1 //===- llvm/Support/Unix/Path.cpp - Unix Path Implementation -----*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the Unix specific portion of the Path class.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 //=== WARNING: Implementation here must contain only generic UNIX code that
16 //=== is guaranteed to work on *all* UNIX variants.
17 //===----------------------------------------------------------------------===//
26 #ifdef HAVE_SYS_MMAN_H
29 #ifdef HAVE_SYS_STAT_H
40 # define NAMLEN(dirent) strlen((dirent)->d_name)
42 # define dirent direct
43 # define NAMLEN(dirent) (dirent)->d_namlen
45 # include <sys/ndir.h>
60 #include <mach-o/dyld.h>
64 #if defined(__GNU__) && !defined(MAXPATHLEN)
65 # define MAXPATHLEN 4096
68 // Put in a hack for Cygwin which falsely reports that the mkdtemp function
69 // is available when it is not.
75 inline bool lastIsSlash(const std::string& path) {
76 return !path.empty() && path[path.length() - 1] == '/';
84 const char sys::PathSeparator = ':';
86 StringRef Path::GetEXESuffix() {
90 Path::Path(StringRef p)
93 Path::Path(const char *StrStart, unsigned StrLen)
94 : path(StrStart, StrLen) {}
97 Path::operator=(StringRef that) {
98 path.assign(that.data(), that.size());
103 Path::isValid() const {
104 // Empty paths are considered invalid here.
105 // This code doesn't check MAXPATHLEN because there's no need. Nothing in
106 // LLVM manipulates Paths with fixed-sizes arrays, and if the OS can't
107 // handle names longer than some limit, it'll report this on demand using
109 return !path.empty();
113 Path::isAbsolute(const char *NameStart, unsigned NameLen) {
117 return NameStart[0] == '/';
121 Path::isAbsolute() const {
124 return path[0] == '/';
128 Path::GetTemporaryDirectory(std::string *ErrMsg) {
129 #if defined(HAVE_MKDTEMP)
130 // The best way is with mkdtemp but that's not available on many systems,
131 // Linux and FreeBSD have it. Others probably won't.
132 char pathname[] = "/tmp/llvm_XXXXXX";
133 if (0 == mkdtemp(pathname)) {
135 std::string(pathname) + ": can't create temporary directory");
138 return Path(pathname);
139 #elif defined(HAVE_MKSTEMP)
140 // If no mkdtemp is available, mkstemp can be used to create a temporary file
141 // which is then removed and created as a directory. We prefer this over
142 // mktemp because of mktemp's inherent security and threading risks. We still
143 // have a slight race condition from the time the temporary file is created to
144 // the time it is re-created as a directoy.
145 char pathname[] = "/tmp/llvm_XXXXXX";
147 if (-1 == (fd = mkstemp(pathname))) {
149 std::string(pathname) + ": can't create temporary directory");
153 ::unlink(pathname); // start race condition, ignore errors
154 if (-1 == ::mkdir(pathname, S_IRWXU)) { // end race condition
156 std::string(pathname) + ": can't create temporary directory");
159 return Path(pathname);
160 #elif defined(HAVE_MKTEMP)
161 // If a system doesn't have mkdtemp(3) or mkstemp(3) but it does have
162 // mktemp(3) then we'll assume that system (e.g. AIX) has a reasonable
163 // implementation of mktemp(3) and doesn't follow BSD 4.3's lead of replacing
164 // the XXXXXX with the pid of the process and a letter. That leads to only
165 // twenty six temporary files that can be generated.
166 char pathname[] = "/tmp/llvm_XXXXXX";
167 char *TmpName = ::mktemp(pathname);
170 std::string(TmpName) + ": can't create unique directory name");
173 if (-1 == ::mkdir(TmpName, S_IRWXU)) {
175 std::string(TmpName) + ": can't create temporary directory");
178 return Path(TmpName);
180 // This is the worst case implementation. tempnam(3) leaks memory unless its
181 // on an SVID2 (or later) system. On BSD 4.3 it leaks. tmpnam(3) has thread
182 // issues. The mktemp(3) function doesn't have enough variability in the
183 // temporary name generated. So, we provide our own implementation that
184 // increments an integer from a random number seeded by the current time. This
185 // should be sufficiently unique that we don't have many collisions between
186 // processes. Generally LLVM processes don't run very long and don't use very
187 // many temporary files so this shouldn't be a big issue for LLVM.
