Apache / PHP on Windows crashes with regular expression

Question

I have discovered that the following PHP code crashes when it runs on a Windows Apache server:

<?php

$test = "0,4,447,11329,316,415,142,5262,6001,9527,11259,11236,1140,9770,9177,9043,11090,268,5270,9907,9196,10226,".
        "9399,1476,9368,6929,1659,11074,10869,8774,739,11344,10482,6399,8910,8997,11198,137,10148,10371,5128,767,2483,".
        "211,9973,10726,9299,778,11157,9497,275,9418,11141,241,5288,11324,776,10960,9289,7944,222,10686,11428,9723,".
        "10615,11399,9869,9083,10180,10043,9957,9387,9215,8869,9667,10174,10902,6607,9282,1259,395,10411,152,9344,8949,".
        "10923,8976,11042,11519,10704,10979,216,10044,9201,1721,5831,881,9721,1757,11054,1335,6151,9526,9081,111,498,".
        "2960,438,5313,206,318,10820,8192,6039,9161,11012,1717,1360,10757,4314,11280,9647,9346,10546,11006,9553,10365,".
        "6148,10565,4532,2776,4124,8853,6145,478,4539,540,9981,726,7186,11122,324,10524,1139,7900,9581,6869,1724,10851,".
        "10059,10018,11032,1290,3818,782,796,917,8740,6935,11439,10799,10948,249,2068,8778,6289,295,2766,9425,791,309,".
        "4753,10418,771,260,10835,10441,6434,10164,10475,10842,9013,11224,2247,8972,2141,2078,2152,475,9077,6291,10285,".
        "8067,753,6660,10889,431,2503,6007,9180,810,11447,2461,3689,7104,10150,10921,895,10598,747,10570,305,4497,11055,".
        "11496,10938,10722,8761,10086,11482,6780,6685,6918,10286,10659,9996,4074,9118,907,5192,283,2230,8884,6966,".
        "8820,8132,3598,9599,6796,11257,7049,5992,8637,4168,9017,7950,7165,10721,10037,1071,8044,759,11429,6380,".
        "10239,1593,9455,9704,10357,6737,2958,4051,9754,6564,11407,8716,7485,1528,6857,7406,9579,7259,1609,7820,".
        "4448,10289,1123,7005,8123,9316,914,9655,5280,9710,7822,510,10795,10476,8706,6160,8248,6978,9300,10643,".
        "7106,10250,519,7860,4733,904,8773,4714,8695,8633,6105,3312,11548,9580,10389,4886,4587,513,8485,4606,".
        "6471,581,526,637,3523,3772,3153,9336,9120,7633,3755,10087,524,10015,8563,556,1230,570,3652,569"
        .",8473,10209,3886,573,5363,4715,3865,9452,1218,7066,575,577,4724,7655"
        ;

$hest = preg_match('/^\d+(?:,\d+)+$/', $test);
var_dump($hest);
?>

Interestingly enough, the code works if the string is a little shorter - if I comment out the last part (line) of the string, the code runs and works as expected.

However, as the code is presented here, the code causes Apache to crash and respawn - no error is logged. If I try to run the code directly through PHP it works as expected, so the problem is somehow related to Apache.

I have experimented with the "pcre.recursion_limit" setting, and found that if I lower this to 689 it doesn't die, but instead the preg_match() call fails with a PREG_RECURSION_LIMIT_ERROR. For higher values of pcre.recursion_limit Apache dies.

Apache 2.2.11 PHP 5.3.0

I have also tested this on a Debian server where I do NOT see the error, and with different versions of PHP and Apache on Windows, where the error occurs as described above, so it seems that it may be Windows related.

Have anyone seens this before? I would really like a hint of some sort!

-- Anders

UPDATE: I is because of a "bug" in Apache for Windows - well, not really a bug, but it seems that Apache is compiled with a small stacksize, and that causes this error in PCRE when PHP uses it. See http://bugs.php.net/bug.php?id=47689

Answer 1

A crash normally means you've hit a PHP bug. It happens now and then. Unless you're able to upgrade your PHP installation, you'll have to rewrite your code. Anyway, the high recursion level suggests that you should not be using regular expressions or, at least, you should first split the input string in smaller pieces.

Answer 2

It's always a good idea to lower "pcre.recursion_limit", because the default high value can corrupt the process stack (see http://php.net/manual/en/pcre.configuration.php) - this is exactly what happens with your mod_php install. Since preg functions don't throw an error when recursion/backtracking limits are reached, it may be useful to have a wrapper like

function match($re, $text) {
    preg_match($re, $text, $m);
    if(preg_last_error())
         trigger_error("preg: " . preg_last_error());
    return $m;
 }

At least, this lets you know when something goes wrong.

Besides that, try to simplify your patterns when possible, for example /^\d[\d,]*\d$/ does the same as above, but with zero recursion.