You might want think more about this then the peep hole approach. Rather then running things in userland and relying on user id isolation; a better way is to write a kernel driver that exposes a very specific and limited interface. GPIOs are directly toggling hardware which means there are more then thing simple "privileges" - - Can the pins be reconfigured to conflict (i.e. BBB is an Output and the thing connected is an output) and is the hardware tolerant of that or will things die?
- Is there anything bad that will happen if the pin floats (i.e. a normally output pin is switched to an input)? Will it pickup noise and let things go crazy? Just merely changing permissions can leave a dangling problem. On Monday, February 08, 2016 10:33:03 [email protected] wrote: > This is John, the guy with the question Drew to above on the Adafruit > board. (Thanks Drew for starting this conversation here.) > > The problem is, this is part of a commercial application. My code will be > reading and writing to the GPIOs and doing various things, communicating > data to remote systems... I anticipate adding a fair bit of complexity. I'd > hate to make a bug that blew something important away or something like > that. I am generally leery of unnecessary privilege escalation. > > So I could > > - sudo my entire application, which would be the same as running as root; > - partition my application and have it invoke the code that interfaces > directly to the GPIOs as sudo in some way (via shell?); > - go by way of the PRUs; > - change permissions; > - some other way. > > I did have partial success changing the permissions on relevant files, ala > > https://github.com/cnobile2012/RobotControl/tree/master/contrib > > which uses udev to set up permissions (not that I understand udev either.) > > I call my success partial because I can get a test LED to turn on, but I > can't get it to flash using PWM. > > Any ideas why that would be? > > On Sunday, February 7, 2016 at 10:00:01 AM UTC-6, Mike Bell wrote: > > On 02/06/2016 12:51 AM, Brian Anderson wrote: > > > > > > My comments are really to do with what I perceive as best practices on how > > one would approach building systems that are "security conscious". Of > > course, "convenience" may direct us in different directions during > > development. I am not sure what you are trying to imply by "safe" as in > > protecting the GPIOs from misuse. I don't actually see any way to > > accomplish that. What I do think one can do is to be aware of security > > considerations and not unnecessarily present an attack surface that can > > compromise the entire system. > > > > > > *Using sudo seems much less secure as it exposes the application to being > > > >>> exploited for security flaws. And since the application is running as > >>> root, > >>> it has access to everything.* > >> > >> So, we have a device on a system that can potentially cause physical > >> damage to external hardware when something like a wrong GPIO state is > >> toggled, or such. How would sudo be less secure in this context? > > > > It wouldn't. And that is not my point. I am not talking about how to > > protect the GPIOs from "bad behaved" programs that are "trusted" as > > implied > > by the fact that they are running as a normal user in the group that has > > access to those GPIOs. If an application is trusted (is a member of the > > appropriate group or for that matter can sudo), it is a hopeless task to > > protect the GPIOs from misuse. What I am trying to point out is that > > running an app as "root" (sudo, set uid, whatever) exposes a security > > attack vector...a vector that has access to _all_ system resources. I > > would claim that it is an unnecessary exposure...from a security point of > > view. YMMV when it comes to "convenience". > > > >> In fact under certain conditions it would be less safe using groups. > > > > How would an application running at a non-root level using groups to > > access protected resources be less "safe" than an application running as > > root using sudo? > > > >> Also, "root has access to everything" is wrong. Reread what I've written > >> above about running specific commands through sudo. > > > > Errr, an application running as root, by definition, has access to _all_ > > system resources. The fact that you are limiting just a single > > application/user to run sudo doesn't limit the attack surface for that > > application. If your root application is compromised in some way, then > > the > > entire system can be compromised. Running as a normal user does not > > present the same attack surface...its much smaller and sandboxed by the > > kernel. Running as root affords no protection enforcement by the kernel. > > > > ba > > > >> On Fri, Feb 5, 2016 at 6:05 PM, Brian Anderson <[email protected]> wrote: > >>> Err, why? > >>> > >>> Groups are frequently used to restrict access to resources. Android > >>> exploits groups for permissions and to sandbox applications. And the > >>> kernel enforces access. > >>> > >>> Using sudo seems much less secure as it exposes the application to being > >>> exploited for security flaws. And since the application is running as > >>> root, > >>> it has access to everything. > >>> > >>> But maybe I'm missing something? > >>> > >>> ba > >>> > >>> Brian, > > > > This is a great summation of the issue! > > > > Mike -- Hunyue Yau http://www.hy-research.com/ -- For more options, visit http://beagleboard.org/discuss --- You received this message because you are subscribed to the Google Groups "BeagleBoard" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