188 static time_t num = ::time(0);
189 char pathname[MAXPATHLEN];
192 sprintf(pathname, "/tmp/llvm_%010u", unsigned(num));
193 } while ( 0 == access(pathname, F_OK ) );
194 if (-1 == ::mkdir(pathname, S_IRWXU)) {
196 std::string(pathname) + ": can't create temporary directory");
199 return Path(pathname);
204 Path::GetCurrentDirectory() {
205 char pathname[MAXPATHLEN];
206 if (!getcwd(pathname, MAXPATHLEN)) {
207 assert(false && "Could not query current working directory.");
211 return Path(pathname);
214 #if defined(__FreeBSD__) || defined (__NetBSD__) || defined(__Bitrig__) || \
215 defined(__OpenBSD__) || defined(__minix) || defined(__FreeBSD_kernel__) || \
216 defined(__linux__) || defined(__CYGWIN__)
218 test_dir(char buf[PATH_MAX], char ret[PATH_MAX],
219 const char *dir, const char *bin)
223 snprintf(buf, PATH_MAX, "%s/%s", dir, bin);
224 if (realpath(buf, ret) == NULL)
226 if (stat(buf, &sb) != 0)
233 getprogpath(char ret[PATH_MAX], const char *bin)
235 char *pv, *s, *t, buf[PATH_MAX];
237 /* First approach: absolute path. */
239 if (test_dir(buf, ret, "/", bin) == 0)
244 /* Second approach: relative path. */
245 if (strchr(bin, '/') != NULL) {
246 if (getcwd(buf, PATH_MAX) == NULL)
248 if (test_dir(buf, ret, buf, bin) == 0)
253 /* Third approach: $PATH */
254 if ((pv = getenv("PATH")) == NULL)
259 while ((t = strsep(&s, ":")) != NULL) {
260 if (test_dir(buf, ret, t, bin) == 0) {
268 #endif // __FreeBSD__ || __NetBSD__ || __FreeBSD_kernel__
270 /// GetMainExecutable - Return the path to the main executable, given the
271 /// value of argv[0] from program startup.
272 Path Path::GetMainExecutable(const char *argv0, void *MainAddr) {
273 #if defined(__APPLE__)
274 // On OS X the executable path is saved to the stack by dyld. Reading it
275 // from there is much faster than calling dladdr, especially for large
276 // binaries with symbols.
277 char exe_path[MAXPATHLEN];
278 uint32_t size = sizeof(exe_path);
279 if (_NSGetExecutablePath(exe_path, &size) == 0) {
280 char link_path[MAXPATHLEN];
281 if (realpath(exe_path, link_path))
282 return Path(link_path);
284 #elif defined(__FreeBSD__) || defined (__NetBSD__) || defined(__Bitrig__) || \
285 defined(__OpenBSD__) || defined(__minix) || defined(__FreeBSD_kernel__)
286 char exe_path[PATH_MAX];
288 if (getprogpath(exe_path, argv0) != NULL)
289 return Path(exe_path);
290 #elif defined(__linux__) || defined(__CYGWIN__)
291 char exe_path[MAXPATHLEN];
292 StringRef aPath("/proc/self/exe");
293 if (sys::fs::exists(aPath)) {
294 // /proc is not always mounted under Linux (chroot for example).
295 ssize_t len = readlink(aPath.str().c_str(), exe_path, sizeof(exe_path));
297 return Path(StringRef(exe_path, len));
299 // Fall back to the classical detection.
300 if (getprogpath(exe_path, argv0) != NULL)
301 return Path(exe_path);
303 #elif defined(HAVE_DLFCN_H)
304 // Use dladdr to get executable path if available.
306 int err = dladdr(MainAddr, &DLInfo);
310 // If the filename is a symlink, we need to resolve and return the location of
311 // the actual executable.
312 char link_path[MAXPATHLEN];
313 if (realpath(DLInfo.dli_fname, link_path))
314 return Path(link_path);
316 #error GetMainExecutable is not implemented on this host yet.
