1 //===--- FileManager.cpp - File System Probing and Caching ----------------===//
2 //
3 // The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements the FileManager interface.
11 //
12 //===----------------------------------------------------------------------===//
13 //
14 // TODO: This should index all interesting directories with dirent calls.
15 // getdirentries ?
16 // opendir/readdir_r/closedir ?
17 //
18 //===----------------------------------------------------------------------===//
19
20 #include "clang/Basic/FileManager.h"
21 #include "clang/Basic/FileSystemStatCache.h"
22 #include "llvm/ADT/SmallString.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 <map>
29 #include <set>
30 #include <string>
31 #include <system_error>
32
33 using namespace clang;
34
35 /// NON_EXISTENT_DIR - A special value distinct from null that is used to
36 /// represent a dir name that doesn't exist on the disk.
37 #define NON_EXISTENT_DIR reinterpret_cast<DirectoryEntry*>((intptr_t)-1)
38
39 /// NON_EXISTENT_FILE - A special value distinct from null that is used to
40 /// represent a filename that doesn't exist on the disk.
41 #define NON_EXISTENT_FILE reinterpret_cast<FileEntry*>((intptr_t)-1)
42
43 //===----------------------------------------------------------------------===//
44 // Common logic.
45 //===----------------------------------------------------------------------===//
46
FileManager(const FileSystemOptions & FSO,IntrusiveRefCntPtr<vfs::FileSystem> FS)47 FileManager::FileManager(const FileSystemOptions &FSO,
48 IntrusiveRefCntPtr<vfs::FileSystem> FS)
49 : FS(FS), FileSystemOpts(FSO),
50 SeenDirEntries(64), SeenFileEntries(64), NextFileUID(0) {
51 NumDirLookups = NumFileLookups = 0;
52 NumDirCacheMisses = NumFileCacheMisses = 0;
53
54 // If the caller doesn't provide a virtual file system, just grab the real
55 // file system.
56 if (!FS)
57 this->FS = vfs::getRealFileSystem();
58 }
59
~FileManager()60 FileManager::~FileManager() {
61 for (unsigned i = 0, e = VirtualFileEntries.size(); i != e; ++i)
62 delete VirtualFileEntries[i];
63 for (unsigned i = 0, e = VirtualDirectoryEntries.size(); i != e; ++i)
64 delete VirtualDirectoryEntries[i];
65 }
66
addStatCache(std::unique_ptr<FileSystemStatCache> statCache,bool AtBeginning)67 void FileManager::addStatCache(std::unique_ptr<FileSystemStatCache> statCache,
68 bool AtBeginning) {
69 assert(statCache && "No stat cache provided?");
70 if (AtBeginning || !StatCache.get()) {
71 statCache->setNextStatCache(std::move(StatCache));
72 StatCache = std::move(statCache);
73 return;
74 }
75
76 FileSystemStatCache *LastCache = StatCache.get();
77 while (LastCache->getNextStatCache())
78 LastCache = LastCache->getNextStatCache();
79
80 LastCache->setNextStatCache(std::move(statCache));
81 }
82
removeStatCache(FileSystemStatCache * statCache)83 void FileManager::removeStatCache(FileSystemStatCache *statCache) {
84 if (!statCache)
85 return;
86
87 if (StatCache.get() == statCache) {
88 // This is the first stat cache.
89 StatCache = StatCache->takeNextStatCache();
90 return;
91 }
92
93 // Find the stat cache in the list.
94 FileSystemStatCache *PrevCache = StatCache.get();
95 while (PrevCache && PrevCache->getNextStatCache() != statCache)
96 PrevCache = PrevCache->getNextStatCache();
97
98 assert(PrevCache && "Stat cache not found for removal");
99 PrevCache->setNextStatCache(statCache->takeNextStatCache());
100 }
101
clearStatCaches()102 void FileManager::clearStatCaches() {
103 StatCache.reset();
104 }
105
106 /// \brief Retrieve the directory that the given file name resides in.
