3 ms·

I've always assumed any device or program I don't control will bypass anything I tell it to use and tunnel all it's evil traffic. It does feel the world is mov

I've always assumed any device or program I don't control will bypass anything I tell it to use and tunnel all it's evil traffic. It does feel the world is moving away from a multi-level network to run everything over TLS/TCP (and probably eventually mainly TLS/UDP), taking away the power from me as a network and device owner, and giving it to the developers

nine_k 6y ago
If only the OS could terminate TLS and allow to filter decrypted traffic locally.
- jhoechtl 6y ago
  Isn't this SSL offloading and done by companies to introspect the traffic?
  - mrkstu 6y ago
    That only works as long as the CA trusted by the user's computer is controlled by the box trying to do this.
    - Spivak 6y ago
      Which shouldn't be a problem since it's the box itself doing the decryption. No need to MiTM since it's the endpoint.
- saurik 6y ago
  This is prevented by certificate pinning.
  - mox1 6y ago
    You could easily add APIs for cert pinning at the OS level. Or provide the details needed to verify with the OS -> Program call return.
  - Fnoord 6y ago
    Yep, and HTST (if you don't trust that you can use an Firefox addon to perform certificate pinning). I believe HTST got introduced in the wake of the DigiNotar debacle (listen Darknet Diaries #3). Mitmproxy also made it easy. I guess this Mozilla Firefox change deprecates HTTPS Everywhere. Either way, we've come a long way. For people in the Dutch security world, DigiNotar was a known joke. I knew about their terrible security (and the implications) back around 2000-2004. Although there have been more vulnerabilities too, such as Hearthbleed and POODLE.
    - jlgaddis 6y ago
      > HTST I'm not familiar with HTST so I went looking but didn't find anything related to the web and HTTP, so I'm guessing you meant HSTS?
      - Fnoord 6y ago
        Yes, thanks for the correction.
  - LinuxBender 6y ago
    Only on a handful of domains. Most site admins are fearful of HPKP. The mitm proxy I am using to post this message only has to exclude a small handful of domains from mitm decrypting. A bunch of google domains, eff.org, paypal, a few mozilla domain names, dropbox and android. To my surprise, no traditional banks, no money management or stock trading sites, no "secure" portals used by law firms to share sensitive files. The sites I would expect to use pinning are the ones that do not. On a side note, I would love to see a system that adds out-of-band validation of an entity. e.g. My bank should have a QR code behind glass on the wall that I can scan, import a key and further validate their site at the application layer, above and beyond HTTPS and CA certs.
- iso1631 6y ago
  I was thinking more of devices I don't control - IOT stuff that requires internet access to function (like a box to watch netflix). Clearly anything running on my machine is fine as it's under my control
  - shawnz 6y ago
    So between the user and the network operator, who should have the final say? It seems like you're saying, "it should be the user when I'm the user and it should be the network operator when I'm the network operator". But I don't think we can have it both ways.
    - iso1631 6y ago
      Generally the policy should be 1) The ISP says what should be used 2) the lan either accepts that, or the operator puts their own values 3) the device either accepts that, or the operator puts their own values Same as object inheritence. An ISP says "use this DNS server it's nearer", the network operator says "no thanks, I'll run my own with an upstream of google as I don't like your NXDOMAIN injects", the user says "actually I'll use cloudflare because I don't trust any of you" That's all fine, and gives the power to the user. Exceptionally if a device doesn't allow the network settings to be overwritten for some technical reason, it should simply accept what the lan sends it.
      - Spivak 6y ago
        I think there will be some contention over 1 and 2 because we're at the point where both of these entities might be trusted (at home and work) but are inherently untrustworthy. And so browsers are saying that the path should be. 1) The browser developers say what should be used. 2) The user either accepts that or puts their own values. Ideally I think it should actually be something like 1) The OS vendor say what should be used. 2) The user either accepts that or puts their own values. 2.1) Lan operators can offer DNS servers which the user can (always) accept or (always) reject with reject being the default. But this whole mess with browser DoH is because OS vendors are slow to move on this. Google Android does it kinda, Windows has it in beta, Linux works but it's a little DIY.
- skissane 6y ago
  IBM z/OS has an interesting feature: AT-TLS (Application Transparent TLS). An app uses the OS sockets API to create plaintext sockets, and the OS adds TLS to them (based on policies configured by the sysadmin) transparent to the application. (There are IOCTLs that apps can call to discover this is going on, turn it on/off, configure it, etc, but the whole idea is you can add TLS support to some legacy app without needing any code changes, so adding those IOCTL calls to your app is totally optional.) In some parallel universe, TLS would have been part of TCP not a separate protocol and the Berkeley sockets implementation in the OS kernel would handle all encryption, certificate validation, etc, on behalf of applications. AT-TLS is a bit like visiting that parallel universe
  - remexre 6y ago
    I think this is actually making a comeback for the server room; https://www.kernel.org/doc/html/latest/networking/tls.html https://www.kernel.org/doc/html/latest/networking/tls.html for example. I _think_ the point of these is hardware TLS accelerators.
    - skissane 6y ago
      Linux kernel TLS, at least at the moment, only does the symmetric encryption part in the kernel, and the TLS handshake, certificate validation, has to be provided by user space. This is different from z/OS AT-TLS in which the OS does the TLS handshake, certificate validation, etc as well. (Strictly speaking, I'm not sure if AT-TLS actually is implemented in the OS kernel – or "nucleus" to use the z/OS terminology – it may actually be implemented somewhere else, say in the C library. I know some of it is actually implemented by a daemon called PAGENT. But, from the application programmers viewpoint, it is provided by the OS, however exactly the OS provides it.)

