Date: Fri, 16 May 2003 09:45:56 -0700 (PDT) From: Don Lewis <truckman@FreeBSD.org> To: bde@zeta.org.au Cc: current@FreeBSD.org Subject: Re: CFR: fifo_open()/fifo_close() patch Message-ID: <200305161646.h4GGjuM7058624@gw.catspoiler.org> In-Reply-To: <20030516233858.U12541@gamplex.bde.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 16 May, Bruce Evans wrote:
> On Fri, 16 May 2003, Don Lewis wrote:
>
>> There are a few problems in the fifo_open() and fifo_close()
>> implementations.
>> ...
>> This patch makes the following changes:
>>
>> Create fifo_inactive() and free the fifo data structures there
>> instead of in fifo_close() to eliminate the need for fifo_open()
>> call fifo_close() in some of the failure cases. This also
>> eliminates the need for the vrefcnt() call in fifo_close().
>>
>> Protect fip->fi_{readers,writers} with the vnode interlock in both
>> fifo_open() and fifo_close().
>>
>> Convert from tsleep() to msleep() using the vnode interlock to
>> eliminate the race condition.
>
> Why not just lock the vnode in fifo_close()? RELENG[2-4] seems to have
> the same bug. I cannot be fixed there using the vnode interlock.
That is probably the proper fix for RELENG4 where finer-grained locking
isn't needed because of Giant. I'd still probably move the resource
deallocation to fifo_inactive().
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200305161646.h4GGjuM7058624>