107 /// Filename can point to either a real file or a virtual file.
getDirectoryFromFile(FileManager & FileMgr,StringRef Filename,bool CacheFailure)108 static const DirectoryEntry *getDirectoryFromFile(FileManager &FileMgr,
109 StringRef Filename,
110 bool CacheFailure) {
111 if (Filename.empty())
112 return nullptr;
113
114 if (llvm::sys::path::is_separator(Filename[Filename.size() - 1]))
115 return nullptr; // If Filename is a directory.
116
117 StringRef DirName = llvm::sys::path::parent_path(Filename);
118 // Use the current directory if file has no path component.
119 if (DirName.empty())
120 DirName = ".";
121
122 return FileMgr.getDirectory(DirName, CacheFailure);
123 }
124
125 /// Add all ancestors of the given path (pointing to either a file or
126 /// a directory) as virtual directories.
addAncestorsAsVirtualDirs(StringRef Path)127 void FileManager::addAncestorsAsVirtualDirs(StringRef Path) {
128 StringRef DirName = llvm::sys::path::parent_path(Path);
129 if (DirName.empty())
130 return;
131
132 auto &NamedDirEnt =
133 *SeenDirEntries.insert(std::make_pair(DirName, nullptr)).first;
134
135 // When caching a virtual directory, we always cache its ancestors
136 // at the same time. Therefore, if DirName is already in the cache,
137 // we don't need to recurse as its ancestors must also already be in
138 // the cache.
139 if (NamedDirEnt.second)
140 return;
141
142 // Add the virtual directory to the cache.
143 DirectoryEntry *UDE = new DirectoryEntry;
144 UDE->Name = NamedDirEnt.first().data();
145 NamedDirEnt.second = UDE;
146 VirtualDirectoryEntries.push_back(UDE);
147
148 // Recursively add the other ancestors.
149 addAncestorsAsVirtualDirs(DirName);
150 }
151
getDirectory(StringRef DirName,bool CacheFailure)152 const DirectoryEntry *FileManager::getDirectory(StringRef DirName,
153 bool CacheFailure) {
154 // stat doesn't like trailing separators except for root directory.
155 // At least, on Win32 MSVCRT, stat() cannot strip trailing '/'.
156 // (though it can strip '\\')
157 if (DirName.size() > 1 &&
158 DirName != llvm::sys::path::root_path(DirName) &&
159 llvm::sys::path::is_separator(DirName.back()))
160 DirName = DirName.substr(0, DirName.size()-1);
161 #ifdef LLVM_ON_WIN32
162 // Fixing a problem with "clang C:test.c" on Windows.
163 // Stat("C:") does not recognize "C:" as a valid directory
164 std::string DirNameStr;
165 if (DirName.size() > 1 && DirName.back() == ':' &&
166 DirName.equals_lower(llvm::sys::path::root_name(DirName))) {
167 DirNameStr = DirName.str() + '.';
168 DirName = DirNameStr;
169 }
170 #endif
171
172 ++NumDirLookups;
173 auto &NamedDirEnt =
174 *SeenDirEntries.insert(std::make_pair(DirName, nullptr)).first;
175
176 // See if there was already an entry in the map. Note that the map
177 // contains both virtual and real directories.
178 if (NamedDirEnt.second)
179 return NamedDirEnt.second == NON_EXISTENT_DIR ? nullptr
180 : NamedDirEnt.second;
181
182 ++NumDirCacheMisses;
183
184 // By default, initialize it to invalid.
185 NamedDirEnt.second = NON_EXISTENT_DIR;
186
187 // Get the null-terminated directory name as stored as the key of the
188 // SeenDirEntries map.
189 const char *InterndDirName = NamedDirEnt.first().data();
190
191 // Check to see if the directory exists.
192 FileData Data;
193 if (getStatValue(InterndDirName, Data, false, nullptr /*directory lookup*/)) {
194 // There's no real directory at the given path.
195 if (!CacheFailure)
196 SeenDirEntries.erase(DirName);
197 return nullptr;
198 }
199
200 // It exists. See if we have already opened a directory with the
201 // same inode (this occurs on Unix-like systems when one dir is
202 // symlinked to another, for example) or the same path (on
203 // Windows).
