Merge pull request #4659 from libgit2/ethomson/submodule-0_27

Backport fixes for CVE 2018-11235

CHANGELOG.md

-v0.27 + 1
+v0.27.1
 ---------
 
-### Changes or improvements
+This is a security release fixing insufficient validation of submodule names
+(CVE-2018-11235, reported by Etienne Stalmans) and disallows `.gitmodules` files
+as symlinks.
 
-### API additions
+While submodule names come from the untrusted ".gitmodules" file, we blindly
+append the name to "$GIT_DIR/modules" to construct the final path of the
+submodule repository. In case the name contains e.g. "../", an adversary would
+be able to escape your repository and write data at arbitrary paths. In
+accordance with git, we now enforce some rules for submodule names which will
+cause libgit2 to ignore these malicious names.
 
-### API removals
+Adding a symlink as `.gitmodules` into the index from the workdir or checking
+out such files is not allowed as this can make a Git implementation write
+outside of the repository and bypass the `fsck` checks for CVE-2018-11235.
 
-### Breaking API changes
+libgit2 is not susceptible to CVE-2018-11233.
 
 v0.27
 ---------

include/git2/version.h

@@ -7,10 +7,10 @@
 #ifndef INCLUDE_git_version_h__
 #define INCLUDE_git_version_h__
 
-#define LIBGIT2_VERSION "0.27.0"
+#define LIBGIT2_VERSION "0.27.1"
 #define LIBGIT2_VER_MAJOR 0
 #define LIBGIT2_VER_MINOR 27
-#define LIBGIT2_VER_REVISION 0
+#define LIBGIT2_VER_REVISION 1
 #define LIBGIT2_VER_PATCH 0
 
 #define LIBGIT2_SOVERSION 27

src/checkout.c

@@ -1276,14 +1276,14 @@ static int checkout_verify_paths(
 	unsigned int flags = GIT_PATH_REJECT_WORKDIR_DEFAULTS;
 
 	if (action & CHECKOUT_ACTION__REMOVE) {
-		if (!git_path_isvalid(repo, delta->old_file.path, flags)) {
+		if (!git_path_isvalid(repo, delta->old_file.path, delta->old_file.mode, flags)) {
 			giterr_set(GITERR_CHECKOUT, "cannot remove invalid path '%s'", delta->old_file.path);
 			return -1;
 		}
 	}
 
 	if (action & ~CHECKOUT_ACTION__REMOVE) {
-		if (!git_path_isvalid(repo, delta->new_file.path, flags)) {
+		if (!git_path_isvalid(repo, delta->new_file.path, delta->new_file.mode, flags)) {
 			giterr_set(GITERR_CHECKOUT, "cannot checkout to invalid path '%s'", delta->new_file.path);
 			return -1;
 		}

src/index.c

@@ -884,11 +884,13 @@ static int index_entry_create(
 	git_index_entry **out,
 	git_repository *repo,
 	const char *path,
+	struct stat *st,
 	bool from_workdir)
 {
 	size_t pathlen = strlen(path), alloclen;
 	struct entry_internal *entry;
 	unsigned int path_valid_flags = GIT_PATH_REJECT_INDEX_DEFAULTS;
+	uint16_t mode = 0;
 
 	/* always reject placing `.git` in the index and directory traversal.
 	 * when requested, disallow platform-specific filenames and upgrade to
@@ -896,8 +898,10 @@ static int index_entry_create(
 	 */
 	if (from_workdir)
 		path_valid_flags |= GIT_PATH_REJECT_WORKDIR_DEFAULTS;
+	if (st)
+		mode = st->st_mode;
 
-	if (!git_path_isvalid(repo, path, path_valid_flags)) {
+	if (!git_path_isvalid(repo, path, mode, path_valid_flags)) {
 		giterr_set(GITERR_INDEX, "invalid path: '%s'", path);
 		return -1;
 	}
@@ -922,15 +926,35 @@ static int index_entry_init(
 {
 	int error = 0;
 	git_index_entry *entry = NULL;
+	git_buf path = GIT_BUF_INIT;
 	struct stat st;
 	git_oid oid;
+	git_repository *repo;
 
 	if (INDEX_OWNER(index) == NULL)
 		return create_index_error(-1,
 			"could not initialize index entry. "
 			"Index is not backed up by an existing repository.");
 