321 bool Path::getMagicNumber(std::string &Magic, unsigned len) const {
322 assert(len < 1024 && "Request for magic string too long");
324 int fd = ::open(path.c_str(), O_RDONLY);
327 ssize_t bytes_read = ::read(fd, Buf, len);
329 if (ssize_t(len) != bytes_read)
331 Magic.assign(Buf, len);
336 Path::exists() const {
337 return 0 == access(path.c_str(), F_OK );
341 Path::isDirectory() const {
343 if (0 != stat(path.c_str(), &buf))
345 return ((buf.st_mode & S_IFMT) == S_IFDIR) ? true : false;
349 Path::isSymLink() const {
351 if (0 != lstat(path.c_str(), &buf))
353 return S_ISLNK(buf.st_mode);
358 Path::canRead() const {
359 return 0 == access(path.c_str(), R_OK);
363 Path::canWrite() const {
364 return 0 == access(path.c_str(), W_OK);
368 Path::isRegularFile() const {
369 // Get the status so we can determine if it's a file or directory
372 if (0 != stat(path.c_str(), &buf))
375 if (S_ISREG(buf.st_mode))
382 Path::canExecute() const {
383 if (0 != access(path.c_str(), R_OK | X_OK ))
386 if (0 != stat(path.c_str(), &buf))
388 if (!S_ISREG(buf.st_mode))
394 PathWithStatus::getFileStatus(bool update, std::string *ErrStr) const {
395 if (!fsIsValid || update) {
397 if (0 != stat(path.c_str(), &buf)) {
398 MakeErrMsg(ErrStr, path + ": can't get status of file");
401 status.fileSize = buf.st_size;
402 status.modTime.fromEpochTime(buf.st_mtime);
403 status.mode = buf.st_mode;
404 status.user = buf.st_uid;
405 status.group = buf.st_gid;
406 status.uniqueID = uint64_t(buf.st_ino);
407 status.isDir = S_ISDIR(buf.st_mode);
408 status.isFile = S_ISREG(buf.st_mode);
414 static bool AddPermissionBits(const Path &File, int bits) {
415 // Get the umask value from the operating system. We want to use it
416 // when changing the file's permissions. Since calling umask() sets
417 // the umask and returns its old value, we must call it a second
418 // time to reset it to the user's preference.
419 int mask = umask(0777); // The arg. to umask is arbitrary.
420 umask(mask); // Restore the umask.
422 // Get the file's current mode.
424 if (0 != stat(File.c_str(), &buf))
426 // Change the file to have whichever permissions bits from 'bits'
427 // that the umask would not disable.
428 if ((chmod(File.c_str(), (buf.st_mode | (bits & ~mask)))) == -1)
433 bool Path::makeReadableOnDisk(std::string* ErrMsg) {
434 if (!AddPermissionBits(*this, 0444))
435 return MakeErrMsg(ErrMsg, path + ": can't make file readable");
439 bool Path::makeWriteableOnDisk(std::string* ErrMsg) {
440 if (!AddPermissionBits(*this, 0222))
441 return MakeErrMsg(ErrMsg, path + ": can't make file writable");
446 Path::getDirectoryContents(std::set<Path>& result, std::string* ErrMsg) const {
447 DIR* direntries = ::opendir(path.c_str());
449 return MakeErrMsg(ErrMsg, path + ": can't open directory");
451 std::string dirPath = path;
452 if (!lastIsSlash(dirPath))
456 struct dirent* de = ::readdir(direntries);
457 for ( ; de != 0; de = ::readdir(direntries)) {
458 if (de->d_name[0] != '.') {
459 Path aPath(dirPath + (const char*)de->d_name);
461 if (0 != lstat(aPath.path.c_str(), &st)) {
462 if (S_ISLNK(st.st_mode))
463 continue; // dangling symlink -- ignore
464 return MakeErrMsg(ErrMsg,
465 aPath.path + ": can't determine file object type");
467 result.insert(aPath);
471 closedir(direntries);
476 Path::set(StringRef a_path) {
484 Path::appendComponent(StringRef name) {
487 if (!lastIsSlash(path))
494 Path::eraseComponent() {
495 size_t slashpos = path.rfind('/',path.size());
496 if (slashpos == 0 || slashpos == std::string::npos) {
500 if (slashpos == path.size() - 1)
501 slashpos = path.rfind('/',slashpos-1);
502 if (slashpos == std::string::npos) {
506 path.erase(slashpos);
511 Path::eraseSuffix() {
512 size_t dotpos = path.rfind('.',path.size());
513 size_t slashpos = path.rfind('/',path.size());
514 if (dotpos != std::string::npos) {
515 if (slashpos == std::string::npos || dotpos > slashpos+1) {
516 path.erase(dotpos, path.size()-dotpos);
523 static bool createDirectoryHelper(char* beg, char* end, bool create_parents) {
525 if (access(beg, R_OK | W_OK) == 0)
528 if (create_parents) {
532 for (; c != beg; --c)
535 // Recurse to handling the parent directory.
537 bool x = createDirectoryHelper(beg, c, create_parents);
540 // Return if we encountered an error.
548 return mkdir(beg, S_IRWXU | S_IRWXG) != 0;
552 Path::createDirectoryOnDisk( bool create_parents, std::string* ErrMsg ) {
553 // Get a writeable copy of the path name
554 std::string pathname(path);
556 // Null-terminate the last component
557 size_t lastchar = path.length() - 1 ;
559 if (pathname[lastchar] != '/')
562 pathname[lastchar] = '\0';
564 if (createDirectoryHelper(&pathname[0], &pathname[lastchar], create_parents))
565 return MakeErrMsg(ErrMsg, pathname + ": can't create directory");
571 Path::createTemporaryFileOnDisk(bool reuse_current, std::string* ErrMsg) {
572 // Make this into a unique file name
573 if (makeUnique( reuse_current, ErrMsg ))
577 int fd = ::open(path.c_str(), O_WRONLY|O_CREAT|O_TRUNC, 0666);
579 return MakeErrMsg(ErrMsg, path + ": can't create temporary file");
585 Path::eraseFromDisk(bool remove_contents, std::string *ErrStr) const {
586 // Get the status so we can determine if it's a file or directory.
