-
Notifications
You must be signed in to change notification settings - Fork 0
AuthHowTo
External authenticators can be written in almost any language. I've seen examples in C, Perl, Python, Java, Ruby and sh. The sample authenticators in the 'test' directory are in Perl. The 'pwauth' authenticator is in ANSI C. The example code fragments in this document are in C.
It should be noted that there is one compelling advantage to using a compiled language like C for the authenticator. Authenticators are run anew for every HTTP request that needs authentication. If your page references 5 images that are also in the protected directory, then your authenticator will be run six times every time the page is loaded. Interpreted programs are slow to start up. Every time you run a Perl program, the whole huge Perl interpreter needs to be run and it needs to read and compile your script file before it can start executing it. By contrast, if you write it in C, then each invocation requires only that the rather small compiled program be loaded and run. So the performance difference is likely to be substantial. And though C programs are certainly harder to write and debug, in most cases it isn't going to be that big a program, so the extra effort isn't that much.
If the authenticator is a script rather than a compiled program, it normally has to start with a "'#!/bin/sh'" or "'#!/usr/bin/perl'" type directive. Scripts without such directives may get interpreted by the shell, or may just not work, depending on your installation.
The authenticator program should be written with great care because it runs as a privileged user and handles privileged data. A poorly written authenticator could substantially compromise the security of your system. You get points for paranoia. Some notes:
-
Don't make any assumptions about the length of the login names and passwords given by the user. I think Apache will never pass you ones that are longer than 8192 characters, but don't depend this. C programmers should check very carefully for buffer overflows.
-
Don't make assumptions about the content of the login and password strings. For example, if you are using them in an SQL query, do proper checking and/or quoting to insure that nobody is doing SQL injection.
-
Think about locking. It is possible to get lots of hits at your website very fast, so there may be many programs simultaneously reading your authentication database, plus updates may be going on at the same time. Probably some form of locking is needed to make all this work right.
-
If you are programming in C, think about core dumps. On some systems core dump files can be publicly readable. A core dump from your authenticator is likely to contain the user's plain text password, and may include large chunks of your password database that may have been in buffers. For C programs on most versions of Unix, it is possible to disable core dumps by doing something like:
rlim.rlim_cur = rlim.rlim_max = 0;
(void)setrlimit(RLIMIT_CORE, &rlim);
It may not hurt to spend a little time looking at the features of the pwauth authenticator, which is the most secure external authenticator that I have written.
Authenticators communicate their result by the exit status code they return.
A value of 0 indicates that the password is correct. Other values indicate
that the password is incorrect, or something else is wrong. It can be
useful to return different numeric error codes for different kinds of errors, say 1 for "no such login", 2 for "bad password", 3 for "cannot read password database", etc. These
will be logged in the Apache error log file, and can be helpful in diagnosing
problems. Mod_auth*_external
does not have any provision for
returning textual error messages from the external authenticator. You might
be able to use syslog() for this. This might be improved in future releases.
Returned error codes should not be negative. Negative values are used internally in 'mod_authnz_external' to indicate problems launching your program or retrieving the returned status code after it exits.
Messages written to standard error will go into the Apache error log. This can be useful for error logging and debugging. You should not write out the password in any production configuration.
How the external authentication program gets its arguments depends on
the method used. The method used is determined by the 'SetExternalAuthMethod
'
command in your Apache configuration file. You need implement only the
method that you plan to use in your configuration, which should almost always be the "pipe" method.
In the "pipe" method, the arguments are read from standard input. The user name will be on the first line, and the password will be on the second. Here's a typical chunk of C code to read that:
main()
{
char user[100], password[100], *p;
if (fgets(user, sizeof(user), stdin) == NULL) exit(2);
if ((p= strchr(user, '\n')) == NULL) exit(4)
*p= '\0';
if (fgets(password, sizeof(password), stdin) == NULL) exit(3);
if ((p= strchr(password, '\n')) == NULL) exit(5)
*p= '\0';
if (check_password(user, password) == OK)
exit(0); /* Good Password */
else
exit(1); /* Incorrect Password */
}
Here we simply read two lines from stdin, being careful not to allow buffer overflows and stripping off trailing newlines.
We assume "check_password()
" is some function that checks the validity of a
password and returns 'OK' if it is good.
Note that we exit with different non-zero error codes in different error cases. This will be helpful for debugging, as those values will be logged when authentication fails, giving you some clue as to what went wrong. It'd really be better for check_password() to return more detailed error codes, but I wanted to keep the example simple.
The "checkpassword" method is identical to the "pipe" method, except that the user name and password are terminated by NUL ('\0'
) characters instead of newline characters, and they must be read from file descriptor 3 instead of standard input. Documentation for the checkpassword interface is at http://cr.yp.to/checkpwd.html.
In the "environment" method, the arguments are passed in environment variables. The user id and the clear-text password are passed in the USER and PASS environment variables respectively.
Note that the environment method has fundamental security weaknesses, and should probably not be used. Use the pipe method instead.
A typical chunk of C code to authenticate with the environment method might be like:
main()
{
char *user, *password;
if ((user= getenv("USER")) == NULL) exit(2);
if ((password= getenv("PASS")) == NULL) exit(3);
if (check_password(user, password) == OK)
exit(0); /* Good Password */
else
exit(1); /* Incorrect Password */
}
Security is generally less of a issue with group authenicators, since they are not handling any data as sensitive as clear-text passwords. They are only passed a user name (presumably already authenticated), and a list of group names. They exit with status code 0 if that user is in one of those groups, and a non-zero code otherwise.
In versions of mod_auth_external
before 2.1.8, external authenticators were always passed just one group name. If the Apache "require group
" directive listed more than one group, then the external authenticator would be called once with each group name, which could be inefficient if you have a large number of groups. Mod_auth_external
will still behave this way if you issue the "GroupExternalManyAtOnce off
" directive.
