Date: Mon, 25 Jul 2005 01:23:22 +0000 From: "Christian S.J. Peron" <csjp@FreeBSD.org> To: Robert Watson <rwatson@FreeBSD.org> Cc: cvs-src@FreeBSD.org, src-committers@FreeBSD.org, cvs-all@FreeBSD.org Subject: Re: cvs commit: src/sys/net bpf.c bpfdesc.h Message-ID: <20050725012322.GB54033@freefall.freebsd.org> In-Reply-To: <20050725001935.B48825@fledge.watson.org> References: <200507241721.j6OHLImZ032073@repoman.freebsd.org> <20050725001935.B48825@fledge.watson.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, Jul 25, 2005 at 12:24:44AM +0100, Robert Watson wrote: > > Looks like you hold bpf_mtx over calls to SYSCTL_OUT(), which may sleep if > it is required to write to a user memory page that is not in memory. > This can result in a lot of nasty things, including deadlock, odd lock > orders (especially if the page fault results in a signal being delivered > to a process), etc. In general, monitoring code of this sort needs to > store its output into a temporary kernel buffer and then copy that out, or > it needs to drop mutexes and accept race conditions. Generally the former > is easier to program, and the latter uses less kernel memory. > > Also, because the bpf_mtx isn't held between the first and second tests of > bpf_bpfd_cnt, a race can occur resulting in a panic when the kassert > fails, if the count is elevated before the call to hold the mutex, and not > once the mutex is released by the other thread. Does the kassert actually > add value here? In the unusual event of a race, you do a slightly more > expensive list walk, but only in rare cases. With the incorrect > KASSERT(), you panic instead. > > Robert N M Watson Robert, good catch, I will fix this up ASAP. -- Christian S.J. Peron csjp@FreeBSD.ORG FreeBSD Committer
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050725012322.GB54033>