virtual()

(PHP 4, PHP 5, PHP 7)

执行 Apache 子请求

说明

virtual(string $filename): bool

virtual()是一个 Apache 特有函数，类似于mod_include中的<!--#include virtual...-->。它执行一个 Apache 子请求。可用于包含一个 CGI 脚本或.shtml文件，或任何其它可通过 Apache 解析的请求。注意对一个 CGI 脚本，该脚本生成合法的 CGI 头，至少必须生成Content-Type头。

为运行子请求，所有缓冲将中止并刷新至浏览器，包括头信息。

、仅在将 PHP 安装为 Apache 模块时，支持此函数。

参数

$filename

virtual命令将执行的文件

返回值

成功执行 virtual 命令，或失败时返回FALSE。

更新日志

范例

示例请看apache_note()。

注释

Warning

查询字符串可被传递至被包含文件，但是$_GET是拷贝于父文件，仅有$_SERVER['QUERY_STRING']将填充传递入的查询字符串。且此查询字符串只在使用 Apache 2 时被填充。此请求文件将不会显示在 Apache 访问日志中。

Note:

在被请求文件中设置的环境变量在原请求文件中不可见。

参见

• apache_note()取得或设置 apache 请求记录

Problem with most of the scripts posted below is that virtual() flushes the pending headers before making the subrequest. Requesting an image with virtual() still returns a text/html type document.
A workaround is to set the content-type first. But that requires getting the content-type first.
I'm using the following script for now. A disadvantage is that Apache makes 2 subrequests.
<?PHP
$file = '/resources/7z.gif';
$file_info = apache_lookup_uri($file);
header('content-type: ' . $file_info -> content_type);
virtual($file);
die();
?>

I have a header that's include()'ed on each of my pages. I then wanted to add a Perl script to that header file (header.php) via the virtual() command. 
Since my header is used by documents in my /www folder along with other folders inside that (and inside those), and virtual() seems to take only relative paths, I had to write some code to dynamically get the path to the perl script.
Hope this helps some ppl out:
 $cwd = getcwd();
 $script_name = "cgi-bin/perl_script.pl";
 $count = substr_count($cwd, '/');
 $count = $count - 3;
 // get rid of extra absolute paths since my directory is /home/user/www
 // Add additional path information
 for($i = 1; $i <= $count; $i++){
  $script_name = "../".$script_name;   
 }
 virtual($script_name);

This documentation is not clear enough. Parameter $filename is not a filename as on the filesystem, but a URI. It can be absolute, starting with /, or relative to URI that involved the PHP script which called virtual(). (I.e. if the PHP script that calls virtual() is invoked via PHP require/require_once/include/include_once mechanism and it passes a relative URI to virtual(), then that URI must be relative to the URI of the topmost PHP script on the inclusion stack.)
Not sure how relative URIs work if the request which calls virtual() was processed through Apache rewrite rules.

You can use virtual() to implement your own dispatcher/auth handler in an efficient and effective way.
For instance if you have a bunch of images you would like to be served statically by Apache (its job after all), but with a more com
plex access pattern than mod_access allows you to do (say a MySQL lookup with your app logic), try this simple Apache rule:
 Order Allow,Deny
 Allow from env=PHP_ALLOW
Then in your PHP script, before sending any content or header:
<?php
$image = "/some/URL/path/test.png";
if (client_may_view_image($image)) {
 apache_setenv('PHP_ALLOW', '1');
 if (virtual($image))
  exit(0);
 echo "Ops, failed to fetched granted image $image (hammer your webmaster).\n";
} else
 echo "Sorry buddy, you're not allowed in here.\n";
?>
Of course very Apache-ish, but it's much more efficient and uniform to rely on Apache rather than passthru() and mime_content_type()
 hacks : it does the path lookup and auth/security audit as the admin expects, use the best static serving it can (think 'sendfile')
 and you can even chain your request with another embedded script eg. in mod_perl.

when php is installed as an apache module, this works pretty well for writing your own php preprocessor/information logger. For example, requests to any URI underneath pre.php will first be executed by pre.php, then returned to the user.
<?
$docroot    = $_SERVER['DOCUMENT_ROOT'];
$script_root  = str_replace( basename($_SERVER['SCRIPT_NAME']),'',$_SERVER['SCRIPT_NAME'] );
$script_ext   = substr( $_SERVER['SCRIPT_NAME'], strrpos( $_SERVER['SCRIPT_NAME'],'.' ) );
$fakework_root = $script_root.basename( $_SERVER['SCRIPT_NAME'] ).'/';
$framework_root = $script_root.'_'.basename( $_SERVER['SCRIPT_NAME'], $script_ext ).'/';
$frequest_root = dirname( $framework_root.substr( $_SERVER['PATH_INFO'], 1 )).'/';
$frequest_name = basename( $_SERVER['PATH_INFO'] );
$frequest_ext  = (strrpos($frequest_name,'.')===FALSE ? FALSE : strtolower(substr( $frequest_name, ( strrpos( $frequest_name, '.' )+1 ) ) ) );
$frequest_full = $frequest_root.$frequest_name;
$doc_frequest  = $docroot.$frequest_full;
$doc_framework  = $docroot.$framework_root;
$DO_PARSE = in_array( $frequest_ext, $chk_exts );
if( $DO_PARSE )
{
  $tmpfname = tempnam( $doc_framework.'tmp', 'aj_' ).($frequest_ext? ('.'.$frequest_ext) : '');
  if( ($to_parse=@file_get_contents($doc_frequest))===FALSE )
    $to_parse="404";
  $tmpvname = str_replace( $docroot, '', $tmpfname ); 
  $tmpvname = str_replace( '\\\\', '/', $tmpvname ); 
// - - - - - - - - - - - - - - - - - - - - - - - - - - -
// Do processing of data stored in $to_parse
// - - - - - - - - - - - - - - - - - - - - - - - - - - -
  $to_parse = striptags( $to_parse );
// - - - - - - - - - - - - - - - - - - - - - - - - - - -
  $handle = fopen($tmpfname, "w");
  fwrite($handle, $to_parse);
  fclose($handle);
  @virtual( $tmpvname.$getvars );
  unlink( $tmpfname );
}
else
  @virtual( $frequest_full.$getvars );
?>
So all files in http://server/sub/pre.php/path/ are really located in http://server/sub/_pre/path/
All this needs is some kind of caching mechanism.
But yeah, this could be modified to add watermarks with the image functions, convert to xml with Tidy, check for extensions better with mimeTypes, proxy content with cURL, validate $_SERVER['HTTP_REFERER'] or $_SERVER['HTTP_USER_AGENT'], etc etc
This gives you much more over than the auto_prepend_file, and auto_append_file, for certain functionality
The key is the virtual function _because_ it delivers the modified content with an apache subrequest.

