[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [dm-devel] Latest dm-userspace kernel code



From: Dan Smith <danms us ibm com>
Subject: Re: [dm-devel] Latest dm-userspace kernel code
Date: Fri, 29 Sep 2006 19:39:46 -0700

> FT> - What is the purpose of dmu_request->deps? I guess that it is for
> FT> preventing I/O being reordered. if so, again, you don't need to do
> FT> that.
> 
> Well, the problem is that a block copy would overwrite other requests.
> I think that while a copy is happening, you have to be careful to not
> write into that block so that your writes are not lost.
> 
> Am I over-thinking it?

Oh, I see. I think that you are right. That should not happen.


> FT> - How does the current kernel/user interface support multiple
> FT> destination feature?
> 
> I'm not sure what you mean here.  Can you elaborate?

I thought that dm-userpsace supports a feature to map a single request
to multiple devices for kinda raid. For that, I guess that struct
dmu_msg_map_response includes several sets of dst_maj and dst_min. Or
you might use 'linked' dmu_msg_map_response, which enables a
user-space daemon to send multiple dmu_msg_map_responses for one
dmu_msg_map_request (if a dmu_msg_map_response is marked 'linked',
dm-userspace module clones the bio and does not free struct
the dmu_request).


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]