-	if (index_entry_create(&entry, INDEX_OWNER(index), rel_path, true) < 0)
+	/*
+	 * FIXME: this is duplicated with the work in
+	 * git_blob__create_from_paths. It should accept an optional stat
+	 * structure so we can pass in the one we have to do here.
+	 */
+	repo = INDEX_OWNER(index);
+	if (git_repository__ensure_not_bare(repo, "create blob from file") < 0)
+		return GIT_EBAREREPO;
+
+	if (git_buf_joinpath(&path, git_repository_workdir(repo), rel_path) < 0)
+		return -1;
+
+	error = git_path_lstat(path.ptr, &st);
+	git_buf_free(&path);
+
+	if (error < 0)
+		return error;
+
+	if (index_entry_create(&entry, INDEX_OWNER(index), rel_path, &st, true) < 0)
 		return -1;
 
 	/* write the blob to disk and get the oid and stat info */
@@ -1016,7 +1040,7 @@ static int index_entry_dup(
 	git_index *index,
 	const git_index_entry *src)
 {
-	if (index_entry_create(out, INDEX_OWNER(index), src->path, false) < 0)
+	if (index_entry_create(out, INDEX_OWNER(index), src->path, NULL, false) < 0)
 		return -1;
 
 	index_entry_cpy(*out, src);
@@ -1038,7 +1062,7 @@ static int index_entry_dup_nocache(
 	git_index *index,
 	const git_index_entry *src)
 {
-	if (index_entry_create(out, INDEX_OWNER(index), src->path, false) < 0)
+	if (index_entry_create(out, INDEX_OWNER(index), src->path, NULL, false) < 0)
 		return -1;
 
 	index_entry_cpy_nocache(*out, src);
@@ -1461,9 +1485,6 @@ static int add_repo_as_submodule(git_index_entry **out, git_index *index, const
 	struct stat st;
 	int error;
 
-	if (index_entry_create(&entry, INDEX_OWNER(index), path, true) < 0)
-		return -1;
-
 	if ((error = git_buf_joinpath(&abspath, git_repository_workdir(repo), path)) < 0)
 		return error;
 
@@ -1472,6 +1493,9 @@ static int add_repo_as_submodule(git_index_entry **out, git_index *index, const
 		return -1;
 	}
 
+	if (index_entry_create(&entry, INDEX_OWNER(index), path, &st, true) < 0)
+		return -1;
+
 	git_index_entry__init_from_stat(entry, &st, !index->distrust_filemode);
 
 	if ((error = git_repository_open(&sub, abspath.ptr)) < 0)
@@ -2965,7 +2989,7 @@ static int read_tree_cb(
 	if (git_buf_joinpath(&path, root, tentry->filename) < 0)
 		return -1;
 
-	if (index_entry_create(&entry, INDEX_OWNER(data->index), path.ptr, false) < 0)
+	if (index_entry_create(&entry, INDEX_OWNER(data->index), path.ptr, NULL, false) < 0)
 		return -1;
 
 	entry->mode = tentry->attr;

src/path.c

@@ -1561,18 +1561,31 @@ static int32_t next_hfs_char(const char **in, size_t *len)
 	return 0; /* NULL byte -- end of string */
 }
 
-static bool verify_dotgit_hfs(const char *path, size_t len)
+static bool verify_dotgit_hfs_generic(const char *path, size_t len, const char *needle, size_t needle_len)
 {
-	if (next_hfs_char(&path, &len) != '.' ||
-		next_hfs_char(&path, &len) != 'g' ||
-		next_hfs_char(&path, &len) != 'i' ||
-		next_hfs_char(&path, &len) != 't' ||
-		next_hfs_char(&path, &len) != 0)
+	size_t i;
+	char c;
+
+	if (next_hfs_char(&path, &len) != '.')
+		return true;
+
+	for (i = 0; i < needle_len; i++) {
+		c = next_hfs_char(&path, &len);
+		if (c != needle[i])
+			return true;
+	}
+
+	if (next_hfs_char(&path, &len) != '\0')
 		return true;
 
 	return false;
 }
 
+static bool verify_dotgit_hfs(const char *path, size_t len)
+{
+	return verify_dotgit_hfs_generic(path, len, "git", CONST_STRLEN("git"));
+}
+
 GIT_INLINE(bool) verify_dotgit_ntfs(git_repository *repo, const char *path, size_t len)
 {
 	git_buf *reserved = git_repository__reserved_names_win32;
@@ -1608,6 +1621,57 @@ GIT_INLINE(bool) verify_dotgit_ntfs(git_repository *repo, const char *path, size
 	return false;
 }
 
