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::GetTemporaryDirectory(std::string *ErrMsg) {
114 #if defined(HAVE_MKDTEMP)
115 // The best way is with mkdtemp but that's not available on many systems,
116 // Linux and FreeBSD have it. Others probably won't.
117 char pathname[] = "/tmp/llvm_XXXXXX";
118 if (0 == mkdtemp(pathname)) {
120 std::string(pathname) + ": can't create temporary directory");
123 return Path(pathname);
124 #elif defined(HAVE_MKSTEMP)
125 // If no mkdtemp is available, mkstemp can be used to create a temporary file
126 // which is then removed and created as a directory. We prefer this over
127 // mktemp because of mktemp's inherent security and threading risks. We still
128 // have a slight race condition from the time the temporary file is created to
129 // the time it is re-created as a directoy.
130 char pathname[] = "/tmp/llvm_XXXXXX";
132 if (-1 == (fd = mkstemp(pathname))) {
134 std::string(pathname) + ": can't create temporary directory");
138 ::unlink(pathname); // start race condition, ignore errors
139 if (-1 == ::mkdir(pathname, S_IRWXU)) { // end race condition
141 std::string(pathname) + ": can't create temporary directory");
144 return Path(pathname);
145 #elif defined(HAVE_MKTEMP)
146 // If a system doesn't have mkdtemp(3) or mkstemp(3) but it does have
147 // mktemp(3) then we'll assume that system (e.g. AIX) has a reasonable
148 // implementation of mktemp(3) and doesn't follow BSD 4.3's lead of replacing
149 // the XXXXXX with the pid of the process and a letter. That leads to only
150 // twenty six temporary files that can be generated.
151 char pathname[] = "/tmp/llvm_XXXXXX";
152 char *TmpName = ::mktemp(pathname);
155 std::string(TmpName) + ": can't create unique directory name");
158 if (-1 == ::mkdir(TmpName, S_IRWXU)) {
160 std::string(TmpName) + ": can't create temporary directory");
163 return Path(TmpName);
165 // This is the worst case implementation. tempnam(3) leaks memory unless its
166 // on an SVID2 (or later) system. On BSD 4.3 it leaks. tmpnam(3) has thread
167 // issues. The mktemp(3) function doesn't have enough variability in the
168 // temporary name generated. So, we provide our own implementation that
169 // increments an integer from a random number seeded by the current time. This
170 // should be sufficiently unique that we don't have many collisions between
171 // processes. Generally LLVM processes don't run very long and don't use very
172 // many temporary files so this shouldn't be a big issue for LLVM.
173 static time_t num = ::time(0);
174 char pathname[MAXPATHLEN];
177 sprintf(pathname, "/tmp/llvm_%010u", unsigned(num));
178 } while ( 0 == access(pathname, F_OK ) );
179 if (-1 == ::mkdir(pathname, S_IRWXU)) {
181 std::string(pathname) + ": can't create temporary directory");
184 return Path(pathname);
189 Path::GetCurrentDirectory() {
190 char pathname[MAXPATHLEN];
191 if (!getcwd(pathname, MAXPATHLEN)) {
192 assert(false && "Could not query current working directory.");
196 return Path(pathname);
199 #if defined(__FreeBSD__) || defined (__NetBSD__) || defined(__Bitrig__) || \
200 defined(__OpenBSD__) || defined(__minix) || defined(__FreeBSD_kernel__) || \
201 defined(__linux__) || defined(__CYGWIN__)
203 test_dir(char buf[PATH_MAX], char ret[PATH_MAX],
204 const char *dir, const char *bin)
208 snprintf(buf, PATH_MAX, "%s/%s", dir, bin);
209 if (realpath(buf, ret) == NULL)
211 if (stat(buf, &sb) != 0)
218 getprogpath(char ret[PATH_MAX], const char *bin)
220 char *pv, *s, *t, buf[PATH_MAX];
222 /* First approach: absolute path. */
224 if (test_dir(buf, ret, "/", bin) == 0)
229 /* Second approach: relative path. */
230 if (strchr(bin, '/') != NULL) {
231 if (getcwd(buf, PATH_MAX) == NULL)
233 if (test_dir(buf, ret, buf, bin) == 0)
238 /* Third approach: $PATH */
239 if ((pv = getenv("PATH")) == NULL)
244 while ((t = strsep(&s, ":")) != NULL) {
245 if (test_dir(buf, ret, t, bin) == 0) {
253 #endif // __FreeBSD__ || __NetBSD__ || __FreeBSD_kernel__
255 /// GetMainExecutable - Return the path to the main executable, given the
256 /// value of argv[0] from program startup.
