On Thu, Jun 05, 2014 at 03:55:40AM +0200, Paolo Bonzini wrote:
Il 05/06/2014 00:33, Hani Benhabiles ha scritto:
IIUC, what this does is ensure that the other side gets a FIN before it gets
a RST. Is this correct?
Yes. Without shutdown(), this could be reproduced (unreliably) on multiple
tries. This is done in nbd_client_close() too, for the same reasons AFAICT.
Actually, nbd_client_close() is different because it's an abortive close of
the socket. nbd_client_close() doesn't care about FIN vs. RST, it does the
shutdown to force all the requests to fail (with either an error for writes,
or a short read if they're receiving). This will cause a flurry of
nbd_client_put() calls soon after nbd_clint_close() returns, until the last
reference is dropped and the socket is closed.
I see, thanks for the explanation.
I'll apply the patch.
Will you apply it directly or should I resend it in v3 ?