Win32: support long paths

Windows paths are typically limited to MAX_PATH = 260 characters, even
though the underlying NTFS file system supports paths up to 32,767 chars.
This limitation is also evident in Windows Explorer, cmd.exe and many
other applications (including IDEs).

Particularly annoying is that most Windows APIs return bogus error codes
if a relative path only barely exceeds MAX_PATH in conjunction with the
current directory, e.g. ERROR_PATH_NOT_FOUND / ENOENT instead of the
infinitely more helpful ERROR_FILENAME_EXCED_RANGE / ENAMETOOLONG.

Many Windows wide char APIs support longer than MAX_PATH paths through the
file namespace prefix ('\\?\' or '\\?\UNC\') followed by an absolute path.
Notable exceptions include functions dealing with executables and the
current directory (CreateProcess, LoadLibrary, Get/SetCurrentDirectory) as
well as the entire shell API (ShellExecute, SHGetSpecialFolderPath...).

Introduce a handle_long_path function to check the length of a specified
path properly (and fail with ENAMETOOLONG), and to optionally expand long
paths using the '\\?\' file namespace prefix. Short paths will not be
modified, so we don't need to worry about device names (NUL, CON, AUX).

Contrary to MSDN docs, the GetFullPathNameW function doesn't seem to be
limited to MAX_PATH (at least not on Win7), so we can use it to do the
heavy lifting of the conversion (translate '/' to '\', eliminate '.' and
'..', and make an absolute path).

Add long path error checking to xutftowcs_path for APIs with hard MAX_PATH
limit.

Add a new MAX_LONG_PATH constant and xutftowcs_long_path function for APIs
that support long paths.

While improved error checking is always active, long paths support must be
explicitly enabled via 'core.longpaths' option. This is to prevent end
users to shoot themselves in the foot by checking out files that Windows
Explorer, cmd/bash or their favorite IDE cannot handle.

Test suite:
Test the case is when the full pathname length of a dir is close
to 260 (MAX_PATH).
Bug report and an original reproducer by Andrey Rogozhnikov:
msysgit#122 (comment)

[jes: adjusted test number to avoid conflicts]

Thanks-to: Martin W. Kirst <[email protected]>
Thanks-to: Doug Kelly <[email protected]>
Signed-off-by: Karsten Blees <[email protected]>
Original-test-by: Andrey Rogozhnikov <[email protected]>
Signed-off-by: Stepan Kasal <[email protected]>
Signed-off-by: Johannes Schindelin <[email protected]>

Author: dscho

Documentation/config.txt

@@ -766,6 +766,13 @@ core.fscache::
 Git for Windows uses this to bulk-read and cache lstat data of entire
 directories (instead of doing lstat file by file).
 
+core.longpaths::
+	Enable long path (> 260) support for builtin commands in Git for
+	Windows. This is disabled by default, as long paths are not supported
+	by Windows Explorer, cmd.exe and the Git for Windows tool chain
+	(msys, bash, tcl, perl...). Only enable this if you know what you're
+	doing and are prepared to live with a few quirks.
+
 core.createObject::
 	You can set this to 'link', in which case a hardlink followed by
 	a delete of the source are used to make sure that object creation

cache.h

@@ -731,6 +731,8 @@ extern enum hide_dotfiles_type hide_dotfiles;
 
 extern int core_fscache;
 
+extern int core_long_paths;
+
 enum branch_track {
 	BRANCH_TRACK_UNSPECIFIED = -1,
 	BRANCH_TRACK_NEVER = 0,

compat/mingw.c

@@ -205,8 +205,8 @@ static int ask_yes_no_if_possible(const char *format, ...)
 int mingw_unlink(const char *pathname)
 {
 	int ret, tries = 0;
-	wchar_t wpathname[MAX_PATH];
-	if (xutftowcs_path(wpathname, pathname) < 0)
+	wchar_t wpathname[MAX_LONG_PATH];
+	if (xutftowcs_long_path(wpathname, pathname) < 0)
 		return -1;
 
