svn commit: r320659 - head/usr.sbin/nfsuserd

[Konstantin Belousov]

On Wed, Jul 05, 2017 at 11:58:56AM +0000, Rick Macklem wrote:
> Well, I know nothing about jails, so I can't answer either of these.
> All I can tell you is that 127.0.0.1 no longer works as localhost when
> jails are run for some situations which I don't understand and that
> breaks the nfsuserd upcall. (Something about 127.0.0.1 being replaced
> by the first IP# for the net interface when jails are running?)
When a jailed process uses a local address, the address is effectively
replaced by the jailed address, as you noted above.

> 
> The patches in PR#205193 change nfsuserd to use a local socket,
> which fixes this problem. However, the original poster of the PR had
> hangs when using the patches which was never resolved (I couldn't
> reproduce the hang).
So consider the issue not reproducable and test it in wild by committing
to HEAD.  Ability to do such kind of experiment is one of the reasons why
do we have HEAD and separate release branches.


> --> It has been a couple of years, but I recall that the hang was in the
>       unix socket code, so I believed it was caused by having mutiple
>       processes trying to use the socket concurrently.
>       --> The original poster never confirmed if using a single daemon
>             avoided the hang.
> As such I am stuck until someone can test the patches? rick