[linux-cifs-client] Re: sendmsg blocking with sendtimeout vs. non-blocking

[Jim Rees]

Jeff Layton wrote:

  It sounds like what we should do for CIFS is the same thing. Set the
  buffer sizes but make sure the SOCK_*BUF_LOCK bits are cleared so that
  they can grow as needed.

You have to be careful about when you set them.  The old nfs code does it in
the wrong place, just before the first read.  That's too late.  You have to
do it just after accept().

  Does the kernel ever shrink these buffers?

Not now, but it could in the future.  Yet another reason this should be an
actual socket (or tcp) call with well defined semantics.

By the way, I'm pretty sure you never have to set sk_sndbuf, as the tcp code
will grow the send buffer as needed.