On Thu, Feb 19, 2015 at 09:30:40PM +0100, Alexander Bluhm wrote:
> On Wed, Feb 18, 2015 at 12:14:15PM +0100, Matthieu Herrb wrote:
> > Feb 18 12:09:59 castor /bsd: splassert: rtrequest1: want 5 have 0
> > Feb 18 12:09:59 castor /bsd: Starting stack trace...
> > Feb 18 12:09:59 castor /bsd: splassert_check() at splassert_check+0x78
> > Feb 18 12:09:59 castor /bsd: rtrequest1() at rtrequest1+0x5e
> > Feb 18 12:09:59 castor /bsd: nd6_prefix_offlink() at
> > nd6_prefix_offlink+0x1bf
> > Feb 18 12:09:59 castor /bsd: pfxlist_onlink_check() at
> > pfxlist_onlink_check+0x25e
> > Feb 18 12:09:59 castor /bsd: in6_control() at in6_control+0x894
> > Feb 18 12:09:59 castor /bsd: ifioctl() at ifioctl+0x175
> > Feb 18 12:09:59 castor /bsd: sys_ioctl() at sys_ioctl+0x169
> > Feb 18 12:09:59 castor /bsd: syscall() at syscall+0x297
> > Feb 18 12:09:59 castor /bsd: --- syscall (number 54) ---
> > Feb 18 12:09:59 castor /bsd: end of kernel
> > Feb 18 12:09:59 castor /bsd: end trace frame: 0xc8115948400, count:
> > 249
> > Feb 18 12:09:59 castor /bsd: 0xc8115715cda:
> > Feb 18 12:09:59 castor /bsd: End of stack trace.
> > Feb 18 12:10:00 castor /bsd: carp0: state transition: BACKUP -> MASTER
>
> Most calls to pfxlist_onlink_check() are protected by splsoftnet.
> Only the path in your trace does not set it. So I suggest to set
> splsoftnet() in in6_control(). I have included the dohooks() as
> this is done in IPv4. While there I have moved some splsoftnet()
> hiding in the declarations to the beginning of the code.
>
> ok?
This fixes the issue (which was reproducible) for me. so ok as far as
I understand the issue.
>
> bluhm
>
> Index: netinet6/in6.c
> ===================================================================
> RCS file: /data/mirror/openbsd/cvs/src/sys/netinet6/in6.c,v
> retrieving revision 1.152
> diff -u -p -r1.152 in6.c
> --- netinet6/in6.c 27 Jan 2015 10:34:27 -0000 1.152
> +++ netinet6/in6.c 19 Feb 2015 18:47:06 -0000
> @@ -552,6 +552,7 @@ in6_control(struct socket *so, u_long cm
> pr->ndpr_refcnt++;
> }
>
> + s = splsoftnet();
> /*
> * this might affect the status of autoconfigured addresses,
> * that is, this address might make other addresses detached.
> @@ -559,6 +560,7 @@ in6_control(struct socket *so, u_long cm
> pfxlist_onlink_check();
>
> dohooks(ifp->if_addrhooks, 0);
> + splx(s);
> break;
> }
>
> Index: netinet6/nd6_rtr.c
> ===================================================================
> RCS file: /data/mirror/openbsd/cvs/src/sys/netinet6/nd6_rtr.c,v
> retrieving revision 1.97
> diff -u -p -r1.97 nd6_rtr.c
> --- netinet6/nd6_rtr.c 27 Jan 2015 03:17:36 -0000 1.97
> +++ netinet6/nd6_rtr.c 19 Feb 2015 17:39:18 -0000
> @@ -707,10 +707,10 @@ defrouter_reset(void)
> void
> defrouter_select(void)
> {
> - int s = splsoftnet();
> struct nd_defrouter *dr, *selected_dr = NULL, *installed_dr = NULL;
> struct rtentry *rt = NULL;
> struct llinfo_nd6 *ln = NULL;
> + int s = splsoftnet();
>
> /*
> * This function should be called only when acting as an autoconfigured
> @@ -1139,12 +1139,13 @@ prelist_update(struct nd_prefix *new, st
> struct ifaddr *ifa;
> struct ifnet *ifp = new->ndpr_ifp;
> struct nd_prefix *pr;
> - int s = splsoftnet();
> - int error = 0;
> + int s, error = 0;
> int tempaddr_preferred = 0, autoconf = 0, statique = 0;
> int auth;
> struct in6_addrlifetime lt6_tmp;
> char addr[INET6_ADDRSTRLEN];
> +
> + s = splsoftnet();
>
> auth = 0;
> if (m) {
>
--
Matthieu Herrb