257 Path Path::GetMainExecutable(const char *argv0, void *MainAddr) {
258 #if defined(__APPLE__)
259 // On OS X the executable path is saved to the stack by dyld. Reading it
260 // from there is much faster than calling dladdr, especially for large
261 // binaries with symbols.
262 char exe_path[MAXPATHLEN];
263 uint32_t size = sizeof(exe_path);
264 if (_NSGetExecutablePath(exe_path, &size) == 0) {
265 char link_path[MAXPATHLEN];
266 if (realpath(exe_path, link_path))
267 return Path(link_path);
269 #elif defined(__FreeBSD__) || defined (__NetBSD__) || defined(__Bitrig__) || \
270 defined(__OpenBSD__) || defined(__minix) || defined(__FreeBSD_kernel__)
271 char exe_path[PATH_MAX];
273 if (getprogpath(exe_path, argv0) != NULL)
274 return Path(exe_path);
275 #elif defined(__linux__) || defined(__CYGWIN__)
276 char exe_path[MAXPATHLEN];
277 StringRef aPath("/proc/self/exe");
278 if (sys::fs::exists(aPath)) {
279 // /proc is not always mounted under Linux (chroot for example).
280 ssize_t len = readlink(aPath.str().c_str(), exe_path, sizeof(exe_path));
282 return Path(StringRef(exe_path, len));
284 // Fall back to the classical detection.
285 if (getprogpath(exe_path, argv0) != NULL)
286 return Path(exe_path);
288 #elif defined(HAVE_DLFCN_H)
289 // Use dladdr to get executable path if available.
291 int err = dladdr(MainAddr, &DLInfo);
295 // If the filename is a symlink, we need to resolve and return the location of
296 // the actual executable.
297 char link_path[MAXPATHLEN];
298 if (realpath(DLInfo.dli_fname, link_path))
299 return Path(link_path);
301 #error GetMainExecutable is not implemented on this host yet.
307 Path::exists() const {
308 return 0 == access(path.c_str(), F_OK );
312 Path::isDirectory() const {
314 if (0 != stat(path.c_str(), &buf))
316 return ((buf.st_mode & S_IFMT) == S_IFDIR) ? true : false;
320 Path::isSymLink() const {
322 if (0 != lstat(path.c_str(), &buf))
324 return S_ISLNK(buf.st_mode);
328 Path::isRegularFile() const {
329 // Get the status so we can determine if it's a file or directory
332 if (0 != stat(path.c_str(), &buf))
335 if (S_ISREG(buf.st_mode))
342 Path::canExecute() const {
343 if (0 != access(path.c_str(), R_OK | X_OK ))
346 if (0 != stat(path.c_str(), &buf))
348 if (!S_ISREG(buf.st_mode))
354 PathWithStatus::getFileStatus(bool update, std::string *ErrStr) const {
355 if (!fsIsValid || update) {
357 if (0 != stat(path.c_str(), &buf)) {
358 MakeErrMsg(ErrStr, path + ": can't get status of file");
361 status.fileSize = buf.st_size;
362 status.modTime.fromEpochTime(buf.st_mtime);
363 status.mode = buf.st_mode;
364 status.user = buf.st_uid;
365 status.group = buf.st_gid;
366 status.isDir = S_ISDIR(buf.st_mode);
367 status.isFile = S_ISREG(buf.st_mode);
373 static bool AddPermissionBits(const Path &File, int bits) {
374 // Get the umask value from the operating system. We want to use it
375 // when changing the file's permissions. Since calling umask() sets
376 // the umask and returns its old value, we must call it a second
377 // time to reset it to the user's preference.
378 int mask = umask(0777); // The arg. to umask is arbitrary.
