This manual is by no means complete, usable, readable, comprehensible, or error free.
If you have any corrections, additions or constructive comments, please report them to the aide website issue tracker, or create a pull request on the website's repository.
This document was originally written by Rami Lehti <rammer@cs.tut.fi> with additions made by Marc Haber <mh+aide-manual@zugschlus.de> , Richard van den Berg <richard@vdberg.org> and Hannes von Haugwitz .
AIDE (Advanced Intrusion Detection Environment) is an intrusion detection program. More specifically a file integrity checker.
AIDE constructs a database of the files specified in aide.conf, AIDE's configuration file. The AIDE database stores various file attributes including: file type, permissions, inode number, user, group, file size, mtime and ctime, atime, growing size, number of links and link name. AIDE also creates a cryptographic checksum or hash of each file using one or a combination of the following message digest algorithms: sha1, sha256, sha512, md5, rmd160, tiger, haval, crc32 (gost and whirlpool can be compiled in if mhash support is available). Additionally, the attributes acl, xattr, selinux and e2fsattrs can be used when explicitly enabled during compile time.
Typically, a system administrator will create an AIDE database on a new system before it is brought onto the network. This first AIDE database is a snapshot of the system in it's normal state and the yardstick by which all subsequent updates and changes will be measured. The database should contain information about key system binaries, libraries, header files, all files that are expected to remain the same over time. The database probably should not contain information about files which change frequently like log files, mail spools, proc filesystems, user's home directories, or temporary directories.
After a break-in, an administrator may begin by examining the system using system tools like ls, ps, netstat, and who --- the very tools most likely to be trojaned. Imagine that ls has been doctored to not show any file named "sniffedpackets.log" and that ps and netstat have been rewritten to not show any information for a process named "sniffdaemond". Even an administrator who had previously printed out on paper the dates and sizes of these key system files can not be certain by comparison that they have not been modified in some way. File dates and sizes can be manipulated, some better root-kits make this trivial.
While it is possible to manipulate file dates and sizes, it is much more difficult to manipulate a single cryptographic checksum like md5, and exponentially more difficult to manipulate each of the entire array of checksums that AIDE supports. By rerunning AIDE after a break-in, a system administrator can quickly identify changes to key files and have a fairly high degree of confidence as to the accuracy of these findings.
Unfortunately, AIDE can not provide absolute sureness about change in files. Like any other system file, AIDE's binary and/or database can also be altered.
After you have installed all the necessary software do
./configure;make;make install in the main AIDE
directory of the unpacked source tree. You should carefully think
about the configuration and what a possible hacker can do if
he/her/they/it has root access.
Before you can compile AIDE you must have certain things:
Please check to see if there are mirrors available.
Once you have the source code of AIDE you should unpack it. If you
have GNU tar then the command is tar zxvf
aide-<VERSION_NUMBER>.tar.gz
It is highly recommended to verify the signature of your downloaded source code. You can either verify the source tarball or the git tag.
To check the supplied signature with GnuPG:
$ gpg --verify aide-<VERSION_NUMBER>.tar.gz.asc
To validate the gpg signature of the git tag:
$ git verify-tag v<VERSION_NUMBER>
Since v0.16a2 the key used for signing is the GnuPG key of Hannes von Haugwitz (the current maintainer of AIDE).
If you do not have that key, you can get it from one of the well known PGP key servers.
You have to make sure that the key you install is not a faked one. You can do this with reasonable assurance by comparing the output of
$ gpg --fingerprint 0x<KEYID>
with the fingerprint published elsewhere.
Next you must use the configure script found in AIDE's source code package to configure the compilation process.
There are several options you can select to configure. You can find out
what options are available with ./configure --help
command. Most of the time you do not need to give any options.
You can just use configure without any parameters.
If you want to change the directory where AIDE is installed you can
use --prefix option. For example ./configure --prefix=/usr
The compilation is done by simply typing make. You can
now type make install to install the binary and the
manual pages. The binary however should be installed on read-only
media or in some other tamperproof place. Also the databases should
be kept somewhere where a possible intruder cannot change them.
Next you have to create a configuration file. You can find more documentation for this in aide.conf(5) manual page.
There are three types of lines in aide.conf:
Lines beginning with # are ignored as comments.
Here is an example configuration.
