From 2bf3ce088d05c873f3c134a944acdc57b8657fdc Mon Sep 17 00:00:00 2001 From: Robert Watson Date: Sun, 7 Mar 2010 14:13:59 +0000 Subject: [PATCH] Add comment in tcp_discardcb() talking about how we don't, but should, address TCP races relating to not calling tcp_drain() on stopped callouts. Discussed with: bz --- sys/netinet/tcp_subr.c | 15 +++++++++++++-- 1 file changed, 13 insertions(+), 2 deletions(-) diff --git a/sys/netinet/tcp_subr.c b/sys/netinet/tcp_subr.c index d0b2e0e29c5a..39755f5d8eb1 100644 --- a/sys/netinet/tcp_subr.c +++ b/sys/netinet/tcp_subr.c @@ -835,8 +835,19 @@ tcp_discardcb(struct tcpcb *tp) INP_WLOCK_ASSERT(inp); /* - * Make sure that all of our timers are stopped before we - * delete the PCB. + * Make sure that all of our timers are stopped before we delete the + * PCB. + * + * XXXRW: Really, we would like to use callout_drain() here in order + * to avoid races experienced in tcp_timer.c where a timer is already + * executing at this point. However, we can't, both because we're + * running in a context where we can't sleep, and also because we + * hold locks required by the timers. What we instead need to do is + * test to see if callout_drain() is required, and if so, defer some + * portion of the remainder of tcp_discardcb() to an asynchronous + * context that can callout_drain() and then continue. Some care + * will be required to ensure that no further processing takes place + * on the tcpcb, even though it hasn't been freed (a flag?). */ callout_stop(&tp->t_timers->tt_rexmt); callout_stop(&tp->t_timers->tt_persist);