379 umask(mask); // Restore the umask.
381 // Get the file's current mode.
383 if (0 != stat(File.c_str(), &buf))
385 // Change the file to have whichever permissions bits from 'bits'
386 // that the umask would not disable.
387 if ((chmod(File.c_str(), (buf.st_mode | (bits & ~mask)))) == -1)
392 bool Path::makeReadableOnDisk(std::string* ErrMsg) {
393 if (!AddPermissionBits(*this, 0444))
394 return MakeErrMsg(ErrMsg, path + ": can't make file readable");
398 bool Path::makeWriteableOnDisk(std::string* ErrMsg) {
399 if (!AddPermissionBits(*this, 0222))
400 return MakeErrMsg(ErrMsg, path + ": can't make file writable");
405 Path::getDirectoryContents(std::set<Path>& result, std::string* ErrMsg) const {
406 DIR* direntries = ::opendir(path.c_str());
408 return MakeErrMsg(ErrMsg, path + ": can't open directory");
410 std::string dirPath = path;
411 if (!lastIsSlash(dirPath))
415 struct dirent* de = ::readdir(direntries);
416 for ( ; de != 0; de = ::readdir(direntries)) {
417 if (de->d_name[0] != '.') {
418 Path aPath(dirPath + (const char*)de->d_name);
420 if (0 != lstat(aPath.path.c_str(), &st)) {
421 if (S_ISLNK(st.st_mode))
422 continue; // dangling symlink -- ignore
423 return MakeErrMsg(ErrMsg,
424 aPath.path + ": can't determine file object type");
426 result.insert(aPath);
430 closedir(direntries);
435 Path::set(StringRef a_path) {
443 Path::appendComponent(StringRef name) {
446 if (!lastIsSlash(path))
453 Path::eraseComponent() {
454 size_t slashpos = path.rfind('/',path.size());
455 if (slashpos == 0 || slashpos == std::string::npos) {
459 if (slashpos == path.size() - 1)
460 slashpos = path.rfind('/',slashpos-1);
461 if (slashpos == std::string::npos) {
465 path.erase(slashpos);
470 Path::eraseSuffix() {
471 size_t dotpos = path.rfind('.',path.size());
472 size_t slashpos = path.rfind('/',path.size());
473 if (dotpos != std::string::npos) {
474 if (slashpos == std::string::npos || dotpos > slashpos+1) {
475 path.erase(dotpos, path.size()-dotpos);
482 static bool createDirectoryHelper(char* beg, char* end, bool create_parents) {
484 if (access(beg, R_OK | W_OK) == 0)
487 if (create_parents) {
491 for (; c != beg; --c)
494 // Recurse to handling the parent directory.
496 bool x = createDirectoryHelper(beg, c, create_parents);
499 // Return if we encountered an error.
507 return mkdir(beg, S_IRWXU | S_IRWXG) != 0;
511 Path::createDirectoryOnDisk( bool create_parents, std::string* ErrMsg ) {
512 // Get a writeable copy of the path name
513 std::string pathname(path);
515 // Null-terminate the last component
516 size_t lastchar = path.length() - 1 ;
518 if (pathname[lastchar] != '/')
521 pathname[lastchar] = '\0';
523 if (createDirectoryHelper(&pathname[0], &pathname[lastchar], create_parents))
524 return MakeErrMsg(ErrMsg, pathname + ": can't create directory");
530 Path::createTemporaryFileOnDisk(bool reuse_current, std::string* ErrMsg) {
531 // Make this into a unique file name
532 if (makeUnique( reuse_current, ErrMsg ))
536 int fd = ::open(path.c_str(), O_WRONLY|O_CREAT|O_TRUNC, 0666);
538 return MakeErrMsg(ErrMsg, path + ": can't create temporary file");
544 Path::eraseFromDisk(bool remove_contents, std::string *ErrStr) const {
545 // Get the status so we can determine if it's a file or directory.
