Fix end-of-line bounds handling.

The existing implementation misses ending lines and as a result, fails
to match certain protocols (SMTP for example, which needs to see at
least 3 commain/response matches).  It appears from the commit history
that an attempt was made (end-1) to prevent reads past payload length.

This can be ensured by simply not reading any payload that is < 3 bytes.

The updated logic for this loop is:
- Payload length is >= 3 bytes, or return.
- Loop over payload, compare for EOL (CR + NL) sequence.
- If found, process string.
- If index 'a' plus two is less than payload length, increment 'a' by
  one and continue.
- Loop return always increments index 'a' by one.
This commit is contained in:
Darryl Sokoloski
2018-07-20 13:51:17 -04:00
parent d9c963061b
commit fbf6bd2be8
+3 -5
View File
@@ -4456,7 +4456,6 @@ void ndpi_parse_packet_line_info(struct ndpi_detection_module_struct *ndpi_struc
{
u_int32_t a;
struct ndpi_packet_struct *packet = &flow->packet;
u_int16_t end = packet->payload_packet_len - 1;
if(packet->packet_lines_parsed_complete != 0)
return;
@@ -4497,15 +4496,14 @@ void ndpi_parse_packet_line_info(struct ndpi_detection_module_struct *ndpi_struc
packet->http_response.len = 0;
packet->http_num_headers=0;
if((packet->payload_packet_len == 0)
|| (packet->payload == NULL)
|| (end == 0))
if((packet->payload_packet_len < 3)
|| (packet->payload == NULL))
return;
packet->line[packet->parsed_lines].ptr = packet->payload;
packet->line[packet->parsed_lines].len = 0;
for(a = 0; a < end-1 /* This because get_u_int16_t(packet->payload, a) reads 2 bytes */; a++) {
for(a = 0; a < packet->payload_packet_len; a++) {
if(get_u_int16_t(packet->payload, a) == ntohs(0x0d0a)) { /* If end of line char sequence CR+NL "\r\n", process line */
packet->line[packet->parsed_lines].len = (u_int16_t)(((unsigned long) &packet->payload[a]) - ((unsigned long) packet->line[packet->parsed_lines].ptr));