 	/* read-only files cannot be removed */
@@ -235,7 +235,7 @@ static int is_dir_empty(const wchar_t *wpath)
 {
 	WIN32_FIND_DATAW findbuf;
 	HANDLE handle;
-	wchar_t wbuf[MAX_PATH + 2];
+	wchar_t wbuf[MAX_LONG_PATH + 2];
 	wcscpy(wbuf, wpath);
 	wcscat(wbuf, L"\\*");
 	handle = FindFirstFileW(wbuf, &findbuf);
@@ -256,8 +256,8 @@ static int is_dir_empty(const wchar_t *wpath)
 int mingw_rmdir(const char *pathname)
 {
 	int ret, tries = 0;
-	wchar_t wpathname[MAX_PATH];
-	if (xutftowcs_path(wpathname, pathname) < 0)
+	wchar_t wpathname[MAX_LONG_PATH];
+	if (xutftowcs_long_path(wpathname, pathname) < 0)
 		return -1;
 
 	while ((ret = _wrmdir(wpathname)) == -1 && tries < ARRAY_SIZE(delay)) {
@@ -332,9 +332,12 @@ static int set_hidden_flag(const wchar_t *path, int set)
 int mingw_mkdir(const char *path, int mode)
 {
 	int ret;
-	wchar_t wpath[MAX_PATH];
-	if (xutftowcs_path(wpath, path) < 0)
+	wchar_t wpath[MAX_LONG_PATH];
+	/* CreateDirectoryW path limit is 248 (MAX_PATH - 8.3 file name) */
+	if (xutftowcs_path_ex(wpath, path, MAX_LONG_PATH, -1, 248,
+			core_long_paths) < 0)
 		return -1;
+
 	ret = _wmkdir(wpath);
 	if (!ret && needs_hiding(path))
 		return set_hidden_flag(wpath, 1);
@@ -346,7 +349,7 @@ int mingw_open (const char *filename, int oflags, ...)
 	va_list args;
 	unsigned mode;
 	int fd;
-	wchar_t wfilename[MAX_PATH];
+	wchar_t wfilename[MAX_LONG_PATH];
 
 	va_start(args, oflags);
 	mode = va_arg(args, int);
@@ -355,7 +358,7 @@ int mingw_open (const char *filename, int oflags, ...)
 	if (filename && !strcmp(filename, "/dev/null"))
 		filename = "nul";
 
-	if (xutftowcs_path(wfilename, filename) < 0)
+	if (xutftowcs_long_path(wfilename, filename) < 0)
 		return -1;
 	fd = _wopen(wfilename, oflags, mode);
 
@@ -412,10 +415,10 @@ FILE *mingw_fopen (const char *filename, const char *otype)
 {
 	int hide = needs_hiding(filename);
 	FILE *file;
-	wchar_t wfilename[MAX_PATH], wotype[4];
+	wchar_t wfilename[MAX_LONG_PATH], wotype[4];
 	if (filename && !strcmp(filename, "/dev/null"))
 		filename = "nul";
-	if (xutftowcs_path(wfilename, filename) < 0 ||
+	if (xutftowcs_long_path(wfilename, filename) < 0 ||
 		xutftowcs(wotype, otype, ARRAY_SIZE(wotype)) < 0)
 		return NULL;
 	if (hide && !access(filename, F_OK) && set_hidden_flag(wfilename, 0)) {
@@ -432,10 +435,10 @@ FILE *mingw_freopen (const char *filename, const char *otype, FILE *stream)
 {
 	int hide = needs_hiding(filename);
 	FILE *file;
-	wchar_t wfilename[MAX_PATH], wotype[4];
+	wchar_t wfilename[MAX_LONG_PATH], wotype[4];
 	if (filename && !strcmp(filename, "/dev/null"))
 		filename = "nul";
-	if (xutftowcs_path(wfilename, filename) < 0 ||
+	if (xutftowcs_long_path(wfilename, filename) < 0 ||
 		xutftowcs(wotype, otype, ARRAY_SIZE(wotype)) < 0)
 		return NULL;
 	if (hide && !access(filename, F_OK) && set_hidden_flag(wfilename, 0)) {
@@ -489,25 +492,32 @@ ssize_t mingw_write(int fd, const void *buf, size_t len)
 
