iproute2 proposal

[Denys Fedoryshchenko <denys@visp.net.lb>]

[-- Attachment #1: Type: text/plain, Size: 975 bytes --]

While using iproute2 in batch mode i notice that adding 11362 rules, on system 
with 500+ interfaces takes up to 1minute on Xeon 3.0 Ghz.
After "oprofiling" i got result:

CPU: P4 / Xeon with 2 hyper-threads, speed 2992.84 MHz (estimated)
Counted GLOBAL_POWER_EVENTS events (time during which processor is not 
stopped) with a unit mask of 0x01 (mandatory) count 100000
samples  %        image name               symbol name
86166    57.3473  tc                       ll_remember_index
48103    32.0147  tc                       parse_rtattr
8497      5.6551  tc                       rtnl_dump_filter
1751      1.1654  [vdso] (tgid:22801 range:0xb7f66000-0xb7f67000) (no symbols)
1085      0.7221  tc                       tc_calc_xmittime
672       0.4472  tc                       .plt

I did small patch, and voila! batch finished in 2 seconds successfully!

Proof of concept patch is attached. Not sure if it is not breaking any other 
things. Waiting for any comments.

[-- Attachment #2: iproute2-proposal.patch --]
[-- Type: text/x-diff, Size: 594 bytes --]

diff -Naur iproute2-original/lib/ll_map.c iproute2-patched/lib/ll_map.c
--- iproute2-original/lib/ll_map.c	2008-08-05 19:15:56.000000000 +0000
+++ iproute2-patched/lib/ll_map.c	2008-08-05 22:22:36.000000000 +0000
@@ -23,6 +23,7 @@
 #include "ll_map.h"
 
 extern unsigned int if_nametoindex (const char *);
+static int ifacecache=0;
 
 struct idxmap
 {
@@ -160,6 +161,10 @@
 
 int ll_init_map(struct rtnl_handle *rth)
 {
+	if (ifacecache == 1)
+	    return 0;
+	ifacecache = 1;
+
 	if (rtnl_wilddump_request(rth, AF_UNSPEC, RTM_GETLINK) < 0) {
 		perror("Cannot send dump request");
 		exit(1);