Newer versions of mod_auth_external
and mod_authnz_external
will pass all group names, separated by spaces. There will only be multiple calls if more than one "require group" (or "require external-group" in Apache 2.4) directive applies to the same program (e.g., if different parent directories contain such directives in their .htaccess
files - for efficiency, this should be avoided). The list of group names is passed in exactly as they appear on the "require group" directive - if your program can't handle multiple spaces between group names, don't put them there.
Arguments are passed in a manner similar to password authenticators. The method used is determined by the 'SetExternalGroupMethod
' command in your Apache configuration file.
In the "pipe" method, the arguments are read from standard input. The user name will be on the first line, and the group name will be on the second. Here's a typical chunk of C code to read that:
main()
{
char user[100], groups[100], *group, *p;
if (fgets(user, sizeof(user), stdin) == NULL) exit(2);
if ((p= strchr(user, '\n')) == NULL) exit(4)
*p= '\0';
if (fgets(groups, sizeof(groups), stdin) == NULL) exit(3);
if ((p= strchr(groups, '\n')) == NULL) exit(5)
*p= '\0';
group= strtok(groups, " ");
while (group != NULL)
{
if (check_group(user, group) == OK)
exit(0); /* User is in group */
group= strtok(NULL, " ");
}
exit(1); /* User is not in any group */
}
We simply read two lines from stdin, being careful not to allow buffer overflows and stripping off trailing newlines. We loop through all groups, checking each.
Here "check_group()" is some function that looks in your database to see if user is in group and returns 'OK' if he is.
Mod_auth_external will happily try to do group authentication via the checkpassword method, piping NUL ('\0'
) terminated user and group names to the child process's file descriptor 3, but this isn't actually allowed for in the checkpassword protocol specification, so I don't recommend it.
In the "environment" method, the arguments are passed in environment variables. The user id and the group names are passed in the USER and GROUP environment variables respectively. A typical chunk of C code to fetch the arguments and check each group might be like:
main()
{
char *user, *groups, *group;
if ((user= getenv("USER")) == NULL) exit(2);
if ((groups= getenv("GROUP")) == NULL) exit(3);
group= strtok(groups, " ");
while (group != NULL)
{
if (check_group(user, group) == OK)
exit(0); /* User is in group */
group= strtok(NULL, " ");
}
exit(1); /* User is not in any group */
}
In all cases (pipe or environment method, password or group authentication), the following additional environment variables will be supplied to the authenticator:
Environment Variable | Description |
---|---|
AUTHTYPE | either "PASS" or "GROUP" depending on whether we are doing password or group authentication. This is handy if you are using one program to do both. |
CONTEXT | a string whose value is set by an "AuthExternalContext" directive in the .htaccess file or "<Directory> " block for the directory. This can be used to select different authentication behaviors in different directories. It is undefined if there is no "AuthExternalContext" directive. |
IP | the client's ip-address. |
HOST | the client's host name, if Apache has "HostnameLookups On". |
PATH | the httpd's path environment variable. |
COOKIE | all cookie values passed in by the client. |
HTTP_HOST | the server's host name, as given in the HTTP request. May be useful if you have multiple virtual hosts sharing an authenticator. |
URI | the document requested. This is the URL including any extra path information, but not including the hostname or any CGI arguments. |
These may be useful for logging, or you may want to accept logins from certain users only if they are connecting from certain locations or requesting certain documents.
Note that if you have configured Apache with "HostnameLookups Off
" then HOST will usually not be set. If you really want host names, either turn on HostnameLookups
or do your own gethostbyaddr()
calls from the authenticator when HOST is not defined. Note that if the user is coming from an unresolvable IP, then host name lookups can be very slow.
Note that using IP addresses to track a user through your site is not reliable. Users of services like AOL and WebTV use proxy servers, so that their IP addresses appear to change constantly since each request my come through a different proxy. A single user's requests for successive pages, or for different images on the same page may all come from different IP addresses.
The PATH environment variable passed to the authenticator is just whatever PATH was in effect when Apache was launched, and may differ if the server was launched automatically during a reboot or manually by an admin. Probably your program should set its own PATH if it needs one.
The COOKIE environment variable contains all cookies set in the current request. This has the same format as the HTTP_COOKIES ("key=val;key=val
") passed to a CGI program. This should be used with caution. Cookies come from the user's computer and might have been created, editted or deleted by the user rather than your website. This severely limits their use for authentication. It is not possible to set cookies from an authentication module, including session cookies such as those used in PHP or other systems.
The URI variable is there because various people want it. Mostly it is useful not for authentication ("who is this person?") but for access control ("is this person permitted to do this?"), and good design usually dictates separating those functions. Strictly speaking, an authenticator is not the right place to be doing access control. Furthermore, in mod_authnz_external 3.3.0 and later requests to new URI's within the same authentication configuration may not always trigger a re-authentication. This will never happen for sub-requests (like an internal redirection or a server-side-include) and may not happen for other requests if you use mod_authz_socache. So really the URI should only be used for logging, not for anything that effects whether you accept the request or not.
You should test you authenticator in isolation before trying to run it from mod_authnz_external
. For an authenticator that takes input from a pipe, I typically issue the following command from 'csh' or 'tcsh':
./myauth; echo $status
or from 'sh' or 'bash':
./myauth; echo $?
Then I type in the login and password, one per line. The correct status code should come back right after I enter the password, zero for valid logins, other values for invalid ones.
- List of Available Authenticators
- Historical License and Version Notes
- A Brief Explanation of the Apache Authn/z Architecture
- Links to Related Software
- Ideas for Future Improvements to Mod_authnz_external