547 if (0 != stat(path.c_str(), &buf)) {
548 MakeErrMsg(ErrStr, path + ": can't get status of file");
552 // Note: this check catches strange situations. In all cases, LLVM should
553 // only be involved in the creation and deletion of regular files. This
554 // check ensures that what we're trying to erase is a regular file. It
555 // effectively prevents LLVM from erasing things like /dev/null, any block
556 // special file, or other things that aren't "regular" files.
557 if (S_ISREG(buf.st_mode)) {
558 if (unlink(path.c_str()) != 0)
559 return MakeErrMsg(ErrStr, path + ": can't destroy file");
563 if (!S_ISDIR(buf.st_mode)) {
564 if (ErrStr) *ErrStr = "not a file or directory";
568 if (remove_contents) {
569 // Recursively descend the directory to remove its contents.
570 std::string cmd = "/bin/rm -rf " + path;
571 if (system(cmd.c_str()) != 0) {
572 MakeErrMsg(ErrStr, path + ": failed to recursively remove directory.");
578 // Otherwise, try to just remove the one directory.
579 std::string pathname(path);
580 size_t lastchar = path.length() - 1;
581 if (pathname[lastchar] == '/')
582 pathname[lastchar] = '\0';
584 pathname[lastchar+1] = '\0';
586 if (rmdir(pathname.c_str()) != 0)
587 return MakeErrMsg(ErrStr, pathname + ": can't erase directory");
592 Path::renamePathOnDisk(const Path& newName, std::string* ErrMsg) {
593 if (0 != ::rename(path.c_str(), newName.c_str()))
594 return MakeErrMsg(ErrMsg, std::string("can't rename '") + path + "' as '" +
595 newName.str() + "'");
600 Path::setStatusInfoOnDisk(const FileStatus &si, std::string *ErrStr) const {
602 utb.actime = si.modTime.toPosixTime();
603 utb.modtime = utb.actime;
604 if (0 != ::utime(path.c_str(),&utb))
605 return MakeErrMsg(ErrStr, path + ": can't set file modification time");
606 if (0 != ::chmod(path.c_str(),si.mode))
607 return MakeErrMsg(ErrStr, path + ": can't set mode");
612 Path::makeUnique(bool reuse_current, std::string* ErrMsg) {
614 if (reuse_current && (fs::exists(path, Exists) || !Exists))
615 return false; // File doesn't exist already, just use it!
617 // Append an XXXXXX pattern to the end of the file for use with mkstemp,
618 // mktemp or our own implementation.
619 // This uses std::vector instead of SmallVector to avoid a dependence on
620 // libSupport. And performance isn't critical here.
621 std::vector<char> Buf;
622 Buf.resize(path.size()+8);
623 char *FNBuffer = &Buf[0];
624 path.copy(FNBuffer,path.size());
626 if (!fs::is_directory(path, isdir) && isdir)
627 strcpy(FNBuffer+path.size(), "/XXXXXX");
629 strcpy(FNBuffer+path.size(), "-XXXXXX");
631 #if defined(HAVE_MKSTEMP)
633 if ((TempFD = mkstemp(FNBuffer)) == -1)
634 return MakeErrMsg(ErrMsg, path + ": can't make unique filename");
636 // We don't need to hold the temp file descriptor... we will trust that no one
637 // will overwrite/delete the file before we can open it again.
643 // By default mkstemp sets the mode to 0600, so update mode bits now.
644 AddPermissionBits (*this, 0666);
645 #elif defined(HAVE_MKTEMP)
646 // If we don't have mkstemp, use the old and obsolete mktemp function.
647 if (mktemp(FNBuffer) == 0)
648 return MakeErrMsg(ErrMsg, path + ": can't make unique filename");
653 // Okay, looks like we have to do it all by our lonesome.
654 static unsigned FCounter = 0;
655 // Try to initialize with unique value.
656 if (FCounter == 0) FCounter = ((unsigned)getpid() & 0xFFFF) << 8;
657 char* pos = strstr(FNBuffer, "XXXXXX");
659 if (++FCounter > 0xFFFFFF) {
660 return MakeErrMsg(ErrMsg,
661 path + ": can't make unique filename: too many files");
663 sprintf(pos, "%06X", FCounter);
666 // POSSIBLE SECURITY BUG: An attacker can easily guess the name and exploit
671 } // end llvm namespace