On 06/08/17(Sun) 17:04, Florian Obser wrote: > On Sun, Aug 06, 2017 at 10:29:09AM -0400, Martin Pieuchot wrote: > > On 03/08/17(Thu) 09:13, Florian Obser wrote: > > > > > > as mpi pointed out in "nd6 address expiration & NET_LOCK() contention" > > > we run nd6_expire every second. That seems a bit silly considering > > > that we normally have a pltime of a day. > > > > > > With a bit of math we can work out when the timer should fire when we > > > set pltime/vltime and when we walk the list in nd6_expire. > > > > > > This intentionally only lowers the timeout in in6_update_ifa(). That > > > means in the case of no privacy addresses and default pltime/vltime > > > and no router advertisements getting lost the timer fires every week > > > and notices that it has nothing to do. I think that is acceptable > > > since it fires 600.000 times less... > > > > > > Tests, comments, OKs? > > > > I like it, comments below. > > next try, but I probably missunderstood your comments.
A bit, more comments below :) > If I try to printf-debug this in nd6_expire_timer_update() Are you trying to print something at an invalid address? > panic(ffff8000221a6480,ffff8000221bd800,ffff8000221b9000,ffffffff8197bca5,0,fff > f8000221bd698) at panic+0xab > __mtx_enter(6,9791cee0ccf8fbcd,ffffff013f60f800,ffff8000221b9000,15000,9791cee0 > ccf8fbcd) at __mtx_enter+0x62 Here trap() calls printf() which tries to grab a mutex. However the current CPU is already holding this mutex because it faulted *inside* printf(). Now __mtx_enter() will want to call panic() which wants to call printf(), hence the loop. > trap() at trap+0x51f > --- trap (number 6) --- > strlen(ffff800000106290,ffff8000003e8c00,ffff8000221bdda0,1,ffffffff81a558f6,97 > 91cee0ccf8fbcd) at strlen+0x20 strlen() faulted means that you gave something invalid to printf(). > printf(0,9791cee0ccf8fbcd,1,ffff8000221bd9b0,ffff8000221bda20,3000000010) at > pr > intf+0x6b > in6_update_ifa(ffff8000221bdda0,0,ffff800000106290,8080691a,0,9791cee0ccf8fbcd) > at in6_update_ifa+0xa5f > in6_ioctl(6,8080691a,ffff800000106290,ffff8000221a6480,ffff8000221bdda0,ffff800 > 000106290) at in6_ioctl+0x4ed > ifioctl(ffff8000221a6480,ffff8000221bdda0,8004667d,ffffff013f2a4bc0,8080691a,97 > 91cee0ccf8fbcd) at ifioctl+0x5e0 > soo_ioctl(80,ffffff01207d3850,8004667d,ffff8000221bde90,ffff8000221bdda0,ffffff > ff81ab4200) at soo_ioctl+0x163 > sys_ioctl(ffff8000221bdf20,360,ffff8000221a6480,0,7f7ffffe0370,9791cee0ccf8fbcd > ) at sys_ioctl+0x353 > syscall() at syscall+0x251 > --- syscall (number 54) --- > end of kernel > end trace frame: 0x7f7ffffe0310, count: -303 > diff --git sys/netinet6/in6.c sys/netinet6/in6.c > index cafdd9fe36f..6f37a0a3418 100644 > --- sys/netinet6/in6.c > +++ sys/netinet6/in6.c > @@ -497,6 +497,7 @@ in6_update_ifa(struct ifnet *ifp, struct in6_aliasreq > *ifra, > struct in6_addrlifetime *lt; > struct in6_multi_mship *imm; > struct rtentry *rt; > + time_t expire_time = -1; > char addr[INET6_ADDRSTRLEN]; > > NET_ASSERT_LOCKED(); > @@ -686,6 +687,10 @@ in6_update_ifa(struct ifnet *ifp, struct in6_aliasreq > *ifra, > */ > ia6->ia6_flags = ifra->ifra_flags; > > + nd6_next_expire_time(ia6, &expire_time); > + > + nd6_expire_timer_update(expire_time, 0); I was suggesting to put everything in one function.