204 DirectoryEntry &UDE = UniqueRealDirs[Data.UniqueID];
205
206 NamedDirEnt.second = &UDE;
207 if (!UDE.getName()) {
208 // We don't have this directory yet, add it. We use the string
209 // key from the SeenDirEntries map as the string.
210 UDE.Name = InterndDirName;
211 }
212
213 return &UDE;
214 }
215
getFile(StringRef Filename,bool openFile,bool CacheFailure)216 const FileEntry *FileManager::getFile(StringRef Filename, bool openFile,
217 bool CacheFailure) {
218 ++NumFileLookups;
219
220 // See if there is already an entry in the map.
221 auto &NamedFileEnt =
222 *SeenFileEntries.insert(std::make_pair(Filename, nullptr)).first;
223
224 // See if there is already an entry in the map.
225 if (NamedFileEnt.second)
226 return NamedFileEnt.second == NON_EXISTENT_FILE ? nullptr
227 : NamedFileEnt.second;
228
229 ++NumFileCacheMisses;
230
231 // By default, initialize it to invalid.
232 NamedFileEnt.second = NON_EXISTENT_FILE;
233
234 // Get the null-terminated file name as stored as the key of the
235 // SeenFileEntries map.
236 const char *InterndFileName = NamedFileEnt.first().data();
237
238 // Look up the directory for the file. When looking up something like
239 // sys/foo.h we'll discover all of the search directories that have a 'sys'
240 // subdirectory. This will let us avoid having to waste time on known-to-fail
241 // searches when we go to find sys/bar.h, because all the search directories
242 // without a 'sys' subdir will get a cached failure result.
243 const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename,
244 CacheFailure);
245 if (DirInfo == nullptr) { // Directory doesn't exist, file can't exist.
246 if (!CacheFailure)
247 SeenFileEntries.erase(Filename);
248
249 return nullptr;
250 }
251
252 // FIXME: Use the directory info to prune this, before doing the stat syscall.
253 // FIXME: This will reduce the # syscalls.
254
255 // Nope, there isn't. Check to see if the file exists.
256 std::unique_ptr<vfs::File> F;
257 FileData Data;
258 if (getStatValue(InterndFileName, Data, true, openFile ? &F : nullptr)) {
259 // There's no real file at the given path.
260 if (!CacheFailure)
261 SeenFileEntries.erase(Filename);
262
263 return nullptr;
264 }
265
266 assert((openFile || !F) && "undesired open file");
267
268 // It exists. See if we have already opened a file with the same inode.
269 // This occurs when one dir is symlinked to another, for example.
270 FileEntry &UFE = UniqueRealFiles[Data.UniqueID];
271
272 NamedFileEnt.second = &UFE;
273
274 // If the name returned by getStatValue is different than Filename, re-intern
275 // the name.
276 if (Data.Name != Filename) {
277 auto &NamedFileEnt =
278 *SeenFileEntries.insert(std::make_pair(Data.Name, nullptr)).first;
279 if (!NamedFileEnt.second)
280 NamedFileEnt.second = &UFE;
281 else
282 assert(NamedFileEnt.second == &UFE &&
283 "filename from getStatValue() refers to wrong file");
284 InterndFileName = NamedFileEnt.first().data();
285 }
286
287 if (UFE.isValid()) { // Already have an entry with this inode, return it.
288
289 // FIXME: this hack ensures that if we look up a file by a virtual path in
290 // the VFS that the getDir() will have the virtual path, even if we found
291 // the file by a 'real' path first. This is required in order to find a
292 // module's structure when its headers/module map are mapped in the VFS.
293 // We should remove this as soon as we can properly support a file having
294 // multiple names.
295 if (DirInfo != UFE.Dir && Data.IsVFSMapped)
296 UFE.Dir = DirInfo;
297
298 // Always update the name to use the last name by which a file was accessed.
299 // FIXME: Neither this nor always using the first name is correct; we want
300 // to switch towards a design where we return a FileName object that
301 // encapsulates both the name by which the file was accessed and the
302 // corresponding FileEntry.
