On Wednesday, March 27, 2013 08:38:07 PM Burn Alting wrote: > All, > > Has anyone considered allowing an includeConfig statement for > audit.rules (or auditd.conf if need be)? > > The action would be to, at that point in the parse (or the end of the > file, if auditd.conf holds the directive), open the nominated directory > and any files within, and parse them. > > The idea is to allow for localization of audit. At an enterprise level > one would deploy the common, corporate set of rules > in /etc/audit/audit.rules. Should a local system need additional rules > such as tailored file watches, workstation or capability specific > monitoring, these could appear in files in the includeConfig directory. > That way, distribution mechanisms such as puppet, rpm satellite server, > apt repositories, etc can maintain the corporate set of rules without > changing localized configurations on updates. Sorry for the late reply, been out a bit and am trying to catch up on email. Well...have you heard of SCAP? Its a whole framework for assessing the security posture of a system based on open standards to prevent vendor lockin. It can also allow for continuous monitoring, boot up attestation via TNC, patch management, and we are working on some basic level of remediation. More info about SCAP can be found at these sites: http://scap.nist.gov/ http://makingsecuritymeasurable.mitre.org/ We have an openscap project http://www.open-scap.org/ There is an SCAP Security Guide which will become a STIG: https://fedorahosted.org/scap-security-guide/ And its being integrated into various systems management tools. The reason I mention this is that currently there is no way that you could evaluate audit rules from an SCAP based tool with a decomposed set of audit rules. The OVAL interpreter is the limitation. It does not have any method of being able to smartly assemble the collective audit rules to assess if the system is in compliance. In the audit system, the order of rules matters and that is one of the problems OVAL would have to be specified and coded to understand. So with SCAP in mind, the options seem to be: 1) Build a rule compiler that assembles a master audit.rules file from sources but only 1 set of rules are loaded. 2) Request a change in OVAL 5.11 to support this kind of setup. Sometime around 2014 NIST should have it approved and content developers can use it. This means holding off the functionality a bit because we can't allow audit configurations that cannot be monitored. 3) ?? (Any other ideas) OVAL has limited capability for reading file formats. Changes in capability have a long lead time. Audit configuration is very important to be able to assess from SCAP. For example, the DISA STIG and USGCB would mandate it. I think someone is working on a DSS-PCI profile which would also include some form of audit rule tests. In my opinion, the ability to assess the audit system's compliance has to take SCAP into account and solutions to allow drop in audit rule additions ought to fit within those limitations. I would imagine the same set of people that care about audit rules are nearly the same set of people that care about SCAP. -Steve

On Wednesday, April 03, 2013 09:37:23 PM Burn Alting wrote: > Thanks for these comments Steve. > > I will come up with a solution based on option one. Perhaps along the > line of a script (to suit both auditd.init and auditd.service) that > a. Looks for a known directory, say /etc/audit/auditd.rules I was thinking something like /etc/audit/rules.d/ or something ending in '.d' under the audit directory so that selinux labels are the same. > b. If not empty, it will concatenate all files of the form xxx.rules, > stripping comments and blank lines. The xxx will determine sort. Sure. I think some people prefix numbers to the name to help guide the ordering like udev. > c. If the resultant file is not empty, it can > replace /etc/audit/audit.rules. Sure. The question is should it do that always on start? What about reload? Should it replace it only if its changed? (writing to /etc/audit/audit.rules is an auditable event...we probably want to minimize that.) Thanks, -Steve > On Tue, 2013-04-02 at 14:03 -0400, Steve Grubb wrote: >> On Wednesday, March 27, 2013 08:38:07 PM Burn Alting wrote: >>> All, >>> >>> Has anyone considered allowing an includeConfig statement for >>> audit.rules (or auditd.conf if need be)? >>> >>> The action would be to, at that point in the parse (or the end of the >>> file, if auditd.conf holds the directive), open the nominated directory >>> and any files within, and parse them. >>> >>> The idea is to allow for localization of audit. At an enterprise level >>> one would deploy the common, corporate set of rules >>> in /etc/audit/audit.rules. Should a local system need additional rules >>> such as tailored file watches, workstation or capability specific >>> monitoring, these could appear in files in the includeConfig directory. >>> That way, distribution mechanisms such as puppet, rpm satellite server, >>> apt repositories, etc can maintain the corporate set of rules without >>> changing localized configurations on updates. >> Sorry for the late reply, been out a bit and am trying to catch up on >> email. >> >> Well...have you heard of SCAP? Its a whole framework for assessing the >> security posture of a system based on open standards to prevent vendor >> lockin. It can also allow for continuous monitoring, boot up attestation >> via TNC, patch management, and we are working on some basic level of >> remediation. >> >> More info about SCAP can be found at these sites: >> http://scap.nist.gov/ >> http://makingsecuritymeasurable.mitre.org/ >> >> We have an openscap project >> http://www.open-scap.org/ >> >> There is an SCAP Security Guide which will become a STIG: >> https://fedorahosted.org/scap-security-guide/ >> >> And its being integrated into various systems management