[Devel] Re: [RFC v14-rc2][PATCH 16/29] A new file type (CR_FD_OBJREF) for a file descriptor already setup

[Serge E. Hallyn]

Quoting Oren Laadan (orenl at cs.columbia.edu):
> While file pointers are shared objects, they may share an underlying
> object themselves. For instance, file pointers of both ends of a pipe
> that share the same pipe inode. In this case, the shared entity to
> handle is the inode that is shared among two file pointers (e.g read-
> and write- ends). In this sort of "nested sharing" we need only save
> the underlying object once (upon first encounter) on checkpoint, and
> restore it only once during restart.
> 
> To checkpoint a file descriptor of this sort, we first lookup the
> inode in the hash table:
> 
> If not found, it is the first encounter of this inode. Here, Besides
> the file descriptor data, we also (a) register the inode in the hash
> and save the corresponding 'objref' of this inode in '->fd_objref' of
> the file descriptor. We then also (b) save the inode data, as per the
> inode type (this is not implemented in this patch, as it depends on
> the object). The file descriptor type will indicate the type of that
> object (e.g. for a pipe, when supported, CR_FD_PIPE).
> 
> If found, it is the second encounter of this inode, e.g. in the case
> of a pipe, as we hit the other end of the same pipe. At this point we
> need only record the reference ('objref') to the inode that we had
> saved before, and the file descriptor type is changed to CR_FD_OBJREF.
> 
> The logic during restart is similar: the '->fd_objref' is looked up in
> the hash table. Unlike checkpoint, during restart the object that is
> placed (and sought) in the hash table is the _file_ pointer, rather
> than the _inode_.
> 
> If not found, it is the first encounter of this inode. Therefore we
> (a) restore the inode data. Specifically, we construct a matching
> object and end up with multiple file pointers (e.g. if the object is a
> pipe, we will have both read- and write- ends). One of those is used
> for the file descriptor in question; the other(s) will be deposited in
> the hash table, to be retrieved and used later on. We also (b) register
> the newly created inode in the hash table using the given 'objref'.
> 
> If found, then we can skip the setup of the underlying object that
> is represented by the inode.
> 
> The type CR_FD_OBJREF indicates, on restart, that the corresponding
> file descriptor is already setup and registered in the hash under the
> '->fd_objref' that it had been assigned.
> 
> The next two patches use CR_FD_OBJREF to implement support for pipes.
> 
> Changelog[v14]:
>   - Introduce patch
> 
> Signed-off-by: Oren Laadan <orenl at cs.columbia.edu>

Acked-by: Serge Hallyn <serue at us.ibm.com>

-serge
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers