blob: 74cd2f295be60f4784944b36647f4f51c462fd41 [file] [log] [blame]
Inna Palantff3f07a2019-07-11 16:15:26 -07001//===--- FileManager.cpp - File System Probing and Caching ----------------===//
2//
Chih-Hung Hsieh08600532019-12-19 15:55:38 -08003// 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
Inna Palantff3f07a2019-07-11 16:15:26 -07006//
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"
Inna Palantff3f07a2019-07-11 16:15:26 -070021#include "llvm/ADT/STLExtras.h"
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +020022#include "llvm/ADT/SmallString.h"
23#include "llvm/ADT/Statistic.h"
24#include "llvm/Config/llvm-config.h"
Inna Palantff3f07a2019-07-11 16:15:26 -070025#include "llvm/Support/FileSystem.h"
26#include "llvm/Support/MemoryBuffer.h"
27#include "llvm/Support/Path.h"
28#include "llvm/Support/raw_ostream.h"
29#include <algorithm>
30#include <cassert>
31#include <climits>
32#include <cstdint>
33#include <cstdlib>
34#include <string>
35#include <utility>
36
37using namespace clang;
38
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +020039#define DEBUG_TYPE "file-search"
40
41ALWAYS_ENABLED_STATISTIC(NumDirLookups, "Number of directory lookups.");
42ALWAYS_ENABLED_STATISTIC(NumFileLookups, "Number of file lookups.");
43ALWAYS_ENABLED_STATISTIC(NumDirCacheMisses,
44 "Number of directory cache misses.");
45ALWAYS_ENABLED_STATISTIC(NumFileCacheMisses, "Number of file cache misses.");
46
Inna Palantff3f07a2019-07-11 16:15:26 -070047//===----------------------------------------------------------------------===//
48// Common logic.
49//===----------------------------------------------------------------------===//
50
51FileManager::FileManager(const FileSystemOptions &FSO,
52 IntrusiveRefCntPtr<llvm::vfs::FileSystem> FS)
53 : FS(std::move(FS)), FileSystemOpts(FSO), SeenDirEntries(64),
54 SeenFileEntries(64), NextFileUID(0) {
Inna Palantff3f07a2019-07-11 16:15:26 -070055 // If the caller doesn't provide a virtual file system, just grab the real
56 // file system.
57 if (!this->FS)
58 this->FS = llvm::vfs::getRealFileSystem();
59}
60
61FileManager::~FileManager() = default;
62
63void FileManager::setStatCache(std::unique_ptr<FileSystemStatCache> statCache) {
64 assert(statCache && "No stat cache provided?");
65 StatCache = std::move(statCache);
66}
67
68void FileManager::clearStatCache() { StatCache.reset(); }
69
70/// Retrieve the directory that the given file name resides in.
71/// Filename can point to either a real file or a virtual file.
Chris Wailese3116c42021-07-13 14:40:48 -070072static llvm::Expected<DirectoryEntryRef>
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +020073getDirectoryFromFile(FileManager &FileMgr, StringRef Filename,
74 bool CacheFailure) {
Inna Palantff3f07a2019-07-11 16:15:26 -070075 if (Filename.empty())
Chris Wailese3116c42021-07-13 14:40:48 -070076 return llvm::errorCodeToError(
77 make_error_code(std::errc::no_such_file_or_directory));
Inna Palantff3f07a2019-07-11 16:15:26 -070078
79 if (llvm::sys::path::is_separator(Filename[Filename.size() - 1]))
Chris Wailese3116c42021-07-13 14:40:48 -070080 return llvm::errorCodeToError(make_error_code(std::errc::is_a_directory));
Inna Palantff3f07a2019-07-11 16:15:26 -070081
82 StringRef DirName = llvm::sys::path::parent_path(Filename);
83 // Use the current directory if file has no path component.
84 if (DirName.empty())
85 DirName = ".";
86
Chris Wailese3116c42021-07-13 14:40:48 -070087 return FileMgr.getDirectoryRef(DirName, CacheFailure);
Inna Palantff3f07a2019-07-11 16:15:26 -070088}
89
90/// Add all ancestors of the given path (pointing to either a file or
91/// a directory) as virtual directories.
92void FileManager::addAncestorsAsVirtualDirs(StringRef Path) {
93 StringRef DirName = llvm::sys::path::parent_path(Path);
94 if (DirName.empty())
95 DirName = ".";
96
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +020097 auto &NamedDirEnt = *SeenDirEntries.insert(
98 {DirName, std::errc::no_such_file_or_directory}).first;
Inna Palantff3f07a2019-07-11 16:15:26 -070099
100 // When caching a virtual directory, we always cache its ancestors
101 // at the same time. Therefore, if DirName is already in the cache,
102 // we don't need to recurse as its ancestors must also already be in
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800103 // the cache (or it's a known non-virtual directory).