303 UFE.Name = InterndFileName;
304
305 return &UFE;
306 }
307
308 // Otherwise, we don't have this file yet, add it.
309 UFE.Name = InterndFileName;
310 UFE.Size = Data.Size;
311 UFE.ModTime = Data.ModTime;
312 UFE.Dir = DirInfo;
313 UFE.UID = NextFileUID++;
314 UFE.UniqueID = Data.UniqueID;
315 UFE.IsNamedPipe = Data.IsNamedPipe;
316 UFE.InPCH = Data.InPCH;
317 UFE.File = std::move(F);
318 UFE.IsValid = true;
319 return &UFE;
320 }
321
322 const FileEntry *
getVirtualFile(StringRef Filename,off_t Size,time_t ModificationTime)323 FileManager::getVirtualFile(StringRef Filename, off_t Size,
324 time_t ModificationTime) {
325 ++NumFileLookups;
326
327 // See if there is already an entry in the map.
328 auto &NamedFileEnt =
329 *SeenFileEntries.insert(std::make_pair(Filename, nullptr)).first;
330
331 // See if there is already an entry in the map.
332 if (NamedFileEnt.second && NamedFileEnt.second != NON_EXISTENT_FILE)
333 return NamedFileEnt.second;
334
335 ++NumFileCacheMisses;
336
337 // By default, initialize it to invalid.
338 NamedFileEnt.second = NON_EXISTENT_FILE;
339
340 addAncestorsAsVirtualDirs(Filename);
341 FileEntry *UFE = nullptr;
342
343 // Now that all ancestors of Filename are in the cache, the
344 // following call is guaranteed to find the DirectoryEntry from the
345 // cache.
346 const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename,
347 /*CacheFailure=*/true);
348 assert(DirInfo &&
349 "The directory of a virtual file should already be in the cache.");
350
351 // Check to see if the file exists. If so, drop the virtual file
352 FileData Data;
353 const char *InterndFileName = NamedFileEnt.first().data();
354 if (getStatValue(InterndFileName, Data, true, nullptr) == 0) {
355 Data.Size = Size;
356 Data.ModTime = ModificationTime;
357 UFE = &UniqueRealFiles[Data.UniqueID];
358
359 NamedFileEnt.second = UFE;
360
361 // If we had already opened this file, close it now so we don't
362 // leak the descriptor. We're not going to use the file
363 // descriptor anyway, since this is a virtual file.
364 if (UFE->File)
365 UFE->closeFile();
366
367 // If we already have an entry with this inode, return it.
368 if (UFE->isValid())
369 return UFE;
370
371 UFE->UniqueID = Data.UniqueID;
372 UFE->IsNamedPipe = Data.IsNamedPipe;
373 UFE->InPCH = Data.InPCH;
374 }
375
376 if (!UFE) {
377 UFE = new FileEntry();
378 VirtualFileEntries.push_back(UFE);
379 NamedFileEnt.second = UFE;
380 }
381
382 UFE->Name = InterndFileName;
383 UFE->Size = Size;
384 UFE->ModTime = ModificationTime;
385 UFE->Dir = DirInfo;
386 UFE->UID = NextFileUID++;
387 UFE->File.reset();
388 return UFE;
389 }
390
FixupRelativePath(SmallVectorImpl<char> & path) const391 void FileManager::FixupRelativePath(SmallVectorImpl<char> &path) const {
392 StringRef pathRef(path.data(), path.size());
393
394 if (FileSystemOpts.WorkingDir.empty()
395 || llvm::sys::path::is_absolute(pathRef))
396 return;
397
398 SmallString<128> NewPath(FileSystemOpts.WorkingDir);
399 llvm::sys::path::append(NewPath, pathRef);
400 path = NewPath;
401 }
402
403 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
getBufferForFile(const FileEntry * Entry,bool isVolatile,bool ShouldCloseOpenFile)404 FileManager::getBufferForFile(const FileEntry *Entry, bool isVolatile,
405 bool ShouldCloseOpenFile) {
406 uint64_t FileSize = Entry->getSize();
407 // If there's a high enough chance that the file have changed since we
408 // got its size, force a stat before opening it.
