| 1 | //===--- FileManager.cpp - File System Probing and Caching ----------------===// |
| 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 FileManager interface. |
| 10 | // |
| 11 | //===----------------------------------------------------------------------===// |
| 12 | // |
| 13 | // TODO: This should index all interesting directories with dirent calls. |
| 14 | // getdirentries ? |
| 15 | // opendir/readdir_r/closedir ? |
| 16 | // |
| 17 | //===----------------------------------------------------------------------===// |
| 18 | |
| 19 | #include "clang/Basic/FileManager.h" |
| 20 | #include "clang/Basic/FileSystemStatCache.h" |
| 21 | #include "llvm/ADT/SmallString.h" |
| 22 | #include "llvm/ADT/Statistic.h" |
| 23 | #include "llvm/Config/llvm-config.h" |
| 24 | #include "llvm/Support/FileSystem.h" |
| 25 | #include "llvm/Support/MemoryBuffer.h" |
| 26 | #include "llvm/Support/Path.h" |
| 27 | #include "llvm/Support/raw_ostream.h" |
| 28 | #include <cassert> |
| 29 | #include <climits> |
| 30 | #include <cstdint> |
| 31 | #include <cstdlib> |
| 32 | #include <optional> |
| 33 | #include <string> |
| 34 | #include <utility> |
| 35 | |
| 36 | using namespace clang; |
| 37 | |
| 38 | #define DEBUG_TYPE "file-search" |
| 39 | |
| 40 | //===----------------------------------------------------------------------===// |
| 41 | // Common logic. |
| 42 | //===----------------------------------------------------------------------===// |
| 43 | |
| 44 | FileManager::FileManager(const FileSystemOptions &FSO, |
| 45 | IntrusiveRefCntPtr<llvm::vfs::FileSystem> FS) |
| 46 | : FS(std::move(FS)), FileSystemOpts(FSO), SeenDirEntries(64), |
| 47 | SeenFileEntries(64), NextFileUID(0) { |
| 48 | // If the caller doesn't provide a virtual file system, just grab the real |
| 49 | // file system. |
| 50 | if (!this->FS) |
| 51 | this->FS = llvm::vfs::getRealFileSystem(); |
| 52 | } |
| 53 | |
| 54 | FileManager::~FileManager() = default; |
| 55 | |
| 56 | void FileManager::setStatCache(std::unique_ptr<FileSystemStatCache> statCache) { |
| 57 | assert(statCache && "No stat cache provided?" ); |
| 58 | StatCache = std::move(statCache); |
| 59 | } |
| 60 | |
| 61 | void FileManager::clearStatCache() { StatCache.reset(); } |
| 62 | |
| 63 | /// Retrieve the directory that the given file name resides in. |
| 64 | /// Filename can point to either a real file or a virtual file. |
| 65 | static llvm::Expected<DirectoryEntryRef> |
| 66 | getDirectoryFromFile(FileManager &FileMgr, StringRef Filename, |
| 67 | bool CacheFailure) { |
| 68 | if (Filename.empty()) |
| 69 | return llvm::errorCodeToError( |
| 70 | EC: make_error_code(e: std::errc::no_such_file_or_directory)); |
| 71 | |
| 72 | if (llvm::sys::path::is_separator(value: Filename[Filename.size() - 1])) |
| 73 | return llvm::errorCodeToError(EC: make_error_code(e: std::errc::is_a_directory)); |
| 74 | |
| 75 | StringRef DirName = llvm::sys::path::parent_path(path: Filename); |
| 76 | // Use the current directory if file has no path component. |
| 77 | if (DirName.empty()) |
| 78 | DirName = "." ; |
| 79 | |
| 80 | return FileMgr.getDirectoryRef(DirName, CacheFailure); |
| 81 | } |
| 82 | |
| 83 | DirectoryEntry *&FileManager::getRealDirEntry(const llvm::vfs::Status &Status) { |
| 84 | assert(Status.isDirectory() && "The directory should exist!" ); |
| 85 | // See if we have already opened a directory with the |
| 86 | // same inode (this occurs on Unix-like systems when one dir is |
| 87 | // symlinked to another, for example) or the same path (on |
| 88 | // Windows). |
| 89 | DirectoryEntry *&UDE = UniqueRealDirs[Status.getUniqueID()]; |
| 90 | |
| 91 | if (!UDE) { |
| 92 | // We don't have this directory yet, add it. We use the string |
| 93 | // key from the SeenDirEntries map as the string. |
| 94 | UDE = new (DirsAlloc.Allocate()) DirectoryEntry(); |
| 95 | } |
| 96 | return UDE; |
| 97 | } |
| 98 | |
| 99 | /// Add all ancestors of the given path (pointing to either a file or |
