| Commit message (Collapse) | Author | Age |
|---|
| ... | |
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Return the "classification-by-ip" as protocol results only if no other
results are available.
In particular, never return something like
"protocol_by_port/protocol_by_ip" (i.e. `NTP/Apple`,
BitTorrent/GoogleCloud`, `Zoom/AWS`) because this kind of classification
is quite confusing, if not plainly wrong.
Notes:
* the information about "classification-by-ip" is always available, so
no information is lost with this change;
* in the unit tests, the previous classifications with confidence
`NDPI_CONFIDENCE_DPI_PARTIAL` were wrong, as noted in #1957
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The goal if to correlate the right request-response pair, exporting
metadata from only one transaction (for example, the right url & return
state pair)
As a nice side effect, the code should be much cleaner, but that is a
matter of taste.
Two differences respect to the previous code:
* as it happens in the CI, if in the flow there are only one response
(before) and one request (after), only the metadata of the response are
saved/exported
* for performance reasons, we don't call `ndpi_parse_packet_line_info()`
anymore for ALL packets triggering the HTTP dissector, but only for the
packets that we already know belong to an HTTP flow. This is the reason
for the changes in RTSP/SOAP/... code
|
| |
|
|
|
| |
For a lot of protocols, reduce the number of packets after which the
protocols dissector gives up.
The values are quite arbitary, tring to not impact on classification
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This piece of code has multiple problems:
* nDPI is able to detect some TCP protocols even with mid-flows (i.e.
without the initial packets of the session); TLS is the most
significative example
* since e6b332aa4a1399e33df68998cf8351bccaee3fc4 it is perfectly valid
to not pass the TCP Handshake packets to nDPI
* in any case, we shouldn't call `ndpi_detection_giveup()`. That
function is usually called by the application and we end up calling it
twice in some cases.
The simple solution is to completely remove that code: process these
kinds of flows like everyone else.
Note that the application can always avoid to pass to nDPI any TCP flows
without the initial handshake; the flow managemnt is always up to the
application.
Looking at the CI results, some rare flows are now processed significantly
longer. As a follow-up we could look into that.
|
| | |
|
| | |
|
| |
|
|
| |
Signed-off-by: lns <matzeton@googlemail.com>
Signed-off-by: Toni Uhlig <matzeton@googlemail.com>
|
| |
|
|
| |
Signed-off-by: lns <matzeton@googlemail.com>
Signed-off-by: Toni Uhlig <matzeton@googlemail.com>
|
| |
|
|
| |
Signed-off-by: lns <matzeton@googlemail.com>
Signed-off-by: Toni Uhlig <matzeton@googlemail.com>
|
|
|
Extend internal unit tests to handle multiple configurations.
As some examples, add tests about:
* disabling some protocols
* disabling Ookla aggressiveness
Every configurations data is stored in a dedicated directory under
`tests\cfgs`
|