409 if (isVolatile)
410 FileSize = -1;
411
412 const char *Filename = Entry->getName();
413 // If the file is already open, use the open file descriptor.
414 if (Entry->File) {
415 auto Result =
416 Entry->File->getBuffer(Filename, FileSize,
417 /*RequiresNullTerminator=*/true, isVolatile);
418 // FIXME: we need a set of APIs that can make guarantees about whether a
419 // FileEntry is open or not.
420 if (ShouldCloseOpenFile)
421 Entry->closeFile();
422 return Result;
423 }
424
425 // Otherwise, open the file.
426
427 if (FileSystemOpts.WorkingDir.empty())
428 return FS->getBufferForFile(Filename, FileSize,
429 /*RequiresNullTerminator=*/true, isVolatile);
430
431 SmallString<128> FilePath(Entry->getName());
432 FixupRelativePath(FilePath);
433 return FS->getBufferForFile(FilePath, FileSize,
434 /*RequiresNullTerminator=*/true, isVolatile);
435 }
436
437 llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
getBufferForFile(StringRef Filename)438 FileManager::getBufferForFile(StringRef Filename) {
439 if (FileSystemOpts.WorkingDir.empty())
440 return FS->getBufferForFile(Filename);
441
442 SmallString<128> FilePath(Filename);
443 FixupRelativePath(FilePath);
444 return FS->getBufferForFile(FilePath.c_str());
445 }
446
447 /// getStatValue - Get the 'stat' information for the specified path,
448 /// using the cache to accelerate it if possible. This returns true
449 /// if the path points to a virtual file or does not exist, or returns
450 /// false if it's an existent real file. If FileDescriptor is NULL,
451 /// do directory look-up instead of file look-up.
getStatValue(const char * Path,FileData & Data,bool isFile,std::unique_ptr<vfs::File> * F)452 bool FileManager::getStatValue(const char *Path, FileData &Data, bool isFile,
453 std::unique_ptr<vfs::File> *F) {
454 // FIXME: FileSystemOpts shouldn't be passed in here, all paths should be
455 // absolute!
456 if (FileSystemOpts.WorkingDir.empty())
457 return FileSystemStatCache::get(Path, Data, isFile, F,StatCache.get(), *FS);
458
459 SmallString<128> FilePath(Path);
460 FixupRelativePath(FilePath);
461
462 return FileSystemStatCache::get(FilePath.c_str(), Data, isFile, F,
463 StatCache.get(), *FS);
464 }
465
getNoncachedStatValue(StringRef Path,vfs::Status & Result)466 bool FileManager::getNoncachedStatValue(StringRef Path,
467 vfs::Status &Result) {
468 SmallString<128> FilePath(Path);
469 FixupRelativePath(FilePath);
470
471 llvm::ErrorOr<vfs::Status> S = FS->status(FilePath.c_str());
472 if (!S)
473 return true;
474 Result = *S;
475 return false;
476 }
477
invalidateCache(const FileEntry * Entry)478 void FileManager::invalidateCache(const FileEntry *Entry) {
479 assert(Entry && "Cannot invalidate a NULL FileEntry");
480
481 SeenFileEntries.erase(Entry->getName());
482
483 // FileEntry invalidation should not block future optimizations in the file
484 // caches. Possible alternatives are cache truncation (invalidate last N) or
485 // invalidation of the whole cache.