588 if (0 != stat(path.c_str(), &buf)) {
589 MakeErrMsg(ErrStr, path + ": can't get status of file");
593 // Note: this check catches strange situations. In all cases, LLVM should
594 // only be involved in the creation and deletion of regular files. This
595 // check ensures that what we're trying to erase is a regular file. It
596 // effectively prevents LLVM from erasing things like /dev/null, any block
597 // special file, or other things that aren't "regular" files.
598 if (S_ISREG(buf.st_mode)) {
599 if (unlink(path.c_str()) != 0)
600 return MakeErrMsg(ErrStr, path + ": can't destroy file");
604 if (!S_ISDIR(buf.st_mode)) {
605 if (ErrStr) *ErrStr = "not a file or directory";
609 if (remove_contents) {
610 // Recursively descend the directory to remove its contents.
611 std::string cmd = "/bin/rm -rf " + path;
612 if (system(cmd.c_str()) != 0) {
613 MakeErrMsg(ErrStr, path + ": failed to recursively remove directory.");
619 // Otherwise, try to just remove the one directory.
620 std::string pathname(path);
621 size_t lastchar = path.length() - 1;
622 if (pathname[lastchar] == '/')
623 pathname[lastchar] = '\0';
625 pathname[lastchar+1] = '\0';
627 if (rmdir(pathname.c_str()) != 0)
628 return MakeErrMsg(ErrStr, pathname + ": can't erase directory");
633 Path::renamePathOnDisk(const Path& newName, std::string* ErrMsg) {
634 if (0 != ::rename(path.c_str(), newName.c_str()))
635 return MakeErrMsg(ErrMsg, std::string("can't rename '") + path + "' as '" +
636 newName.str() + "'");
641 Path::setStatusInfoOnDisk(const FileStatus &si, std::string *ErrStr) const {
643 utb.actime = si.modTime.toPosixTime();
644 utb.modtime = utb.actime;
645 if (0 != ::utime(path.c_str(),&utb))
646 return MakeErrMsg(ErrStr, path + ": can't set file modification time");
647 if (0 != ::chmod(path.c_str(),si.mode))
648 return MakeErrMsg(ErrStr, path + ": can't set mode");
653 Path::makeUnique(bool reuse_current, std::string* ErrMsg) {
655 if (reuse_current && (fs::exists(path, Exists) || !Exists))
656 return false; // File doesn't exist already, just use it!
658 // Append an XXXXXX pattern to the end of the file for use with mkstemp,
659 // mktemp or our own implementation.
660 // This uses std::vector instead of SmallVector to avoid a dependence on
661 // libSupport. And performance isn't critical here.
662 std::vector<char> Buf;
663 Buf.resize(path.size()+8);
664 char *FNBuffer = &Buf[0];
665 path.copy(FNBuffer,path.size());
667 if (!fs::is_directory(path, isdir) && isdir)
668 strcpy(FNBuffer+path.size(), "/XXXXXX");
670 strcpy(FNBuffer+path.size(), "-XXXXXX");
672 #if defined(HAVE_MKSTEMP)
674 if ((TempFD = mkstemp(FNBuffer)) == -1)
675 return MakeErrMsg(ErrMsg, path + ": can't make unique filename");
677 // We don't need to hold the temp file descriptor... we will trust that no one
678 // will overwrite/delete the file before we can open it again.
684 // By default mkstemp sets the mode to 0600, so update mode bits now.
685 AddPermissionBits (*this, 0666);
686 #elif defined(HAVE_MKTEMP)
687 // If we don't have mkstemp, use the old and obsolete mktemp function.
688 if (mktemp(FNBuffer) == 0)
689 return MakeErrMsg(ErrMsg, path + ": can't make unique filename");
694 // Okay, looks like we have to do it all by our lonesome.
695 static unsigned FCounter = 0;
696 // Try to initialize with unique value.
697 if (FCounter == 0) FCounter = ((unsigned)getpid() & 0xFFFF) << 8;
698 char* pos = strstr(FNBuffer, "XXXXXX");
700 if (++FCounter > 0xFFFFFF) {
701 return MakeErrMsg(ErrMsg,
702 path + ": can't make unique filename: too many files");
704 sprintf(pos, "%06X", FCounter);
707 // POSSIBLE SECURITY BUG: An attacker can easily guess the name and exploit
712 } // end llvm namespace