104 if (NamedDirEnt.second)
Inna Palantff3f07a2019-07-11 16:15:26 -0700105 return;
106
107 // Add the virtual directory to the cache.
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200108 auto UDE = std::make_unique<DirectoryEntry>();
Inna Palantff3f07a2019-07-11 16:15:26 -0700109 UDE->Name = NamedDirEnt.first();
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200110 NamedDirEnt.second = *UDE.get();
Inna Palantff3f07a2019-07-11 16:15:26 -0700111 VirtualDirectoryEntries.push_back(std::move(UDE));
112
113 // Recursively add the other ancestors.
114 addAncestorsAsVirtualDirs(DirName);
115}
116
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200117llvm::Expected<DirectoryEntryRef>
118FileManager::getDirectoryRef(StringRef DirName, bool CacheFailure) {
Inna Palantff3f07a2019-07-11 16:15:26 -0700119 // stat doesn't like trailing separators except for root directory.
120 // At least, on Win32 MSVCRT, stat() cannot strip trailing '/'.
121 // (though it can strip '\\')
122 if (DirName.size() > 1 &&
123 DirName != llvm::sys::path::root_path(DirName) &&
124 llvm::sys::path::is_separator(DirName.back()))
125 DirName = DirName.substr(0, DirName.size()-1);
126#ifdef _WIN32
127 // Fixing a problem with "clang C:test.c" on Windows.
128 // Stat("C:") does not recognize "C:" as a valid directory
129 std::string DirNameStr;
130 if (DirName.size() > 1 && DirName.back() == ':' &&
Chris Wailesbcf972c2021-10-21 11:03:28 -0700131 DirName.equals_insensitive(llvm::sys::path::root_name(DirName))) {
Inna Palantff3f07a2019-07-11 16:15:26 -0700132 DirNameStr = DirName.str() + '.';
133 DirName = DirNameStr;
134 }
135#endif
136
137 ++NumDirLookups;
Inna Palantff3f07a2019-07-11 16:15:26 -0700138
139 // See if there was already an entry in the map. Note that the map
140 // contains both virtual and real directories.
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200141 auto SeenDirInsertResult =
142 SeenDirEntries.insert({DirName, std::errc::no_such_file_or_directory});
143 if (!SeenDirInsertResult.second) {
144 if (SeenDirInsertResult.first->second)
Chris Wailese3116c42021-07-13 14:40:48 -0700145 return DirectoryEntryRef(*SeenDirInsertResult.first);
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200146 return llvm::errorCodeToError(SeenDirInsertResult.first->second.getError());
147 }
Inna Palantff3f07a2019-07-11 16:15:26 -0700148
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800149 // We've not seen this before. Fill it in.
Inna Palantff3f07a2019-07-11 16:15:26 -0700150 ++NumDirCacheMisses;
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800151 auto &NamedDirEnt = *SeenDirInsertResult.first;
152 assert(!NamedDirEnt.second && "should be newly-created");
Inna Palantff3f07a2019-07-11 16:15:26 -0700153
154 // Get the null-terminated directory name as stored as the key of the
155 // SeenDirEntries map.
156 StringRef InterndDirName = NamedDirEnt.first();
157
158 // Check to see if the directory exists.
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800159 llvm::vfs::Status Status;
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200160 auto statError = getStatValue(InterndDirName, Status, false,
161 nullptr /*directory lookup*/);
162 if (statError) {
Inna Palantff3f07a2019-07-11 16:15:26 -0700163 // There's no real directory at the given path.
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200164 if (CacheFailure)
165 NamedDirEnt.second = statError;
166 else
Inna Palantff3f07a2019-07-11 16:15:26 -0700167 SeenDirEntries.erase(DirName);
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200168 return llvm::errorCodeToError(statError);
Inna Palantff3f07a2019-07-11 16:15:26 -0700169 }
170
171 // It exists. See if we have already opened a directory with the
172 // same inode (this occurs on Unix-like systems when one dir is
173 // symlinked to another, for example) or the same path (on
174 // Windows).
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800175 DirectoryEntry &UDE = UniqueRealDirs[Status.getUniqueID()];
Inna Palantff3f07a2019-07-11 16:15:26 -0700176
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200177 NamedDirEnt.second = UDE;
Inna Palantff3f07a2019-07-11 16:15:26 -0700178 if (UDE.getName().empty()) {
179 // We don't have this directory yet, add it. We use the string
180 // key from the SeenDirEntries map as the string.
