On Wed, 24 Oct 2001, Jonathan Lemon wrote:
> Hmm. The way the revamped console code works is this:
>
> cn_devopen() calls vn_open() to open the device. If this is not a
> VCHR device, then it is closed. Otherwise, the vnode is stashed in
> cnd->cnd_vp.
>
> When the device is closed though cnclose(), it walks through a list
> of console devices, and if cnd_vp != NULL, calls vn_close() for that
> vnode, and then NULLs out cnd_vp. (drops the reference.)
>
> My understanding is that vn_open/vn_close will track the count of
> outstanding references to the vnode, so this should be safe to do.
The reference counts were inadequate for handling controlling terminals
in the old version. I don't see how this can work now. You do extra
opens. This will prevent controlling ttys going away properly. E.g.,
suppose sio0 is a console device and is completely closed. The following
sequence of operations should leave it a non-controlling tty:
sio0's ref count
open sio0 1
make sio0 a ctty 2
open cn 3
close sio0 2
lose sio0 as a ctty 1
sioclose() will not be called here since the reference count is > 0,
so ctty-related stuff in its tty struct will not be cleared, and things
break.
Similarly with sio0 and cn reversed.
The old code had kludges to handle one of these cases. I had fixes
for the other one. These are now invalid (sigh). I didn't commit
them because there seemed to be a lot of races nearby. Perhaps the
panic is caused a race. The races are related to stashing of pointers
related to controlling terminals. The lifetime of these pointers is
unclear.
Bruce
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message