tools. The reason >> I mention this is that currently there is no way that you could evaluate >> audit rules from an SCAP based tool with a decomposed set of audit rules. >> The OVAL interpreter is the limitation. It does not have any method of >> being able to smartly assemble the collective audit rules to assess if >> the system is in compliance. In the audit system, the order of rules >> matters and that is one of the problems OVAL would have to be specified >> and coded to understand. >> >> So with SCAP in mind, the options seem to be: >> >> 1) Build a rule compiler that assembles a master audit.rules file from >> sources but only 1 set of rules are loaded. >> 2) Request a change in OVAL 5.11 to support this kind of setup. Sometime >> around 2014 NIST should have it approved and content developers can use >> it. >> This means holding off the functionality a bit because we can't allow >> audit >> configurations that cannot be monitored. >> 3) ?? (Any other ideas) >> >> OVAL has limited capability for reading file formats. Changes in >> capability >> have a long lead time. Audit configuration is very important to be able to >> assess from SCAP. For example, the DISA STIG and USGCB would mandate it. I >> think someone is working on a DSS-PCI profile which would also include >> some >> form of audit rule tests. >> >> In my opinion, the ability to assess the audit system's compliance has to >> take SCAP into account and solutions to allow drop in audit rule >> additions ought to fit within those limitations. I would imagine the same >> set of people that care about audit rules are nearly the same set of >> people that care about SCAP. >> >> -Steve -- Linux-audit mailing list Linux-audit(a)redhat.com https://www.redhat.com/mailman/listinfo/linux-audit

Steve, I'll take your recommendations on board and, Kevin, I'll look at Canonical's methods of file building. I'll also check the limitations in the number of rules loadable which auditctl mentions. I believe, by offering a rule building capability, we indirectly introduce a risk of increasing the rule set size. Kevin, I think to incorporate your recommendations would be a contrib element that can 'manage' a file in /etc/audit/rules.d. I'll take this into consideration as I document the file nomenclature in the rules directory. Will author all this on the weekend. Rgds Burn On Wed, 2013-04-03 at 09:19 -0400, Boyce, Kevin P. (AS) wrote: > I think this is a worthwhile effort. You might have a look at how the > Canonical folks do things like this, in particular the update-grub > script, uses a bunch of files in a ".d" directory and build the actual > config file (/boot/grub/grub.cfg). > > On another note I will take the opportunity to introduce some feature > creep. At one point I had written a cron script for my environment that > rebuilt the snare.conf file every week and restart the audit daemon. > Additionally, this script would take in a list of the names of packages > you were interested in auditing. For example passwd-0.77-4.el6_2.2 (by > name passwd) and wireshark-1.2.15-2.el6_2.1 (by name wireshark). The > script would then query the package manager to see if it was installed. > If it is, it would list all files provided by the package, filter out > the executables from the libraries from the config files from the > documentation. Then it would generate a rule for each type of file. > Config files and libraries were added to the rule list looking for > failure to write or change the file. Executables would be added to the > rule list looking for success or failure to execute the file. > > The reason for all of this was that in a large environment with many > users with root privilege you don't always know what software would be > installed on a system. If at some point someone had added wireshark > (via rpm) to a system you know it will get audited after that. The > other benefit is that it make the generation of rules sort of agnostic > with regard to the architecture of the system; the package manager > handles telling you the location of the files you are interested in for > any given package. > > I don't know if this sort of thing has value to anyone else, but I > thought I'd throw it out there as a suggestion anyway. > > Kevin > > On 04/03/2013 07:42 AM, Steve Grubb wrote: > > On Wednesday, April 03, 2013 09:37:23 PM Burn Alting wrote: > >> Thanks for these comments Steve. > >> > >> I will come up with a solution based on option one. Perhaps along the > >> line of a script (to suit both auditd.init and auditd.service) that > >> a. Looks for a known directory, say /etc/audit/auditd.rules > > I was thinking something like /etc/audit/rules.d/ or something ending in '.d' > > under the audit directory so that selinux labels are the same. > > > >> b. If not empty, it will concatenate all files of the form xxx.rules, > >> stripping comments and blank lines. The xxx will determine sort. > > Sure. I think some people prefix numbers to the name to help guide the ordering > > like udev. > > > > > >> c. If the resultant file is not empty, it can > >> replace /etc/audit/audit.rules. > > Sure. The question is should it do that always on start? What about reload? > > Should it replace it only if its changed? (writing to /etc/audit/audit.rules > > is an auditable event...we probably want to minimize that.) > > > > Thanks, > > -Steve > > > > > >> On Tue, 2013-04-02 at 14:03 -0400, Steve Grubb wrote: > >>> On Wednesday, March 27, 2013 08:38:07 PM Burn Alting wrote: > >>>> All, > >>>> > >>>> Has anyone considered allowing an includeConfig statement for > >>>> audit.rules (or auditd.conf if need be)? > >>>> > >>>> The action would be to, at that point in the parse (or the end of the > >>>> file, if auditd.conf holds the directive), open the nominated directory > >>>> and any files within, and parse them. > >>>> > >>>> The idea is to allow for localization of audit. At an enterprise