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/Support/FileSystem.h"
24#include "llvm/Support/MemoryBuffer.h"
25#include "llvm/Support/raw_ostream.h"
26#include "llvm/Support/Path.h"
27#include "llvm/Support/system_error.h"
28#include "llvm/Config/llvm-config.h"
29#include <map>
30#include <set>
31#include <string>
32
33// FIXME: This is terrible, we need this for ::close.
34#if !defined(_MSC_VER) && !defined(__MINGW32__)
35#include <unistd.h>
36#include <sys/uio.h>
37#else
38#include <io.h>
39#endif
40using namespace clang;
41
42// FIXME: Enhance libsystem to support inode and other fields.
43#include <sys/stat.h>
44
45/// NON_EXISTENT_DIR - A special value distinct from null that is used to
46/// represent a dir name that doesn't exist on the disk.
47#define NON_EXISTENT_DIR reinterpret_cast<DirectoryEntry*>((intptr_t)-1)
48
49/// NON_EXISTENT_FILE - A special value distinct from null that is used to
50/// represent a filename that doesn't exist on the disk.
51#define NON_EXISTENT_FILE reinterpret_cast<FileEntry*>((intptr_t)-1)
52
53
54FileEntry::~FileEntry() {
55  // If this FileEntry owns an open file descriptor that never got used, close
56  // it.
57  if (FD != -1) ::close(FD);
58}
59
60//===----------------------------------------------------------------------===//
61// Windows.
62//===----------------------------------------------------------------------===//
63
64#ifdef LLVM_ON_WIN32
65
66namespace {
67  static std::string GetFullPath(const char *relPath) {
68    char *absPathStrPtr = _fullpath(NULL, relPath, 0);
69    assert(absPathStrPtr && "_fullpath() returned NULL!");
70
71    std::string absPath(absPathStrPtr);
72
73    free(absPathStrPtr);
74    return absPath;
75  }
76}
77
78class FileManager::UniqueDirContainer {
79  /// UniqueDirs - Cache from full path to existing directories/files.
80  ///
81  llvm::StringMap<DirectoryEntry> UniqueDirs;
82
83public:
84  /// getDirectory - Return an existing DirectoryEntry with the given
85  /// name if there is already one; otherwise create and return a
86  /// default-constructed DirectoryEntry.
87  DirectoryEntry &getDirectory(const char *Name,
88                               const struct stat & /*StatBuf*/) {
89    std::string FullPath(GetFullPath(Name));
90    return UniqueDirs.GetOrCreateValue(FullPath).getValue();
91  }
92
93  size_t size() const { return UniqueDirs.size(); }
94};
95
96class FileManager::UniqueFileContainer {
97  /// UniqueFiles - Cache from full path to existing directories/files.
98  ///
99  llvm::StringMap<FileEntry, llvm::BumpPtrAllocator> UniqueFiles;
100
101public:
102  /// getFile - Return an existing FileEntry with the given name if
103  /// there is already one; otherwise create and return a
104  /// default-constructed FileEntry.
105  FileEntry &getFile(const char *Name, const struct stat & /*StatBuf*/) {
106    std::string FullPath(GetFullPath(Name));
107
108    // Lowercase string because Windows filesystem is case insensitive.
109    FullPath = StringRef(FullPath).lower();
110    return UniqueFiles.GetOrCreateValue(FullPath).getValue();
111  }
112
113  size_t size() const { return UniqueFiles.size(); }
114
115  void erase(const FileEntry *Entry) {
116    std::string FullPath(GetFullPath(Entry->getName()));
117
118    // Lowercase string because Windows filesystem is case insensitive.
119    FullPath = StringRef(FullPath).lower();
120    UniqueFiles.erase(FullPath);
121  }
122};
123
124//===----------------------------------------------------------------------===//
125// Unix-like Systems.
126//===----------------------------------------------------------------------===//
127
128#else
129
130class FileManager::UniqueDirContainer {
131  /// UniqueDirs - Cache from ID's to existing directories/files.
