Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Sat, 28 Aug 2021 01:22:41 +0200
From: Ismael Luceno <ismael@...ev.co.uk>
To: musl@...ts.openwall.com
Cc: Rich Felker <dalias@...c.org>,
	Ismael Luceno <ismael@...ev.co.uk>
Subject: [PATCH v6 3/3] glob: implement GLOB_ALTDIRFUNC et al

Signed-off-by: Ismael Luceno <ismael@...ev.co.uk>
---

Notes:
    Changes since v5:
    
    - Copy glob_t, which should solve UB & modifying it on non-successful
      return
    - Removed glob wrapper for POSIX sources
    - Keep static check as a static function (doesn't emit any code)
    
    Changes since v4:
    
    - Rebased on top of do_glob context patch
    - Added alternate definitions of glob_t for BSD/GNU vs POSIX
    - Added static check to ensure the memory layout of both glob_t
      versions matches plus a wrapper for the POSIX glob function
    
    Changes since v3:
    
    - Wrap pointers used by GLOB_ALTDIRFUNC on the header to protect from
      namespace contamination. GNU and BSD code shouldn't be using
      "dirent", so it should be safe.
    
    Changes since v2:
    
    - Rebased
    
    Changes since v1:
    
    - Avoid overwriting the function pointers in glob_t
    - Wrap {open,read}dir too

 include/glob.h   | 16 +++++++++-
 src/regex/glob.c | 83 ++++++++++++++++++++++++++++++++++++++++++------
 2 files changed, 89 insertions(+), 10 deletions(-)

diff --git a/include/glob.h b/include/glob.h
index 0ff70bdfeef2..fa9d04146826 100644
--- a/include/glob.h
+++ b/include/glob.h
@@ -11,12 +11,25 @@ extern "C" {
 
 #include <bits/alltypes.h>
 
+#if defined(_BSD_SOURCE) || defined(_GNU_SOURCE)
+#define __dirent dirent
+#define __stat stat
+#endif
+
+struct __dirent;
+struct __stat;
+
 typedef struct {
 	size_t gl_pathc;
 	char **gl_pathv;
 	size_t gl_offs;
 	int __dummy1;
-	void *__dummy2[5];
+
+	void (*gl_closedir)(void *);
+	struct __dirent *(*gl_readdir)(void *);
+	void *(*gl_opendir)(const char *);
+	int (*gl_lstat)(const char *__restrict, struct __stat *__restrict);
+	int (*gl_stat)(const char *__restrict, struct __stat *__restrict);
 } glob_t;
 
 int  glob(const char *__restrict, int, int (*)(const char *, int), glob_t *__restrict);
@@ -31,6 +44,7 @@ void globfree(glob_t *);
 #define GLOB_NOESCAPE 0x40
 #define	GLOB_PERIOD   0x80
 
+#define GLOB_ALTDIRFUNC  0x0200
 #define GLOB_NOMAGIC     0x0800
 #define GLOB_TILDE       0x1000
 #define GLOB_TILDE_CHECK 0x4000
diff --git a/src/regex/glob.c b/src/regex/glob.c
index 9491eaeef266..d73a19856afc 100644
--- a/src/regex/glob.c
+++ b/src/regex/glob.c
@@ -36,6 +36,13 @@ struct glob_ctx {
 	struct match **tail;
 	int flags;
 	int (*errfunc)(const char *path, int err);
+
+	/* for GLOB_ALTDIRFUNC */
+	void (*closedir)(void *);
+	struct dirent *(*readdir)(void *);
+	void *(*opendir)(const char *);
+	int (*lstat)(const char *restrict, struct stat *restrict);
+	int (*stat)(const char *restrict, struct stat *restrict);
 };
 
 static int do_glob(char *buf, size_t pos, int type, char *pat, const struct glob_ctx *restrict ctx)
@@ -106,11 +113,11 @@ static int do_glob(char *buf, size_t pos, int type, char *pat, const struct glob
 		 * or if that fails, use lstat for determining existence to
 		 * avoid false negatives in the case of broken symlinks. */
 		struct stat st;
-		if ((ctx->flags & GLOB_MARK) && (!type||type==DT_LNK) && !stat(buf, &st)) {
+		if ((ctx->flags & GLOB_MARK) && (!type||type==DT_LNK) && !ctx->stat(buf, &st)) {
 			if (S_ISDIR(st.st_mode)) type = DT_DIR;
 			else type = DT_REG;
 		}
-		if (!type && lstat(buf, &st)) {
+		if (!type && ctx->lstat(buf, &st)) {
 			if (errno!=ENOENT && (ctx->errfunc(buf, errno) || (ctx->flags & GLOB_ERR)))
 				return GLOB_ABORTED;
 			return 0;
@@ -130,7 +137,7 @@ static int do_glob(char *buf, size_t pos, int type, char *pat, const struct glob
 			saved_sep = '\\';
 		}
 	}
-	DIR *dir = opendir(pos ? buf : ".");
+	DIR *dir = ctx->opendir(pos ? buf : ".");
 	if (!dir) {
 		if (ctx->errfunc(buf, errno) || (ctx->flags & GLOB_ERR))
 			return GLOB_ABORTED;
@@ -138,7 +145,7 @@ static int do_glob(char *buf, size_t pos, int type, char *pat, const struct glob
 	}
 	int old_errno = errno;
 	struct dirent *de;
-	while (errno=0, de=readdir(dir)) {
+	while (errno=0, de=ctx->readdir(dir)) {
 		/* Quickly skip non-directories when there's pattern left. */
 		if (p2 && de->d_type && de->d_type!=DT_DIR && de->d_type!=DT_LNK)
 			continue;
@@ -165,13 +172,13 @@ static int do_glob(char *buf, size_t pos, int type, char *pat, const struct glob
 		if (p2) *p2 = saved_sep;
 		int r = do_glob(buf, pos+l, de->d_type, p2 ? p2 : "", ctx);
 		if (r) {
-			closedir(dir);
+			ctx->closedir(dir);
 			return r;
 		}
 	}
 	int readerr = errno;
 	if (p2) *p2 = saved_sep;
-	closedir(dir);
+	ctx->closedir(dir);
 	if (readerr && (ctx->errfunc(buf, errno) || (ctx->flags & GLOB_ERR)))
 		return GLOB_ABORTED;
 	errno = old_errno;
@@ -230,14 +237,23 @@ static int expand_tilde(char **pat, char *buf, size_t *pos)
 	return 0;
 }
 