 int mingw_access(const char *filename, int mode)
 {
-	wchar_t wfilename[MAX_PATH];
-	if (xutftowcs_path(wfilename, filename) < 0)
+	wchar_t wfilename[MAX_LONG_PATH];
+	if (xutftowcs_long_path(wfilename, filename) < 0)
 		return -1;
 	/* X_OK is not supported by the MSVCRT version */
 	return _waccess(wfilename, mode & ~X_OK);
 }
 
+/* cached length of current directory for handle_long_path */
+static int current_directory_len = 0;
+
 int mingw_chdir(const char *dirname)
 {
+	int result;
 	wchar_t wdirname[MAX_PATH];
+	/* SetCurrentDirectoryW doesn't support long paths */
 	if (xutftowcs_path(wdirname, dirname) < 0)
 		return -1;
-	return _wchdir(wdirname);
+	result = _wchdir(wdirname);
+	current_directory_len = GetCurrentDirectoryW(0, NULL);
+	return result;
 }
 
 int mingw_chmod(const char *filename, int mode)
 {
-	wchar_t wfilename[MAX_PATH];
-	if (xutftowcs_path(wfilename, filename) < 0)
+	wchar_t wfilename[MAX_LONG_PATH];
+	if (xutftowcs_long_path(wfilename, filename) < 0)
 		return -1;
 	return _wchmod(wfilename, mode);
 }
@@ -555,8 +565,8 @@ static int has_valid_directory_prefix(wchar_t *wfilename)
 static int do_lstat(int follow, const char *file_name, struct stat *buf)
 {
 	WIN32_FILE_ATTRIBUTE_DATA fdata;
-	wchar_t wfilename[MAX_PATH];
-	if (xutftowcs_path(wfilename, file_name) < 0)
+	wchar_t wfilename[MAX_LONG_PATH];
+	if (xutftowcs_long_path(wfilename, file_name) < 0)
 		return -1;
 
 	if (GetFileAttributesExW(wfilename, GetFileExInfoStandard, &fdata)) {
@@ -705,8 +715,8 @@ int mingw_utime (const char *file_name, const struct utimbuf *times)
 	FILETIME mft, aft;
 	int fh, rc;
 	DWORD attrs;
-	wchar_t wfilename[MAX_PATH];
-	if (xutftowcs_path(wfilename, file_name) < 0)
+	wchar_t wfilename[MAX_LONG_PATH];
+	if (xutftowcs_long_path(wfilename, file_name) < 0)
 		return -1;
 
 	/* must have write permission */
@@ -754,6 +764,7 @@ unsigned int sleep (unsigned int seconds)
 char *mingw_mktemp(char *template)
 {
 	wchar_t wtemplate[MAX_PATH];
+	/* we need to return the path, thus no long paths here! */
 	if (xutftowcs_path(wtemplate, template) < 0)
 		return NULL;
 	if (!_wmktemp(wtemplate))
@@ -1104,6 +1115,7 @@ static pid_t mingw_spawnve_fd(const char *cmd, const char **argv, char **deltaen
 	si.hStdOutput = winansi_get_osfhandle(fhout);
 	si.hStdError = winansi_get_osfhandle(fherr);
 
+	/* executables and the current directory don't support long paths */
 	if (xutftowcs_path(wcmd, cmd) < 0)
 		return -1;
 	if (dir && xutftowcs_path(wdir, dir) < 0)
@@ -1679,8 +1691,9 @@ int mingw_rename(const char *pold, const char *pnew)
 {
 	DWORD attrs, gle;
 	int tries = 0;
-	wchar_t wpold[MAX_PATH], wpnew[MAX_PATH];
-	if (xutftowcs_path(wpold, pold) < 0 || xutftowcs_path(wpnew, pnew) < 0)
+	wchar_t wpold[MAX_LONG_PATH], wpnew[MAX_LONG_PATH];
+	if (xutftowcs_long_path(wpold, pold) < 0 ||
+	    xutftowcs_long_path(wpnew, pnew) < 0)
 		return -1;
 