level > >>>> one would deploy the common, corporate set of rules > >>>> in /etc/audit/audit.rules. Should a local system need additional rules > >>>> such as tailored file watches, workstation or capability specific > >>>> monitoring, these could appear in files in the includeConfig directory. > >>>> That way, distribution mechanisms such as puppet, rpm satellite server, > >>>> apt repositories, etc can maintain the corporate set of rules without > >>>> changing localized configurations on updates. > >>> Sorry for the late reply, been out a bit and am trying to catch up on > >>> email. > >>> > >>> Well...have you heard of SCAP? Its a whole framework for assessing the > >>> security posture of a system based on open standards to prevent vendor > >>> lockin. It can also allow for continuous monitoring, boot up attestation > >>> via TNC, patch management, and we are working on some basic level of > >>> remediation. > >>> > >>> More info about SCAP can be found at these sites: > >>> http://scap.nist.gov/ > >>> http://makingsecuritymeasurable.mitre.org/ > >>> > >>> We have an openscap project > >>> http://www.open-scap.org/ > >>> > >>> There is an SCAP Security Guide which will become a STIG: > >>> https://fedorahosted.org/scap-security-guide/ > >>> > >>> And its being integrated into various systems management tools. The reason > >>> I mention this is that currently there is no way that you could evaluate > >>> audit rules from an SCAP based tool with a decomposed set of audit rules. > >>> The OVAL interpreter is the limitation. It does not have any method of > >>> being able to smartly assemble the collective audit rules to assess if > >>> the system is in compliance. In the audit system, the order of rules > >>> matters and that is one of the problems OVAL would have to be specified > >>> and coded to understand. > >>> > >>> So with SCAP in mind, the options seem to be: > >>> > >>> 1) Build a rule compiler that assembles a master audit.rules file from > >>> sources but only 1 set of rules are loaded. > >>> 2) Request a change in OVAL 5.11 to support this kind of setup. Sometime > >>> around 2014 NIST should have it approved and content developers can use > >>> it. > >>> This means holding off the functionality a bit because we can't allow > >>> audit > >>> configurations that cannot be monitored. > >>> 3) ?? (Any other ideas) > >>> > >>> OVAL has limited capability for reading file formats. Changes in > >>> capability > >>> have a long lead time. Audit configuration is very important to be able to > >>> assess from SCAP. For example, the DISA STIG and USGCB would mandate it. I > >>> think someone is working on a DSS-PCI profile which would also include > >>> some > >>> form of audit rule tests. > >>> > >>> In my opinion, the ability to assess the audit system's compliance has to > >>> take SCAP into account and solutions to allow drop in audit rule > >>> additions ought to fit within those limitations. I would imagine the same > >>> set of people that care about audit rules are nearly the same set of > >>> people that care about SCAP. > >>> > >>> -Steve > > -- > > Linux-audit mailing list > > Linux-audit(a)redhat.com > > https://www.redhat.com/mailman/listinfo/linux-audit > > -- > Linux-audit mailing list > Linux-audit(a)redhat.com > https://www.redhat.com/mailman/listinfo/linux-audit -- Linux-audit mailing list Linux-audit(a)redhat.com https://www.redhat.com/mailman/listinfo/linux-audit

On Sunday, April 07, 2013 09:16:46 PM Burn Alting wrote: > Please find attached my patch on this matter. Thanks for taking this on. > I essence, /etc/audit/audit.rules is now formed from files (.rules > suffixed) within /etc/audit/rules.d. The new script /sbin/augenrules is > executed by from either startup script, /etc/init.d/auditd > or /usr/lib/systemd/system/auditd.service before calling auditctl. One issue that I am concerned about is how this feature gets added to existing setups. For example, someone may have a /etc/audit/audit.rules file, then upgrade and if there is an empty shipped policy in /etc/audit/audit.d, it will erase the installed rules. So, I think we should have an /etc/sysconfig option that enables augenrules so that an admin has to do something to turn this on thus preventing automatic deletion of rules. For systemd, I think we want to ship the service file with the ExecStartPost line commented out which then requires an admin to take an action to enable. We really don't want unexpected things to happen during an upgrade. > The generated file ensures > - the last processed -D directive without an option, if present, is > emitted on the first line In generating rules, we should always start with -D. I can't imagine not having it. > - the last processed -b directive, if present, is emitted on the second > line We probably want the largest in all the processed files. > - the last processed -f directive, if present, is emitted on the third > line We probably want the largest here, too. > - the last processed -e directive, if present, is emitted as the last > line. I was thinking that if any of the files try to ask for it to be immutable, then it should go at the end. > The file, /etc/audit/audit.rules, is only updated if it has changed. > > https://www.redhat.com/mailman/listinfo/linux-audit That is great, because any write could be an auditable event. At some point we also might want to add support for a --check option which does everything except overwrite the final rules. -Steve

On Friday, April 19, 2013 07:23:53 AM Burn Alting wrote: > Steve, > > I will make the changes on the weekend and re-submit. No need, I can take care of it. I just wanted to air the concerns and make sure everyone was in agreement or maybe someone saw another way to solve the problem. I will merge the code today with a couple changes. I am trying to get the audit package ready for another release sometime in the next couple days.