#AIDE conf # Here are all the things we can check - these are the default rules # #p: permissions #ftype: file type #i: inode #n: number of links #l: link name #u: user #g: group #s: size #b: block count #m: mtime #a: atime #c: ctime #S: check for growing size #I: ignore changed filename #md5: md5 checksum #sha1: sha1 checksum #sha256: sha256 checksum #sha512: sha512 checksum #rmd160: rmd160 checksum #tiger: tiger checksum #haval: haval checksum #crc32: crc32 checksum #R: p+ftype+i+l+n+u+g+s+m+c+md5 #L: p+ftype+i+l+n+u+g #E: Empty group #>: Growing file p+ftype+l+u+g+i+n+S #The following are available if you have mhash support enabled: #gost: gost checksum #whirlpool: whirlpool checksum #The following are available and added to the default groups R, L and > #only when explicitly enabled using configure: #acl: access control list #selinux SELinux security context #xattrs: extended file attributes #e2fsattrs: file attributes on a second extended file system # You can also create custom rules - my home made rule definition goes like this # MyRule = p+i+n+u+g+s+b+m+c+md5+sha1 # Next decide what directories/files you want in the database /etc p+i+u+g #check only permissions, inode, user and group for etc /bin MyRule # apply the custom rule to the files in bin /sbin MyRule # apply the same custom rule to the files in sbin /var MyRule !/var/log/.* # ignore the log dir it changes too often !/var/spool/.* # ignore spool dirs as they change too often !/var/adm/utmp$ # ignore the file /var/adm/utmp
Here we include files in /etc, /bin and /sbin. We also include /var but ignore /var/log, /var/spool and a single file /var/adm/utmp.
It is generally a good idea to ignore directories that frequently change, unless you want to read long reports. It is good practice to exclude tmp directories, mail spools, log directories, proc filesystems, user's home directories, web content directories, anything that changes regularly. It is also good practice to include all system binaries, libraries, include files, system source files. It will also be a good idea to include directories you don't often look in like /dev /usr/man/.*usr/. Of course you'll want to include as many files as practical, but think about what you include.
One example: If you have a block device whose owner is changing frequently, you can configure aide to just check the attributes that do not normally change (inode, number of links, ctime).
Note that if you are referring to a single file you should add $ to the end of the regexp. This matches to the name of the file exactly and does not include any other files that might have the same beginning. In the example, all filenames beginning with /var/adm/utmp would be ignored if there were no dollar sign at the end of the last line. An intruder could then create a directory called /var/adm/utmp_root_kit and place all the files he/she/they wanted there and they would be ignored by AIDE.
There are several special group definitions to tweak what attributes are printed in the report. First report_force_attrs lists those attributes that are always printed from changed files. For example, if you say
report_force_attrs = u+g
and the size of a file changes, it's user and group id will also be printed in the report. Secondly, report_ignore_added_attrs, report_ignore_removed_attrs and report_ignore_changed_attrs define which attributes to ignore from the report. For example, if you define
report_ignore_changed_attrs = b
and the size of a file changes, it's block count will not be printed in the report, even if it did change as well.
If an attribute is both ignored and forced the attribute is not considered for file change but printed in the final report if the file has been otherwise changed.
Making a config file is a lot of hard work and must be done on a case by case bases. Don't give up simply because you don't get it right the first time around. This section gives you a few hints on how to debug your config.
You can use aide --verbose=255 to generate a lot of debug
output to help you see which files get added and which are discarded.
The following section gives some more information about
AIDE's rule matching algorithm.
Before reading this you should have basic understanding of how regular expressions in general and Perl Compatible Regular Expressions in particular work. There are several good books about this. Several Perl-books also have decent explanations about this subject.
As you already know, aide has three types of selection lines:
The string following the first character is taken as a regular expression matching to a complete filename, including the path. In a regular selection rule, the slash is included in the regular expression. An implicit ^ is added in front of each rule. A group definition follows the regular expression.
When reading the configuration file, aide internally builds a tree that roughly resembles the directory tree to be checked. Each node corresponds to a directory, and each node has one rule list for the associated regular selection lines, one for the associated negative selection lines and one for the associated equals selection lines. If there is no associated rule, the respective list may be empty.
aide tries to place a rule as far down in the tree as possible while
still assuring that it is above all files that it matches. This is
determined by the first "special" regexp character in the rule. For
example, !/proc would be placed in the root node,
!/proc/.* would be placed in the /proc node,
!/var/log/syslog* is placed in the /var/log node and,
finally, !/home/[a-z0-9]+/.bashrc$ is placed in the /home
node.
The algorithm that aide uses for rule matching is described in the following paragraphs. The pseudocode is an adaption from src/gen_list.c.
check_node_for_match(node,filename,first_time)
if (first_time)
check(equals list for this node)
check(regular list for this node)
if (node is not the root node)
check_node_for_match(nodes parent,filename,false)
if (this file is about to be added)
check(negative list for this node)
return (info about whether this file should be added or not and how)
When aide needs to determine whether a file found in the file system is to be checked, it first determines the deepest possible node x to match the current file against (that algorithm is not part of the pseudocode above), and then calls check-node_for_match(x, filename, true). So, the recursion starts at the deepest possible match.
