Page 2 of 5 FirstFirst 1234 ... LastLast
Results 16 to 30 of 64

Thread: TTG and MRV announced for S3 Tivos in November

  1. #16
    Join Date
    Nov 2002
    Posts
    1,076
    Quote Originally Posted by AlphaWolf View Post
    I wonder how much these two nuggets would help:

    http://www.schneier.com/blog/archive...a1_broken.html
    http://nsa.unaligned.org/

    They claim to be able to find any SHA-1 hash collision within a day.
    Since the kernel is signed with ElGamal, it would seem this only could help if you wanted to break the train of trust by attacking an sha1 signed file in the root filesystem. Assuming you succeeded at this, and posted an exploit, all tivo has to do is change the initrd image to use a stronger hash function.

  2. #17
    Join Date
    Aug 2004
    Posts
    4,075
    Quote Originally Posted by bcc View Post
    Since the kernel is signed with ElGamal, it would seem this only could help if you wanted to break the train of trust by attacking an sha1 signed file in the root filesystem. Assuming you succeeded at this, and posted an exploit, all tivo has to do is change the initrd image to use a stronger hash function.
    Isn't the thing that is ElGamal signed an SHA-1 hash of the kernel image? I'm just basing this on ADH's post here. If so, then in theory if you could produce a modified kernel that had the same SHA-1 hash as the original, you could use the original signature. That's essentially what killhdinitrd does, relying on the fact that parts of the kernel image were not included in the hash with the old PROMs. I don't think the brute force or other cracks on SHA-1 give you what you need, but I'm no crypto expert.
    Last edited by Jamie; 09-11-2007 at 02:35 PM.

  3. #18
    Join Date
    Nov 2002
    Posts
    1,076
    Quote Originally Posted by Jamie View Post
    Isn't the thing that is ElGamal signed an SHA-1 hash of the kernel image? I'm just basing this on ADH's post here. If so, then in theory if you could produce a modified kernel that had the same SHA-1 hash as the original, you could use the original signature. That's essentially what killhdinitrd does, relying on the fact that parts of the kernel image were not included in the hash with the old PROMs. I don't think the brute force or other cracks on SHA-1 give you what you need, but I'm no crypto expert.
    I thought that was old information - that old proms just used an sha hash as the kernel signature. If s3 proms are doing an El Gamal signature of *just* the hash, then that leaves the hash as a weak link. Weak enough? Sounds like one needs to put together FPGA hardware for brute-force attacks to run in a practical amount of time, which is harder than modding the PROM IMO.

  4. #19
    Join Date
    Aug 2004
    Posts
    4,075
    Quote Originally Posted by bcc View Post
    I thought that was old information - that old proms just used an sha hash as the kernel signature. If s3 proms are doing an El Gamal signature of *just* the hash, then that leaves the hash as a weak link. Weak enough? Sounds like one needs to put together FPGA hardware for brute-force attacks to run in a practical amount of time, which is harder than modding the PROM IMO.
    I think the info in ADH's post is still current. There's a discussion here of ElGamal and how it is only as secure as the hash function used (SHA-1 in this case).

    I'm sure killhdinitrd was harder to develop than doing one prom mod too. Possible motivations for developing a software hack: for fun. To be a her0: once you have a modified kernel that passes the security checks, anyone can use it without need a prom mod, so you save a lot of other people the work of doing a prom mod.

    Of course, when/if an exploit is developed, there's a pretty good chance it will be closed in the next hardware release.

  5. #20
    Join Date
    Nov 2004
    Posts
    420
    Quote Originally Posted by Jamie View Post
    I think the info in ADH's post is still current. There's a discussion here of ElGamal and how it is only as secure as the hash function used (SHA-1 in this case).
    A signature scheme is only as strong as its weakest link (the hash in this case), but I believe that the Wang et. al. attacks on full-round SHA-1 are collision attacks and not the preimage attacks needed to create rogue kernels that would verify as valid.

