On 06/03/2015 10:17 AM, Redestig, Johan wrote:
> For built in commands seclabel obviously is not a factor (I was referencing 
> rmdir because that is something we usually invoke as an external command 
> since the internal implementation of it does not make any sense).
> 
> There have recently been a couple of new exec users that explicitly set the 
> seclabel. It may just become a pattern others follow. I shared the concern of 
> not unnecessarily specifying the seclabel in the init files, just pointing 
> out that there are legitimate uses of besides the ramdisk limitation, so 
> killing it entirely would have other consequences. And from this point of 
> view exec/service is identical, just difference syntax.

Looking at system/core/rootdir/init.rc, I see these two exec calls:
324:    exec u:r:tzdatacheck:s0 system system -- /system/bin/tzdatacheck
/system/usr/share/zoneinfo /data/misc/zoneinfo
657:    exec - logd log -- /system/bin/logcat -L -b all -v threadtime -v
usec -v printable -D -f /data/misc/logd/logcat -r 64 -n 256

Not clear that the first one needs a seclabel or could just use "-" to
indicate use of the default since policy defines a transition for
tzdatacheck?

Second one is already using the - notation to indicate the default.

We only need a seclabel if we cannot assign the executable a unique file
context that will trigger an automatic domain transition upon exec, as
with rootfs binaries or commands that share a common binary but require
different domains (e.g. if you wanted to run different toybox/toolbox
commands in different domains).

I wouldn't suggest removing seclabel support altogether, but we should
try to minimize its usage to only the cases where it is truly required.



_______________________________________________
Seandroid-list mailing list
[email protected]
To unsubscribe, send email to [email protected].
To get help, send an email containing "help" to 
[email protected].

Reply via email to