6 ms·

Didn't know you could do deep packet inspection on encrypted traffic? Anyone who can shed some light on how they might be doing this? I'm curious.

netneutral 8y ago
You cannot do deep packet inspection, but can identify encrypted packets and then run heuristics and statistical packet and end-point analysis on them. If Telegram traffic has a specific signature or predictable pattern of packet sequences and sizes married with known IP addresses, quite a lot can be done at the packet inspection level (but not « deep » per se).
partyfaker 8y ago
Telegram uses it's own custom MTProto protocol. Although the traffic is encrypted, it is identifiable as Telegram traffic (fixed structure, identifiable prefix for packets). There's an outcry for them to use TLS but I didn't see that they said anything about that. Here's a nice article for more info: https://blog.susanka.eu/how-telegram-obfuscates-its-mtproto-traffic/ https://blog.susanka.eu/how-telegram-obfuscates-its-mtproto-...
- aaomidi 8y ago
  It's not like TLS would've fixed this. Iran is blocking Telegram very easily. If they also used anything with TLS 1.3 or HTTPS with TLS 1.2, the plaintext SNI would've been enough to detect Telegram.
  - lucb1e 8y ago
    Wait, didn't TLS 1.3 have encrypted SNI? I haven't yet read up on the details but iirc that was one of the major new things.
    - kardos 8y ago
      Unfortunately no, but there is an RFC [1] in the works for the more distant future [1] https://tools.ietf.org/html/draft-rescorla-tls-esni-00 https://tools.ietf.org/html/draft-rescorla-tls-esni-00
- codedokode 8y ago
  The MTProto traffic (which can be detected by signature) is wrapped into an obfuscation layer by some clients (including the official one). This layer makes packet look like random bytes. Earlier it has a flaw making all packets the same length and some ISPs started blocking packets by length by later it was fixed. This my comment contains more links about protocol and obfuscation: https://news.ycombinator.com/item?id=17888241 https://news.ycombinator.com/item?id=17888241
gnode 8y ago
Encrypted protocols are often not fully encrypted, and expose enough to determine the nature of the traffic, if not the content. Even TLS, leaving Server Name Indication (SNI) unencrypted. Fortunately there is work underway to encrypt SNI. Also, in the TLS case, domain fronting (asking for e.g. google.com from the discriminated service, to fool filters) has become one tool to evade SNI-based attacks.
- xnyanta 8y ago
  >domain fronting [...] has become one tool to evade SNI-based attacks. Unfortunately, most major CDNs like Google and Amazon have stopped offering domain fronting capabilities. I think Azure is one of the major services that stills supports it however.
  - codedokode 8y ago
    If they would not stop this they could be blocked in Russia completely (because Russian authorities started banning whole networks where Telegram servers or proxies were found).
    - vbezhenar 8y ago
      Microsoft is known to cooperate with governments. I'm sure that they'll stop domain fronting if asked.

netneutral 8y ago

You cannot do deep packet inspection, but can identify encrypted packets and then run heuristics and statistical packet and end-point analysis on them. If Telegram traffic has a specific signature or predictable pattern of packet sequences and sizes married with known IP addresses, quite a lot can be done at the packet inspection level (but not « deep » per se).

partyfaker 8y ago

Telegram uses it's own custom MTProto protocol. Although the traffic is encrypted, it is identifiable as Telegram traffic (fixed structure, identifiable prefix for packets). There's an outcry for them to use TLS but I didn't see that they said anything about that. Here's a nice article for more info: https://blog.susanka.eu/how-telegram-obfuscates-its-mtproto-traffic/ https://blog.susanka.eu/how-telegram-obfuscates-its-mtproto-...

aaomidi 8y ago
It's not like TLS would've fixed this. Iran is blocking Telegram very easily. If they also used anything with TLS 1.3 or HTTPS with TLS 1.2, the plaintext SNI would've been enough to detect Telegram.
- lucb1e 8y ago
  Wait, didn't TLS 1.3 have encrypted SNI? I haven't yet read up on the details but iirc that was one of the major new things.
  - kardos 8y ago
    Unfortunately no, but there is an RFC [1] in the works for the more distant future [1] https://tools.ietf.org/html/draft-rescorla-tls-esni-00 https://tools.ietf.org/html/draft-rescorla-tls-esni-00
codedokode 8y ago
The MTProto traffic (which can be detected by signature) is wrapped into an obfuscation layer by some clients (including the official one). This layer makes packet look like random bytes. Earlier it has a flaw making all packets the same length and some ISPs started blocking packets by length by later it was fixed. This my comment contains more links about protocol and obfuscation: https://news.ycombinator.com/item?id=17888241 https://news.ycombinator.com/item?id=17888241

gnode 8y ago

Encrypted protocols are often not fully encrypted, and expose enough to determine the nature of the traffic, if not the content. Even TLS, leaving Server Name Indication (SNI) unencrypted. Fortunately there is work underway to encrypt SNI. Also, in the TLS case, domain fronting (asking for e.g. google.com from the discriminated service, to fool filters) has become one tool to evade SNI-based attacks.

xnyanta 8y ago
>domain fronting [...] has become one tool to evade SNI-based attacks. Unfortunately, most major CDNs like Google and Amazon have stopped offering domain fronting capabilities. I think Azure is one of the major services that stills supports it however.
- codedokode 8y ago
  If they would not stop this they could be blocked in Russia completely (because Russian authorities started banning whole networks where Telegram servers or proxies were found).
  - vbezhenar 8y ago
    Microsoft is known to cooperate with governments. I'm sure that they'll stop domain fronting if asked.