Messages in this thread | | | Date | Wed, 4 Jun 2008 09:42:58 +0300 (EEST) | From | "Ilpo Järvinen" <> | Subject | Re: [fixed] [patch] Re: [bug] stuck localhost TCP connections, v2.6.26-rc3+ |
| |
On Tue, 3 Jun 2008, Patrick McManus wrote:
> On Wed, 2008-06-04 at 02:22 +0300, Ilpo Järvinen wrote: > == > > -- > > [PATCH] tcp DEFER_ACCEPT: fix racy access to listen_sk > > > > It seems that replacement of DA code also moved parts outside > > of appropriate locking. The Ingo's problem seems to come from > > the fact that two flows could now race in > > (inet_csk_)reqsk_queue_add corrupting the queue. ...This can > > leave dangling socks around which won't resolve themselves > > without stimuli from outside (e.g., external RST would help > > I think).
> do_rcv() clearly has the listening socket locked in the non-DA case, and > in the DA case it is the 'child' ESTABLISHED socket that is locked - > leaving the accept queue unprotected. So simple.
It also well explains why Ingo finally got the KERNEL: assertion (!sk->sk_ack_backlog) which is adjusted in the very same reqsk_queue_add making it and the actual queue come out of sync.
...too bad this has no relevance to Håkon's case, so more digging is necessary... :-)
-- i. | |