 	/*
@@ -1920,9 +1933,9 @@ int link(const char *oldpath, const char *newpath)
 {
 	typedef BOOL (WINAPI *T)(LPCWSTR, LPCWSTR, LPSECURITY_ATTRIBUTES);
 	static T create_hard_link = NULL;
-	wchar_t woldpath[MAX_PATH], wnewpath[MAX_PATH];
-	if (xutftowcs_path(woldpath, oldpath) < 0 ||
-		xutftowcs_path(wnewpath, newpath) < 0)
+	wchar_t woldpath[MAX_LONG_PATH], wnewpath[MAX_LONG_PATH];
+	if (xutftowcs_long_path(woldpath, oldpath) < 0 ||
+	    xutftowcs_long_path(wnewpath, newpath) < 0)
 		return -1;
 
 	if (!create_hard_link) {
@@ -2135,6 +2148,68 @@ static void setup_windows_environment(void)
 		setenv("TERM", "cygwin", 1);
 }
 
+int handle_long_path(wchar_t *path, int len, int max_path, int expand)
+{
+	int result;
+	wchar_t buf[MAX_LONG_PATH];
+
+	/*
+	 * we don't need special handling if path is relative to the current
+	 * directory, and current directory + path don't exceed the desired
+	 * max_path limit. This should cover > 99 % of cases with minimal
+	 * performance impact (git almost always uses relative paths).
+	 */
+	if ((len < 2 || (!is_dir_sep(path[0]) && path[1] != ':')) &&
+	    (current_directory_len + len < max_path))
+		return len;
+
+	/*
+	 * handle everything else:
+	 * - absolute paths: "C:\dir\file"
+	 * - absolute UNC paths: "\\server\share\dir\file"
+	 * - absolute paths on current drive: "\dir\file"
+	 * - relative paths on other drive: "X:file"
+	 * - prefixed paths: "\\?\...", "\\.\..."
+	 */
+
+	/* convert to absolute path using GetFullPathNameW */
+	result = GetFullPathNameW(path, MAX_LONG_PATH, buf, NULL);
+	if (!result) {
+		errno = err_win_to_posix(GetLastError());
+		return -1;
+	}
+
+	/*
+	 * return absolute path if it fits within max_path (even if
+	 * "cwd + path" doesn't due to '..' components)
+	 */
+	if (result < max_path) {
+		wcscpy(path, buf);
+		return result;
+	}
+
+	/* error out if we shouldn't expand the path or buf is too small */
+	if (!expand || result >= MAX_LONG_PATH - 6) {
+		errno = ENAMETOOLONG;
+		return -1;
+	}
+
+	/* prefix full path with "\\?\" or "\\?\UNC\" */
+	if (buf[0] == '\\') {
+		/* ...unless already prefixed */
+		if (buf[1] == '\\' && (buf[2] == '?' || buf[2] == '.'))
+			return len;
+
+		wcscpy(path, L"\\\\?\\UNC\\");
+		wcscpy(path + 8, buf + 2);
+		return result + 6;
+	} else {
+		wcscpy(path, L"\\\\?\\");
+		wcscpy(path + 4, buf);
+		return result + 4;
+	}
+}
+
 /*
  * Disable MSVCRT command line wildcard expansion (__getmainargs called from
  * mingw startup code, see init.c in mingw runtime).
@@ -2226,6 +2301,9 @@ void mingw_startup(void)
 
 	/* initialize Unicode console */
 	winansi_init();
+
+	/* init length of current directory for handle_long_path */
+	current_directory_len = GetCurrentDirectoryW(0, NULL);
 }
 
 int uname(struct utsname *buf)

compat/mingw.h

@@ -437,6 +437,42 @@ int mingw_offset_1st_component(const char *path);
 #include <inttypes.h>
 #endif
 