181 UDE.Name = InterndDirName;
182 }
183
Chris Wailese3116c42021-07-13 14:40:48 -0700184 return DirectoryEntryRef(NamedDirEnt);
Inna Palantff3f07a2019-07-11 16:15:26 -0700185}
186
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200187llvm::ErrorOr<const DirectoryEntry *>
188FileManager::getDirectory(StringRef DirName, bool CacheFailure) {
189 auto Result = getDirectoryRef(DirName, CacheFailure);
190 if (Result)
191 return &Result->getDirEntry();
192 return llvm::errorToErrorCode(Result.takeError());
193}
194
195llvm::ErrorOr<const FileEntry *>
196FileManager::getFile(StringRef Filename, bool openFile, bool CacheFailure) {
197 auto Result = getFileRef(Filename, openFile, CacheFailure);
198 if (Result)
199 return &Result->getFileEntry();
200 return llvm::errorToErrorCode(Result.takeError());
201}
202
203llvm::Expected<FileEntryRef>
204FileManager::getFileRef(StringRef Filename, bool openFile, bool CacheFailure) {
Inna Palantff3f07a2019-07-11 16:15:26 -0700205 ++NumFileLookups;
206
207 // See if there is already an entry in the map.
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200208 auto SeenFileInsertResult =
209 SeenFileEntries.insert({Filename, std::errc::no_such_file_or_directory});
210 if (!SeenFileInsertResult.second) {
211 if (!SeenFileInsertResult.first->second)
212 return llvm::errorCodeToError(
213 SeenFileInsertResult.first->second.getError());
214 // Construct and return and FileEntryRef, unless it's a redirect to another
215 // filename.
Chris Wailese3116c42021-07-13 14:40:48 -0700216 FileEntryRef::MapValue Value = *SeenFileInsertResult.first->second;
217 if (LLVM_LIKELY(Value.V.is<FileEntry *>()))
218 return FileEntryRef(*SeenFileInsertResult.first);
219 return FileEntryRef(*reinterpret_cast<const FileEntryRef::MapEntry *>(
220 Value.V.get<const void *>()));
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200221 }
Inna Palantff3f07a2019-07-11 16:15:26 -0700222
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800223 // We've not seen this before. Fill it in.
Inna Palantff3f07a2019-07-11 16:15:26 -0700224 ++NumFileCacheMisses;
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200225 auto *NamedFileEnt = &*SeenFileInsertResult.first;
226 assert(!NamedFileEnt->second && "should be newly-created");
Inna Palantff3f07a2019-07-11 16:15:26 -0700227
228 // Get the null-terminated file name as stored as the key of the
229 // SeenFileEntries map.
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200230 StringRef InterndFileName = NamedFileEnt->first();
Inna Palantff3f07a2019-07-11 16:15:26 -0700231
232 // Look up the directory for the file. When looking up something like
233 // sys/foo.h we'll discover all of the search directories that have a 'sys'
234 // subdirectory. This will let us avoid having to waste time on known-to-fail
235 // searches when we go to find sys/bar.h, because all the search directories
236 // without a 'sys' subdir will get a cached failure result.
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200237 auto DirInfoOrErr = getDirectoryFromFile(*this, Filename, CacheFailure);
238 if (!DirInfoOrErr) { // Directory doesn't exist, file can't exist.
Chris Wailese3116c42021-07-13 14:40:48 -0700239 std::error_code Err = errorToErrorCode(DirInfoOrErr.takeError());
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200240 if (CacheFailure)
Chris Wailese3116c42021-07-13 14:40:48 -0700241 NamedFileEnt->second = Err;
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200242 else
Inna Palantff3f07a2019-07-11 16:15:26 -0700243 SeenFileEntries.erase(Filename);
244
Chris Wailese3116c42021-07-13 14:40:48 -0700245 return llvm::errorCodeToError(Err);
Inna Palantff3f07a2019-07-11 16:15:26 -0700246 }
Chris Wailese3116c42021-07-13 14:40:48 -0700247 DirectoryEntryRef DirInfo = *DirInfoOrErr;
Inna Palantff3f07a2019-07-11 16:15:26 -0700248
249 // FIXME: Use the directory info to prune this, before doing the stat syscall.
250 // FIXME: This will reduce the # syscalls.
251
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800252 // Check to see if the file exists.
Inna Palantff3f07a2019-07-11 16:15:26 -0700253 std::unique_ptr<llvm::vfs::File> F;
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800254 llvm::vfs::Status Status;
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200255 auto statError = getStatValue(InterndFileName, Status, true,
256 openFile ? &F : nullptr);
257 if (statError) {
Inna Palantff3f07a2019-07-11 16:15:26 -0700258 // There's no real file at the given path.
