parses \p Buffer, which is expected to be in YAML format and 820 /// returns a virtual file system representing its contents. 821 static std::unique_ptr<RedirectingFileSystem> 822 create(std::unique_ptr<MemoryBuffer> Buffer, 823 SourceMgr::DiagHandlerTy DiagHandler, StringRef YAMLFilePath, 824 void *DiagContext, IntrusiveRefCntPtr<FileSystem> ExternalFS); 825 826 /// Redirect each of the remapped files from first to second. 827 static std::unique_ptr<RedirectingFileSystem> 828 create(ArrayRef<std::pair<std::string, std::string>> RemappedFiles, 829 bool UseExternalNames, FileSystem &ExternalFS); 830 831 ErrorOr<Status> status(const Twine &Path) override; 832 ErrorOr<std::unique_ptr<File>> openFileForRead(const Twine &Path) override; 833 834 std::error_code getRealPath(const Twine &Path, 835 SmallVectorImpl<char> &Output) const override; 836 837 llvm::ErrorOr<std::string> getCurrentWorkingDirectory() const override; 838 839 std::error_code setCurrentWorkingDirectory(const Twine &Path) override; 840 841 std::error_code isLocal(const Twine &Path, bool &Result) override; 842 843 std::error_code makeAbsolute(SmallVectorImpl<char> &Path) const override; 844 845 directory_iterator dir_begin(const Twine &Dir, std::error_code &EC) override; 846 847 void setExternalContentsPrefixDir(StringRef PrefixDir); 848 849 StringRef getExternalContentsPrefixDir() const; 850 851 void setFallthrough(bool Fallthrough); 852 853 std::vector<llvm::StringRef> getRoots() const; 854 855 void dump(raw_ostream &OS) const; 856 void dumpEntry(raw_ostream &OS, Entry *E, int NumSpaces = 0) const; 857 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) 858 LLVM_DUMP_METHOD void dump() const; 859 #endif 860 }; 861 862 /// Collect all pairs of <virtual path, real path> entries from the 863 /// \p YAMLFilePath. This is used by the module dependency collector to forward 864 /// the entries into the reproducer output VFS YAML file. 865 void collectVFSFromYAML( 866 std::unique_ptr<llvm::MemoryBuffer> Buffer, 867 llvm::SourceMgr::DiagHandlerTy DiagHandler, StringRef YAMLFilePath, 868 SmallVectorImpl<YAMLVFSEntry> &CollectedEntries, 869 void *DiagContext = nullptr, 870 IntrusiveRefCntPtr<FileSystem> ExternalFS = getRealFileSystem()); 871 872 class YAMLVFSWriter { 873 std::vector<YAMLVFSEntry> Mappings; 874 Optional<bool> IsCaseSensitive; 875 Optional<bool> IsOverlayRelative; 876 Optional<bool> UseExternalNames; 877 std::string OverlayDir; 878 879 void addEntry(StringRef VirtualPath, StringRef RealPath, bool IsDirectory); 880 881 public: 882 YAMLVFSWriter() = default; 883 884 void addFileMapping(StringRef VirtualPath, StringRef RealPath); 885 void addDirectoryMapping(StringRef VirtualPath, StringRef RealPath); 886 setCaseSensitivity(bool CaseSensitive)887 void setCaseSensitivity(bool CaseSensitive) { 888 IsCaseSensitive = CaseSensitive; 889 } 890 setUseExternalNames(bool UseExtNames)891 void setUseExternalNames(bool UseExtNames) { UseExternalNames = UseExtNames; } 892 setOverlayDir(StringRef OverlayDirectory)893 void setOverlayDir(StringRef OverlayDirectory) { 894 IsOverlayRelative = true; 895 OverlayDir.assign(OverlayDirectory.str()); 896 } 897 getMappings()898 const std::vector<YAMLVFSEntry> &getMappings() const { return Mappings; } 899 900 void write(llvm::raw_ostream &OS); 901 }; 902 903 } // namespace vfs 904 } // namespace llvm 905 906 #endif // LLVM_SUPPORT_VIRTUALFILESYSTEM_H 907