|
php.net | support | documentation | report a bug | advanced search | search howto | statistics | random bug | login |
[2009-01-06 14:05 UTC] kulminaator at gmail dot com
Description:
------------
print_r creates output although it was not asked to
The code
$str = print_r($expression, true);
should never produce any output, but it does if php hits the memory limit closely enough.
My memory_limit is set to 128M (may be this has to be the same to reproduce the bug).
Needless to say, this is a security issue and may output information to clients that never should go there by the code.
Reproduce code:
---------------
<?php
// php.ini has memory limit at 128M
$limit = (str_replace('M','',ini_get('memory_limit')))*1024*1024;
print "Memory limit is $limit bytes\n";
$data = str_repeat('x', $limit / 3 );
$x = print_r($data, true);
Expected result:
----------------
Out of memory error or no output at all.
Actual result:
--------------
The huge amount of x-es (xxxxxxxxxxxx about 40 million times) followed by
Fatal error: Allowed memory size of 134217728 bytes exhausted (tried to allocate 44739243 bytes) in /private/tmp/proof_of_concept.php on line 6
PatchesPull RequestsHistoryAllCommentsChangesGit/SVN commits
|
|||||||||||||||||||||||||||||||||||||
Copyright © 2001-2025 The PHP GroupAll rights reserved. |
Last updated: Mon Dec 01 15:00:01 2025 UTC |
After looking around a bit, this extra output of xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx like data is caused by the weird behaviour of ob_start/ob_end and the fact that output buffers are always flushed when a fatal error is coughed up from below. straight from the source of ext/standard/basic_functions.c PHP_FUNCTION(print_r) { zval *var; zend_bool i = 0; if (zend_parse_parameters(ZEND_NUM_ARGS() TSRMLS_CC, "z|b", &var, &i) == FAILURE) { RETURN_FALSE; } if (i) { php_start_ob_buffer (NULL, 0, 1 TSRMLS_CC); } zend_print_zval_r(var, 0 TSRMLS_CC); if (i) { php_ob_get_buffer (return_value TSRMLS_CC); php_end_ob_buffer (0, 0 TSRMLS_CC); } else { RETURN_TRUE; } } Here the zend_print_zval_r causes creating/appending to ob , ob hits the memory limit and bang, the output is created, although it never should be. Can we hope for a fix (either root out the issue of ob being flushed or avoiding ob in the first place ?) or are supposed to remove all sensitive data from ever reaching print_r ?