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200259 if (CacheFailure)
260 NamedFileEnt->second = statError;
261 else
Inna Palantff3f07a2019-07-11 16:15:26 -0700262 SeenFileEntries.erase(Filename);
263
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200264 return llvm::errorCodeToError(statError);
Inna Palantff3f07a2019-07-11 16:15:26 -0700265 }
266
267 assert((openFile || !F) && "undesired open file");
268
269 // It exists. See if we have already opened a file with the same inode.
270 // This occurs when one dir is symlinked to another, for example.
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800271 FileEntry &UFE = UniqueRealFiles[Status.getUniqueID()];
Inna Palantff3f07a2019-07-11 16:15:26 -0700272
Chris Wailese3116c42021-07-13 14:40:48 -0700273 if (Status.getName() == Filename) {
274 // The name matches. Set the FileEntry.
275 NamedFileEnt->second = FileEntryRef::MapValue(UFE, DirInfo);
276 } else {
277 // Name mismatch. We need a redirect. First grab the actual entry we want
278 // to return.
279 auto &Redirection =
280 *SeenFileEntries
281 .insert({Status.getName(), FileEntryRef::MapValue(UFE, DirInfo)})
282 .first;
283 assert(Redirection.second->V.is<FileEntry *>() &&
284 "filename redirected to a non-canonical filename?");
285 assert(Redirection.second->V.get<FileEntry *>() == &UFE &&
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800286 "filename from getStatValue() refers to wrong file");
Chris Wailese3116c42021-07-13 14:40:48 -0700287
288 // Cache the redirection in the previously-inserted entry, still available
289 // in the tentative return value.
290 NamedFileEnt->second = FileEntryRef::MapValue(Redirection);
291
292 // Fix the tentative return value.
293 NamedFileEnt = &Redirection;
Inna Palantff3f07a2019-07-11 16:15:26 -0700294 }
295
Chris Wailese3116c42021-07-13 14:40:48 -0700296 FileEntryRef ReturnedRef(*NamedFileEnt);
Inna Palantff3f07a2019-07-11 16:15:26 -0700297 if (UFE.isValid()) { // Already have an entry with this inode, return it.
298
299 // FIXME: this hack ensures that if we look up a file by a virtual path in
300 // the VFS that the getDir() will have the virtual path, even if we found
301 // the file by a 'real' path first. This is required in order to find a
302 // module's structure when its headers/module map are mapped in the VFS.
303 // We should remove this as soon as we can properly support a file having
304 // multiple names.
Chris Wailese3116c42021-07-13 14:40:48 -0700305 if (&DirInfo.getDirEntry() != UFE.Dir && Status.IsVFSMapped)
306 UFE.Dir = &DirInfo.getDirEntry();
Inna Palantff3f07a2019-07-11 16:15:26 -0700307
Chris Wailese3116c42021-07-13 14:40:48 -0700308 // Always update LastRef to the last name by which a file was accessed.
309 // FIXME: Neither this nor always using the first reference is correct; we
310 // want to switch towards a design where we return a FileName object that
Inna Palantff3f07a2019-07-11 16:15:26 -0700311 // encapsulates both the name by which the file was accessed and the
312 // corresponding FileEntry.
Chris Wailese3116c42021-07-13 14:40:48 -0700313 // FIXME: LastRef should be removed from FileEntry once all clients adopt
314 // FileEntryRef.
315 UFE.LastRef = ReturnedRef;
Inna Palantff3f07a2019-07-11 16:15:26 -0700316
Chris Wailese3116c42021-07-13 14:40:48 -0700317 return ReturnedRef;
Inna Palantff3f07a2019-07-11 16:15:26 -0700318 }
319
320 // Otherwise, we don't have this file yet, add it.
Chris Wailese3116c42021-07-13 14:40:48 -0700321 UFE.LastRef = ReturnedRef;
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800322 UFE.Size = Status.getSize();
323 UFE.ModTime = llvm::sys::toTimeT(Status.getLastModificationTime());
Chris Wailese3116c42021-07-13 14:40:48 -0700324 UFE.Dir = &DirInfo.getDirEntry();
Inna Palantff3f07a2019-07-11 16:15:26 -0700325 UFE.UID = NextFileUID++;
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800326 UFE.UniqueID = Status.getUniqueID();
327 UFE.IsNamedPipe = Status.getType() == llvm::sys::fs::file_type::fifo_file;
Inna Palantff3f07a2019-07-11 16:15:26 -0700328 UFE.File = std::move(F);
329 UFE.IsValid = true;
330
331 if (UFE.File) {
332 if (auto PathName = UFE.File->getName())
333 fillRealPathName(&UFE, *PathName);
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800334 } else if (!openFile) {
335 // We should still fill the path even if we aren't opening the file.
