First impressions of lua(jit) performance in Suricata

Today I decided to look into the potential performance of the luajit keyword a bit. It’s important to know if this can perform at reasonable speeds so that we can actually use it in real deployments. Even if we can’t the feature may still be appealing though, for offline pcap analysis.

So far, the results are rather encouraging.

First, I added 2 buffers today: http.uri, which contains the normalized uri (same buffer as the http_uri content modifier inspects) and http.request_line, which is the request line given to us by libhtp. This contains method, separators, uri, HTTP version.

Next I created 5 rules. A pure Lua rule (1), a Lua rule with content prefilter (2), a Lua rule with content and pcre prefilter (3), a pcre rule with content prefilter (4) and a pure pcre rule (5).

alert http any any -> any any (msg:"LUAJIT HTTP POST test, pure lua"; luajit:test2.lua; sid:1;)
alert http any any -> any any (msg:"LUAJIT HTTP POST test, content prefilter"; content:"POST"; http_method; content:".php"; http_uri; luajit:test2.lua; sid:2;)
alert http any any -> any any (msg:"LUAJIT HTTP POST test, pcre prefilter"; content:"POST"; http_method; content:".php"; http_uri; pcre:"/^POST\s+\/.*\.php\s+HTTP\/1\.0\r\n/m"; luajit:test2.lua; sid:3;)
alert http any any -> any any (msg:"LUAJIT HTTP POST test, pcre no lua"; content:"POST"; http_method; content:".php"; http_uri; pcre:"/^POST\s+\/.*\.php\s+HTTP\/1\.0\r\n/m"; sid:4;)
alert http any any -> any any (msg:"LUAJIT HTTP POST test, pure pcre"; pcre:"/^POST\s+\/.*\.php\s+HTTP\/1\.0\r\n/m"; sid:5;)

and the following Lua script:

function init (args)
    local needs = {}
    needs["http.request_line"] = tostring(true)
    return needs
end

-- match if packet and payload both contain HTTP
function match(args)
    a = tostring(args["http.request_line"])
    if #a > 0 then
        if a:find("^POST%s+/.*%.php%s+HTTP/1.0$") then
            return 1
        end
    end
  
    return 0
end

return 0

The script does a pattern match (regex even) against the request line, something I’d consider quite expensive.

So, how does this perform? Here are the rule perf stats:

   Num      Rule        Avg Ticks   Avg Match   Avg No Match
  -------- ------------ ----------- ----------- -------------- 
  1        5            12113.53    7198.08     12114.28   
  2        3            11638.15    39842.23    9424.83    
  3        2            10682.71    35497.08    10194.56   
  4        1            8812.31     15841.85    8807.01    
  5        4            8536.46     20074.97    7630.97 

Pure pcre rules are bad, we all know that, but they end up being most expensive in this test which surprises me. The pure Lua rule is quite a bit cheaper and even ends up below the prefilted Lua rules. Only the content+pcre (no Lua) rule is slightly faster.

So far things look rather good for the lua keyword. Who knows, maybe it can even be used on live traffic.

The work continues! :)

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s