132  std::map<std::pair<dev_t, ino_t>, DirectoryEntry> UniqueDirs;
133
134public:
135  /// getDirectory - Return an existing DirectoryEntry with the given
136  /// ID's if there is already one; otherwise create and return a
137  /// default-constructed DirectoryEntry.
138  DirectoryEntry &getDirectory(const char * /*Name*/,
139                               const struct stat &StatBuf) {
140    return UniqueDirs[std::make_pair(StatBuf.st_dev, StatBuf.st_ino)];
141  }
142
143  size_t size() const { return UniqueDirs.size(); }
144};
145
146class FileManager::UniqueFileContainer {
147  /// UniqueFiles - Cache from ID's to existing directories/files.
148  std::set<FileEntry> UniqueFiles;
149
150public:
151  /// getFile - Return an existing FileEntry with the given ID's if
152  /// there is already one; otherwise create and return a
153  /// default-constructed FileEntry.
154  FileEntry &getFile(const char * /*Name*/, const struct stat &StatBuf) {
155    return
156      const_cast<FileEntry&>(
157                    *UniqueFiles.insert(FileEntry(StatBuf.st_dev,
158                                                  StatBuf.st_ino,
159                                                  StatBuf.st_mode)).first);
160  }
161
162  size_t size() const { return UniqueFiles.size(); }
163
164  void erase(const FileEntry *Entry) { UniqueFiles.erase(*Entry); }
165};
166
167#endif
168
169//===----------------------------------------------------------------------===//
170// Common logic.
171//===----------------------------------------------------------------------===//
172
173FileManager::FileManager(const FileSystemOptions &FSO)
174  : FileSystemOpts(FSO),
175    UniqueRealDirs(*new UniqueDirContainer()),
176    UniqueRealFiles(*new UniqueFileContainer()),
177    SeenDirEntries(64), SeenFileEntries(64), NextFileUID(0) {
178  NumDirLookups = NumFileLookups = 0;
179  NumDirCacheMisses = NumFileCacheMisses = 0;
180}
181
182FileManager::~FileManager() {
183  delete &UniqueRealDirs;
184  delete &UniqueRealFiles;
185  for (unsigned i = 0, e = VirtualFileEntries.size(); i != e; ++i)
186    delete VirtualFileEntries[i];
187  for (unsigned i = 0, e = VirtualDirectoryEntries.size(); i != e; ++i)
188    delete VirtualDirectoryEntries[i];
189}
190
191void FileManager::addStatCache(FileSystemStatCache *statCache,
192                               bool AtBeginning) {
193  assert(statCache && "No stat cache provided?");
194  if (AtBeginning || StatCache.get() == 0) {
195    statCache->setNextStatCache(StatCache.take());
196    StatCache.reset(statCache);
197    return;
198  }
199
200  FileSystemStatCache *LastCache = StatCache.get();
201  while (LastCache->getNextStatCache())
202    LastCache = LastCache->getNextStatCache();
203
204  LastCache->setNextStatCache(statCache);
205}
206
207void FileManager::removeStatCache(FileSystemStatCache *statCache) {
208  if (!statCache)
209    return;
210
211  if (StatCache.get() == statCache) {
212    // This is the first stat cache.
213    StatCache.reset(StatCache->takeNextStatCache());
214    return;
215  }
216
217  // Find the stat cache in the list.
218  FileSystemStatCache *PrevCache = StatCache.get();
219  while (PrevCache && PrevCache->getNextStatCache() != statCache)
220    PrevCache = PrevCache->getNextStatCache();
221
222  assert(PrevCache && "Stat cache not found for removal");
223  PrevCache->setNextStatCache(statCache->getNextStatCache());
224}
225
226void FileManager::clearStatCaches() {
227  StatCache.reset(0);
228}
229
230/// \brief Retrieve the directory that the given file name resides in.
231/// Filename can point to either a real file or a virtual file.
