> With it, the protocol in the client daemon will ack the kernel _before_
> it gets an ack from the remote server. That should help relieve the
> deadlock. Things then go like this:
>
> kernel runs low on memory
> kernel flushes buffers to device drivers under pressure
> nbd client daemon sends to the net
> * client acks kernel and releases buffers in kernel
> server on the _same machine_ receives request over the net
> server tries to write request to disk
> server process needs buffers to write to
> * server gets buffers released by client in kernel
>
> At least, potentially. If I recall right, there's still a deadlock
> window in-kernel, but it's small. I don't recall the details. Oh -
> well, the request still hangs around in the driver until the client
> daemon acks .. maybe the client daemon can't ack without being swapped
> in first, and that'd be deadlock.
>
> Well, there's a "-s" flag (for swap devices) that does an mlockall()
> that might take care of that. So "-a -s" might do it. Really needs
> a "-aa" option (release write request in kernel asap).
Well, with mlockall(), I'd believe it could be made to work. Okay.
Pavel
-- (about SSSCA) "I don't say this lightly. However, I really think that the U.S. no longer is classifiable as a democracy, but rather as a plutocracy." --hpa - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/