486 UniqueRealFiles.erase(Entry->getUniqueID());
487 }
488
489
GetUniqueIDMapping(SmallVectorImpl<const FileEntry * > & UIDToFiles) const490 void FileManager::GetUniqueIDMapping(
491 SmallVectorImpl<const FileEntry *> &UIDToFiles) const {
492 UIDToFiles.clear();
493 UIDToFiles.resize(NextFileUID);
494
495 // Map file entries
496 for (llvm::StringMap<FileEntry*, llvm::BumpPtrAllocator>::const_iterator
497 FE = SeenFileEntries.begin(), FEEnd = SeenFileEntries.end();
498 FE != FEEnd; ++FE)
499 if (FE->getValue() && FE->getValue() != NON_EXISTENT_FILE)
500 UIDToFiles[FE->getValue()->getUID()] = FE->getValue();
501
502 // Map virtual file entries
503 for (SmallVectorImpl<FileEntry *>::const_iterator
504 VFE = VirtualFileEntries.begin(), VFEEnd = VirtualFileEntries.end();
505 VFE != VFEEnd; ++VFE)
506 if (*VFE && *VFE != NON_EXISTENT_FILE)
507 UIDToFiles[(*VFE)->getUID()] = *VFE;
508 }
509
modifyFileEntry(FileEntry * File,off_t Size,time_t ModificationTime)510 void FileManager::modifyFileEntry(FileEntry *File,
511 off_t Size, time_t ModificationTime) {
512 File->Size = Size;
513 File->ModTime = ModificationTime;
514 }
515
516 /// Remove '.' path components from the given absolute path.
517 /// \return \c true if any changes were made.
518 // FIXME: Move this to llvm::sys::path.
removeDotPaths(SmallVectorImpl<char> & Path)519 bool FileManager::removeDotPaths(SmallVectorImpl<char> &Path) {
520 using namespace llvm::sys;
521
522 SmallVector<StringRef, 16> ComponentStack;
523 StringRef P(Path.data(), Path.size());
524
525 // Skip the root path, then look for traversal in the components.
526 StringRef Rel = path::relative_path(P);
527 bool AnyDots = false;
528 for (StringRef C : llvm::make_range(path::begin(Rel), path::end(Rel))) {
529 if (C == ".") {
530 AnyDots = true;
531 continue;
532 }
533 ComponentStack.push_back(C);
534 }
535
536 if (!AnyDots)
537 return false;
538
539 SmallString<256> Buffer = path::root_path(P);
540 for (StringRef C : ComponentStack)
541 path::append(Buffer, C);
542
543 Path.swap(Buffer);
544 return true;
545 }
546
getCanonicalName(const DirectoryEntry * Dir)547 StringRef FileManager::getCanonicalName(const DirectoryEntry *Dir) {
548 // FIXME: use llvm::sys::fs::canonical() when it gets implemented
549 llvm::DenseMap<const DirectoryEntry *, llvm::StringRef>::iterator Known
550 = CanonicalDirNames.find(Dir);
551 if (Known != CanonicalDirNames.end())
552 return Known->second;
553
554 StringRef CanonicalName(Dir->getName());
555
556 #ifdef LLVM_ON_UNIX
557 char CanonicalNameBuf[PATH_MAX];
558 if (realpath(Dir->getName(), CanonicalNameBuf)) {
559 unsigned Len = strlen(CanonicalNameBuf);
560 char *Mem = static_cast<char *>(CanonicalNameStorage.Allocate(Len, 1));
561 memcpy(Mem, CanonicalNameBuf, Len);
562 CanonicalName = StringRef(Mem, Len);
563 }
564 #else
565 SmallString<256> CanonicalNameBuf(CanonicalName);
566 llvm::sys::fs::make_absolute(CanonicalNameBuf);
567 llvm::sys::path::native(CanonicalNameBuf);
568 removeDotPaths(CanonicalNameBuf);
569 #endif
570
571 CanonicalDirNames.insert(std::make_pair(Dir, CanonicalName));
572 return CanonicalName;
573 }
574
PrintStats() const575 void FileManager::PrintStats() const {
576 llvm::errs() << "\n*** File Manager Stats:\n";
577 llvm::errs() << UniqueRealFiles.size() << " real files found, "
578 << UniqueRealDirs.size() << " real dirs found.\n";
579 llvm::errs() << VirtualFileEntries.size() << " virtual files found, "
580 << VirtualDirectoryEntries.size() << " virtual dirs found.\n";
581 llvm::errs() << NumDirLookups << " dir lookups, "
582 << NumDirCacheMisses << " dir cache misses.\n";
583 llvm::errs() << NumFileLookups << " file lookups, "
584 << NumFileCacheMisses << " file cache misses.\n";
585
586 //llvm::errs() << PagesMapped << BytesOfPagesMapped << FSLookups;
587 }
588