Date: Sun, 10 Feb 2019 12:03:14 +0100 (CET) From: sthaug@nethelp.no To: freebsd-current@freebsd.org Cc: bz@freebsd.org Subject: 64-bit integer overflow computing user CPU time in calcru1() in kern_resource.c Message-ID: <20190210.120314.255774056.sthaug@nethelp.no>
next in thread | raw e-mail | index | archive | help
There is a 64-bit integer overflow computing user cpu time in calcru1() in kern_resource.c. This was discovered because CPU statistics from the PowerDNS-recursor name server stopped working (essentially, got "stuck") after a while: time_t milliseconds 1547818832 301274008.418503 1547822864 301784302.665002 1547826896 302310096.107672 1547830928 302844638.859146 1547834960 303381189.070208 1547838992 303924399.662413 1547843024 304477529.572919 1547847056 305025750.193424 1547851088 305544141.140036 1547855120 306001630.092938 1547859152 306153010.535298 1547863184 306141696.000000 1547867216 306141696.000000 1547871248 306141696.000000 1547875280 306141696.000000 1547879312 306141696.000000 1547883344 306141696.000000 1547887376 306141696.000000 Note that the number just stops increasing beyond 1547863184. I complained about this on the Pdns-user mailing list, https://mailman.powerdns.com/pipermail/pdns-users/2019-January/025739.html and received help from Bert Hubert of PowerDNS to find https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=227689 and it definitely looks like this is the bug causing the disappearing CPU statistics graphs. I fixed the problem by following the link from the FreeBSD bug above to https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=76972 which has an attachment (kern_resource.c.patch.txt) at https://bz-attachments.freebsd.org/attachment.cgi?id=50537 After adding this patch and rebooting, the problem has not reoccurred. I assume Bjorn Zeeb's patch has indeed fixed the problem. Note that the problem was discovered in 11.2-STABLE r338949 - however, looking at https://svnweb.freebsd.org/base/head/sys/kern/kern_resource.c?view=markup exactly the same user CPU time code seems to be present in HEAD, so I assume the same overflow is also present. I sent a message about this problem on the FreeBSD-stable mailing list recently, https://lists.freebsd.org/pipermail/freebsd-stable/2019-February/090523.html but with no reaction there I'm now trying FreeBSD-current. I'm hoping for bz's patch to be applied to HEAD, and at some point an MFC to 11.2-STABLE. Steinar Haug, Nethelp consulting, sthaug@nethelp.no
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20190210.120314.255774056.sthaug>