232static const DirectoryEntry *getDirectoryFromFile(FileManager &FileMgr,
233                                                  StringRef Filename,
234                                                  bool CacheFailure) {
235  if (Filename.empty())
236    return NULL;
237
238  if (llvm::sys::path::is_separator(Filename[Filename.size() - 1]))
239    return NULL;  // If Filename is a directory.
240
241  StringRef DirName = llvm::sys::path::parent_path(Filename);
242  // Use the current directory if file has no path component.
243  if (DirName.empty())
244    DirName = ".";
245
246  return FileMgr.getDirectory(DirName, CacheFailure);
247}
248
249/// Add all ancestors of the given path (pointing to either a file or
250/// a directory) as virtual directories.
251void FileManager::addAncestorsAsVirtualDirs(StringRef Path) {
252  StringRef DirName = llvm::sys::path::parent_path(Path);
253  if (DirName.empty())
254    return;
255
256  llvm::StringMapEntry<DirectoryEntry *> &NamedDirEnt =
257    SeenDirEntries.GetOrCreateValue(DirName);
258
259  // When caching a virtual directory, we always cache its ancestors
260  // at the same time.  Therefore, if DirName is already in the cache,
261  // we don't need to recurse as its ancestors must also already be in
262  // the cache.
263  if (NamedDirEnt.getValue())
264    return;
265
266  // Add the virtual directory to the cache.
267  DirectoryEntry *UDE = new DirectoryEntry;
268  UDE->Name = NamedDirEnt.getKeyData();
269  NamedDirEnt.setValue(UDE);
270  VirtualDirectoryEntries.push_back(UDE);
271
272  // Recursively add the other ancestors.
273  addAncestorsAsVirtualDirs(DirName);
274}
275
276const DirectoryEntry *FileManager::getDirectory(StringRef DirName,
277                                                bool CacheFailure) {
278  // stat doesn't like trailing separators except for root directory.
279  // At least, on Win32 MSVCRT, stat() cannot strip trailing '/'.
280  // (though it can strip '\\')
281  if (DirName.size() > 1 &&
282      DirName != llvm::sys::path::root_path(DirName) &&
283      llvm::sys::path::is_separator(DirName.back()))
284    DirName = DirName.substr(0, DirName.size()-1);
285
286  ++NumDirLookups;
287  llvm::StringMapEntry<DirectoryEntry *> &NamedDirEnt =
288    SeenDirEntries.GetOrCreateValue(DirName);
289
290  // See if there was already an entry in the map.  Note that the map
291  // contains both virtual and real directories.
292  if (NamedDirEnt.getValue())
293    return NamedDirEnt.getValue() == NON_EXISTENT_DIR
294              ? 0 : NamedDirEnt.getValue();
295
296  ++NumDirCacheMisses;
297
298  // By default, initialize it to invalid.
299  NamedDirEnt.setValue(NON_EXISTENT_DIR);
300
301  // Get the null-terminated directory name as stored as the key of the
302  // SeenDirEntries map.
303  const char *InterndDirName = NamedDirEnt.getKeyData();
304
305  // Check to see if the directory exists.
306  struct stat StatBuf;
307  if (getStatValue(InterndDirName, StatBuf, 0/*directory lookup*/)) {
308    // There's no real directory at the given path.
309    if (!CacheFailure)
310      SeenDirEntries.erase(DirName);
311    return 0;
312  }
313
314  // It exists.  See if we have already opened a directory with the
315  // same inode (this occurs on Unix-like systems when one dir is
316  // symlinked to another, for example) or the same path (on
317  // Windows).
318  DirectoryEntry &UDE = UniqueRealDirs.getDirectory(InterndDirName, StatBuf);
319
320  NamedDirEnt.setValue(&UDE);
321  if (!UDE.getName()) {
322    // We don't have this directory yet, add it.  We use the string
323    // key from the SeenDirEntries map as the string.
324    UDE.Name  = InterndDirName;
325  }
326
327  return &UDE;
328}
329
330const FileEntry *FileManager::getFile(StringRef Filename, bool openFile,
331                                      bool CacheFailure) {
332  ++NumFileLookups;
333
334  // See if there is already an entry in the map.
