Posted on January 11, 2018 at 14:07gonzalez.laurent wrote:
2/ security downgrade from level2 uses probing. Are there any generic purpose MCU protected against probing ? Anyway, I agreed that the encoding of RDP is problematic.
***************************************** -end-of-quote ************************
Yes, the simple 'Any other combination is RDP2' would have made 'amateur' breaking of the protection close to impossible. Decapping a chip is easy enough, exposing to UV too.
***************************************** -restart-of-quote ************************
3/ debug interface exploit looks really unreliable (how to sort FW from garbage?), even the author only says 'practically feasible'.
The debug interface reports ERROR (permission denied, try again) or OK (you got your data, store it, and go to the next location). So this is completely easy.
The fact that the debug interface exploit works, together with openocd only supporting lock level 1, probably means that lots of people are not protected, when they thought they were.....
These guys are just some students that have tried a bunch of tricks to break this and report on the three tricks that worked. They probably aslo tried a few things that didn't work. But with such sloppy design (the encoding of lock-level-2 is wrong, everybody can see that!) it is likely that someone else may find another bug.
For example, it is clear that these guys are not good at controlling the power to their target. They claim 45 bytes per minute. So that means 11 succesful 32-bit attempts per minute, or about 22 raw tries. That translates to about 3 seconds per attempt. With a mosfet I can apply power to a CPU in less than a microsecond. Hold the reset line for a microsecond, and then release. Now start doing the SWD stuff, and a few microseconds later you have your results. Several hundred or thousand attempts per second should be possible. But if they use a mechanical relay to power up the device, then have to wait for big capacitors to drain then that can lead to a cycle time of several seconds.
But indeed. The result is that you can read a lockec STM32F0 at AT LEAST 45 bytes per second. But unless you're starting a business doing this and have tens of clients a day, that is not worth the further engineering effort.
