Date: Thu, 2 Apr 2015 14:23:00 +0000 (UTC) From: Gleb Smirnoff <glebius@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r280989 - head/sys/netinet Message-ID: <201504021423.t32EN0Qu098229@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: glebius Date: Thu Apr 2 14:22:59 2015 New Revision: 280989 URL: https://svnweb.freebsd.org/changeset/base/280989 Log: Provide a comment explaining issues with the counter(9) trick, so that people won't copy and paste it blindly. Prodded by: ian Sponsored by: Nginx, Inc. Modified: head/sys/netinet/ip_id.c Modified: head/sys/netinet/ip_id.c ============================================================================== --- head/sys/netinet/ip_id.c Thu Apr 2 13:51:06 2015 (r280988) +++ head/sys/netinet/ip_id.c Thu Apr 2 14:22:59 2015 (r280989) @@ -254,6 +254,20 @@ ip_fillid(struct ip *ip) ip->ip_id = ip_randomid(); else { counter_u64_add(V_ip_id, 1); + /* + * There are two issues about this trick, to be kept in mind. + * 1) We can migrate between counter_u64_add() and next + * line, and grab counter from other CPU, resulting in too + * quick ID reuse. This is tolerable in our particular case, + * since probability of such event is much lower then reuse + * of ID due to legitimate overflow, that at modern Internet + * speeds happens all the time. + * 2) We are relying on the fact that counter(9) is based on + * UMA_ZONE_PCPU uma(9) zone. We also take only last + * sixteen bits of a counter, so we don't care about the + * fact that machines with 32-bit word update their counters + * not atomically. + */ ip->ip_id = htons((*(uint64_t *)zpcpu_get(V_ip_id)) & 0xffff); } }
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201504021423.t32EN0Qu098229>