335  llvm::StringMapEntry<FileEntry *> &NamedFileEnt =
336    SeenFileEntries.GetOrCreateValue(Filename);
337
338  // See if there is already an entry in the map.
339  if (NamedFileEnt.getValue())
340    return NamedFileEnt.getValue() == NON_EXISTENT_FILE
341                 ? 0 : NamedFileEnt.getValue();
342
343  ++NumFileCacheMisses;
344
345  // By default, initialize it to invalid.
346  NamedFileEnt.setValue(NON_EXISTENT_FILE);
347
348  // Get the null-terminated file name as stored as the key of the
349  // SeenFileEntries map.
350  const char *InterndFileName = NamedFileEnt.getKeyData();
351
352  // Look up the directory for the file.  When looking up something like
353  // sys/foo.h we'll discover all of the search directories that have a 'sys'
354  // subdirectory.  This will let us avoid having to waste time on known-to-fail
355  // searches when we go to find sys/bar.h, because all the search directories
356  // without a 'sys' subdir will get a cached failure result.
357  const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename,
358                                                       CacheFailure);
359  if (DirInfo == 0) {  // Directory doesn't exist, file can't exist.
360    if (!CacheFailure)
361      SeenFileEntries.erase(Filename);
362
363    return 0;
364  }
365
366  // FIXME: Use the directory info to prune this, before doing the stat syscall.
367  // FIXME: This will reduce the # syscalls.
368
369  // Nope, there isn't.  Check to see if the file exists.
370  int FileDescriptor = -1;
371  struct stat StatBuf;
372  if (getStatValue(InterndFileName, StatBuf, &FileDescriptor)) {
373    // There's no real file at the given path.
374    if (!CacheFailure)
375      SeenFileEntries.erase(Filename);
376
377    return 0;
378  }
379
380  if (FileDescriptor != -1 && !openFile) {
381    close(FileDescriptor);
382    FileDescriptor = -1;
383  }
384
385  // It exists.  See if we have already opened a file with the same inode.
386  // This occurs when one dir is symlinked to another, for example.
387  FileEntry &UFE = UniqueRealFiles.getFile(InterndFileName, StatBuf);
388
389  NamedFileEnt.setValue(&UFE);
390  if (UFE.getName()) { // Already have an entry with this inode, return it.
391    // If the stat process opened the file, close it to avoid a FD leak.
392    if (FileDescriptor != -1)
393      close(FileDescriptor);
394
395    return &UFE;
396  }
397
398  // Otherwise, we don't have this directory yet, add it.
399  // FIXME: Change the name to be a char* that points back to the
400  // 'SeenFileEntries' key.
401  UFE.Name    = InterndFileName;
402  UFE.Size    = StatBuf.st_size;
403  UFE.ModTime = StatBuf.st_mtime;
404  UFE.Dir     = DirInfo;
405  UFE.UID     = NextFileUID++;
406  UFE.FD      = FileDescriptor;
407  return &UFE;
408}
409
410const FileEntry *
411FileManager::getVirtualFile(StringRef Filename, off_t Size,
412                            time_t ModificationTime) {
413  ++NumFileLookups;
414
415  // See if there is already an entry in the map.
416  llvm::StringMapEntry<FileEntry *> &NamedFileEnt =
417    SeenFileEntries.GetOrCreateValue(Filename);
418
419  // See if there is already an entry in the map.
420  if (NamedFileEnt.getValue() && NamedFileEnt.getValue() != NON_EXISTENT_FILE)
421    return NamedFileEnt.getValue();
422
423  ++NumFileCacheMisses;
424
425  // By default, initialize it to invalid.
426  NamedFileEnt.setValue(NON_EXISTENT_FILE);
427
428  addAncestorsAsVirtualDirs(Filename);
429  FileEntry *UFE = 0;
430
431  // Now that all ancestors of Filename are in the cache, the
432  // following call is guaranteed to find the DirectoryEntry from the
433  // cache.
