Re: [Libguestfs] [PATCH nbdkit] server: Implement extents/can_extents calls for plugins and filters.

On Tue, Mar 12, 2019 at 01:23:37PM +0100, Martin Kletzander wrote: > >+=head2 C<.extents> > >+ > >+ int zero (void *handle, uint32_t count, uint64_t offset, uint32_t flags, > > s/zero/extents/ Oops. > >+ if (f->filter.extents) > >+ return f->filter.extents (&next_ops, &nxdata, handle, > >+ count, offset, flags, > >+ nr_extents, extents, err); > >+ else > >+ return f->backend.next->extents (f->backend.next, conn, > >+ count, offset, flags, > >+ nr_extents, extents, err); > > Do I understand it correctly that if a filter does not support > extents, then its function will only be applied on the allocated > blocks? It makes sense, but I feel like it would be nice to have > that mentioned somewhere. No. The code snippet above allows filters to intercept the extents call and modify it (either on the way in, or the way out, or most likely in this case both). The conditional ‘if (f->filter.extents)’ is checking if the filter has an extents callback, which is the callback that the filter has to provide to do this interception. I think most filters will need to be adjusted by adding an extents callback. Certainly any filter which deals with offsets (most obviously nbdkit-offset-filter, nbdkit-partition-filter) will need to implement the .extents call and both adjust the offset parameter that is passed down, and examine the list of extents being returned through the filter and adjust the offsets inside that list. This is the broad reason why we don't offer a stable API for filters (unlike plugins). Nor do we advise that it's a good idea to distribute filters separately from nbdkit. If an old filter was used with a plugin which was using the extents API then the filter would basically corrupt data.

> Other than that it looks OK to me, I'll try to cook up a test filter > to test this, but mainly as a fun exercise for myself. So note at the moment this patch doesn't really do anything. nbdkit server still doesn't support block status, so neither of the functions will ever be called. This patch was just for discussion about what the plugin API might look like.

Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com virt-df lists disk usage of guests without needing to install any software inside the virtual machine. Supports Linux and Windows. http://people.redhat.com/~rjones/virt-df/