336 fillRealPathName(&UFE, InterndFileName);
Inna Palantff3f07a2019-07-11 16:15:26 -0700337 }
Chris Wailese3116c42021-07-13 14:40:48 -0700338 return ReturnedRef;
Inna Palantff3f07a2019-07-11 16:15:26 -0700339}
340
Chris Wailese3116c42021-07-13 14:40:48 -0700341llvm::Expected<FileEntryRef> FileManager::getSTDIN() {
342 // Only read stdin once.
343 if (STDIN)
344 return *STDIN;
345
346 std::unique_ptr<llvm::MemoryBuffer> Content;
347 if (auto ContentOrError = llvm::MemoryBuffer::getSTDIN())
348 Content = std::move(*ContentOrError);
349 else
350 return llvm::errorCodeToError(ContentOrError.getError());
351
352 STDIN = getVirtualFileRef(Content->getBufferIdentifier(),
353 Content->getBufferSize(), 0);
354 FileEntry &FE = const_cast<FileEntry &>(STDIN->getFileEntry());
355 FE.Content = std::move(Content);
356 FE.IsNamedPipe = true;
357 return *STDIN;
358}
359
360const FileEntry *FileManager::getVirtualFile(StringRef Filename, off_t Size,
361 time_t ModificationTime) {
362 return &getVirtualFileRef(Filename, Size, ModificationTime).getFileEntry();
363}
364
365FileEntryRef FileManager::getVirtualFileRef(StringRef Filename, off_t Size,
366 time_t ModificationTime) {
Inna Palantff3f07a2019-07-11 16:15:26 -0700367 ++NumFileLookups;
368
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800369 // See if there is already an entry in the map for an existing file.
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200370 auto &NamedFileEnt = *SeenFileEntries.insert(
371 {Filename, std::errc::no_such_file_or_directory}).first;
372 if (NamedFileEnt.second) {
Chris Wailese3116c42021-07-13 14:40:48 -0700373 FileEntryRef::MapValue Value = *NamedFileEnt.second;
374 if (LLVM_LIKELY(Value.V.is<FileEntry *>()))
375 return FileEntryRef(NamedFileEnt);
376 return FileEntryRef(*reinterpret_cast<const FileEntryRef::MapEntry *>(
377 Value.V.get<const void *>()));
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200378 }
Inna Palantff3f07a2019-07-11 16:15:26 -0700379
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800380 // We've not seen this before, or the file is cached as non-existent.
Inna Palantff3f07a2019-07-11 16:15:26 -0700381 ++NumFileCacheMisses;
Inna Palantff3f07a2019-07-11 16:15:26 -0700382 addAncestorsAsVirtualDirs(Filename);
383 FileEntry *UFE = nullptr;
384
385 // Now that all ancestors of Filename are in the cache, the
386 // following call is guaranteed to find the DirectoryEntry from the
Chris Wailesbcf972c2021-10-21 11:03:28 -0700387 // cache. A virtual file can also have an empty filename, that could come
388 // from a source location preprocessor directive with an empty filename as
389 // an example, so we need to pretend it has a name to ensure a valid directory
390 // entry can be returned.
391 auto DirInfo = expectedToOptional(getDirectoryFromFile(
392 *this, Filename.empty() ? "." : Filename, /*CacheFailure=*/true));
Inna Palantff3f07a2019-07-11 16:15:26 -0700393 assert(DirInfo &&
394 "The directory of a virtual file should already be in the cache.");
395
396 // Check to see if the file exists. If so, drop the virtual file
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800397 llvm::vfs::Status Status;
Inna Palantff3f07a2019-07-11 16:15:26 -0700398 const char *InterndFileName = NamedFileEnt.first().data();
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200399 if (!getStatValue(InterndFileName, Status, true, nullptr)) {
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800400 UFE = &UniqueRealFiles[Status.getUniqueID()];
401 Status = llvm::vfs::Status(
402 Status.getName(), Status.getUniqueID(),
403 llvm::sys::toTimePoint(ModificationTime),
404 Status.getUser(), Status.getGroup(), Size,
405 Status.getType(), Status.getPermissions());
Inna Palantff3f07a2019-07-11 16:15:26 -0700406
Chris Wailese3116c42021-07-13 14:40:48 -0700407 NamedFileEnt.second = FileEntryRef::MapValue(*UFE, *DirInfo);
Inna Palantff3f07a2019-07-11 16:15:26 -0700408
409 // If we had already opened this file, close it now so we don't
410 // leak the descriptor. We're not going to use the file
411 // descriptor anyway, since this is a virtual file.
412 if (UFE->File)
413 UFE->closeFile();
414
415 // If we already have an entry with this inode, return it.
Chris Wailese3116c42021-07-13 14:40:48 -0700416 //
417 // FIXME: Surely this should add a reference by the new name, and return
418 // it instead...