    Software-only exploits also lower the barrier to entry, which can develop more interest in the hobby (though it also lets in the riff-raff... you know, those people with unimaginative usernames. )

  6. #21
    Join Date
    Aug 2004
    Posts
    4,075
    Quote Originally Posted by tivo4mevo View Post
    ... I believe that the Wang et. al. attacks on full-round SHA-1 are collision attacks and not the preimage attacks needed to create rogue kernels that would verify as valid.
    This more recent work might be more relevant, although I still don't think we are there yet.

    Some of this ground was covered back last year: link.

    Guess I should go shopping for a new handle....

  7. #22
    Join Date
    Nov 2004
    Posts
    420
    Still looks to be collision and not preimage from a cursory read (though the new developments allow a block of the message to be well-formatted), but my jest did not come through quite right. No slight intended; it was mean to be self-referential, as handles including "tivo" seem to be a dime a dozen and because I was part of the riffraff ushered in by killhdinitrd.

    I'm shopping for a new handle... Count Von Count sounds about right.

  8. #23
    Join Date
    Aug 2004
    Posts
    4,075
    Quote Originally Posted by tivo4mevo View Post
    Still looks to be collision and not preimage from a cursory read (though the new developments allow a block of the message to be well-formatted), but my jest did not come through quite right. No slight intended; it was mean to be self-referential, as handles including "tivo" seem to be a dime a dozen and because I was part of the riffraff ushered in by killhdinitrd.

    I'm shopping for a new handle... Count Von Count sounds about right.
    Yes, it is a collision attack, not a preimage attack, so perhaps not useful.
    Last edited by Jamie; 09-11-2007 at 05:05 PM.

  9. #24
    Join Date
    Jan 2002
    Location
    Sonoran Desert
    Posts
    2,829
    Quote Originally Posted by bcc View Post
    I thought that was old information - that old proms just used an sha hash as the kernel signature. If s3 proms are doing an El Gamal signature of *just* the hash, then that leaves the hash as a weak link. Weak enough? Sounds like one needs to put together FPGA hardware for brute-force attacks to run in a practical amount of time, which is harder than modding the PROM IMO.
    AFAIK this is standard practice in any digital document signing. Hash the document, then sign the hash. Unless my understanding is wrong...I dunno. I read a short guide to cryptography once that is targeted at people who have no prior knowledge of cryptography and it went over the basic concepts in like 5 pages or so.

    My thought here was that we could create just enough code to boot off of and redirect to an arbitrary spot where we could put another kernel, then from there add junk data at the end of that to cause a collision with the signed hash. Essentially this would become a partial password match, as opposed to just starting with nothing, and would of course require significantly more memory to pull off, but not much more processing power I would venture to guess.
    Last edited by AlphaWolf; 09-12-2007 at 08:19 PM.
    Before PMing me: Iím not your personal tech support. If you have a question, ask in public so I don't have to repeat if somebody else asks. If you want images or slices, use emule. I will ignore all support PMs.

    Sponsor a vegetarian! I have taken the pledge, how about you?

  10. #25
    Join Date
    Nov 2002
    Posts
    1,076
    Quote Originally Posted by AlphaWolf View Post
    AFAIK this is standard practice in any digital document signing. Hash the document, then sign the hash. Unless my understanding is wrong...I dunno. I read a short guide to cryptography once that is targeted at people who have no prior knowledge of cryptography and it went over the basic concepts in like 5 pages or so.
    In crypto systems such as kerberos, signing is usually done on tuples that include more than just 1 thing so I wouldn't assume the signature is just over the hash. References I saw just indicated that the prom was doing SHA1{Kernel} or El Gamal{Kernel} not El Gamal{SHA1{Kernel}}. I've been fortunate enough to not need to disassemble the prom myself so I'll just trust that the prom is indeed doing El Gamal{SHA1{Kernel}} on the s3 boxes.
    Quote Originally Posted by AlphaWolf View Post
    My thought here was that we could create just enough code to boot off of and redirect to an arbitrary spot where we could put another kernel, then from there add junk data at the end of that to cause a collision with the signed hash. Essentially this would become a partial password match, as opposed to just starting with nothing, and would of course require significantly more memory to pull off, but not much more processing power I would venture to guess.
    Well after I publicly mentioned how the kernel load address could be tampered with, as it was outside the hash, tivo blocked that hole. I've verified that it can no longer be tampered with directly, and it sounds like the state of the art of sha1 cracking is not up to what you're suggesting (the preimage point jamie made).

