[CRIU] [RFC PATCH 0/5] introduce --enable-fs cli option

[Pavel Emelyanov]

On 04/09/2015 08:46 PM, Oleg Nesterov wrote:
> Hello, and sorry for delay.
> 
> Almost the same as V1, just some cosmetics changes. Seems to work.
> 
> I decided to keep mnt_entry->fsname and avoid the "clever" games with
> dump/resore auto fsnames once. Because:
> 
> 	1. I am lazy and you seem to agree with ->fsname ;)
> 
> 	2. Your protobuf/mnt.proto is stupid (wrong afaics) and this
> 	   means that I need to write more changes...

Can you elaborate on this? :)

> 	   Why do we expose "enum fstype" to protobufs at all? Please
> 	   note that typeof(mnt_entry->fstype) is uint32, not "fstype".

Yes, we have this fact mentioned on the "what's bad with v1 images" page.
We want it to be the "fstype".

> Pavel, I didn't convert fstypes[] into the list as you suggested. Can
> I do this in a separate patch?

Sure!

> Because personally I think this makes
> no sense and this (trivial) code will look worse, so we can discuss
> this simple change separately, including the cosmetic issues like the
> necessary ctor for non-auto entries.

But current implementation doesn't require this array to be bigger than
it actually is -- you don't add custom fsnames in it.

> TODO: resolve_source() && FSTYPE__AUTO. I'll resend my hack later, if/
> when you accept this series. This needs another (more) discussion too.

TODO2: Add a test (even trivial) for this functionality. Please. All this
stuff is quite easy to break, so any help in catching simple mistakes are
really helpful.

> Plus support this option (and skip-mnt) in RPC, but this looks simple.
> 
> Oleg.
> 
> .
> 

-- Pavel