434  const DirectoryEntry *DirInfo = getDirectoryFromFile(*this, Filename,
435                                                       /*CacheFailure=*/true);
436  assert(DirInfo &&
437         "The directory of a virtual file should already be in the cache.");
438
439  // Check to see if the file exists. If so, drop the virtual file
440  int FileDescriptor = -1;
441  struct stat StatBuf;
442  const char *InterndFileName = NamedFileEnt.getKeyData();
443  if (getStatValue(InterndFileName, StatBuf, &FileDescriptor) == 0) {
444    // If the stat process opened the file, close it to avoid a FD leak.
445    if (FileDescriptor != -1)
446      close(FileDescriptor);
447
448    StatBuf.st_size = Size;
449    StatBuf.st_mtime = ModificationTime;
450    UFE = &UniqueRealFiles.getFile(InterndFileName, StatBuf);
451
452    NamedFileEnt.setValue(UFE);
453
454    // If we had already opened this file, close it now so we don't
455    // leak the descriptor. We're not going to use the file
456    // descriptor anyway, since this is a virtual file.
457    if (UFE->FD != -1) {
458      close(UFE->FD);
459      UFE->FD = -1;
460    }
461
462    // If we already have an entry with this inode, return it.
463    if (UFE->getName())
464      return UFE;
465  }
466
467  if (!UFE) {
468    UFE = new FileEntry();
469    VirtualFileEntries.push_back(UFE);
470    NamedFileEnt.setValue(UFE);
471  }
472
473  UFE->Name    = InterndFileName;
474  UFE->Size    = Size;
475  UFE->ModTime = ModificationTime;
476  UFE->Dir     = DirInfo;
477  UFE->UID     = NextFileUID++;
478  UFE->FD      = -1;
479  return UFE;
480}
481
482void 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(pathRef))
487    return;
488
489  SmallString<128> NewPath(FileSystemOpts.WorkingDir);
490  llvm::sys::path::append(NewPath, pathRef);
491  path = NewPath;
492}
493
494llvm::MemoryBuffer *FileManager::
495getBufferForFile(const FileEntry *Entry, std::string *ErrorStr,
496                 bool isVolatile) {
497  OwningPtr<llvm::MemoryBuffer> Result;
498  llvm::error_code ec;
499
500  uint64_t FileSize = Entry->getSize();
501  // If there's a high enough chance that the file have changed since we
502  // got its size, force a stat before opening it.
503  if (isVolatile)
504    FileSize = -1;
505
506  const char *Filename = Entry->getName();
507  // If the file is already open, use the open file descriptor.
508  if (Entry->FD != -1) {
509    ec = llvm::MemoryBuffer::getOpenFile(Entry->FD, Filename, Result, FileSize);
510    if (ErrorStr)
511      *ErrorStr = ec.message();
512
513    close(Entry->FD);
514    Entry->FD = -1;
515    return Result.take();
516  }
517
518  // Otherwise, open the file.
519
520  if (FileSystemOpts.WorkingDir.empty()) {
521    ec = llvm::MemoryBuffer::getFile(Filename, Result, FileSize);
522    if (ec && ErrorStr)
523      *ErrorStr = ec.message();
524    return Result.take();
525  }
526
527  SmallString<128> FilePath(Entry->getName());
528  FixupRelativePath(FilePath);
529  ec = llvm::MemoryBuffer::getFile(FilePath.str(), Result, FileSize);
530  if (ec && ErrorStr)
531    *ErrorStr = ec.message();
532  return Result.take();
533}
534
535llvm::MemoryBuffer *FileManager::
536getBufferForFile(StringRef Filename, std::string *ErrorStr) {
537  OwningPtr<llvm::MemoryBuffer> Result;
538  llvm::error_code ec;
539  if (FileSystemOpts.WorkingDir.empty()) {
540    ec = llvm::MemoryBuffer::getFile(Filename, Result);
541    if (ec && ErrorStr)
542      *ErrorStr = ec.message();
543    return Result.take();
544  }
545
546  SmallString<128> FilePath(Filename);
547  FixupRelativePath(FilePath);
548  ec = llvm::MemoryBuffer::getFile(FilePath.c_str(), Result);
549  if (ec && ErrorStr)
550    *ErrorStr = ec.message();
551  return Result.take();
552}
553
554/// getStatValue - Get the 'stat' information for the specified path,
555/// using the cache to accelerate it if possible.  This returns true
556/// if the path points to a virtual file or does not exist, or returns
557/// false if it's an existent real file.  If FileDescriptor is NULL,
558/// do directory look-up instead of file look-up.
