MEDIUM: h2: properly check and deduplicate the content-length header in HTX

When producing an HTX message, we can't rely on the next-level H1 parser
to check and deduplicate the content-length header, so we have to do it
while parsing a message. The algorithm is the exact same as used for H1
messages.
diff --git a/include/common/h2.h b/include/common/h2.h
index e068427..acc80c0 100644
--- a/include/common/h2.h
+++ b/include/common/h2.h
@@ -155,6 +155,7 @@
 /* various protocol processing functions */
 
 int h2_make_h1_request(struct http_hdr *list, char *out, int osize, unsigned int *msgf);
+int h2_parse_cont_len_header(unsigned int *msgf, struct ist *value, unsigned long long *body_len);
 int h2_make_htx_request(struct http_hdr *list, struct htx *htx, unsigned int *msgf);
 int h2_make_htx_response(struct http_hdr *list, struct htx *htx, unsigned int *msgf);
 
diff --git a/src/h2.c b/src/h2.c
index 8830752..7c6cbee 100644
--- a/src/h2.c
+++ b/src/h2.c
@@ -320,6 +320,78 @@
 	return -1;
 }
 
+/* Parse the Content-Length header field of an HTTP/2 request. The function
+ * checks all possible occurrences of a comma-delimited value, and verifies
+ * if any of them doesn't match a previous value. It returns <0 if a value
+ * differs, 0 if the whole header can be dropped (i.e. already known), or >0
+ * if the value can be indexed (first one). In the last case, the value might
+ * be adjusted and the caller must only add the updated value.
+ */
+int h2_parse_cont_len_header(unsigned int *msgf, struct ist *value, unsigned long long *body_len)
+{
+	char *e, *n;
+	unsigned long long cl;
+	int not_first = !!(*msgf & H2_MSGF_BODY_CL);
+	struct ist word;
+
+	word.ptr = value->ptr - 1; // -1 for next loop's pre-increment
+	e = value->ptr + value->len;
+
+	while (++word.ptr < e) {
+		/* skip leading delimitor and blanks */
+		if (unlikely(HTTP_IS_LWS(*word.ptr)))
+			continue;
+
+		/* digits only now */
+		for (cl = 0, n = word.ptr; n < e; n++) {
+			unsigned int c = *n - '0';
+			if (unlikely(c > 9)) {
+				/* non-digit */
+				if (unlikely(n == word.ptr)) // spaces only
+					goto fail;
+				break;
+			}
+			if (unlikely(cl > ULLONG_MAX / 10ULL))
+				goto fail; /* multiply overflow */
+			cl = cl * 10ULL;
+			if (unlikely(cl + c < cl))
+				goto fail; /* addition overflow */
+			cl = cl + c;
+		}
+
+		/* keep a copy of the exact cleaned value */
+		word.len = n - word.ptr;
+
+		/* skip trailing LWS till next comma or EOL */
+		for (; n < e; n++) {
+			if (!HTTP_IS_LWS(*n)) {
+				if (unlikely(*n != ','))
+					goto fail;
+				break;
+			}
+		}
+
+		/* if duplicate, must be equal */
+		if (*msgf & H2_MSGF_BODY_CL && cl != *body_len)
+			goto fail;
+
+		/* OK, store this result as the one to be indexed */
+		*msgf |= H2_MSGF_BODY_CL;
+		*body_len = cl;
+		*value = word;
+		word.ptr = n;
+	}
+	/* here we've reached the end with a single value or a series of
+	 * identical values, all matching previous series if any. The last
+	 * parsed value was sent back into <value>. We just have to decide
+	 * if this occurrence has to be indexed (it's the first one) or
+	 * silently skipped (it's not the first one)
+	 */
+	return !not_first;
+ fail:
+	return -1;
+}
+
 /* Prepare the request line into <htx> from pseudo headers stored in <phdr[]>.
  * <fields> indicates what was found so far. This should be called once at the
  * detection of the first general header field or at the end of the request if
@@ -424,6 +496,7 @@
 	int i;
 	struct htx_sl *sl = NULL;
 	unsigned int sl_flags = 0;
+	unsigned long long body_len;
 
 	lck = ck = -1; // no cookie for now
 	fields = 0;
@@ -475,10 +548,14 @@
 		if (isteq(list[idx].n, ist("host")))
 			fields |= H2_PHDR_FND_HOST;
 
+		if (isteq(list[idx].n, ist("content-length"))) {
+			ret = h2_parse_cont_len_header(msgf, &list[idx].v, &body_len);
+			if (ret < 0)
+				goto fail;
+
-		if ((*msgf & (H2_MSGF_BODY|H2_MSGF_BODY_TUNNEL|H2_MSGF_BODY_CL)) == H2_MSGF_BODY &&
-		    isteq(list[idx].n, ist("content-length"))) {
-			*msgf |= H2_MSGF_BODY_CL;
 			sl_flags |= HTX_SL_F_CLEN;
+			if (ret == 0)
+				continue; // skip this duplicate
 		}
 
 		/* these ones are forbidden in requests (RFC7540#8.1.2.2) */
@@ -646,6 +723,7 @@
 	int i;
 	struct htx_sl *sl = NULL;
 	unsigned int sl_flags = 0;
+	unsigned long long body_len;
 
 	fields = 0;
 	for (idx = 0; list[idx].n.len != 0; idx++) {
@@ -693,10 +771,14 @@
 			fields |= H2_PHDR_FND_NONE;
 		}
 
+		if (isteq(list[idx].n, ist("content-length"))) {
+			ret = h2_parse_cont_len_header(msgf, &list[idx].v, &body_len);
+			if (ret < 0)
+				goto fail;
+
-		if ((*msgf & (H2_MSGF_BODY|H2_MSGF_BODY_TUNNEL|H2_MSGF_BODY_CL)) == H2_MSGF_BODY &&
-		    isteq(list[idx].n, ist("content-length"))) {
-			*msgf |= H2_MSGF_BODY_CL;
 			sl_flags |= HTX_SL_F_CLEN;
+			if (ret == 0)
+				continue; // skip this duplicate
 		}
 
 		/* these ones are forbidden in responses (RFC7540#8.1.2.2) */