Re: [PATCH] virt-aa-helper: allow common riscv64 loader paths

On 9/29/22 23:43, Christian Ehrhardt wrote: > On Thu, Sep 29, 2022 at 11:30 PM Jim Fehlig <jfehlig(a)suse.com&gt; wrote: >> >> On 9/28/22 06:45, christian.ehrhardt(a)canonical.com wrote: >>> From: Christian Ehrhardt <christian.ehrhardt(a)canonical.com&gt; >>> >>> Riscv64 usually uses u-boot as external -kernel and a loader from >>> the open implementation of RISC-V SBI. The paths for those binaries >>> as packaged in Debian and Ubuntu are in paths which are usually >>> forbidden to be added by the user under /usr/lib... >> >> Do you know if the path is configurable? > > Not on the libvirt stage, here it is > a) whatever users specify as kernel/loader in guest XML > b) (our) qemu 7.0 delivers it in the default paths for firmwares (that > are already allowed in src/security/apparmor/libvirt-qemu, but you'd > not want virt-aa-helper to complain on it still) > >> Are distros free to put those binaries where they choose? > > Well, in Debian/Ubuntu we'd still obey the FHS rules, but to some > extent yes it could be in other places. > >> E.g. /usr/libexec or similar? > > The libexec folks are usually fedora/RH which do not care about > apparmor that much, so we have historically not added those in code to > keep it readable (static rules use @libexecdir@ replaced at build > time). > > AFAICS you show to copy out the u-boot from the image [1] > Which does not end up in a predictable path-prefix that I could add here. > > If you tell me that on Suse the paths for these binaries are different > I'm more than happy to spin a v2 that also has your paths. > Or you provide a follow up with that content once/if you've found a > path that you'll regularly need. Thanks for the info. Let's go with the latter option. I'll send a follow up patch if/when needed.

Regards, Jim