Re: [PATCH v2 07/15] libxl: convert libxl_domain_destroy to an async op

[Roger Pau Monne <roger.pau@citrix.com>]

Ian Jackson wrote:

[...]

I will apply the block of changes that I've removed from this reply.

>> +static void device_remove_callback(libxl__egc *egc, libxl__ao_device *aorm)
>> +{
>> +    STATE_AO_GC(aorm->ao);
>> +    libxl__devices_remove_state *drs = CONTAINER_OF(aorm->base, *drs, aorm);
>> +    char *be_path = libxl__device_backend_path(gc, aorm->dev);
>> +    char *fe_path = libxl__device_frontend_path(gc, aorm->dev);
>> +    xs_transaction_t t = 0;
>> +    int rc = 0, last = 1;
>> +
>> +retry_transaction:
>> +    t = xs_transaction_start(CTX->xsh);
>> +    if (aorm->action == DEVICE_DISCONNECT) {
>> +        libxl__xs_path_cleanup(gc, t, fe_path);
>> +        libxl__xs_path_cleanup(gc, t, be_path);
>> +    }
>> +    if (!xs_transaction_end(CTX->xsh, t, 0)) {
>> +        if (errno == EAGAIN)
>> +            goto retry_transaction;
>> +        else {
>> +            rc = ERROR_FAIL;
>> +            goto out;
>> +        }
>> +    }
>> +
>> +out:
>> +    rc = libxl__ao_device_check_last(gc, aorm, drs->aorm,
>> +                                     drs->num_devices,&last);
>> +    if (last)
>> +        drs->callback(egc, drs, rc);
>> +    return;
>
> I don't understand why this is here in this patch.  Surely it belongs
> in the previous one ?

Not really, because libxl__destroy_devices is called from 
libxl_domain_destroy, which does not ao until this patch, so this 
callback is the libxl__devices_destroy one (which was not async in 
previous patches), since it called libxl__device_destroy instead of 
libxl__initiate_device_remove.

Thanks for the review.