| 100 | /// a directory) as virtual directories. |
| 101 | void FileManager::addAncestorsAsVirtualDirs(StringRef Path) { |
| 102 | StringRef DirName = llvm::sys::path::parent_path(path: Path); |
| 103 | if (DirName.empty()) |
| 104 | DirName = "." ; |
| 105 | |
| 106 | auto &NamedDirEnt = *SeenDirEntries.insert( |
| 107 | KV: {DirName, std::errc::no_such_file_or_directory}).first; |
| 108 | |
| 109 | // When caching a virtual directory, we always cache its ancestors |
| 110 | // at the same time. Therefore, if DirName is already in the cache, |
| 111 | // we don't need to recurse as its ancestors must also already be in |
| 112 | // the cache (or it's a known non-virtual directory). |
| 113 | if (NamedDirEnt.second) |
| 114 | return; |
| 115 | |
| 116 | // Check to see if the directory exists. |
| 117 | llvm::vfs::Status Status; |
| 118 | auto statError = |
| 119 | getStatValue(Path: DirName, Status, isFile: false, F: nullptr /*directory lookup*/); |
| 120 | if (statError) { |
| 121 | // There's no real directory at the given path. |
| 122 | // Add the virtual directory to the cache. |
| 123 | auto *UDE = new (DirsAlloc.Allocate()) DirectoryEntry(); |
| 124 | NamedDirEnt.second = *UDE; |
| 125 | VirtualDirectoryEntries.push_back(Elt: UDE); |
| 126 | } else { |
| 127 | // There is the real directory |
| 128 | DirectoryEntry *&UDE = getRealDirEntry(Status); |
| 129 | NamedDirEnt.second = *UDE; |
| 130 | } |
| 131 | |
| 132 | // Recursively add the other ancestors. |
| 133 | addAncestorsAsVirtualDirs(Path: DirName); |
| 134 | } |
| 135 | |
| 136 | llvm::Expected<DirectoryEntryRef> |
| 137 | FileManager::getDirectoryRef(StringRef DirName, bool CacheFailure) { |
| 138 | // stat doesn't like trailing separators except for root directory. |
| 139 | // At least, on Win32 MSVCRT, stat() cannot strip trailing '/'. |
| 140 | // (though it can strip '\\') |
| 141 | if (DirName.size() > 1 && |
| 142 | DirName != llvm::sys::path::root_path(path: DirName) && |
| 143 | llvm::sys::path::is_separator(value: DirName.back())) |
| 144 | DirName = DirName.drop_back(); |
| 145 | std::optional<std::string> DirNameStr; |
| 146 | if (is_style_windows(S: llvm::sys::path::Style::native)) { |
| 147 | // Fixing a problem with "clang C:test.c" on Windows. |
| 148 | // Stat("C:") does not recognize "C:" as a valid directory |
| 149 | if (DirName.size() > 1 && DirName.back() == ':' && |
| 150 | DirName.equals_insensitive(RHS: llvm::sys::path::root_name(path: DirName))) { |
| 151 | DirNameStr = DirName.str() + '.'; |
| 152 | DirName = *DirNameStr; |
| 153 | } |
| 154 | } |
| 155 | |
| 156 | ++NumDirLookups; |
| 157 | |
| 158 | // See if there was already an entry in the map. Note that the map |
| 159 | // contains both virtual and real directories. |
| 160 | auto SeenDirInsertResult = |
| 161 | SeenDirEntries.insert(KV: {DirName, std::errc::no_such_file_or_directory}); |
| 162 | if (!SeenDirInsertResult.second) { |
| 163 | if (SeenDirInsertResult.first->second) |
| 164 | return DirectoryEntryRef(*SeenDirInsertResult.first); |
| 165 | return llvm::errorCodeToError(EC: SeenDirInsertResult.first->second.getError()); |
| 166 | } |
| 167 | |
| 168 | // We've not seen this before. Fill it in. |
| 169 | ++NumDirCacheMisses; |
| 170 | auto &NamedDirEnt = *SeenDirInsertResult.first; |
| 171 | assert(!NamedDirEnt.second && "should be newly-created" ); |
| 172 | |
| 173 | // Get the null-terminated directory name as stored as the key of the |
| 174 | // SeenDirEntries map. |
| 175 | StringRef InterndDirName = NamedDirEnt.first(); |
| 176 | |
| 177 | // Check to see if the directory exists. |
| 178 | llvm::vfs::Status Status; |
| 179 | auto statError = getStatValue(Path: InterndDirName, Status, isFile: false, |
| 180 | F: nullptr /*directory lookup*/); |
| 181 | if (statError) { |
| 182 | // There's no real directory at the given path. |
| 183 | if (CacheFailure) |
| 184 | NamedDirEnt.second = statError; |
