diff mbox series

unleak paths allocated in "diff --no-index"

Message ID xmqqilm579hc.fsf@gitster.g (mailing list archive)
State New, archived
Headers show
Series unleak paths allocated in "diff --no-index" | expand

Commit Message

Junio C Hamano Sept. 2, 2022, 9:27 p.m. UTC
"git diff --no-index" codepath starts with the two elements in
argv[] and munges them into two paths to be compared, stored in a
separate path[] arrays.  The munging is implemented in a rather
haphazard way, sometimes overwriting old version with a new copy,
and sometimes a constant string assigned to path[], making it
impossible to release the resources properly:

 * A single dash "-" from the command line is a special signal that
   the standard input is used for the side to be compared, and is
   internally replaced with a copy of string "-" at a known address.

 * When run in a subdirectory, full paths to the two paths are
   allocated and placed in path[].

 * After the above happens, when comparing a file with a directory,
   the directory side is replaced with the path to a file in the
   directory with the same name as the file.

This was perfectly fine for just two strings that are pathnames used
during the lifetime of the program and cleaned up upon program exit,
but it gets in the way when leak sanitizer is in effect.  The third
step can be losing the full path that was allocated in the second
step, but it is not easy to tell if its input is an allocated piece
of memory to begin with.

Loosen the earlier two steps a bit so that elements of the path[]
array that come to the directory/file comparison code are either the
singleton "-" or an allocated piece of memory.  Use that knowledge
in the third step to release an allocated piece of memory when it
replaces the path to a directory with the path to a file in that
directory, and also at the end to release the two elements of the
path[] array as needed.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
---

 * Because we never pass NULL to the new helper, !path check is not
   strictly necessary, but it is conventional for free()-like
   functions to take NULL and safely become a no-op.

 diff-no-index.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)
diff mbox series

Patch

diff --git c/diff-no-index.c w/diff-no-index.c
index 9a8b09346b..2770b7d15a 100644
--- c/diff-no-index.c
+++ w/diff-no-index.c
@@ -208,6 +208,14 @@  static void append_basename(struct strbuf *path, const char *dir, const char *fi
 	strbuf_addstr(path, tail ? tail + 1 : file);
 }
 
+static void free_allocated_path(const char *path)
+{
+	if (!path ||
+	    (path == file_from_standard_input))
+		return;
+	free((char *)path);
+}
+
 /*
  * DWIM "diff D F" into "diff D/F F" and "diff F D" into "diff F D/F"
  * Note that we append the basename of F to D/, so "diff a/b/file D"
@@ -226,9 +234,11 @@  static void fixup_paths(const char **path, struct strbuf *replacement)
 		return;
 	if (isdir0) {
 		append_basename(replacement, path[0], path[1]);
+		free_allocated_path(path[0]);
 		path[0] = replacement->buf;
 	} else {
 		append_basename(replacement, path[1], path[0]);
+		free_allocated_path(path[1]);
 		path[1] = replacement->buf;
 	}
 }
@@ -274,6 +284,8 @@  int diff_no_index(struct rev_info *revs,
 			p = file_from_standard_input;
 		else if (prefix)
 			p = prefix_filename(prefix, p);
+		else
+			p = xstrdup(p);
 		paths[i] = p;
 	}
 
@@ -294,13 +306,18 @@  int diff_no_index(struct rev_info *revs,
 	setup_diff_pager(&revs->diffopt);
 	revs->diffopt.flags.exit_with_status = 1;
 
-	if (queue_diff(&revs->diffopt, paths[0], paths[1]))
+	if (queue_diff(&revs->diffopt, paths[0], paths[1])) {
+		free_allocated_path(paths[0]);
+		free_allocated_path(paths[1]);
 		return 1;
+	}
 	diff_set_mnemonic_prefix(&revs->diffopt, "1/", "2/");
 	diffcore_std(&revs->diffopt);
 	diff_flush(&revs->diffopt);
 
 	strbuf_release(&replacement);
+	free_allocated_path(paths[0]);
+	free_allocated_path(paths[1]);
 
 	/*
 	 * The return code for --no-index imitates diff(1):