core: Fix ap_varbuf_regsub() crash on an empty result

* server/util.c (regsub_core): Do not terminate the result when nothing
  is added and the varbuf has no buffer yet; vb->buf is then the shared
  read-only empty string.  This crashed mod_substitute when a regular
  expression rule replaced a whole line with an empty string.

* test/modules/filters/test_004_substitute_empty.py: New test.

PR: 70012

Author: arturobernalg

changes-entries/varbuf-regsub-empty.txt

@@ -0,0 +1,5 @@
+  *) core: Fix a crash in ap_varbuf_regsub() when the result is empty and
+     the buffer has not been allocated yet.  This made mod_substitute crash
+     when a regular expression rule replaced a whole line with an empty
+     string.  PR 70012.
+     [Arturo Bernal]

server/util.c

@@ -421,6 +421,11 @@ static apr_status_t regsub_core(apr_pool_t *p, char **result,
     else {
         if (vb->strlen == AP_VARBUF_UNKNOWN)
             vb->strlen = strlen(vb->buf);
+        if (!len && !vb->avail) {
+            /* Nothing to add and no buffer yet, vb->buf is the shared
+             * empty string which must not be written to. */
+            return APR_SUCCESS;
+        }
         ap_varbuf_grow(vb, vb->strlen + len);
         dst = vb->buf + vb->strlen;
         vb->strlen += len;

test/modules/filters/test_004_substitute_empty.py

@@ -0,0 +1,92 @@
+import os
+
+import pytest
+
+from pyhttpd.conf import HttpdConf
+
+# A substitution may legitimately produce zero bytes.  Later rules must still
+# see the (now empty) line and continue to work on the following lines.
+MODES = {
+    'regex_default': '',
+    'regex_flatten': 'f',
+    'regex_quick': 'q',
+}
+
+DOC = "## remove this completely\nfoo\n"
+
+
+class TestSubstituteEmptyResult:
+
+    @pytest.fixture(autouse=True, scope='class')
+    def _class_scope(self, env):
+        doc_dir = os.path.join(env.server_dir, "htdocs", "test1")
+        with open(os.path.join(doc_dir, "doc.html"), "w") as f:
+            f.write(DOC)
+
+    def configure(self, env, rules):
+        conf = HttpdConf(env, extras={
+            f"test1.{env.http_tld}": f"""
+            <Location "/">
+                AddOutputFilterByType SUBSTITUTE text/html
+                {rules}
+            </Location>
+            """,
+        })
+        conf.add_vhost_test1()
+        conf.install()
+        assert env.apache_restart() == 0
+
+    def get(self, env, path):
+        return env.curl_get(env.mkurl("http", "test1", path))
+
+    # A rule that empties a whole line, followed by another rule.
+    @pytest.mark.parametrize("mode", list(MODES), ids=list(MODES))
+    def test_filters_004_01(self, env, mode):
+        flags = MODES[mode]
+        self.configure(env, f"""
+                Substitute "s/^##.*//{flags}"
+                Substitute "s/foo/bar/{flags}"
+        """)
+        r = self.get(env, "/doc.html")
+        assert r.response, f"{mode}: no response"
+        assert r.response["status"] == 200
+        assert r.response["body"].decode() == "\nbar\n"
+
+    # The same with a literal (non-regex) pattern.
+    @pytest.mark.parametrize("flags", ["nf", "nq"], ids=["literal_flatten", "literal_quick"])
+    def test_filters_004_05(self, env, flags):
+        self.configure(env, f"""
+                Substitute "s/## remove this completely//{flags}"
+                Substitute "s/foo/bar/{flags}"
+        """)
+        r = self.get(env, "/doc.html")
+        assert r.response, f"{flags}: no response"
+        assert r.response["status"] == 200
+        assert r.response["body"].decode() == "\nbar\n"
+
+    # Control: the emptying rule on its own.
+    def test_filters_004_02(self, env):
+        self.configure(env, 'Substitute "s/^##.*//"')
+        r = self.get(env, "/doc.html")
+        assert r.response, "no response"
+        assert r.response["body"].decode() == "\nfoo\n"
+
+    # Control: the first rule does not match, the second one does.
+    def test_filters_004_03(self, env):
+        self.configure(env, """
+                Substitute "s/^nomatch.*//"
+                Substitute "s/foo/bar/"
+        """)
+        r = self.get(env, "/doc.html")
+        assert r.response, "no response"
+        assert r.response["body"].decode() == "## remove this completely\nbar\n"
+
+    # Control: the first rule replaces the line with something non-empty.
+    def test_filters_004_04(self, env):
+        self.configure(env, """
+                Substitute "s/^##.*/xx/"
+                Substitute "s/foo/bar/"
+        """)
+        r = self.get(env, "/doc.html")
+        assert r.response, "no response"
+        assert r.response["body"].decode() == "xx\nbar\n"