| 185 | else |
| 186 | SeenDirEntries.erase(Key: DirName); |
| 187 | return llvm::errorCodeToError(EC: statError); |
| 188 | } |
| 189 | |
| 190 | // It exists. |
| 191 | DirectoryEntry *&UDE = getRealDirEntry(Status); |
| 192 | NamedDirEnt.second = *UDE; |
| 193 | |
| 194 | return DirectoryEntryRef(NamedDirEnt); |
| 195 | } |
| 196 | |
| 197 | llvm::Expected<FileEntryRef> FileManager::getFileRef(StringRef Filename, |
| 198 | bool openFile, |
| 199 | bool CacheFailure, |
| 200 | bool IsText) { |
| 201 | ++NumFileLookups; |
| 202 | |
| 203 | // See if there is already an entry in the map. |
| 204 | auto SeenFileInsertResult = |
| 205 | SeenFileEntries.insert(KV: {Filename, std::errc::no_such_file_or_directory}); |
| 206 | if (!SeenFileInsertResult.second) { |
| 207 | if (!SeenFileInsertResult.first->second) |
| 208 | return llvm::errorCodeToError( |
| 209 | EC: SeenFileInsertResult.first->second.getError()); |
| 210 | return FileEntryRef(*SeenFileInsertResult.first); |
| 211 | } |
| 212 | |
| 213 | // We've not seen this before. Fill it in. |
| 214 | ++NumFileCacheMisses; |
| 215 | auto *NamedFileEnt = &*SeenFileInsertResult.first; |
| 216 | assert(!NamedFileEnt->second && "should be newly-created" ); |
| 217 | |
| 218 | // Get the null-terminated file name as stored as the key of the |
| 219 | // SeenFileEntries map. |
| 220 | StringRef InterndFileName = NamedFileEnt->first(); |
| 221 | |
| 222 | // Look up the directory for the file. When looking up something like |
| 223 | // sys/foo.h we'll discover all of the search directories that have a 'sys' |
| 224 | // subdirectory. This will let us avoid having to waste time on known-to-fail |
| 225 | // searches when we go to find sys/bar.h, because all the search directories |
| 226 | // without a 'sys' subdir will get a cached failure result. |
| 227 | auto DirInfoOrErr = getDirectoryFromFile(FileMgr&: *this, Filename, CacheFailure); |
| 228 | if (!DirInfoOrErr) { // Directory doesn't exist, file can't exist. |
| 229 | std::error_code Err = errorToErrorCode(Err: DirInfoOrErr.takeError()); |
| 230 | if (CacheFailure) |
| 231 | NamedFileEnt->second = Err; |
| 232 | else |
| 233 | SeenFileEntries.erase(Key: Filename); |
| 234 | |
| 235 | return llvm::errorCodeToError(EC: Err); |
| 236 | } |
| 237 | DirectoryEntryRef DirInfo = *DirInfoOrErr; |
| 238 | |
| 239 | // FIXME: Use the directory info to prune this, before doing the stat syscall. |
| 240 | // FIXME: This will reduce the # syscalls. |
| 241 | |
| 242 | // Check to see if the file exists. |
| 243 | std::unique_ptr<llvm::vfs::File> F; |
| 244 | llvm::vfs::Status Status; |
| 245 | auto statError = getStatValue(Path: InterndFileName, Status, isFile: true, |
| 246 | F: openFile ? &F : nullptr, IsText); |
| 247 | if (statError) { |
| 248 | // There's no real file at the given path. |
| 249 | if (CacheFailure) |
| 250 | NamedFileEnt->second = statError; |
| 251 | else |
| 252 | SeenFileEntries.erase(Key: Filename); |
| 253 | |
| 254 | return llvm::errorCodeToError(EC: statError); |
| 255 | } |
| 256 | |
| 257 | assert((openFile || !F) && "undesired open file" ); |
| 258 | |
| 259 | // It exists. See if we have already opened a file with the same inode. |
| 260 | // This occurs when one dir is symlinked to another, for example. |
| 261 | FileEntry *&UFE = UniqueRealFiles[Status.getUniqueID()]; |
| 262 | bool ReusingEntry = UFE != nullptr; |
| 263 | if (!UFE) |
| 264 | UFE = new (FilesAlloc.Allocate()) FileEntry(); |
| 265 | |
| 266 | if (!Status.ExposesExternalVFSPath || Status.getName() == Filename) { |
| 267 | // Use the requested name. Set the FileEntry. |
| 268 | NamedFileEnt->second = FileEntryRef::MapValue(*UFE, DirInfo); |
| 269 | } else { |
| 270 | // Name mismatch. We need a redirect. First grab the actual entry we want |
| 271 | // to return. |
| 272 | // |
| 273 | // This redirection logic intentionally leaks the external name of a |
| 274 | // redirected file that uses 'use-external-name' in \a |