I saw the note above about the length of the query string... but didn't know what it was, so have altered the code so it can post to the script.
Probly only works on nix systems as it makes use of the echo function...
This code also will look evaluate the result, so you can get cgi to dynamically create PHP (probly best to watch out that posted variables do not include script!)
<?
$CGISCRIPT="./cgi-bin/cgiscript.cgi";
 // preparing the arguments passed to this PHP page 
$QSTRING = $QUERY_STRING; 
foreach ($HTTP_POST_VARS as $header=> $value ){
 if($QSTRING==""){
  $QSTRING = $header.'='.urlencode($value); 
 }else{
  $QSTRING = $QSTRING.'&'.$header.'='.urlencode($value); 
 }
}
 
putenv('REQUEST_METHOD=POST'); 
putenv('CONTENT_TYPE=application/x-www-form-urlencoded');
putenv('CONTENT_LENGTH='.strlen($QSTRING));
putenv('QUERY_STRING='.$QSTRING); 
unset($return_array); 
exec('echo "'.$QSTRING.'"| '.$CGISCRIPT, $return_array, $return_val); 
//The 1st line of my script was "Content...." ... so remove it!
$firstline=array_shift($return_array);
//evaluate the code
eval('?>'.implode($return_array,''));
?>

Another way of passing arguments:
If you have some CGI programs that depend on some libraries where you can't change the source code (in my case an online payment library), you can pass the arguments by changing some environment variables.
Of course the CGI program has to get the GET/POST variables in the usual manner.
It simulates, more or less, a direct call from the server to a CGI program:
// preparing the arguments passed to this PHP page
$QSTRING = $QUERY_STRING;
// pay attention to the maximum length of the QUERY string.
while (list ($header, $value) = each ($HTTP_POST_VARS)){
 if (empty($QSTRING))
  $QSTRING = $header.'='.$value;
 else
  $QSTRING = $QSTRING.'&'.$header.'='.$value;
} 
putenv('REQUEST_METHOD=GET');
putenv('QUERY_STRING='.$QSTRING);
unset($return_array);
exec('my_CGI', $return_array, $return_val);
Now you can parse the output of 'my_CGI' in return_array.

If you want to pass all post and get values to the cgi script you can use this code:
<?php
$QSTRING = $QUERY_STRING;
while (list ($header, $value) = each ($HTTP_POST_VARS)) 
{
  $QSTRING = $QSTRING.'&'.$header.'='.$value;
}
virtual($script.'?'.$QSTRING);
?>
It takes all the values of $HTTP_POST_VARS and appends them in the proper format to the values you get in $QUERY_STRING

Virtual returns the HTTP entity header after the requested file, when it's the first output to the page.
The work-around to prevent seeing the header is, of course, to output something (such as echo " "; ) before calling virtual.

Note that QUERY_STRING seems to get inherited, so to make a virtual request WITHOUT one, one needs to explicitly append a "?" to the URL of the sub-request (to cause the creation of a "null" query string). Of course, if the desired URL has its own query string, that will override and an additional "?" should not be appended.
This was with PHP 4.4.7 (released May 2007).

Here's an update to tomwk's code:
function safe_virtual( $filename )
{
  $curDir = getcwd();
  virtual ( $filename );
  chdir( $curDir );
}
This is better if you've already changed your current directory to be something other than your script's directory. It works for PHP4 and above.

If you are having problems using virtual include becuase of files being stored in different directories, a root-relative path will make things much easier:
virtual ("/root directory/directory/filename.htm/");
where root directory is the root directory of your site (ask your sys admin if you dont know what it is) Don't include the protocol or host name.
This will also allow you to move your files around your site without having to redirect your includes which is *very* helpfull

While the virtual() function has it's promising sides, there exists issues when using it in relation to a cache system such as eAccellerator. The issue becomes that the first time you load with a virtual file, it will seem to work fine. But once the cache is put into play, the virtual call will end up returning nothing at all, and basically returning a blank page.
Please realize that this is not an issue with virtual(), but instead an issue of the cacheing application. Should anyone else have a similar issue, hopefully this will shed some light onto this subject.
Jonathon Hibbard