Try to have a faster classification, on first packet; use standard extra
dissection data path for sub-classification, metadata extraction and
monitoring.
STUN caches:
* use the proper confidence value
* lookup into the caches only once per flow, after having found a proper
STUN classification
Add identification of Telegram VoIP calls.
Fix compilation on Windows.
"dirent.h" file has been taken from https://github.com/tronkko/dirent/
Fix Python bindings
Fix some warnings with x86_64-w64-mingw32-gcc:
```
protocols/dns.c: In function ‘ndpi_search_dns’:
protocols/dns.c:775:41: error: cast from pointer to integer of different size [-Werror=pointer-to-int-cast]
775 | unsigned long first_element_len = (unsigned long)dot - (unsigned long)_hostname;
| ^
protocols/dns.c:775:62: error: cast from pointer to integer of different size [-Werror=pointer-to-int-cast]
775 | unsigned long first_element_len = (unsigned long)dot - (unsigned long)_hostname;
|
```
```
In file included from ndpi_bitmap64.c:31:
third_party/include/binaryfusefilter.h: In function ‘binary_fuse8_hash’:
third_party/include/binaryfusefilter.h:160:32: error: left shift count >= width of type [-Werror=shift-count-overflow]
160 | uint64_t hh = hash & ((1UL << 36) - 1);
```
```
In function ‘ndpi_match_custom_category’,
inlined from ‘ndpi_fill_protocol_category.part.0’ at ndpi_main.c:7056:16:
ndpi_main.c:3419:3: error: ‘strncpy’ specified bound depends on the length of the source argument [-Werror=stringop-overflow=]
3419 | strncpy(buf, name, name_len);
```
approach for handling Internet domain names.
For switching back to Aho-Corasick it is necessary to edit
ndpi-typedefs.h and uncomment the line
// #define USE_LEGACY_AHO_CORASICK
[1] With Aho-Corasick
$ ./example/ndpiReader -G ./lists/ -i tests/pcap/ookla.pcap | grep Memory
nDPI Memory statistics:
nDPI Memory (once): 37.34 KB
Flow Memory (per flow): 960 B
Actual Memory: 33.09 MB
Peak Memory: 33.09 MB
[2] With the new algorithm
$ ./example/ndpiReader -G ./lists/ -i tests/pcap/ookla.pcap | grep Memory
nDPI Memory statistics:
nDPI Memory (once): 37.31 KB
Flow Memory (per flow): 960 B
Actual Memory: 7.42 MB
Peak Memory: 7.42 MB
In essence from ~33 MB to ~7 MB
This new algorithm will enable larger lists to be loaded (e.g. top 1M domans
https://s3-us-west-1.amazonaws.com/umbrella-static/index.html)
In ./lists there are file names that are named as <category>_<string>.list
With -G ndpiReader can load all of them at startup
The option NSID (RFC5001) is used by Google DNS to report the
airport code of the metro where the DNS query is handled.
This option is quite rare, but the added overhead in DNS code is pretty
much zero for "normal" DNS traffic
Use two separate lists:
* one for the ingress nodes, which triggers a ProtonVPN classification
* one for the egress nodes, which triggers the
`NDPI_ANONYMOUS_SUBSCRIBER` risk
Add a command line option (to `ndpiReader`) to easily test IP/port
matching.
Add another example of custom rule.
A fully encrypted session is a flow where every bytes of the
payload is encrypted in an attempt to “look like nothing”.
The heuristic needs only the very first packet of the flow.
See: https://www.usenix.org/system/files/sec23fall-prepub-234-wu-mingshi.pdf
A basic, but generic, inplementation of the popcpunt alg has been added
Some notes:
* libinjection: according to https://github.com/libinjection/libinjection/issues/44,
it seems NULL characters are valid in the input string;
* RTP: `rtp_get_stream_type()` is called only for RTP packets; if you
want to tell RTP from RTCP you should use `is_rtp_or_rtcp()`;
* TLS: unnecessary check; we already make the same check just above, at
the beginning of the `while` loop
Regardless of the name, the removed trace doesn't contain meaningful
Hangout traffic.
Remove last piece of sub-classifiction based only on ip addresses.
Try avoiding false positives: look for 3 RTP packets before classifing
the flow as such.
Add a generic function `is_rtp_or_rtcp()` to identify RTP/RTCP packets also
in other dissectors (see 3608ab01b commit message for an example)
In some networks, there are some anomalous TCP flows where the smallest
ACK packets have some kind of zero padding.
It looks like the IP and TCP headers in those frames wrongly consider the
0x00 Ethernet padding bytes as part of the TCP payload.
While this kind of packets is perfectly valid per-se, in some conditions
they might be treated by the TCP reassembler logic as (partial) overlaps,
deceiving the classification engine.
Add an heuristic to detect these packets and to ignore them, allowing
correct detection/classification.
This heuristic is configurable. Default value:
* in the library, it is disabled
* in `ndpiReader` and in the fuzzers, it is enabled (to ease testing)
Credit to @vel21ripn for the initial patch.
Close#1946