| 275 | // vfs::RedirectionFileSystem. This allows clang to report the external |
| 276 | // name to users (in diagnostics) and to tools that don't have access to |
| 277 | // the VFS (in debug info and dependency '.d' files). |
| 278 | // |
| 279 | // FIXME: This is pretty complex and has some very complicated interactions |
| 280 | // with the rest of clang. It's also inconsistent with how "real" |
| 281 | // filesystems behave and confuses parts of clang expect to see the |
| 282 | // name-as-accessed on the \a FileEntryRef. |
| 283 | // |
| 284 | // A potential plan to remove this is as follows - |
| 285 | // - Update callers such as `HeaderSearch::findUsableModuleForHeader()` |
| 286 | // to explicitly use the `getNameAsRequested()` rather than just using |
| 287 | // `getName()`. |
| 288 | // - Add a `FileManager::getExternalPath` API for explicitly getting the |
| 289 | // remapped external filename when there is one available. Adopt it in |
| 290 | // callers like diagnostics/deps reporting instead of calling |
| 291 | // `getName()` directly. |
| 292 | // - Switch the meaning of `FileEntryRef::getName()` to get the requested |
| 293 | // name, not the external name. Once that sticks, revert callers that |
| 294 | // want the requested name back to calling `getName()`. |
| 295 | // - Update the VFS to always return the requested name. This could also |
| 296 | // return the external name, or just have an API to request it |
| 297 | // lazily. The latter has the benefit of making accesses of the |
| 298 | // external path easily tracked, but may also require extra work than |
| 299 | // just returning up front. |
| 300 | // - (Optionally) Add an API to VFS to get the external filename lazily |
| 301 | // and update `FileManager::getExternalPath()` to use it instead. This |
| 302 | // has the benefit of making such accesses easily tracked, though isn't |
| 303 | // necessarily required (and could cause extra work than just adding to |
| 304 | // eg. `vfs::Status` up front). |
| 305 | auto &Redirection = |
| 306 | *SeenFileEntries |
| 307 | .insert(KV: {Status.getName(), FileEntryRef::MapValue(*UFE, DirInfo)}) |
| 308 | .first; |
| 309 | assert(isa<FileEntry *>(Redirection.second->V) && |
| 310 | "filename redirected to a non-canonical filename?" ); |
| 311 | assert(cast<FileEntry *>(Redirection.second->V) == UFE && |
| 312 | "filename from getStatValue() refers to wrong file" ); |
| 313 | |
| 314 | // Cache the redirection in the previously-inserted entry, still available |
| 315 | // in the tentative return value. |
| 316 | NamedFileEnt->second = FileEntryRef::MapValue(Redirection, DirInfo); |
| 317 | } |
| 318 | |
| 319 | FileEntryRef ReturnedRef(*NamedFileEnt); |
| 320 | if (ReusingEntry) { // Already have an entry with this inode, return it. |
| 321 | return ReturnedRef; |
| 322 | } |
| 323 | |
| 324 | // Otherwise, we don't have this file yet, add it. |
| 325 | UFE->Size = Status.getSize(); |
| 326 | UFE->ModTime = llvm::sys::toTimeT(TP: Status.getLastModificationTime()); |
| 327 | UFE->Dir = &DirInfo.getDirEntry(); |
| 328 | UFE->UID = NextFileUID++; |
| 329 | UFE->UniqueID = Status.getUniqueID(); |
| 330 | UFE->IsNamedPipe = Status.getType() == llvm::sys::fs::file_type::fifo_file; |
| 331 | UFE->IsDeviceFile = |
| 332 | Status.getType() == llvm::sys::fs::file_type::character_file; |
| 333 | UFE->File = std::move(F); |
| 334 | |
| 335 | if (UFE->File) { |
| 336 | if (auto PathName = UFE->File->getName()) |
| 337 | fillRealPathName(UFE, FileName: *PathName); |
| 338 | } else if (!openFile) { |
| 339 | // We should still fill the path even if we aren't opening the file. |
| 340 | fillRealPathName(UFE, FileName: InterndFileName); |
| 341 | } |
| 342 | return ReturnedRef; |
| 343 | } |
| 344 | |
| 345 | llvm::Expected<FileEntryRef> FileManager::getSTDIN() { |
| 346 | // Only read stdin once. |
| 347 | if (STDIN) |
| 348 | return *STDIN; |
| 349 | |