+GIT_INLINE(bool) only_spaces_and_dots(const char *path)
+{
+	const char *c = path;
+
+	for (;; c++) {
+		if (*c == '\0')
+			return true;
+		if (*c != ' ' && *c != '.')
+			return false;
+	}
+
+	return true;
+}
+
+GIT_INLINE(bool) verify_dotgit_ntfs_generic(const char *name, size_t len, const char *dotgit_name, size_t dotgit_len, const char *shortname_pfix)
+{
+	int i, saw_tilde;
+
+	if (name[0] == '.' && len >= dotgit_len &&
+	    !strncasecmp(name + 1, dotgit_name, dotgit_len)) {
+		return !only_spaces_and_dots(name + dotgit_len + 1);
+	}
+
+	/* Detect the basic NTFS shortname with the first six chars */
+	if (!strncasecmp(name, dotgit_name, 6) && name[6] == '~' &&
+	    name[7] >= '1' && name[7] <= '4')
+		return !only_spaces_and_dots(name + 8);
+
+	/* Catch fallback names */
+	for (i = 0, saw_tilde = 0; i < 8; i++) {
+		if (name[i] == '\0') {
+			return true;
+		} else if (saw_tilde) {
+			if (name[i] < '0' || name[i] > '9')
+				return true;
+		} else if (name[i] == '~') {
+			if (name[i+1] < '1' || name[i+1]  > '9')
+				return true;
+			saw_tilde = 1;
+		} else if (i >= 6) {
+			return true;
+		} else if (name[i] < 0) {
+			return true;
+		} else if (git__tolower(name[i]) != shortname_pfix[i]) {
+			return true;
+		}
+	}
+
+	return !only_spaces_and_dots(name + i);
+}
+
 GIT_INLINE(bool) verify_char(unsigned char c, unsigned int flags)
 {
 	if ((flags & GIT_PATH_REJECT_BACKSLASH) && c == '\\')
@@ -1636,6 +1700,24 @@ GIT_INLINE(bool) verify_char(unsigned char c, unsigned int flags)
 }
 
 /*
+ * Return the length of the common prefix between str and prefix, comparing them
+ * case-insensitively (must be ASCII to match).
+ */
+GIT_INLINE(size_t) common_prefix_icase(const char *str, size_t len, const char *prefix)
+{
+	size_t count = 0;
+
+	while (len >0 && tolower(*str) == tolower(*prefix)) {
+		count++;
+		str++;
+		prefix++;
+		len--;
+	}
+
+	return count;
+}
+
+/*
  * We fundamentally don't like some paths when dealing with user-inputted
  * strings (in checkout or ref names): we don't want dot or dot-dot
  * anywhere, we want to avoid writing weird paths on Windows that can't
@@ -1648,6 +1730,7 @@ static bool verify_component(
 	git_repository *repo,
 	const char *component,
 	size_t len,
+	uint16_t mode,
 	unsigned int flags)
 {
 	if (len == 0)
@@ -1680,26 +1763,38 @@ static bool verify_component(
 			return false;
 	}
 
-	if (flags & GIT_PATH_REJECT_DOT_GIT_HFS &&
-		!verify_dotgit_hfs(component, len))
-		return false;
+	if (flags & GIT_PATH_REJECT_DOT_GIT_HFS) {
+		if (!verify_dotgit_hfs(component, len))
+			return false;
+		if (S_ISLNK(mode) && git_path_is_hfs_dotgit_modules(component, len))
+			return false;
+	}
 
-	if (flags & GIT_PATH_REJECT_DOT_GIT_NTFS &&
-		!verify_dotgit_ntfs(repo, component, len))
-		return false;
+	if (flags & GIT_PATH_REJECT_DOT_GIT_NTFS) {
+		if (!verify_dotgit_ntfs(repo, component, len))
+			return false;
+		if (S_ISLNK(mode) && git_path_is_ntfs_dotgit_modules(component, len))
+			return false;
+	}
 
 	/* don't bother rerunning the `.git` test if we ran the HFS or NTFS
 	 * specific tests, they would have already rejected `.git`.
 	 */
 	if ((flags & GIT_PATH_REJECT_DOT_GIT_HFS) == 0 &&
-		(flags & GIT_PATH_REJECT_DOT_GIT_NTFS) == 0 &&
-		(flags & GIT_PATH_REJECT_DOT_GIT_LITERAL) &&
-		len == 4 &&
-		component[0] == '.' &&
-		(component[1] == 'g' || component[1] == 'G') &&
-		(component[2] == 'i' || component[2] == 'I') &&
-		(component[3] == 't' || component[3] == 'T'))
-		return false;
+	    (flags & GIT_PATH_REJECT_DOT_GIT_NTFS) == 0 &&
+	    (flags & GIT_PATH_REJECT_DOT_GIT_LITERAL)) {
+		if (len >= 4 &&
+		    component[0] == '.' &&
+		    (component[1] == 'g' || component[1] == 'G') &&
+		    (component[2] == 'i' || component[2] == 'I') &&
+		    (component[3] == 't' || component[3] == 'T')) {
+			if (len == 4)
+				return false;
+
+			if (S_ISLNK(mode) && common_prefix_icase(component, len, ".gitmodules") == len)
+				return false;
+		}
+	    }
 