Inna Palantff3f07a2019-07-11 16:15:26 -0700419 if (UFE->isValid())
Chris Wailese3116c42021-07-13 14:40:48 -0700420 return FileEntryRef(NamedFileEnt);
Inna Palantff3f07a2019-07-11 16:15:26 -0700421
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800422 UFE->UniqueID = Status.getUniqueID();
423 UFE->IsNamedPipe = Status.getType() == llvm::sys::fs::file_type::fifo_file;
424 fillRealPathName(UFE, Status.getName());
425 } else {
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200426 VirtualFileEntries.push_back(std::make_unique<FileEntry>());
Inna Palantff3f07a2019-07-11 16:15:26 -0700427 UFE = VirtualFileEntries.back().get();
Chris Wailese3116c42021-07-13 14:40:48 -0700428 NamedFileEnt.second = FileEntryRef::MapValue(*UFE, *DirInfo);
Inna Palantff3f07a2019-07-11 16:15:26 -0700429 }
430
Chris Wailese3116c42021-07-13 14:40:48 -0700431 UFE->LastRef = FileEntryRef(NamedFileEnt);
Inna Palantff3f07a2019-07-11 16:15:26 -0700432 UFE->Size = Size;
433 UFE->ModTime = ModificationTime;
Chris Wailese3116c42021-07-13 14:40:48 -0700434 UFE->Dir = &DirInfo->getDirEntry();
Inna Palantff3f07a2019-07-11 16:15:26 -0700435 UFE->UID = NextFileUID++;
436 UFE->IsValid = true;
437 UFE->File.reset();
Chris Wailese3116c42021-07-13 14:40:48 -0700438 return FileEntryRef(NamedFileEnt);
Inna Palantff3f07a2019-07-11 16:15:26 -0700439}
440
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200441llvm::Optional<FileEntryRef> FileManager::getBypassFile(FileEntryRef VF) {
442 // Stat of the file and return nullptr if it doesn't exist.
443 llvm::vfs::Status Status;
444 if (getStatValue(VF.getName(), Status, /*isFile=*/true, /*F=*/nullptr))
445 return None;
446
Chris Wailese3116c42021-07-13 14:40:48 -0700447 if (!SeenBypassFileEntries)
448 SeenBypassFileEntries = std::make_unique<
449 llvm::StringMap<llvm::ErrorOr<FileEntryRef::MapValue>>>();
450
451 // If we've already bypassed just use the existing one.
452 auto Insertion = SeenBypassFileEntries->insert(
453 {VF.getName(), std::errc::no_such_file_or_directory});
454 if (!Insertion.second)
455 return FileEntryRef(*Insertion.first);
456
457 // Fill in the new entry from the stat.
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200458 BypassFileEntries.push_back(std::make_unique<FileEntry>());
459 const FileEntry &VFE = VF.getFileEntry();
460 FileEntry &BFE = *BypassFileEntries.back();
Chris Wailese3116c42021-07-13 14:40:48 -0700461 Insertion.first->second = FileEntryRef::MapValue(BFE, VF.getDir());
462 BFE.LastRef = FileEntryRef(*Insertion.first);
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200463 BFE.Size = Status.getSize();
464 BFE.Dir = VFE.Dir;
465 BFE.ModTime = llvm::sys::toTimeT(Status.getLastModificationTime());
466 BFE.UID = NextFileUID++;
467 BFE.IsValid = true;
Chris Wailese3116c42021-07-13 14:40:48 -0700468
469 // Save the entry in the bypass table and return.
470 return FileEntryRef(*Insertion.first);
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200471}
472
Inna Palantff3f07a2019-07-11 16:15:26 -0700473bool FileManager::FixupRelativePath(SmallVectorImpl<char> &path) const {
474 StringRef pathRef(path.data(), path.size());
475
476 if (FileSystemOpts.WorkingDir.empty()
477 || llvm::sys::path::is_absolute(pathRef))
478 return false;
479
480 SmallString<128> NewPath(FileSystemOpts.WorkingDir);
481 llvm::sys::path::append(NewPath, pathRef);
482 path = NewPath;
483 return true;
484}
485
486bool FileManager::makeAbsolutePath(SmallVectorImpl<char> &Path) const {
487 bool Changed = FixupRelativePath(Path);
488
489 if (!llvm::sys::path::is_absolute(StringRef(Path.data(), Path.size()))) {
490 FS->makeAbsolute(Path);
491 Changed = true;
492 }
493
494 return Changed;
495}
496
497void FileManager::fillRealPathName(FileEntry *UFE, llvm::StringRef FileName) {
498 llvm::SmallString<128> AbsPath(FileName);
499 // This is not the same as `VFS::getRealPath()`, which resolves symlinks
500 // but can be very expensive on real file systems.