| 350 | std::unique_ptr<llvm::MemoryBuffer> Content; |
| 351 | if (auto ContentOrError = llvm::MemoryBuffer::getSTDIN()) |
| 352 | Content = std::move(*ContentOrError); |
| 353 | else |
| 354 | return llvm::errorCodeToError(EC: ContentOrError.getError()); |
| 355 | |
| 356 | STDIN = getVirtualFileRef(Filename: Content->getBufferIdentifier(), |
| 357 | Size: Content->getBufferSize(), ModificationTime: 0); |
| 358 | FileEntry &FE = const_cast<FileEntry &>(STDIN->getFileEntry()); |
| 359 | FE.Content = std::move(Content); |
| 360 | FE.IsNamedPipe = true; |
| 361 | return *STDIN; |
| 362 | } |
| 363 | |
| 364 | void FileManager::trackVFSUsage(bool Active) { |
| 365 | FS->visit(Callback: [Active](llvm::vfs::FileSystem &FileSys) { |
| 366 | if (auto *RFS = dyn_cast<llvm::vfs::RedirectingFileSystem>(Val: &FileSys)) |
| 367 | RFS->setUsageTrackingActive(Active); |
| 368 | }); |
| 369 | } |
| 370 | |
| 371 | const FileEntry *FileManager::getVirtualFile(StringRef Filename, off_t Size, |
| 372 | time_t ModificationTime) { |
| 373 | return &getVirtualFileRef(Filename, Size, ModificationTime).getFileEntry(); |
| 374 | } |
| 375 | |
| 376 | FileEntryRef FileManager::getVirtualFileRef(StringRef Filename, off_t Size, |
| 377 | time_t ModificationTime) { |
| 378 | ++NumFileLookups; |
| 379 | |
| 380 | // See if there is already an entry in the map for an existing file. |
| 381 | auto &NamedFileEnt = *SeenFileEntries.insert( |
| 382 | KV: {Filename, std::errc::no_such_file_or_directory}).first; |
| 383 | if (NamedFileEnt.second) { |
| 384 | FileEntryRef::MapValue Value = *NamedFileEnt.second; |
| 385 | if (LLVM_LIKELY(isa<FileEntry *>(Value.V))) |
| 386 | return FileEntryRef(NamedFileEnt); |
| 387 | return FileEntryRef(*cast<const FileEntryRef::MapEntry *>(Val&: Value.V)); |
| 388 | } |
| 389 | |
| 390 | // We've not seen this before, or the file is cached as non-existent. |
| 391 | ++NumFileCacheMisses; |
| 392 | addAncestorsAsVirtualDirs(Path: Filename); |
| 393 | FileEntry *UFE = nullptr; |
| 394 | |
| 395 | // Now that all ancestors of Filename are in the cache, the |
| 396 | // following call is guaranteed to find the DirectoryEntry from the |
| 397 | // cache. A virtual file can also have an empty filename, that could come |
| 398 | // from a source location preprocessor directive with an empty filename as |
| 399 | // an example, so we need to pretend it has a name to ensure a valid directory |
| 400 | // entry can be returned. |
| 401 | auto DirInfo = expectedToOptional(E: getDirectoryFromFile( |
| 402 | FileMgr&: *this, Filename: Filename.empty() ? "." : Filename, /*CacheFailure=*/true)); |
| 403 | assert(DirInfo && |
| 404 | "The directory of a virtual file should already be in the cache." ); |
| 405 | |
| 406 | // Check to see if the file exists. If so, drop the virtual file |
| 407 | llvm::vfs::Status Status; |
| 408 | const char *InterndFileName = NamedFileEnt.first().data(); |
| 409 | if (!getStatValue(Path: InterndFileName, Status, isFile: true, F: nullptr)) { |
| 410 | Status = llvm::vfs::Status( |
| 411 | Status.getName(), Status.getUniqueID(), |
| 412 | llvm::sys::toTimePoint(T: ModificationTime), |
| 413 | Status.getUser(), Status.getGroup(), Size, |
| 414 | Status.getType(), Status.getPermissions()); |
| 415 | |
| 416 | auto &RealFE = UniqueRealFiles[Status.getUniqueID()]; |
| 417 | if (RealFE) { |
| 418 | // If we had already opened this file, close it now so we don't |
| 419 | // leak the descriptor. We're not going to use the file |
| 420 | // descriptor anyway, since this is a virtual file. |
| 421 | if (RealFE->File) |
| 422 | RealFE->closeFile(); |
| 423 | // If we already have an entry with this inode, return it. |
| 424 | // |
| 425 | // FIXME: Surely this should add a reference by the new name, and return |
| 426 | // it instead... |
| 427 | NamedFileEnt.second = FileEntryRef::MapValue(*RealFE, *DirInfo); |
| 428 | return FileEntryRef(NamedFileEnt); |
| 429 | } |
| 430 | // File exists, but no entry - create it. |