    Suggest you just practice more with chipquik, it's not that hard really; come on in the water's fine I got it done with a cheap radio shack soldering iron even. It took me a couple tries on the first one before I realized I had to fully clean off the chipquik alloy and re-tin the pads before putting the socket on but after that the process worked pretty well.

  11. #26
    Join Date
    Jan 2002
    Location
    Sonoran Desert
    Posts
    2,829
    Quote Originally Posted by bcc View Post
    Suggest you just practice more with chipquik, it's not that hard really; come on in the water's fine I got it done with a cheap radio shack soldering iron even. It took me a couple tries on the first one before I realized I had to fully clean off the chipquik alloy and re-tin the pads before putting the socket on but after that the process worked pretty well.
    Yeah I am going to regardless of whether or not a soft attack surfaces. I just haven't had the time to mess with it lately. I actually made a video of my prom removal btw, and that part I did just fine. As soon as I get the time to edit that I'll post it.

    It's the getting the socket on there that is giving me hell. Or at least, that is where I lifted the pad on my S1 anyways. I am sure I can repair it, but I am not getting my hopes up as there are other things I possibly damaged in the process. I accidently flicked some chipquik on the CPU, I cleaned it off but some of the pins bent a tad while doing so...no reason to suspect damage, but its still ugly. That and I accidentally trimmed a little of that green stuff off of the top of some of the traces between the pads...they are still plenty conductive though, just hoping that no solder sticks to them upon socket installation.

    Yeah I made quite a mess of things. Ironically my very first prom removal went completely without error on the other hand and looks perfectly clean with no scorch marks or scrapes anywhere. This is the one I already posted pics of, and it looks beautiful. It was the second removal that got ugly.
    Last edited by AlphaWolf; 09-12-2007 at 10:02 PM.
    Before PMing me: Iím not your personal tech support. If you have a question, ask in public so I don't have to repeat if somebody else asks. If you want images or slices, use emule. I will ignore all support PMs.

    Sponsor a vegetarian! I have taken the pledge, how about you?

  12. #27
    Join Date
    Apr 2002
    Posts
    884
    According to TCF software version 9.1.L5 is being rolled out. It's claimed this software has MRV capability but tivo won't enable this capability until some time in the future.

  13. #28
    Join Date
    Jan 2002
    Location
    Sonoran Desert
    Posts
    2,829
    That L5 nomenclature sounds like a beta version.
    Before PMing me: Iím not your personal tech support. If you have a question, ask in public so I don't have to repeat if somebody else asks. If you want images or slices, use emule. I will ignore all support PMs.

    Sponsor a vegetarian! I have taken the pledge, how about you?

  14. #29
    Join Date
    Nov 2002
    Posts
    1,076
    Quote Originally Posted by AlphaWolf View Post
    That L5 nomenclature sounds like a beta version.
    I think "L" means limited release (to unsuspecting end users). Ie the typical limited early rollout. I got 9.1L5 and I'm not in the beta.

  15. #30
    Join Date
    Apr 2002
    Posts
    884
    It's been posted that the features are:

    Enchanced Wishlists
    Multiroom viewing for TiVo HD and S3 (needs to be enabled by TiVo*)
    Crestron Integration for S3
    Improvement in Emergency Alert Handling
    Bug fixes

    According to a tivo rep this is the "measured part of the rollout" After two weeks,if the call center doesn't find problems it will be rolled out to the rest of the units (over several weeks).

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •