In response to the solution posted in the comment below, there are some practical issues with this solution that must be kept in mind and handled by your code. I developed an application using a similar "use-it-once" key to manage sessions and it worked great but we got some complaints about legitimate users getting logged out without reasons. Turns out the problem was not tentative highjacking, it was either:
A- Users double click on links or make 2 clicks very fast. The same key is sent for the 2 clicks because the new key from the first click didn't get to the browser on time for the second one but the session on the server did trash the key for the new one. Thus, the second click causes a termination of the session. (install the LiveHttpHeaders extension on firefox and look at the headers sent when you click twice very fast, you'll see the same cookie sent on both and the new cookie getting back from the server too late).
B- For any given reason, the server experiences a slow down and the response with the new key (which has replaced the old one on the server) is not returned to the browser fast enough. The user gets tired of waiting and clicks somewhere else. He gets logged out because this second click send the old key which won't match the one you have on your server.
Our solution was to set up a grace period where the old key was still valid (the current key and the previous key were both kept at all times, we used 15 seconds as a grace period where the old key could still be used). This has the drawback of increasing the window of time for a person to highjack the session but if you tie the validity of the old key to an IP address and/or user agent string, you still get pretty good session security with very very few undesired session termination.
章 35. cookies
PHP 透明地支持 HTTP cookie。cookie 是一种在远程浏览器端储存数据并以此来跟踪和识别用户的机制。可以用 setcookie() 或 setrawcookie() 函数来设置 cookie。cookie 是 HTTP 标头的一部分，因此 setcookie() 函数必须在其它信息被输出到浏览器前调用，这和对 header() 函数的限制类似。可以使用输出缓冲函数来延迟脚本的输出，直到按需要设置好了所有的 cookie 或者其它 HTTP 标头。
任何从客户端发送的 cookie 将会像 GET 和 POST 数据一样被自动的转换为 PHP 变量，该过程受 register_globals 和 variables_order 两个配置选项的影响。如果希望对一个 cookie 变量设置多个值，则需在 cookie 的名称后加 [] 符号。
在 PHP 4.1.0 及更高版本，自动全局变量数组 $_COOKIE 将总是包含所有从客户端发来的 cookie 数据。在老版本的 PHP 中，当 track_vars 配置选项打开时（此选项自 PHP 4.0.3 后总是打开的），系统为 cookie 设置的数组变量是 $HTTP_COOKIE_VARS。
关于更多细节以及有关浏览器问题的注意事项，参见 setcookie() 和 setrawcookie() 函数。
add a note
User Contributed Notes
kalla_durga at gmail dot com
04-Feb-2006 04:10
04-Feb-2006 04:10
bmorency at jbmlogic dot com
07-Oct-2005 08:14
07-Oct-2005 08:14
In response to the solution posted in the comment below, there are some practical issues with this solution that must be kept in mind and handled by your code. I developed an application using a similar "use-it-once" key to manage sessions and it worked great but we got some complaints about legitimate users getting logged out without reasons. Turns out the problem was not tentative highjacking, it was either:
A- Users double click on links or make 2 clicks very fast. The same key is sent for the 2 clicks because the new key from the first click didn't get to the browser on time for the second one but the session on the server did trash the key for the new one. Thus, the second click causes a termination of the session. (install the LiveHttpHeaders extension on firefox and look at the headers sent when you click twice very fast, you'll see the same cookie sent on both and the new cookie getting back from the server too late).
B- For any given reason, the server experiences a slow down and the response with the new key (which has replaced the old one on the server) is not returned to the browser fast enough. The user gets tired of waiting and clicks somewhere else. He gets logged out because this second click send the old key which won't match the one you have on your server.
Our solution was to set up a grace period where the old key was still valid (the current key and the previous key were both kept at all times, we used 15 seconds as a grace period where the old key could still be used). This has the drawback of increasing the window of time for a person to highjack the session but if you tie the validity of the old key to an IP address and/or user agent string, you still get pretty good session security with very very few undesired session termination.
mega-squall at caramail dot com
24-Feb-2005 05:04
24-Feb-2005 05:04
I found a solution for protecting session ID without tying them to client's IP. Each session ID gives access for only ONE querry. On the next querry, another session ID is generated and stored. If somebody hacks the cookie (or the session ID), the first one of the user and the pirate that will use the cookie will get the second disconnected, because the session ID has been used.
If the user gets disconnected, he will reconnect : as my policy is not to have more than one session ID for each user (sessions entries have a UNIQUE key on the collomn in which is stored user login), every entries for that user gets wiped, a new session ID is generated and stored on users dirve : the pirate gets disconnected. This lets the pirate usually just a few seconds to act. The slower visitors are browsing, the longer is the time pirates get for hacking. Also, if users forget to explicitly end their sessions .... some of my users set timeout longer than 20 minutes !
IMPORTANT NOTE : This disables the ability of using the back button if you send the session ID via POST or GET.
James Olsen
22-Jan-2005 12:14
22-Jan-2005 12:14
Tying the session to the IP of the user is not a good idea. Some users, notably AOL users, are behind a rotating proxy which means their hits to the server will actually be coming from different IP addresses over the duration of their visit. Trying the session to the IP will not work properly for those users.
myfirstname at braincell dot cx
24-Sep-2003 11:47
24-Sep-2003 11:47
[Editor's note: Wilson's comment has been deleted since it didn't contain much useful information, but this note is preserved although its reference is lost]
Just a general comment on Wilton's code snippet: It's generally considered very bad practice to store usernames and/or passwords in cookies, whether or not they're obsfucated. Many spyware programs make a point of stealing cookie contents.
A much better solution would be to either use the PHP built in session handler or create something similar using your own cookie-based session ID. This session ID could be tied to the source IP address or can be timed out as required but since the ID can be expired separately from the authentication criteria the authentication itself is not compromised.
Stuart Livings