| 431 | RealFE = new (FilesAlloc.Allocate()) FileEntry(); |
| 432 | RealFE->UniqueID = Status.getUniqueID(); |
| 433 | RealFE->IsNamedPipe = |
| 434 | Status.getType() == llvm::sys::fs::file_type::fifo_file; |
| 435 | fillRealPathName(UFE: RealFE, FileName: Status.getName()); |
| 436 | |
| 437 | UFE = RealFE; |
| 438 | } else { |
| 439 | // File does not exist, create a virtual entry. |
| 440 | UFE = new (FilesAlloc.Allocate()) FileEntry(); |
| 441 | VirtualFileEntries.push_back(Elt: UFE); |
| 442 | } |
| 443 | |
| 444 | NamedFileEnt.second = FileEntryRef::MapValue(*UFE, *DirInfo); |
| 445 | UFE->Size = Size; |
| 446 | UFE->ModTime = ModificationTime; |
| 447 | UFE->Dir = &DirInfo->getDirEntry(); |
| 448 | UFE->UID = NextFileUID++; |
| 449 | UFE->File.reset(); |
| 450 | return FileEntryRef(NamedFileEnt); |
| 451 | } |
| 452 | |
| 453 | OptionalFileEntryRef FileManager::getBypassFile(FileEntryRef VF) { |
| 454 | // Stat of the file and return nullptr if it doesn't exist. |
| 455 | llvm::vfs::Status Status; |
| 456 | if (getStatValue(Path: VF.getName(), Status, /*isFile=*/true, /*F=*/nullptr)) |
| 457 | return std::nullopt; |
| 458 | |
| 459 | if (!SeenBypassFileEntries) |
| 460 | SeenBypassFileEntries = std::make_unique< |
| 461 | llvm::StringMap<llvm::ErrorOr<FileEntryRef::MapValue>>>(); |
| 462 | |
| 463 | // If we've already bypassed just use the existing one. |
| 464 | auto Insertion = SeenBypassFileEntries->insert( |
| 465 | KV: {VF.getName(), std::errc::no_such_file_or_directory}); |
| 466 | if (!Insertion.second) |
| 467 | return FileEntryRef(*Insertion.first); |
| 468 | |
| 469 | // Fill in the new entry from the stat. |
| 470 | FileEntry *BFE = new (FilesAlloc.Allocate()) FileEntry(); |
| 471 | BypassFileEntries.push_back(Elt: BFE); |
| 472 | Insertion.first->second = FileEntryRef::MapValue(*BFE, VF.getDir()); |
| 473 | BFE->Size = Status.getSize(); |
| 474 | BFE->Dir = VF.getFileEntry().Dir; |
| 475 | BFE->ModTime = llvm::sys::toTimeT(TP: Status.getLastModificationTime()); |
| 476 | BFE->UID = NextFileUID++; |
| 477 | |
| 478 | // Save the entry in the bypass table and return. |
| 479 | return FileEntryRef(*Insertion.first); |
| 480 | } |
| 481 | |
| 482 | bool FileManager::FixupRelativePath(SmallVectorImpl<char> &path) const { |
| 483 | StringRef pathRef(path.data(), path.size()); |
| 484 | |
| 485 | if (FileSystemOpts.WorkingDir.empty() |
| 486 | || llvm::sys::path::is_absolute(path: pathRef)) |
| 487 | return false; |
| 488 | |
| 489 | SmallString<128> NewPath(FileSystemOpts.WorkingDir); |
| 490 | llvm::sys::path::append(path&: NewPath, a: pathRef); |
| 491 | path = NewPath; |
| 492 | return true; |
| 493 | } |
| 494 | |
| 495 | bool FileManager::makeAbsolutePath(SmallVectorImpl<char> &Path) const { |
| 496 | bool Changed = FixupRelativePath(path&: Path); |
| 497 | |
| 498 | if (!llvm::sys::path::is_absolute(path: StringRef(Path.data(), Path.size()))) { |
| 499 | FS->makeAbsolute(Path); |
| 500 | Changed = true; |
| 501 | } |
| 502 | |
| 503 | return Changed; |
| 504 | } |
| 505 | |
| 506 | void FileManager::fillRealPathName(FileEntry *UFE, llvm::StringRef FileName) { |
| 507 | llvm::SmallString<128> AbsPath(FileName); |
| 508 | // This is not the same as `VFS::getRealPath()`, which resolves symlinks |
| 509 | // but can be very expensive on real file systems. |
| 510 | // FIXME: the semantic of RealPathName is unclear, and the name might be |
| 511 | // misleading. We need to clean up the interface here. |
| 512 | makeAbsolutePath(Path&: AbsPath); |
| 513 | llvm::sys::path::remove_dots(path&: AbsPath, /*remove_dot_dot=*/true); |
| 514 | UFE->RealPathName = std::string(AbsPath); |
| 515 | } |
| 516 | |
| 517 | llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> |
| 518 | FileManager::getBufferForFile(FileEntryRef FE, bool isVolatile, |
| 519 | bool RequiresNullTerminator, |
| 520 | std::optional<int64_t> MaybeLimit, bool IsText) { |
| 521 | const FileEntry *Entry = &FE.getFileEntry(); |
