cvs commit: src/sys/kern vfs_subr.c

[Jeff Roberson]

On Sun, 5 Oct 2003, Don Lewis wrote:

> On  5 Oct, Jeff Roberson wrote:
>
> > In vflush we do sleep on the vnode lock.  The flags are LK_EXCLUSIVE |
> > LK_INTERLOCK.  We dont pass in LK_NOWAIT.
>
> Ohhh ...
>
> I always thought we wouldn't sleep unless we specified LK_RETRY.  The
> fine man page isn't terribly helpful about when to use some of the
> flags.
>
> If we hold a reference to an active vnode, can it ever be type VT_NON?
> And if it is of type VT_NON, could it ever morph into some other type
> while we hold the reference?  If not, why would we want to specify the
> LK_RETRY flag for such a vnode?  Does this mean that way too many users
> of vn_lock() are specifying the LK_RETRY flag?
>

I think so, I've actually just mailed kirk about this.  If someone does a
forceclose while you hold a reference it will be cleaned and removed.
Holding either the vi lock or the vnode lock prevents this, however.

Cheers,
Jeff