 	return true;
 }
@@ -1737,6 +1832,7 @@ GIT_INLINE(unsigned int) dotgit_flags(
 bool git_path_isvalid(
 	git_repository *repo,
 	const char *path,
+	uint16_t mode,
 	unsigned int flags)
 {
 	const char *start, *c;
@@ -1750,14 +1846,14 @@ bool git_path_isvalid(
 			return false;
 
 		if (*c == '/') {
-			if (!verify_component(repo, start, (c - start), flags))
+			if (!verify_component(repo, start, (c - start), mode, flags))
 				return false;
 
 			start = c+1;
 		}
 	}
 
-	return verify_component(repo, start, (c - start), flags);
+	return verify_component(repo, start, (c - start), mode, flags);
 }
 
 int git_path_normalize_slashes(git_buf *out, const char *path)
@@ -1775,3 +1871,65 @@ int git_path_normalize_slashes(git_buf *out, const char *path)
 
 	return 0;
 }
+
+static int verify_dotgit_generic(const char *name, size_t len, const char *dotgit_name, size_t dotgit_len, const char *shortname_pfix)
+{
+	if (!verify_dotgit_ntfs_generic(name, len, dotgit_name, dotgit_len, shortname_pfix))
+		return false;
+
+	return verify_dotgit_hfs_generic(name, len, dotgit_name, dotgit_len);
+}
+
+int git_path_is_ntfs_dotgit_modules(const char *name, size_t len)
+{
+	return !verify_dotgit_ntfs_generic(name, len, "gitmodules", CONST_STRLEN("gitmodules"), "gi7eba");
+}
+
+int git_path_is_hfs_dotgit_modules(const char *name, size_t len)
+{
+	return !verify_dotgit_hfs_generic(name, len, "gitmodules", CONST_STRLEN("gitmodules"));
+}
+
+int git_path_is_dotgit_modules(const char *name, size_t len)
+{
+	if (git_path_is_hfs_dotgit_modules(name, len))
+		return 1;
+
+	return git_path_is_ntfs_dotgit_modules(name, len);
+}
+
+int git_path_is_ntfs_dotgit_ignore(const char *name, size_t len)
+{
+	return !verify_dotgit_ntfs_generic(name, len, "gitignore", CONST_STRLEN("gitignore"), "gi250a");
+}
+
+int git_path_is_hfs_dotgit_ignore(const char *name, size_t len)
+{
+	return !verify_dotgit_hfs_generic(name, len, "gitignore", CONST_STRLEN("gitignore"));
+}
+
+int git_path_is_dotgit_ignore(const char *name, size_t len)
+{
+	if (git_path_is_hfs_dotgit_ignore(name, len))
+		return 1;
+
+	return git_path_is_ntfs_dotgit_ignore(name, len);
+}
+
+int git_path_is_hfs_dotgit_attributes(const char *name, size_t len)
+{
+	return !verify_dotgit_hfs_generic(name, len, "gitattributes", CONST_STRLEN("gitattributes"));
+}
+
+int git_path_is_ntfs_dotgit_attributes(const char *name, size_t len)
+{
+	return !verify_dotgit_ntfs_generic(name, len, "gitattributes", CONST_STRLEN("gitattributes"), "gi7d29");
+}
+
+int git_path_is_dotgit_attributes(const char *name, size_t len)
+{
+	if (git_path_is_hfs_dotgit_attributes(name, len))
+		return 1;
+
+	return git_path_is_ntfs_dotgit_attributes(name, len);
+}

src/path.h

@@ -637,6 +637,7 @@ extern int git_path_from_url_or_path(git_buf *local_path_out, const char *url_or
 extern bool git_path_isvalid(
 	git_repository *repo,
 	const char *path,
+	uint16_t mode,
 	unsigned int flags);
 
 /**
@@ -644,4 +645,76 @@ extern bool git_path_isvalid(
  */
 int git_path_normalize_slashes(git_buf *out, const char *path);
 