| 522 | // If the content is living on the file entry, return a reference to it. |
| 523 | if (Entry->Content) |
| 524 | return llvm::MemoryBuffer::getMemBuffer(Ref: Entry->Content->getMemBufferRef()); |
| 525 | |
| 526 | uint64_t FileSize = Entry->getSize(); |
| 527 | |
| 528 | if (MaybeLimit) |
| 529 | FileSize = *MaybeLimit; |
| 530 | |
| 531 | // If there's a high enough chance that the file have changed since we |
| 532 | // got its size, force a stat before opening it. |
| 533 | if (isVolatile || Entry->isNamedPipe()) |
| 534 | FileSize = -1; |
| 535 | |
| 536 | StringRef Filename = FE.getName(); |
| 537 | // If the file is already open, use the open file descriptor. |
| 538 | if (Entry->File) { |
| 539 | auto Result = Entry->File->getBuffer(Name: Filename, FileSize, |
| 540 | RequiresNullTerminator, IsVolatile: isVolatile); |
| 541 | Entry->closeFile(); |
| 542 | return Result; |
| 543 | } |
| 544 | |
| 545 | // Otherwise, open the file. |
| 546 | return getBufferForFileImpl(Filename, FileSize, isVolatile, |
| 547 | RequiresNullTerminator, IsText); |
| 548 | } |
| 549 | |
| 550 | llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> |
| 551 | FileManager::getBufferForFileImpl(StringRef Filename, int64_t FileSize, |
| 552 | bool isVolatile, bool RequiresNullTerminator, |
| 553 | bool IsText) const { |
| 554 | if (FileSystemOpts.WorkingDir.empty()) |
| 555 | return FS->getBufferForFile(Name: Filename, FileSize, RequiresNullTerminator, |
| 556 | IsVolatile: isVolatile, IsText); |
| 557 | |
| 558 | SmallString<128> FilePath(Filename); |
| 559 | FixupRelativePath(path&: FilePath); |
| 560 | return FS->getBufferForFile(Name: FilePath, FileSize, RequiresNullTerminator, |
| 561 | IsVolatile: isVolatile, IsText); |
| 562 | } |
| 563 | |
| 564 | /// getStatValue - Get the 'stat' information for the specified path, |
| 565 | /// using the cache to accelerate it if possible. This returns true |
| 566 | /// if the path points to a virtual file or does not exist, or returns |
| 567 | /// false if it's an existent real file. If FileDescriptor is NULL, |
| 568 | /// do directory look-up instead of file look-up. |
| 569 | std::error_code FileManager::getStatValue(StringRef Path, |
| 570 | llvm::vfs::Status &Status, |
| 571 | bool isFile, |
| 572 | std::unique_ptr<llvm::vfs::File> *F, |
| 573 | bool IsText) { |
| 574 | // FIXME: FileSystemOpts shouldn't be passed in here, all paths should be |
| 575 | // absolute! |
| 576 | if (FileSystemOpts.WorkingDir.empty()) |
| 577 | return FileSystemStatCache::get(Path, Status, isFile, F, Cache: StatCache.get(), |
| 578 | FS&: *FS, IsText); |
| 579 | |
| 580 | SmallString<128> FilePath(Path); |
| 581 | FixupRelativePath(path&: FilePath); |
| 582 | |
| 583 | return FileSystemStatCache::get(Path: FilePath.c_str(), Status, isFile, F, |
| 584 | Cache: StatCache.get(), FS&: *FS, IsText); |
| 585 | } |
| 586 | |
| 587 | std::error_code |
| 588 | FileManager::getNoncachedStatValue(StringRef Path, |
| 589 | llvm::vfs::Status &Result) { |
| 590 | SmallString<128> FilePath(Path); |
| 591 | FixupRelativePath(path&: FilePath); |
| 592 | |
| 593 | llvm::ErrorOr<llvm::vfs::Status> S = FS->status(Path: FilePath.c_str()); |
| 594 | if (!S) |
| 595 | return S.getError(); |
| 596 | Result = *S; |
| 597 | return std::error_code(); |
| 598 | } |
| 599 | |
| 600 | void FileManager::GetUniqueIDMapping( |
| 601 | SmallVectorImpl<OptionalFileEntryRef> &UIDToFiles) const { |
| 602 | UIDToFiles.clear(); |
| 603 | UIDToFiles.resize(N: NextFileUID); |
| 604 | |
| 605 | for (const auto &Entry : SeenFileEntries) { |
| 606 | // Only return files that exist and are not redirected. |
| 607 | if (!Entry.getValue() || !isa<FileEntry *>(Val: Entry.getValue()->V)) |
| 608 | continue; |
| 609 | FileEntryRef FE(Entry); |
| 610 | // Add this file if it's the first one with the UID, or if its name is |
| 611 | // better than the existing one. |
