3 ms·
Providing user control over page faults and using that for a security exploit reminds me of the classic UNIX tale of password checking. A version of UNIX had a
by JoshTriplett 10y ago
Providing user control over page faults and using that for a security exploit reminds me of the classic UNIX tale of password checking. A version of UNIX had a privileged mechanism that would check a password (provided by pointer); it did so character-by-character. It also had a way for userspace processes to handle page faults themselves. So, put a password buffer across two pages, with the page boundary after the first character, and change the first character until you get a page fault. Repeat for each character of the password...
- Aissen 10y agoIt's nice because it allows a slow and controlled timing attack, but you don't need a page fault to do that, you can simply measure elapsed time with high resolution timers, perf or whatever you have on hand :-)
- JoshTriplett 10y agoTrue, a privileged process having data-dependent execution time for secure data can lead to a variety of side-channel attacks. For more complex cases than password validation, it would help if it were easier to write side-channel-resistant algorithms in something other than assembly. Even a C compiler, let alone something higher-level, can introduce side-channel leaks through optimization.
- Animats 10y agoThat wasn't UNIX. It was TOPS-20 on the DEC 36-bit machines.
- shenberg 10y agoThe system mentioned is TENEX (not *nix) according to http://research.microsoft.com/en-us/um/people/blampson/33-Hints/WebPage.html http://research.microsoft.com/en-us/um/people/blampson/33-Hi... (a great paper from '83 that was posted here recently. search for password to find the exact reference there)
- JoshTriplett 10y agoThanks for the correction. The version of the story I saw came from a systems programming book, which attributed it vaguely to some past UNIX system of yore. Glad to have a more accurate reference.