+/**
+ * Check whether a path component corresponds to a .gitmodules file
+ *
+ * @param name the path component to check
+ * @param len the length of `name`
+ */
+extern int git_path_is_dotgit_modules(const char *name, size_t len);
+
+/**
+ * Check whether a path component corresponds to a .gitmodules file in NTFS
+ *
+ * @param name the path component to check
+ * @param len the length of `name`
+ */
+extern int git_path_is_ntfs_dotgit_modules(const char *name, size_t len);
+
+/**
+ * Check whether a path component corresponds to a .gitmodules file in HFS+
+ *
+ * @param name the path component to check
+ * @param len the length of `name`
+ */
+extern int git_path_is_hfs_dotgit_modules(const char *name, size_t len);
+
+/**
+ * Check whether a path component corresponds to a .gitignore file
+ *
+ * @param name the path component to check
+ * @param len the length of `name`
+ */
+extern int git_path_is_dotgit_ignore(const char *name, size_t len);
+
+/**
+ * Check whether a path component corresponds to a .gitignore file in NTFS
+ *
+ * @param name the path component to check
+ * @param len the length of `name`
+ */
+extern int git_path_is_ntfs_dotgit_ignore(const char *name, size_t len);
+
+/**
+ * Check whether a path component corresponds to a .gitignore file in HFS+
+ *
+ * @param name the path component to check
+ * @param len the length of `name`
+ */
+extern int git_path_is_hfs_dotgit_ignore(const char *name, size_t len);
+
+/**
+ * Check whether a path component corresponds to a .gitignore file
+ *
+ * @param name the path component to check
+ * @param len the length of `name`
+ */
+extern int git_path_is_dotgit_attributes(const char *name, size_t len);
+
+/**
+ * Check whether a path component corresponds to a .gitattributes file in NTFS
+ *
+ * @param name the path component to check
+ * @param len the length of `name`
+ */
+extern int git_path_is_ntfs_dotgit_attributes(const char *name, size_t len);
+
+/**
+ * Check whether a path component corresponds to a .gitattributes file in HFS+
+ *
+ * @param name the path component to check
+ * @param len the length of `name`
+ */
+extern int git_path_is_hfs_dotgit_attributes(const char *name, size_t len);
+
 #endif

src/refdb_fs.c

@@ -744,7 +744,7 @@ static int loose_lock(git_filebuf *file, refdb_fs_backend *backend, const char *
 
 	assert(file && backend && name);
 
-	if (!git_path_isvalid(backend->repo, name, GIT_PATH_REJECT_FILESYSTEM_DEFAULTS)) {
+	if (!git_path_isvalid(backend->repo, name, 0, GIT_PATH_REJECT_FILESYSTEM_DEFAULTS)) {
 		giterr_set(GITERR_INVALID, "invalid reference name '%s'", name);
 		return GIT_EINVALIDSPEC;
 	}
@@ -1740,7 +1740,7 @@ static int lock_reflog(git_filebuf *file, refdb_fs_backend *backend, const char
 
 	repo = backend->repo;
 
-	if (!git_path_isvalid(backend->repo, refname, GIT_PATH_REJECT_FILESYSTEM_DEFAULTS)) {
+	if (!git_path_isvalid(backend->repo, refname, 0, GIT_PATH_REJECT_FILESYSTEM_DEFAULTS)) {
 		giterr_set(GITERR_INVALID, "invalid reference name '%s'", refname);
 		return GIT_EINVALIDSPEC;
 	}

src/submodule.c

@@ -169,13 +169,13 @@ static void free_submodule_names(git_strmap *names)
  * TODO: for some use-cases, this might need case-folding on a
  * case-insensitive filesystem
  */
-static int load_submodule_names(git_strmap *out, git_config *cfg)
+static int load_submodule_names(git_strmap *out, git_repository *repo, git_config *cfg)
 {
 	const char *key = "submodule\\..*\\.path";
 	git_config_iterator *iter;
 	git_config_entry *entry;
 	git_buf buf = GIT_BUF_INIT;
-	int rval;
+	int rval, isvalid;
 	int error = 0;
 
 	if ((error = git_config_iterator_glob_new(&iter, cfg, key)) < 0)
@@ -186,16 +186,29 @@ static int load_submodule_names(git_strmap *out, git_config *cfg)
 		fdot = strchr(entry->name, '.');
 		ldot = strrchr(entry->name, '.');
 
+		git_buf_clear(&buf);
 		git_buf_put(&buf, fdot + 1, ldot - fdot - 1);
+		isvalid = git_submodule_name_is_valid(repo, buf.ptr, 0);
+		if (isvalid < 0) {
+			error = isvalid;
+			goto out;
+		}
+		if (!isvalid)
+			continue;
+
 		git_strmap_insert(out, entry->value, git_buf_detach(&buf), &rval);
 		if (rval < 0) {
 			giterr_set(GITERR_NOMEMORY, "error inserting submodule into hash table");
 			return -1;
 		}
 	}
+	if (error == GIT_ITEROVER)
+		error = 0;
 
+out:
+	git_buf_free(&buf);
 	git_config_iterator_free(iter);
-	return 0;
+	return error;
 }
 
 int git_submodule_lookup(
@@ -314,6 +327,28 @@ int git_submodule_lookup(
 	return 0;
 }
 
+int git_submodule_name_is_valid(git_repository *repo, const char *name, int flags)
+{
+	git_buf buf = GIT_BUF_INIT;
+	int error, isvalid;
+
+	if (flags == 0)
+		flags = GIT_PATH_REJECT_FILESYSTEM_DEFAULTS;
+
+	/* Avoid allocating a new string if we can avoid it */
+	if (strchr(name, '\\') != NULL) {
+		if ((error = git_path_normalize_slashes(&buf, name)) < 0)
+			return error;
+	} else {
+		git_buf_attach_notowned(&buf, name, strlen(name));
+	}
+
+	isvalid =  git_path_isvalid(repo, buf.ptr, 0, flags);
+	git_buf_free(&buf);
+
+	return isvalid;
+}
+
 static void submodule_free_dup(void *sm)
 {
 	git_submodule_free(sm);
@@ -359,7 +394,7 @@ static int submodules_from_index(git_strmap *map, git_index *idx, git_config *cf
 	git_strmap *names = 0;
 
 	git_strmap_alloc(&names);
-	if ((error = load_submodule_names(names, cfg)))
+	if ((error = load_submodule_names(names, git_index_owner(idx), cfg)))
 		goto done;
 
 	if ((error = git_iterator_for_index(&i, git_index_owner(idx), idx, NULL)) < 0)
@@ -411,7 +446,7 @@ static int submodules_from_head(git_strmap *map, git_tree *head, git_config *cfg
 	const git_index_entry *entry;
 	git_strmap *names = 0;
 	git_strmap_alloc(&names);
-	if ((error = load_submodule_names(names, cfg)))
+	if ((error = load_submodule_names(names, git_tree_owner(head), cfg)))
 		goto done;
 
 	if ((error = git_iterator_for_tree(&i, head, NULL)) < 0)
@@ -1502,13 +1537,19 @@ static int submodule_update_head(git_submodule *submodule)
 
 int git_submodule_reload(git_submodule *sm, int force)
 {
-	int error = 0;
+	int error = 0, isvalid;
 	git_config *mods;
 
 	GIT_UNUSED(force);
 
 	assert(sm);
 
+	isvalid = git_submodule_name_is_valid(sm->repo, sm->name, 0);
+	if (isvalid <= 0) {
+		/* This should come with a warning, but we've no API for that */
+		return isvalid;
+	}
+
 	if (!git_repository_is_bare(sm->repo)) {
 		/* refresh config data */
 		mods = gitmodules_snapshot(sm->repo);
@@ -1849,7 +1890,7 @@ static int submodule_load_each(const git_config_entry *entry, void *payload)
 	git_strmap *map = data->map;
 	git_buf name = GIT_BUF_INIT;
 	git_submodule *sm;
-	int error;
+	int error, isvalid;
 
 	if (git__prefixcmp(entry->name, "submodule.") != 0)
 		return 0;
@@ -1865,6 +1906,12 @@ static int submodule_load_each(const git_config_entry *entry, void *payload)
 	if ((error = git_buf_set(&name, namestart, property - namestart -1)) < 0)
 		return error;
 
+	isvalid = git_submodule_name_is_valid(data->repo, name.ptr, 0);
+	if (isvalid <= 0) {
+		error = isvalid;
+		goto done;
+	}
+
 	/*
 	 * Now that we have the submodule's name, we can use that to
 	 * figure out whether it's in the map. If it's not, we create

src/submodule.h

@@ -148,4 +148,17 @@ extern int git_submodule_parse_update(
 extern int git_submodule__map(
 	git_repository *repo,
 	git_strmap *map);
+
+/**
+ * Check whether a submodule's name is valid.
+ *
+ * Check the path against the path validity rules, either the filesystem
+ * defaults (like checkout does) or whichever you want to compare against.
+ *
+ * @param repo the repository which contains the submodule
+ * @param name the name to check
+ * @param flags the `GIT_PATH` flags to use for the check (0 to use filesystem defaults)
+ */
+extern int git_submodule_name_is_valid(git_repository *repo, const char *name, int flags);
+
 #endif

src/tree.c

@@ -54,7 +54,7 @@ GIT_INLINE(git_filemode_t) normalize_filemode(git_filemode_t filemode)
 static int valid_entry_name(git_repository *repo, const char *filename)
 {
 	return *filename != '\0' &&
-		git_path_isvalid(repo, filename,
+		git_path_isvalid(repo, filename, 0,
 		GIT_PATH_REJECT_TRAVERSAL | GIT_PATH_REJECT_DOT_GIT | GIT_PATH_REJECT_SLASH);
 }
 

tests/checkout/nasty.c

@@ -364,3 +364,15 @@ void test_checkout_nasty__symlink3(void)
 	test_checkout_passes("refs/heads/symlink3", ".git/foobar");
 }
 
