Logrotate, audit.log, selinux, cron, and ansible

The story

The disk space for /var/log/audit/audit.log tends to get filled up. The audit daemon has an ability to rotate its own logs. See the man page for auditd.conf.

max_log_file             =  100
max_log_file_action      =  rotate

That’s swell and all, until you realize that auditd cannot compress its rotated logs. On a small /var/log/audit mount point, you’ll fill it up with uncompressed logs.

/dev/mapper/os-var_log_audit          2490M 2136M      355M      86% /var/log/audit

So on a RHEL7 system with logrotate, you can adjust logrotate to handle the audit.log file. Now, logrotate is a finicky application. It has caused me many hours of grief in the past.
You would want to set auditd.conf a certain way:

max_log_file             =  0
max_log_file_action      =  ignore

And set /etc/logrotate.d/audit:

/var/log/audit/*.log {
        rotate 30
        minsize 100k
        maxsize 200M
                touch /var/log/audit/audit.log ||:
                chmod 0600 /var/log/audit/audit.log ||:
                service auditd restart

And ensure you’ve got a /etc/cron.weekly/logrotate:


/usr/sbin/logrotate /etc/logrotate.conf
if [ $EXITVALUE != 0 ]; then
    /usr/bin/logger -t logrotate "ALERT exited abnormally with [$EXITVALUE]"
exit 0

After a few days, I learned that my logs were getting filled up so fast, the weekly rotation wasn’t good enough. So I had to place it in my cron.hourly.
And then I learned that it wasn’t running every hour. I spent a few days investigating, and eventually learned that some systems use a specific status file for logrotate. I remember in the past logrotate needs an execution with a -f flag to force the rotation the first time and add a new file to the status file. So if a new file was never force-rotated, it won’t be added to the status file.
My manual logrotate -f command was indeed adding my audit.log log file to the status file, but to the wrong one!
Some of my systems use -s /var/lib/logrotate/logrotate.status but the default is /var/lib/logrotate.status.
So I had to reflect that in my ansible playbook. Actually, I had to write some logic to find the one used by the cronjob and then use that status file.

So I got the correct logrotate status file set up in the ansible playbook. I spent the next week figuring out that logrotate simply couldn’t rotate the file when called from cron. I piped the utility to tee, and also included the -v flag on logrotate. I saw a permission denied.
With the permission issue, I had no choices left by selinux. I had to use the audit.log file to determine that the audit.log file is not readable by logrotate when called by cron.
I finally set captured all the actions performed by logrotate by setting the selinux process context to be permissive:

semanage permissive -a logrotate_t
I let it run, and then had to collect all the actions it performed, and saw what had happened.
{ grep logrotate /var/log/audit/audit.log ; zgrep logrotate /var/log/audit/audit.log.1.gz ; } | audit2why

So I used audit2allow to convert it to an selinux policy.

{ grep logrotate /var/log/audit/audit.log ; zgrep logrotate /var/log/audit/audit.log.1.gz ; } | audit2allow -M logrotate-audit

And then after some searching online, I learned how I can keep the text definition file, and compile the policy from it when I need to:

grep logrotate /var/log/audit/audit.log | audit2allow -m logrotate-audit # saves to logrotate-audit.te
checkmodule -M -m -o logrotate-audit.mod logrotate-audit.te # intermediate step
semodule_package -o logrotate-audit.pp -m logrotate-audit.mod # compiled policy
semodule -i logrotate-audit.pp

The text definition of logrotate-audit policy:

#semodule -i logrotate-audit.pp

module logrotate-audit 1.0;

require {
        type auditd_etc_t;
        type logrotate_t;
        type auditd_log_t;
        class file { create getattr ioctl open read rename setattr unlink write };
        class dir { add_name read remove_name write };

#============= logrotate_t ==============
allow logrotate_t auditd_etc_t:file getattr;
allow logrotate_t auditd_log_t:dir { read write add_name remove_name };
allow logrotate_t auditd_log_t:file { create ioctl open read rename getattr setattr unlink write };

Now, I wrote a master ansible playbook that performs this whole operation, from loading the .te file and compiling it and installing it, to setting logrotate to watch the audit file, and telling auditd to ignore rotating it.
Note: It is outside the scope of this task to ensure that the selinux tools are in place on each server. My environment already ensures package libselinux-python is present on each system, which should bring in all the dependencies of this ansible playbook.

# File: /etc/ansible/books/fix_var-log-audit.yml
# Author: bgstack15
# Startdate: 2018-01-24
# Title: Playbook that Fixes the /var/log/audit Space Issue
# Purpose: Logical Disk Free Space is too low
# History:
# Usage:
#    ansible-playbook -i /etc/ansible/inv/hosts /etc/ansible/configuration/fix_var-log-audit.yml -l hostwithproblem201
#    Use the -l host1,host2 parameter.
# Reference:
#    roles/general_conf/tasks/04_selinux.yml
#    roles/general_conf/tasks/05_auditd.yml
# Improve:
# Documentation:
#    The intention with auditd is to minimize the disk usage of the logs

- hosts: all
  remote_user: ansible_user
  become: yes

    auditd_conf: /etc/audit/auditd.conf
    auditd_log_cleanup_regex: '.*audit\.log\.[0-9]+'
    auditd_log_dir: /var/log/audit
    auditd_logrotate_conf: /etc/logrotate.d/audit


# To make it possible to just drop in files to the files directory and have this module read them automatically, use these two.
#  - name: learn full list of semodules available to install, modular list version
#    shell: warn=no find /etc/ansible/roles/general_conf/files/selinux/ -regex '.*.te' -printf '%f\n' | sed -r -e 's/\.te$//;'
#    register: semodules_list
#    changed_when: false
#    delegate_to: localhost
#    ignore_errors: yes
#  - name: learn semodule versions to install, modular list version
#	shell: warn=no grep -E '^\s*module\s+{{ item }}\s+[0-9\.]+;\s*$' /etc/ansible/roles/general_conf/files/selinux/{{ item }}.te | awk '{print $3*1000;}'
#    register: selinux_pol_versions_target
#    changed_when: false
#    delegate_to: localhost
#    with_items:
#    - "{{ semodules_list.stdout_lines }}"

  - name: learn semodule versions to install, static version
    shell: warn=no grep -E '^\s*module\s+{{ item }}\s+[0-9\.]+;\s*$' /etc/ansible/templates/{{ item }}.te | awk '{print $3*1000;}'
    register: selinux_pol_versions_target
    changed_when: false
    delegate_to: localhost
    - logrotate-audit

  #- debug:
  #    msg: "{{ item.item }} should be {{ item.stdout }}"
  #  with_items:
  #  - "{{ selinux_pol_versions_target.results }}"

  - name: learn current semodule versions
    shell: warn=no semodule --list | awk '$1=="{{ item.item }}" {print $2*1000} END {print "0";}' | head -n1
    register: selinux_pol_versions_current
    changed_when: false
    - "{{ selinux_pol_versions_target.results }}"

  - debug:
      msg: "{{ item.item.item }} is currently {{ item.stdout }} and should be {{ item.item.stdout }}"
    - "{{ selinux_pol_versions_current.results }}"

  #- pause:
  #    prompt: "Does the above look good?........................"

  - name: download selinux modules that need to be installed
      src: "/etc/ansible/templates/{{ item.item.item }}.te"
      dest: "/tmp/{{ item.item.item }}.te"
      mode: 0644
      owner: root
      group: root
      backup: no
      force: yes
    changed_when: false
    - "item.item.stdout > item.stdout"
    - "{{ selinux_pol_versions_current.results }}"

  - name: install selinux modules
    shell: chdir=/tmp warn=no /usr/bin/checkmodule -M -m -o "/tmp/{{ item.item.item }}.mod" "/tmp/{{ item.item.item }}.te" && /usr/bin/semodule_package -m "/tmp/{{ item.item.item }}.mod" -o "/tmp/{{ item.item.item }}.pp" && /usr/sbin/semodule -v -i "/tmp/{{ item.item.item }}.pp"
    - "item.item.stdout > item.stdout"
    - "{{ selinux_pol_versions_current.results }}"

  - name: clean any temporary selinux modules files
      path: "/tmp/{{ item[0].item.item }}.{{ item[1] }}"
      state: absent
    changed_when: false
    - "item[0].item.stdout > item[0].stdout"
    - "{{ selinux_pol_versions_current.results }}"
    - [ 'te', 'pp', 'mod' ]


  # modify auditd.conf which notifies the handler
  - name: auditd does not keep logs
      path: "{{ auditd_conf }}"
      regexp: "{{ item.r }}"
      backrefs: yes
      line: "{{ item.l }}"
      create: no
      state: present
      backup: yes
    #notify: auditd handler
    - { r: '^max_log_file_action.*$', l: 'max_log_file_action      =  ignore' }
    - { r: '^max_log_file\s.*$', l: 'max_log_file             =  0' }

  # tarball and cleanup any existing audit.log.1 files
  - name: list all old auditd logs which need to be compressed and cleaned up
    shell: warn=no find /var/log/audit -regex {{ auditd_log_cleanup_regex }}
    register: cleanup_list
    ignore_errors: yes
    changed_when: cleanup_list.stdout_lines | length > 0

  - name: get archive filename
    shell: warn=no echo "audit.log.{{ ansible_date_time.epoch }}.tgz"
    register: audit_log_tgz
    changed_when: audit_log_tgz.stdout_lines | length != 1

  - name: touch archive file
      path: "{{ auditd_log_dir }}/../{{ audit_log_tgz.stdout }}"
      state: touch
      owner: root
      group: root
      mode: 0600
    when: cleanup_list.stdout_lines | length > 0

  - name: archive and cleanup existing audit.log.1 files
      dest: "{{ auditd_log_dir }}/../{{ audit_log_tgz.stdout }}"
      path: "{{ cleanup_list.stdout_lines }}"
      format: gz
      owner: root
      group: root
      remove: yes
    ignore_errors: yes
    when: cleanup_list.stdout_lines | length > 0

  - name: check for existence of new tarball
      path: "{{ auditd_log_dir }}/../{{ audit_log_tgz.stdout }}"
    ignore_errors: yes
    register: audit_log_tarball

  - name: place audit log tarball in auditd_log_dir
    shell: warn=no /bin/mv "{{ auditd_log_dir }}/../{{ audit_log_tgz.stdout }}" "{{ auditd_log_dir }}/"
    ignore_errors: yes
    - audit_log_tarball.stat.exists is defined
    - audit_log_tarball.stat.exists

  - name: get current size of audit log
      path: "{{ auditd_log_dir }}/audit.log"
    ignore_errors: yes
    register: audit_log_stat

  - name: apply logrotate script for audit
      src: /etc/ansible/templates/etc-logrotate.d-audit
      dest: "{{ auditd_logrotate_conf }}"
      owner: root
      group: root
      mode: 0644
      backup: yes

  - name: learn the logrotate.status file to use, if any
    shell: warn=no grep -rE -- 'bin\/logrotate\>.*(-s|--state)(\s|=)[\/[A-Za-z0-9\.]+\>' /etc/cron.* 2>/dev/null | grep -oE '(-s|--state)(\s|=)[\/[A-Za-z0-9\.]+\>' | sort | uniq | head -n1
    ignore_errors: yes
    changed_when: false
    register: this_logrotate_flag

  - name: show which logrotate.status file to use, if any
      msg: "The status file that will be used is {{ this_logrotate_flag.stdout }}"

  - name: run logrotate
    shell: warn=no /usr/sbin/logrotate {{ this_logrotate_flag.stdout }} -f "{{ auditd_logrotate_conf }}"
    register: run_logrotate
    when: ( cleanup_list.stdout_lines | length > 0 ) or ( audit_log_stat.stat.exists and audit_log_stat.stat.size > 190000000 )



So, logrotate can be configured to rotate the audit log. It just takes a few minutes to configure correctly, after about 2 weeks of research and testing.



  1. http://melikedev.com/2013/08/19/linux-selinux-semodule-compile-pp-module-from-te-file/
  2. https://linux.die.net/man/8/auditd.conf

Personal effort

Hours and hours of my original research
Years of administering RHEL servers with logrotate


Docker cannot write to mounted volume

So you’ve already investigated the permissions, and the selinux context. There are no errors in the audit logs.

And if you’re using a directory like /var/lib/docker/db, it will have context unconfined_u:object_r:container_var_lib_t:s0.

For mounting with -v /var/lib/docker/db/appname:/opt/application/ and it to be readable, you will need a new context.

semanage fcontext -a -t svirt_sandbox_file_t '/var/lib/docker/db(/.*)?'

Configure SELinux to allow Nagios publickey auth

Nagios is a tool for monitoring servers. In a security-minded environment, you need to make allowances for nagios. It operates over ssh using a public key, which SELinux doesn’t like.

One problem that can occur is that the ~nagios/.ssh/authorized_keys file will not have the right selinux context. Fix that with

semanage fcontext -a -t "ssh_home_t" "/var/spool/nagios(/.*)?"
restorecon -RvF /var/spool/nagios

This will make a new rule in selinux for that directory to have a regular ssh-homedir context, so public keys will work properly. If nagios cannot connect passwordlessly, it will throw fits.