[Devel] Re: [PATCH 0/9] namespaces: Introduction

[Eric W. Biederman]

Dave Hansen <haveblue at us.ibm.com> writes:

> On Fri, 2006-05-19 at 08:13 -0700, Andrew Morton wrote:
>> snapshot/restart/migration worry me.  If they require complete
>> serialisation of complex kernel data structures then we have a problem,
>> because it means that any time anyone changes such a structure they need to
>> update (and test) the serialisation.
>
> The idea of actually serializing kernel data structures keeps me up at
> night.  This is especially true when we already have some method of
> exporting these structures to userspace.

Serialization of kernel data structures is a thorny issue, that
we are far enough away from that we don't need to tackle just yet.
I do consider it a failure to export/import things properly if you
need to use the same kernel version.

For the short term all that is interesting from a checkpoint/restart/migration
perspective is that you can have multiple instances of global identifiers,
pids, sysvipc ids, etc.

> However, the proc-maps/mmap approach would require new interfaces to be
> implemented.  There are plenty of attributes not currently readily
> visible to userspace like VM_NONLINEAR, or resources which are normally
> inaccessible to userspace like deleted files.  Those would need extended
> user/kernel interfaces.

Deleted files are accessible through /proc/<pid>/fd.

Eric