-int glob(const char *restrict pat, int flags, int (*errfunc)(const char *path, int err), glob_t *restrict g)
+static void wrap_closedir(void *p) { __closedir(p); }
+static struct dirent *wrap_readdir(void *d) { return __readdir(d); }
+static void *wrap_opendir(const char *path) { return __opendir(path); }
+
+int glob(const char *restrict pat, int flags, int (*errfunc)(const char *path, int err), glob_t *restrict pglob)
 {
 	struct match head = { .next = NULL }, *tail = &head;
 	size_t cnt, i;
-	size_t offs = (flags & GLOB_DOOFFS) ? g->gl_offs : 0;
+	size_t offs;
 	int error = 0;
 	char buf[PATH_MAX];
-	
+	glob_t g[1];
+
+	/* copy avoids UB and prevents modifying pglob on error return */
+	memcpy(g, pglob, sizeof(*g));
+	offs = (flags & GLOB_DOOFFS) ? g->gl_offs : 0;
+
 	if (!errfunc) errfunc = ignore_err;
 
 	if (!(flags & GLOB_APPEND)) {
@@ -259,6 +275,11 @@ int glob(const char *restrict pat, int flags, int (*errfunc)(const char *path, i
 				.tail = &tail,
 				.flags = flags,
 				.errfunc = errfunc,
+				.closedir = (flags & GLOB_ALTDIRFUNC) ? g->gl_closedir : wrap_closedir,
+				.readdir = (flags & GLOB_ALTDIRFUNC) ? g->gl_readdir : wrap_readdir,
+				.opendir = (flags & GLOB_ALTDIRFUNC) ? g->gl_opendir : wrap_opendir,
+				.lstat = (flags & GLOB_ALTDIRFUNC) ? g->gl_lstat : lstat,
+				.stat = (flags & GLOB_ALTDIRFUNC) ? g->gl_stat : stat,
 			};
 			error = do_glob(buf, pos, 0, s, &ctx);
 		}
@@ -311,6 +332,8 @@ int glob(const char *restrict pat, int flags, int (*errfunc)(const char *path, i
 	if (!(flags & GLOB_NOSORT))
 		qsort(g->gl_pathv+offs, cnt, sizeof(char *), sort);
 	
+	if (!error)
+		memcpy(pglob, g, sizeof(*g));
 	return error;
 }
 
@@ -326,3 +349,45 @@ void globfree(glob_t *g)
 
 weak_alias(glob, glob64);
 weak_alias(globfree, globfree64);
+
+/*
+ * Statically check layout of BSD/GNU vs POSIX versions of glob_t.
+ */
+#undef __dirent
+#undef __stat
+struct __dirent;
+struct __stat;
+typedef struct {
+	size_t gl_pathc;
+	char **gl_pathv;
+	size_t gl_offs;
+	int __dummy1;
+
+	void (*gl_closedir)(void *);
+	struct __dirent *(*gl_readdir)(void *);
+	void *(*gl_opendir)(const char *);
+	int (*gl_lstat)(const char *__restrict, struct __stat *__restrict);
+	int (*gl_stat)(const char *__restrict, struct __stat *__restrict);
+} __posix_glob_t;
+
+#define CMP_MEMB(A, B, memb) ( \
+	offsetof(A, memb) == offsetof(B, memb) && \
+	sizeof(((A*)0)->memb) == sizeof(((B*)0)->memb))
+
+static void __test_glob_t(void)
+{
+	/* statically check size and member offsets */
+	switch(1) {
+	case 0:
+	case (sizeof(__posix_glob_t) == sizeof(glob_t) &&
+	      CMP_MEMB(__posix_glob_t, glob_t, gl_pathc) &&
+	      CMP_MEMB(__posix_glob_t, glob_t, gl_pathv) &&
+	      CMP_MEMB(__posix_glob_t, glob_t, gl_offs) &&
+	      CMP_MEMB(__posix_glob_t, glob_t, gl_closedir) &&
+	      CMP_MEMB(__posix_glob_t, glob_t, gl_readdir) &&
+	      CMP_MEMB(__posix_glob_t, glob_t, gl_opendir) &&
+	      CMP_MEMB(__posix_glob_t, glob_t, gl_lstat) &&
+	      CMP_MEMB(__posix_glob_t, glob_t, gl_stat)):
+		break;
+	}
+}
-- 
2.33.0

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.