501 // FIXME: the semantic of RealPathName is unclear, and the name might be
502 // misleading. We need to clean up the interface here.
503 makeAbsolutePath(AbsPath);
504 llvm::sys::path::remove_dots(AbsPath, /*remove_dot_dot=*/true);
ThiƩbaud Weksteene40e7362020-10-28 15:03:00 +0100505 UFE->RealPathName = std::string(AbsPath.str());
Inna Palantff3f07a2019-07-11 16:15:26 -0700506}
507
508llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
ThiƩbaud Weksteene40e7362020-10-28 15:03:00 +0100509FileManager::getBufferForFile(const FileEntry *Entry, bool isVolatile,
510 bool RequiresNullTerminator) {
Chris Wailese3116c42021-07-13 14:40:48 -0700511 // If the content is living on the file entry, return a reference to it.
512 if (Entry->Content)
513 return llvm::MemoryBuffer::getMemBuffer(Entry->Content->getMemBufferRef());
514
Inna Palantff3f07a2019-07-11 16:15:26 -0700515 uint64_t FileSize = Entry->getSize();
516 // If there's a high enough chance that the file have changed since we
517 // got its size, force a stat before opening it.
Chris Wailese3116c42021-07-13 14:40:48 -0700518 if (isVolatile || Entry->isNamedPipe())
Inna Palantff3f07a2019-07-11 16:15:26 -0700519 FileSize = -1;
520
521 StringRef Filename = Entry->getName();
522 // If the file is already open, use the open file descriptor.
523 if (Entry->File) {
ThiƩbaud Weksteene40e7362020-10-28 15:03:00 +0100524 auto Result = Entry->File->getBuffer(Filename, FileSize,
525 RequiresNullTerminator, isVolatile);
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200526 Entry->closeFile();
Inna Palantff3f07a2019-07-11 16:15:26 -0700527 return Result;
528 }
529
530 // Otherwise, open the file.
ThiƩbaud Weksteene40e7362020-10-28 15:03:00 +0100531 return getBufferForFileImpl(Filename, FileSize, isVolatile,
532 RequiresNullTerminator);
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200533}
Inna Palantff3f07a2019-07-11 16:15:26 -0700534
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200535llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>>
536FileManager::getBufferForFileImpl(StringRef Filename, int64_t FileSize,
ThiƩbaud Weksteene40e7362020-10-28 15:03:00 +0100537 bool isVolatile,
538 bool RequiresNullTerminator) {
Inna Palantff3f07a2019-07-11 16:15:26 -0700539 if (FileSystemOpts.WorkingDir.empty())
ThiƩbaud Weksteene40e7362020-10-28 15:03:00 +0100540 return FS->getBufferForFile(Filename, FileSize, RequiresNullTerminator,
541 isVolatile);
Inna Palantff3f07a2019-07-11 16:15:26 -0700542
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200543 SmallString<128> FilePath(Filename);
Inna Palantff3f07a2019-07-11 16:15:26 -0700544 FixupRelativePath(FilePath);
ThiƩbaud Weksteene40e7362020-10-28 15:03:00 +0100545 return FS->getBufferForFile(FilePath, FileSize, RequiresNullTerminator,
546 isVolatile);
Inna Palantff3f07a2019-07-11 16:15:26 -0700547}
548
Inna Palantff3f07a2019-07-11 16:15:26 -0700549/// getStatValue - Get the 'stat' information for the specified path,
550/// using the cache to accelerate it if possible. This returns true
551/// if the path points to a virtual file or does not exist, or returns
552/// false if it's an existent real file. If FileDescriptor is NULL,
553/// do directory look-up instead of file look-up.
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200554std::error_code
555FileManager::getStatValue(StringRef Path, llvm::vfs::Status &Status,
556 bool isFile, std::unique_ptr<llvm::vfs::File> *F) {
Inna Palantff3f07a2019-07-11 16:15:26 -0700557 // FIXME: FileSystemOpts shouldn't be passed in here, all paths should be
558 // absolute!