+void test_checkout_nasty__gitmodules_symlink(void)
+{
+	cl_repo_set_bool(repo, "core.protectHFS", true);
+	test_checkout_fails("refs/heads/gitmodules-symlink", ".gitmodules");
+	cl_repo_set_bool(repo, "core.protectHFS", false);
+
+	cl_repo_set_bool(repo, "core.protectNTFS", true);
+	test_checkout_fails("refs/heads/gitmodules-symlink", ".gitmodules");
+	cl_repo_set_bool(repo, "core.protectNTFS", false);
+
+	test_checkout_fails("refs/heads/gitmodules-symlink", ".gitmodules");
+}

tests/path/dotgit.c

+#include "clar_libgit2.h"
+#include "path.h"
+
+static char *gitmodules_altnames[] = {
+	".gitmodules",
+
+	/*
+	 * Equivalent to the ".git\u200cmodules" string from git but hard-coded
+	 * as a UTF-8 sequence
+	 */
+	".git\xe2\x80\x8cmodules",
+
+	".Gitmodules",
+	".gitmoduleS",
+
+	".gitmodules ",
+	".gitmodules.",
+	".gitmodules  ",
+	".gitmodules. ",
+	".gitmodules .",
+	".gitmodules..",
+	".gitmodules   ",
+	".gitmodules.  ",
+	".gitmodules . ",
+	".gitmodules  .",
+
+	".Gitmodules ",
+	".Gitmodules.",
+	".Gitmodules  ",
+	".Gitmodules. ",
+	".Gitmodules .",
+	".Gitmodules..",
+	".Gitmodules   ",
+	".Gitmodules.  ",
+	".Gitmodules . ",
+	".Gitmodules  .",
+
+	"GITMOD~1",
+	"gitmod~1",
+	"GITMOD~2",
+	"gitmod~3",
+	"GITMOD~4",
+
+	"GITMOD~1 ",
+	"gitmod~2.",
+	"GITMOD~3  ",
+	"gitmod~4. ",
+	"GITMOD~1 .",
+	"gitmod~2   ",
+	"GITMOD~3.  ",
+	"gitmod~4 . ",
+
+	"GI7EBA~1",
+	"gi7eba~9",
+
+	"GI7EB~10",
+	"GI7EB~11",
+	"GI7EB~99",
+	"GI7EB~10",
+	"GI7E~100",
+	"GI7E~101",
+	"GI7E~999",
+	"~1000000",
+	"~9999999",
+};
+
+static char *gitmodules_not_altnames[] = {
+	".gitmodules x",
+	".gitmodules .x",
+
+	" .gitmodules",
+
+	"..gitmodules",
+
+	"gitmodules",
+
+	".gitmodule",
+
+	".gitmodules x ",
+	".gitmodules .x",
+
+	"GI7EBA~",
+	"GI7EBA~0",
+	"GI7EBA~~1",
+	"GI7EBA~X",
+	"Gx7EBA~1",
+	"GI7EBX~1",
+
+	"GI7EB~1",
+	"GI7EB~01",
+	"GI7EB~1",
+};
+
+void test_path_dotgit__dotgit_modules(void)
+{
+	size_t i;
+	cl_assert_equal_i(1, git_path_is_dotgit_modules(".gitmodules", strlen(".gitmodules")));
+	cl_assert_equal_i(1, git_path_is_dotgit_modules(".git\xe2\x80\x8cmodules", strlen(".git\xe2\x80\x8cmodules")));
+
+	for (i = 0; i < ARRAY_SIZE(gitmodules_altnames); i++) {
+		const char *name = gitmodules_altnames[i];
+		if (!git_path_is_dotgit_modules(name, strlen(name)))
+			cl_fail(name);
+	}
+
+	for (i = 0; i < ARRAY_SIZE(gitmodules_not_altnames); i++) {
+		const char *name = gitmodules_not_altnames[i];
+		if (git_path_is_dotgit_modules(name, strlen(name)))
+			cl_fail(name);
+	}
+
+}
+
+void test_path_dotgit__dotgit_modules_symlink(void)
+{
+	cl_assert_equal_b(true, git_path_isvalid(NULL, ".gitmodules", 0, GIT_PATH_REJECT_DOT_GIT_HFS|GIT_PATH_REJECT_DOT_GIT_NTFS));
+	cl_assert_equal_b(false, git_path_isvalid(NULL, ".gitmodules", S_IFLNK, GIT_PATH_REJECT_DOT_GIT_HFS));
+	cl_assert_equal_b(false, git_path_isvalid(NULL, ".gitmodules", S_IFLNK, GIT_PATH_REJECT_DOT_GIT_NTFS));
+}

