Date: Tue, 21 Oct 2003 15:19:28 -0700 (PDT) From: Sam Leffler <sam@FreeBSD.org> To: Perforce Change Reviews <perforce@freebsd.org> Subject: PERFORCE change 40125 for review Message-ID: <200310212219.h9LMJSW9013165@repoman.freebsd.org>
next in thread | raw e-mail | index | archive | help
http://perforce.freebsd.org/chv.cgi?CH=40125 Change 40125 by sam@sam_ebb on 2003/10/21 15:19:09 apply bandaid to locking problem in divert sockets Affected files ... .. //depot/projects/netperf/sys/netinet/ip_divert.c#6 edit Differences ... ==== //depot/projects/netperf/sys/netinet/ip_divert.c#6 (text+ko) ==== @@ -219,6 +219,20 @@ m->m_pkthdr.rcvif->if_unit); } + /* + * XXX sbappendaddr must be protected by Giant until + * we have locking at the socket layer. When entered + * from below we come in w/o Giant and must take it + * here. Unfortunately we cannot tell whether we're + * entering from above (already holding Giant), + * below (potentially without Giant), or otherwise + * (e.g. from tcp_syncache through a timeout) so we + * have to grab it regardless. This causes a LOR with + * the tcp lock, at least, and possibly others. For + * the moment we're ignoring this. Once sockets are + * locked this cruft can be removed. + */ + mtx_lock(&Giant); /* XXX */ /* Put packet on socket queue, if any */ sa = NULL; nport = htons((u_int16_t)port); @@ -240,6 +254,7 @@ INP_UNLOCK(inp); } INP_INFO_RUNLOCK(&divcbinfo); + mtx_unlock(&Giant); /* XXX */ if (sa == NULL) { m_freem(m); ipstat.ips_noproto++;
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200310212219.h9LMJSW9013165>