Very Computer

by **avnindr** » Thu, 21 Jan 1999 04:00:00

Hello,

Can anybody tell me why we are specifying 'close' and 'open'
routines of a STREAMS device driver, are normally processed in read
queue instead of the write queue of the streams queue pair?
Thanks in advance,
Avnindra
--
Surf Usenet at home, on the road, and by email -- always at Talkway.
http://www.talkway.com

by **Doug Hocki** » Fri, 22 Jan 1999 04:00:00

I don't know what the designers thinking was, but it seems reasonable
to have open and close be consistent, both be delivered on either
the read queue or the write queue.

If you need to get at the write queue just use "WR(q)". On older
Unixes it was a macro producing inline code but on newer ones
with multithreaded streams drivers (e.g. UnixWare) it's a function.

Quote:> Can anybody tell me why we are specifying 'close' and 'open'
> routines of a STREAMS device driver, are normally processed in read
> queue instead of the write queue of the streams queue pair?

-- Doug

by **Casper H.S. Dik - Network Security Engine** » Fri, 22 Jan 1999 04:00:00

[[ PLEASE DON'T SEND ME EMAIL COPIES OF POSTINGS ]]

>I don't know what the designers thinking was, but it seems reasonable
>to have open and close be consistent, both be delivered on either
>the read queue or the write queue.
>If you need to get at the write queue just use "WR(q)". On older
>Unixes it was a macro producing inline code but on newer ones
>with multithreaded streams drivers (e.g. UnixWare) it's a function.

Also, when the close is performed, the write queue is flushed and the
STREAMs are closed from the driver to stream head (i.e., in the direction
of the read queue)

Casper
--
Expressed in this posting are my opinions. They are in no way related
to opinions held by my employer, Sun Microsystems.
Statements on Sun products included here are not gospel and may
be fiction rather than truth.

by **Doug Hocki** » Sat, 23 Jan 1999 04:00:00

Quote:> Also, when the close is performed, the write queue is flushed and the
> STREAMs are closed from the driver to stream head (i.e., in the direction
> of the read queue)

Are you sure? I've seen the stream dismantled and the closes
performed in the opposite order from when it was constructed --
from the stream head down (at least on UnixWare).

-- Doug

by **Casper H.S. Dik - Network Security Engine** » Sun, 24 Jan 1999 04:00:00

[[ PLEASE DON'T SEND ME EMAIL COPIES OF POSTINGS ]]

>> Also, when the close is performed, the write queue is flushed and the
>> STREAMs are closed from the driver to stream head (i.e., in the direction
>> of the read queue)
>Are you sure? I've seen the stream dismantled and the closes
>performed in the opposite order from when it was constructed --
>from the stream head down (at least on UnixWare).

The Solaris manual for close(9e) (9e being the section that describes the
entry points of kernel drivers), says:

The close() routines of STREAMS drivers and modules are
called when a stream is dismantled or a module popped. The
steps for dismantling a stream are performed in the follow-
ing order. First, any multiplexor links present are
unlinked and the lower streams are closed. Next, the fol-
lowing steps are performed for each module or driver on the
stream, starting at the head and working toward the tail:

1. The write queue is given a chance to drain.

2. The close() routine is called.

3. The module or driver is removed from the stream.

(On second reading, I might had the order reversed.)

Casper

--
Expressed in this posting are my opinions. They are in no way related
to opinions held by my employer, Sun Microsystems.
Statements on Sun products included here are not gospel and may
be fiction rather than truth.

Very Computer

Streams device driver - open and close routines

Streams device driver - open and close routines

Streams device driver - open and close routines

Streams device driver - open and close routines

Streams device driver - open and close routines

Streams device driver - open and close routines