tests/resources/nasty/.gitted/refs/heads/gitmodules-symlink

+e30b60b120761f44ebd0f0a7b0e9445ce8e11d68

tests/submodule/escape.c

+#include "clar_libgit2.h"
+#include "posix.h"
+#include "path.h"
+#include "submodule_helpers.h"
+#include "fileops.h"
+#include "repository.h"
+
+static git_repository *g_repo = NULL;
+
+void test_submodule_escape__cleanup(void)
+{
+	cl_git_sandbox_cleanup();
+}
+
+#define EVIL_SM_NAME "../../modules/evil"
+#define EVIL_SM_NAME_WINDOWS "..\\\\..\\\\modules\\\\evil"
+#define EVIL_SM_NAME_WINDOWS_UNESC "..\\..\\modules\\evil"
+
+static int find_evil(git_submodule *sm, const char *name, void *payload)
+{
+	int *foundit = (int *) payload;
+
+	GIT_UNUSED(sm);
+
+	if (!git__strcmp(EVIL_SM_NAME, name) ||
+	    !git__strcmp(EVIL_SM_NAME_WINDOWS_UNESC, name))
+		*foundit = true;
+
+	return 0;
+}
+
+void test_submodule_escape__from_gitdir(void)
+{
+	int foundit;
+	git_submodule *sm;
+	git_buf buf = GIT_BUF_INIT;
+	unsigned int sm_location;
+
+	g_repo = setup_fixture_submodule_simple();
+
+	cl_git_pass(git_buf_joinpath(&buf, git_repository_workdir(g_repo), ".gitmodules"));
+	cl_git_rewritefile(buf.ptr,
+			   "[submodule \"" EVIL_SM_NAME "\"]\n"
+			   "    path = testrepo\n"
+			   "    url = ../testrepo.git\n");
+	git_buf_free(&buf);
+
+	/* Find it all the different ways we know about it */
+	foundit = 0;
+	cl_git_pass(git_submodule_foreach(g_repo, find_evil, &foundit));
+	cl_assert_equal_i(0, foundit);
+	cl_git_fail_with(GIT_ENOTFOUND, git_submodule_lookup(&sm, g_repo, EVIL_SM_NAME));
+	/*
+	 * We do know about this as it's in the index and HEAD, but the data is
+	 * incomplete as there is no configured data for it (we pretend it
+	 * doesn't exist). This leaves us with an odd situation but it's
+	 * consistent with what we would do if we did add a submodule with no
+	 * configuration.
+	 */
+	cl_git_pass(git_submodule_lookup(&sm, g_repo, "testrepo"));
+	cl_git_pass(git_submodule_location(&sm_location, sm));
+	cl_assert_equal_i(GIT_SUBMODULE_STATUS_IN_INDEX | GIT_SUBMODULE_STATUS_IN_HEAD, sm_location);
+	git_submodule_free(sm);
+}
+
+void test_submodule_escape__from_gitdir_windows(void)
+{
+	int foundit;
+	git_submodule *sm;
+	git_buf buf = GIT_BUF_INIT;
+	unsigned int sm_location;
+
+	g_repo = setup_fixture_submodule_simple();
+
+	cl_git_pass(git_buf_joinpath(&buf, git_repository_workdir(g_repo), ".gitmodules"));
+	cl_git_rewritefile(buf.ptr,
+			   "[submodule \"" EVIL_SM_NAME_WINDOWS "\"]\n"
+			   "    path = testrepo\n"
+			   "    url = ../testrepo.git\n");
+	git_buf_free(&buf);
+
+	/* Find it all the different ways we know about it */
+	foundit = 0;
+	cl_git_pass(git_submodule_foreach(g_repo, find_evil, &foundit));
+	cl_assert_equal_i(0, foundit);
+	cl_git_fail_with(GIT_ENOTFOUND, git_submodule_lookup(&sm, g_repo, EVIL_SM_NAME_WINDOWS_UNESC));
+	/*
+	 * We do know about this as it's in the index and HEAD, but the data is
+	 * incomplete as there is no configured data for it (we pretend it
+	 * doesn't exist). This leaves us with an odd situation but it's
+	 * consistent with what we would do if we did add a submodule with no
+	 * configuration.
+	 */
+	cl_git_pass(git_submodule_lookup(&sm, g_repo, "testrepo"));
+	cl_git_pass(git_submodule_location(&sm_location, sm));
+	cl_assert_equal_i(GIT_SUBMODULE_STATUS_IN_INDEX | GIT_SUBMODULE_STATUS_IN_HEAD, sm_location);
+	git_submodule_free(sm);
+}