On 10/6/20 6:13 PM, Miklos Szeredi wrote:
On Fri, Sep 25, 2020 at 10:35 AM Pavel Tikhomirov
<ptikhomirov@xxxxxxxxxxxxx> wrote:
Note: In our (Virtuozzo) use case users inside a container can create
"regular" overlayfs mounts without any "index=" option, but we still
want to migrate this containers with CRIU so we set "index=on" as kernel
default so that all the container overlayfs mounts get support of file
handles automatically. With "uuid=off" we want the same thing (to be
able to "copy" container with uuid change) - we would set kernel default
so that all the container overlayfs mounts get "uuid=off" automatically.
I'm not sure I buy that argument for a kernel option. It should
rather be a "container" option in that case, but AFAIK the kernel
doesn't have a concept of a container. I think this needs to be
discussed on the relevant mailing lists.
As of now mainline kernel doesn't support unprivileged overlay mounts,
so I guess this is not an issue. Let's just merge this without the
kernel and the module options.
Virtuozzo kernel does have a "container" concept and we do have
unprivileged overlay mounts to support docker inside Virtuozzo
containers. We don't face any major issues with it. But you are right
it's not mainstream.
Probably a normal user of mainstream kernel also might want to set
index=on+uuid=off by default, so that all their docker containters
automatically support inotifies and survive backing disk uuid change
automaticaly.
I will prepare next patchset version without default.
Thanks,
Miklos
--
Best regards, Tikhomirov Pavel
Software Developer, Virtuozzo.
