In message <200007111440.SAA16159@ms2.inr.ac.ru> you write: > I see no problems though. All input functions assume that they > are called from bh context. If netfilter would not use them > doing output, it would make sense to select only cases > when disabling bh is really required. But it uses, so that > reinject has to be locked always. Nothing to debug. I agree with Alexey. This is once off error. FYI it used to be fine until BR_NETPROTO_LOCK addition. Someone screwed up adding br_read_lock to nf_reinject in 2.3.49 without adding unlock. I forwarded James Morris's fix (in the wrong place) without checking it, which was added in 2.3.50. My bad, Rusty. --- linux-2.4.0-test3-7/net/core/netfilter.c.~1~ Thu Apr 27 12:57:41 2000 +++ linux-2.4.0-test3-7/net/core/netfilter.c Wed Jul 12 07:13:47 2000 @@ -512,7 +512,6 @@ info->indev, info->outdev, &elem, info->okfn); } - br_read_unlock_bh(BR_NETPROTO_LOCK); switch (verdict) { case NF_ACCEPT: @@ -527,6 +526,7 @@ kfree_skb(skb); break; } + br_read_unlock_bh(BR_NETPROTO_LOCK); /* Release those devices we held, or Alexey will kill me. */ if (info->indev) dev_put(info->indev); -- Hacking time. - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.rutgers.edu