| 612 | OptionalFileEntryRef &ExistingFE = UIDToFiles[FE.getUID()]; |
| 613 | if (!ExistingFE || FE.getName() < ExistingFE->getName()) |
| 614 | ExistingFE = FE; |
| 615 | } |
| 616 | } |
| 617 | |
| 618 | StringRef FileManager::getCanonicalName(DirectoryEntryRef Dir) { |
| 619 | return getCanonicalName(Entry: Dir, Name: Dir.getName()); |
| 620 | } |
| 621 | |
| 622 | StringRef FileManager::getCanonicalName(FileEntryRef File) { |
| 623 | return getCanonicalName(Entry: File, Name: File.getName()); |
| 624 | } |
| 625 | |
| 626 | StringRef FileManager::getCanonicalName(const void *Entry, StringRef Name) { |
| 627 | llvm::DenseMap<const void *, llvm::StringRef>::iterator Known = |
| 628 | CanonicalNames.find(Val: Entry); |
| 629 | if (Known != CanonicalNames.end()) |
| 630 | return Known->second; |
| 631 | |
| 632 | // Name comes from FileEntry/DirectoryEntry::getName(), so it is safe to |
| 633 | // store it in the DenseMap below. |
| 634 | StringRef CanonicalName(Name); |
| 635 | |
| 636 | SmallString<256> AbsPathBuf; |
| 637 | SmallString<256> RealPathBuf; |
| 638 | if (!FS->getRealPath(Path: Name, Output&: RealPathBuf)) { |
| 639 | if (is_style_windows(S: llvm::sys::path::Style::native)) { |
| 640 | // For Windows paths, only use the real path if it doesn't resolve |
| 641 | // a substitute drive, as those are used to avoid MAX_PATH issues. |
| 642 | AbsPathBuf = Name; |
| 643 | if (!FS->makeAbsolute(Path&: AbsPathBuf)) { |
| 644 | if (llvm::sys::path::root_name(path: RealPathBuf) == |
| 645 | llvm::sys::path::root_name(path: AbsPathBuf)) { |
| 646 | CanonicalName = RealPathBuf.str().copy(A&: CanonicalNameStorage); |
| 647 | } else { |
| 648 | // Fallback to using the absolute path. |
| 649 | // Simplifying /../ is semantically valid on Windows even in the |
| 650 | // presence of symbolic links. |
| 651 | llvm::sys::path::remove_dots(path&: AbsPathBuf, /*remove_dot_dot=*/true); |
| 652 | CanonicalName = AbsPathBuf.str().copy(A&: CanonicalNameStorage); |
| 653 | } |
| 654 | } |
| 655 | } else { |
| 656 | CanonicalName = RealPathBuf.str().copy(A&: CanonicalNameStorage); |
| 657 | } |
| 658 | } |
| 659 | |
| 660 | CanonicalNames.insert(KV: {Entry, CanonicalName}); |
| 661 | return CanonicalName; |
| 662 | } |
| 663 | |
| 664 | void FileManager::AddStats(const FileManager &Other) { |
| 665 | assert(&Other != this && "Collecting stats into the same FileManager" ); |
| 666 | NumDirLookups += Other.NumDirLookups; |
| 667 | NumFileLookups += Other.NumFileLookups; |
| 668 | NumDirCacheMisses += Other.NumDirCacheMisses; |
| 669 | NumFileCacheMisses += Other.NumFileCacheMisses; |
| 670 | } |
| 671 | |
| 672 | void FileManager::PrintStats() const { |
| 673 | llvm::errs() << "\n*** File Manager Stats:\n" ; |
| 674 | llvm::errs() << UniqueRealFiles.size() << " real files found, " |
| 675 | << UniqueRealDirs.size() << " real dirs found.\n" ; |
| 676 | llvm::errs() << VirtualFileEntries.size() << " virtual files found, " |
| 677 | << VirtualDirectoryEntries.size() << " virtual dirs found.\n" ; |
| 678 | llvm::errs() << NumDirLookups << " dir lookups, " |
| 679 | << NumDirCacheMisses << " dir cache misses.\n" ; |
| 680 | llvm::errs() << NumFileLookups << " file lookups, " |
| 681 | << NumFileCacheMisses << " file cache misses.\n" ; |
| 682 | |
| 683 | getVirtualFileSystem().visit(Callback: [](llvm::vfs::FileSystem &VFS) { |
| 684 | if (auto *T = dyn_cast_or_null<llvm::vfs::TracingFileSystem>(Val: &VFS)) |
| 685 | llvm::errs() << "\n*** Virtual File System Stats:\n" |
| 686 | << T->NumStatusCalls << " status() calls\n" |
| 687 | << T->NumOpenFileForReadCalls << " openFileForRead() calls\n" |
| 688 | << T->NumDirBeginCalls << " dir_begin() calls\n" |
| 689 | << T->NumGetRealPathCalls << " getRealPath() calls\n" |
| 690 | << T->NumExistsCalls << " exists() calls\n" |
| 691 | << T->NumIsLocalCalls << " isLocal() calls\n" ; |
| 692 | }); |
| 693 | |
| 694 | //llvm::errs() << PagesMapped << BytesOfPagesMapped << FSLookups; |
| 695 | } |
| 696 | |