559 if (FileSystemOpts.WorkingDir.empty())
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200560 return FileSystemStatCache::get(Path, Status, isFile, F,
561 StatCache.get(), *FS);
Inna Palantff3f07a2019-07-11 16:15:26 -0700562
563 SmallString<128> FilePath(Path);
564 FixupRelativePath(FilePath);
565
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200566 return FileSystemStatCache::get(FilePath.c_str(), Status, isFile, F,
567 StatCache.get(), *FS);
Inna Palantff3f07a2019-07-11 16:15:26 -0700568}
569
ThiƩbaud Weksteene40e7362020-10-28 15:03:00 +0100570std::error_code
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200571FileManager::getNoncachedStatValue(StringRef Path,
572 llvm::vfs::Status &Result) {
Inna Palantff3f07a2019-07-11 16:15:26 -0700573 SmallString<128> FilePath(Path);
574 FixupRelativePath(FilePath);
575
576 llvm::ErrorOr<llvm::vfs::Status> S = FS->status(FilePath.c_str());
577 if (!S)
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200578 return S.getError();
Inna Palantff3f07a2019-07-11 16:15:26 -0700579 Result = *S;
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200580 return std::error_code();
Inna Palantff3f07a2019-07-11 16:15:26 -0700581}
582
583void FileManager::GetUniqueIDMapping(
584 SmallVectorImpl<const FileEntry *> &UIDToFiles) const {
585 UIDToFiles.clear();
586 UIDToFiles.resize(NextFileUID);
587
588 // Map file entries
Chris Wailese3116c42021-07-13 14:40:48 -0700589 for (llvm::StringMap<llvm::ErrorOr<FileEntryRef::MapValue>,
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200590 llvm::BumpPtrAllocator>::const_iterator
591 FE = SeenFileEntries.begin(),
592 FEEnd = SeenFileEntries.end();
Inna Palantff3f07a2019-07-11 16:15:26 -0700593 FE != FEEnd; ++FE)
Chris Wailese3116c42021-07-13 14:40:48 -0700594 if (llvm::ErrorOr<FileEntryRef::MapValue> Entry = FE->getValue()) {
595 if (const auto *FE = Entry->V.dyn_cast<FileEntry *>())
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200596 UIDToFiles[FE->getUID()] = FE;
597 }
Inna Palantff3f07a2019-07-11 16:15:26 -0700598
599 // Map virtual file entries
600 for (const auto &VFE : VirtualFileEntries)
Chih-Hung Hsieh08600532019-12-19 15:55:38 -0800601 UIDToFiles[VFE->getUID()] = VFE.get();
Inna Palantff3f07a2019-07-11 16:15:26 -0700602}
603
Inna Palantff3f07a2019-07-11 16:15:26 -0700604StringRef FileManager::getCanonicalName(const DirectoryEntry *Dir) {
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200605 llvm::DenseMap<const void *, llvm::StringRef>::iterator Known
606 = CanonicalNames.find(Dir);
607 if (Known != CanonicalNames.end())
Inna Palantff3f07a2019-07-11 16:15:26 -0700608 return Known->second;
609
610 StringRef CanonicalName(Dir->getName());
611
612 SmallString<4096> CanonicalNameBuf;
613 if (!FS->getRealPath(Dir->getName(), CanonicalNameBuf))
Chris Wailesbcf972c2021-10-21 11:03:28 -0700614 CanonicalName = CanonicalNameBuf.str().copy(CanonicalNameStorage);
Inna Palantff3f07a2019-07-11 16:15:26 -0700615
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200616 CanonicalNames.insert({Dir, CanonicalName});
617 return CanonicalName;
618}
619
620StringRef FileManager::getCanonicalName(const FileEntry *File) {
621 llvm::DenseMap<const void *, llvm::StringRef>::iterator Known
622 = CanonicalNames.find(File);
623 if (Known != CanonicalNames.end())
624 return Known->second;
625
626 StringRef CanonicalName(File->getName());
627
628 SmallString<4096> CanonicalNameBuf;
629 if (!FS->getRealPath(File->getName(), CanonicalNameBuf))
Chris Wailesbcf972c2021-10-21 11:03:28 -0700630 CanonicalName = CanonicalNameBuf.str().copy(CanonicalNameStorage);
Jeff Vander Stoep247d86b2020-08-11 14:27:44 +0200631
632 CanonicalNames.insert({File, CanonicalName});
Inna Palantff3f07a2019-07-11 16:15:26 -0700633 return CanonicalName;
634}
635
636void FileManager::PrintStats() const {
637 llvm::errs() << "\n*** File Manager Stats:\n";
638 llvm::errs() << UniqueRealFiles.size() << " real files found, "
639 << UniqueRealDirs.size() << " real dirs found.\n";
640 llvm::errs() << VirtualFileEntries.size() << " virtual files found, "
641 << VirtualDirectoryEntries.size() << " virtual dirs found.\n";
642 llvm::errs() << NumDirLookups << " dir lookups, "
643 << NumDirCacheMisses << " dir cache misses.\n";
644 llvm::errs() << NumFileLookups << " file lookups, "
645 << NumFileCacheMisses << " file cache misses.\n";
646
647 //llvm::errs() << PagesMapped << BytesOfPagesMapped << FSLookups;
648}