Re: [PATCH V12 05/15] block/export: track IOThread references

On Thu, Aug 20, 2026 at 2:32 AM Stefan Hajnoczi <[email protected]> wrote:
>
> On Sat, Aug 15, 2026 at 01:48:19AM +0800, Zhang Chen wrote:
> > @@ -139,7 +140,16 @@ BlockExport *blk_exp_add(BlockExportOptions *export, Error **errp)
> >              goto fail;
> >          }
> >
> > -        new_ctx = iothread_get_aio_context(iothread);
> > +        holder_id = export->id;
> > +        const IOThreadHolder holder = {
> > +            .type = IO_THREAD_HOLDER_KIND_BLOCK_EXPORT,
> > +            .u.block_export.export_id = holder_id,
> > +        };
> > +
> > +        new_ctx = iothread_ref_and_get_aio_context(iothread, &holder);
> > +        multithread_count = 1;
>
> This changes behavior in:
>
>   ret = drv->create(exp, export, multithread_ctxs, multithread_count, errp);
>
> Now export drivers will see drv->create(exp, export, NULL, 1, errp);
>                                                      ^^^^  ^
>
> Before it was drv->create(exp, export, NULL, 0, errp).
>
> fuse_export_create() has an assertion that fails now:
>
>   if (multithread) {
>       ...
>   } else {
>       /* Guaranteed by common export code */
>       assert(mt_count == 0);
>       ...
>   }

Good catch, it looks like the multithread_count only designed for the
multi threads cases.
Will introduce a new iothread_count here.

Yes, iothread_holder_id is redundant because exp->id remains valid
until after the IOThread
references are released. I’ll remove it and use export->id before
BlockExport allocation and
exp->id afterwards.

Thanks
Chen