As it can also be seen, equals selection lines are only checked in the first recursion step, thus providing some kind of speed optimization by reducing the number of necessary regular expression evaluations, which is a quite expensive operation.
There are some side-effects from this algorithm that might seem strange at first. For example if you have the following rules:
/ R
=/var/log/messages$ R+a
!/var/log/messages.*
This is what you might write if you want to check /var/log/messages but not /var/log/messages.0 and /var/log/messages.1 etc. However since the negative selection rules are checked last and .* can match to an empty string /var/log/messages is not added to the database. The following is a more correct way of doing it.
/ R
=/var/log/messages$ R+a
!/var/log/messages\.[0-9]$
Now only messages files ending in number 0-9 are not included in the database. Note an intruder could disguise a rootkit by creating a directory called messages.9. If messages.9 does not already exist that is.
Consider the following rules:
/ n+p+l+i+u+g+s+b+m+c+md5+sha1+rmd160+haval+gost+crc32+tiger
/etc$ n+p+l+i+u+g
/etc/resolv.conf$ n+p+l+u+g
This way, changing /etc/resolv.conf will also report /etc as having their mtime and ctime changed, even if /etc is configured not to be checked for mtime and ctime. The reason is that aide only uses a deepest-match algorithm to find the tree node to search, but a first-match algorithm inside the node. Since /etc is in the / directory, /etc will match the rule for the root directory and ignore the specialized /etc rule.
Rearranging the configuration like this:
/etc/resolv.conf$ n+p+l+u+g
/etc$ n+p+l+i+u+g
/ n+p+l+i+u+g+s+b+m+c+md5+sha1+rmd160+haval+gost+crc32+tiger
will solve the issue. It is generally a good idea to write the most general rules last.
First you must create a database against which future checks are
performed. This should be done immediately after the operating system
and applications have been installed, before the machine is plugged
into a network. You can do this by giving the command
aide --init.
This creates a database that contains all of the files that you
selected in your config file. The newly created database should now be
moved to a secure location such as read-only media. You should also
place the configuration file and the AIDE binary and preferably the
manual pages and this manual on that media also. Please remember to
edit the configuration file so that the input database is read from
that read-only media. The config file should not be kept on the
target machine. The attacker could read the config file and alter it
and if he does alter it he could place his rootkit in a place that
AIDE does not check. So the read-only media should be accessible only
during the check.
Now you are all set to go. You can now check the integrity of the
files. This can be done by giving the command
aide --check.
AIDE now reads the database and compares it to the files found on the
disk. AIDE may find changes in places that you might not expect. For
instance tty devices often change owners and permissions. You may want
to read long reports and that is up to you to decide. But most of us
do not have the time or the inclination read through tons of garbage
every day, so you should trim the config file to include only the
files and attributes of certain files that should not change. But keep
in mind that you should not ignore too much as that leaves you open
for an attack. An intruder might place his/her/its/their root kit in a
directory that you have ignored completely. One good example is
/var/spool/lp or something similar. This is the place that lp daemon
stores its temporary files. You should not ignore it completely
however. You should only ignore the format of files that you lp daemon
keeps creating. And remember to use the $-sign at the end of your
regexps. This stops someone from creating a directory that is ignored
along with its contents.
Now that you have trimmed your config file you should update the
database. This can be done by:
aide --update
The update command also does the same thing as check but it
creates a new database. This database should now be placed on
that read-only media along with the new config file. The check, trim,
update cycle should be repeated as long as necessary. I recommend that
the config file should be reviewed once in a while. The definition of
"a while" depends on your paranoia. Some might want do it daily after
each check. Some might want to do it weekly.
There is usually some drift in the databases. What I mean by drift is that new files are created, config files of applications are edited, tons of small changes pile up until the report becomes unreadable. This can be avoided by updating the database once in a while. I myself run the update every night. But, I don't replace the input database nearly as often. The replacement of the input datbase should always be a manual operation. This should not be automated.
There is also an alternative way of doing this. This method may be
preferable for people that have lots of machines that run aide.
You can run
aide --init
on all of the hosts and move the generated databases to a central host
where you compare different versions of the databases with
aide --compare
This has the benefit of freeing up resources on the monitored
machines.
The AIDE database can be used to find the real names and places of files that have been moved to lost+found directory by fsck.