559bool FileManager::getStatValue(const char *Path, struct stat &StatBuf,
560                               int *FileDescriptor) {
561  // FIXME: FileSystemOpts shouldn't be passed in here, all paths should be
562  // absolute!
563  if (FileSystemOpts.WorkingDir.empty())
564    return FileSystemStatCache::get(Path, StatBuf, FileDescriptor,
565                                    StatCache.get());
566
567  SmallString<128> FilePath(Path);
568  FixupRelativePath(FilePath);
569
570  return FileSystemStatCache::get(FilePath.c_str(), StatBuf, FileDescriptor,
571                                  StatCache.get());
572}
573
574bool FileManager::getNoncachedStatValue(StringRef Path,
575                                        struct stat &StatBuf) {
576  SmallString<128> FilePath(Path);
577  FixupRelativePath(FilePath);
578
579  return ::stat(FilePath.c_str(), &StatBuf) != 0;
580}
581
582void FileManager::invalidateCache(const FileEntry *Entry) {
583  assert(Entry && "Cannot invalidate a NULL FileEntry");
584
585  SeenFileEntries.erase(Entry->getName());
586
587  // FileEntry invalidation should not block future optimizations in the file
588  // caches. Possible alternatives are cache truncation (invalidate last N) or
589  // invalidation of the whole cache.
590  UniqueRealFiles.erase(Entry);
591}
592
593
594void FileManager::GetUniqueIDMapping(
595                   SmallVectorImpl<const FileEntry *> &UIDToFiles) const {
596  UIDToFiles.clear();
597  UIDToFiles.resize(NextFileUID);
598
599  // Map file entries
600  for (llvm::StringMap<FileEntry*, llvm::BumpPtrAllocator>::const_iterator
601         FE = SeenFileEntries.begin(), FEEnd = SeenFileEntries.end();
602       FE != FEEnd; ++FE)
603    if (FE->getValue() && FE->getValue() != NON_EXISTENT_FILE)
604      UIDToFiles[FE->getValue()->getUID()] = FE->getValue();
605
606  // Map virtual file entries
607  for (SmallVector<FileEntry*, 4>::const_iterator
608         VFE = VirtualFileEntries.begin(), VFEEnd = VirtualFileEntries.end();
609       VFE != VFEEnd; ++VFE)
610    if (*VFE && *VFE != NON_EXISTENT_FILE)
611      UIDToFiles[(*VFE)->getUID()] = *VFE;
612}
613
614void FileManager::modifyFileEntry(FileEntry *File,
615                                  off_t Size, time_t ModificationTime) {
616  File->Size = Size;
617  File->ModTime = ModificationTime;
618}
619
620
621void FileManager::PrintStats() const {
622  llvm::errs() << "\n*** File Manager Stats:\n";
623  llvm::errs() << UniqueRealFiles.size() << " real files found, "
624               << UniqueRealDirs.size() << " real dirs found.\n";
625  llvm::errs() << VirtualFileEntries.size() << " virtual files found, "
626               << VirtualDirectoryEntries.size() << " virtual dirs found.\n";
627  llvm::errs() << NumDirLookups << " dir lookups, "
628               << NumDirCacheMisses << " dir cache misses.\n";
629  llvm::errs() << NumFileLookups << " file lookups, "
630               << NumFileCacheMisses << " file cache misses.\n";
631
632  //llvm::errs() << PagesMapped << BytesOfPagesMapped << FSLookups;
633}
634