+/**
+ * Max length of long paths (exceeding MAX_PATH). The actual maximum supported
+ * by NTFS is 32,767 (* sizeof(wchar_t)), but we choose an arbitrary smaller
+ * value to limit required stack memory.
+ */
+#define MAX_LONG_PATH 4096
+
+/**
+ * Handles paths that would exceed the MAX_PATH limit of Windows Unicode APIs.
+ *
+ * With expand == false, the function checks for over-long paths and fails
+ * with ENAMETOOLONG. The path parameter is not modified, except if cwd + path
+ * exceeds max_path, but the resulting absolute path doesn't (e.g. due to
+ * eliminating '..' components). The path parameter must point to a buffer
+ * of max_path wide characters.
+ *
+ * With expand == true, an over-long path is automatically converted in place
+ * to an absolute path prefixed with '\\?\', and the new length is returned.
+ * The path parameter must point to a buffer of MAX_LONG_PATH wide characters.
+ *
+ * Parameters:
+ * path: path to check and / or convert
+ * len: size of path on input (number of wide chars without \0)
+ * max_path: max short path length to check (usually MAX_PATH = 260, but just
+ * 248 for CreateDirectoryW)
+ * expand: false to only check the length, true to expand the path to a
+ * '\\?\'-prefixed absolute path
+ *
+ * Return:
+ * length of the resulting path, or -1 on failure
+ *
+ * Errors:
+ * ENAMETOOLONG if path is too long
+ */
+int handle_long_path(wchar_t *path, int len, int max_path, int expand);
+
 /**
  * Converts UTF-8 encoded string to UTF-16LE.
  *
@@ -495,18 +531,49 @@ static inline int xutftowcs(wchar_t *wcs, const char *utf, size_t wcslen)
 }
 
 /**
- * Simplified file system specific variant of xutftowcsn, assumes output
- * buffer size is MAX_PATH wide chars and input string is \0-terminated,
- * fails with ENAMETOOLONG if input string is too long.
+ * Simplified file system specific wrapper of xutftowcsn and handle_long_path.
+ * Converts ERANGE to ENAMETOOLONG. If expand is true, wcs must be at least
+ * MAX_LONG_PATH wide chars (see handle_long_path).
  */
-static inline int xutftowcs_path(wchar_t *wcs, const char *utf)
+static inline int xutftowcs_path_ex(wchar_t *wcs, const char *utf,
+		size_t wcslen, int utflen, int max_path, int expand)
 {
-	int result = xutftowcsn(wcs, utf, MAX_PATH, -1);
+	int result = xutftowcsn(wcs, utf, wcslen, utflen);
 	if (result < 0 && errno == ERANGE)
 		errno = ENAMETOOLONG;
+	if (result >= 0)
+		result = handle_long_path(wcs, result, max_path, expand);
 	return result;
 }
 
+/**
+ * Simplified file system specific variant of xutftowcsn, assumes output
+ * buffer size is MAX_PATH wide chars and input string is \0-terminated,
+ * fails with ENAMETOOLONG if input string is too long. Typically used for
+ * Windows APIs that don't support long paths, e.g. SetCurrentDirectory,
+ * LoadLibrary, CreateProcess...
+ */
+static inline int xutftowcs_path(wchar_t *wcs, const char *utf)
+{
+	return xutftowcs_path_ex(wcs, utf, MAX_PATH, -1, MAX_PATH, 0);
+}
+
+/* need to re-declare that here as mingw.h is included before cache.h */
+extern int core_long_paths;
+
+/**
+ * Simplified file system specific variant of xutftowcsn for Windows APIs
+ * that support long paths via '\\?\'-prefix, assumes output buffer size is
+ * MAX_LONG_PATH wide chars, fails with ENAMETOOLONG if input string is too
+ * long. The 'core.longpaths' git-config option controls whether the path
+ * is only checked or expanded to a long path.
+ */
+static inline int xutftowcs_long_path(wchar_t *wcs, const char *utf)
+{
+	return xutftowcs_path_ex(wcs, utf, MAX_LONG_PATH, -1, MAX_PATH,
+			core_long_paths);
+}
+
 /**
  * Converts UTF-16LE encoded string to UTF-8.
  *