nine_k 6y ago

If only the OS could terminate TLS and allow to filter decrypted traffic locally.

jhoechtl 6y ago
Isn't this SSL offloading and done by companies to introspect the traffic?
- mrkstu 6y ago
  That only works as long as the CA trusted by the user's computer is controlled by the box trying to do this.
  - Spivak 6y ago
    Which shouldn't be a problem since it's the box itself doing the decryption. No need to MiTM since it's the endpoint.
saurik 6y ago
This is prevented by certificate pinning.
- mox1 6y ago
  You could easily add APIs for cert pinning at the OS level. Or provide the details needed to verify with the OS -> Program call return.
- Fnoord 6y ago
  Yep, and HTST (if you don't trust that you can use an Firefox addon to perform certificate pinning). I believe HTST got introduced in the wake of the DigiNotar debacle (listen Darknet Diaries #3). Mitmproxy also made it easy. I guess this Mozilla Firefox change deprecates HTTPS Everywhere. Either way, we've come a long way. For people in the Dutch security world, DigiNotar was a known joke. I knew about their terrible security (and the implications) back around 2000-2004. Although there have been more vulnerabilities too, such as Hearthbleed and POODLE.
  - jlgaddis 6y ago
    > HTST I'm not familiar with HTST so I went looking but didn't find anything related to the web and HTTP, so I'm guessing you meant HSTS?
    - Fnoord 6y ago
      Yes, thanks for the correction.
- LinuxBender 6y ago
  Only on a handful of domains. Most site admins are fearful of HPKP. The mitm proxy I am using to post this message only has to exclude a small handful of domains from mitm decrypting. A bunch of google domains, eff.org, paypal, a few mozilla domain names, dropbox and android. To my surprise, no traditional banks, no money management or stock trading sites, no "secure" portals used by law firms to share sensitive files. The sites I would expect to use pinning are the ones that do not. On a side note, I would love to see a system that adds out-of-band validation of an entity. e.g. My bank should have a QR code behind glass on the wall that I can scan, import a key and further validate their site at the application layer, above and beyond HTTPS and CA certs.
iso1631 6y ago
I was thinking more of devices I don't control - IOT stuff that requires internet access to function (like a box to watch netflix). Clearly anything running on my machine is fine as it's under my control
- shawnz 6y ago
  So between the user and the network operator, who should have the final say? It seems like you're saying, "it should be the user when I'm the user and it should be the network operator when I'm the network operator". But I don't think we can have it both ways.
  - iso1631 6y ago
    Generally the policy should be 1) The ISP says what should be used 2) the lan either accepts that, or the operator puts their own values 3) the device either accepts that, or the operator puts their own values Same as object inheritence. An ISP says "use this DNS server it's nearer", the network operator says "no thanks, I'll run my own with an upstream of google as I don't like your NXDOMAIN injects", the user says "actually I'll use cloudflare because I don't trust any of you" That's all fine, and gives the power to the user. Exceptionally if a device doesn't allow the network settings to be overwritten for some technical reason, it should simply accept what the lan sends it.
    - Spivak 6y ago
      I think there will be some contention over 1 and 2 because we're at the point where both of these entities might be trusted (at home and work) but are inherently untrustworthy. And so browsers are saying that the path should be. 1) The browser developers say what should be used. 2) The user either accepts that or puts their own values. Ideally I think it should actually be something like 1) The OS vendor say what should be used. 2) The user either accepts that or puts their own values. 2.1) Lan operators can offer DNS servers which the user can (always) accept or (always) reject with reject being the default. But this whole mess with browser DoH is because OS vendors are slow to move on this. Google Android does it kinda, Windows has it in beta, Linux works but it's a little DIY.
skissane 6y ago
IBM z/OS has an interesting feature: AT-TLS (Application Transparent TLS). An app uses the OS sockets API to create plaintext sockets, and the OS adds TLS to them (based on policies configured by the sysadmin) transparent to the application. (There are IOCTLs that apps can call to discover this is going on, turn it on/off, configure it, etc, but the whole idea is you can add TLS support to some legacy app without needing any code changes, so adding those IOCTL calls to your app is totally optional.) In some parallel universe, TLS would have been part of TCP not a separate protocol and the Berkeley sockets implementation in the OS kernel would handle all encryption, certificate validation, etc, on behalf of applications. AT-TLS is a bit like visiting that parallel universe
- remexre 6y ago
  I think this is actually making a comeback for the server room; https://www.kernel.org/doc/html/latest/networking/tls.html https://www.kernel.org/doc/html/latest/networking/tls.html for example. I _think_ the point of these is hardware TLS accelerators.
  - skissane 6y ago
    Linux kernel TLS, at least at the moment, only does the symmetric encryption part in the kernel, and the TLS handshake, certificate validation, has to be provided by user space. This is different from z/OS AT-TLS in which the OS does the TLS handshake, certificate validation, etc as well. (Strictly speaking, I'm not sure if AT-TLS actually is implemented in the OS kernel – or "nucleus" to use the z/OS terminology – it may actually be implemented somewhere else, say in the C library. I know some of it is actually implemented by a daemon called PAGENT. But, from the application programmers viewpoint, it is provided by the OS, however exactly the OS provides it.)