Alex,
This is a little outside my experience.
One assumes the audit_failure variable has been set in the kernel
(kernel/audit.c). Perhaps you can test this.
Given you can get a copy of the kernel source you are running, perhaps
trace through what's happening. Using the messages
before/during/directly after the death of auditd, and what's routing to
dmesg, perhaps you can reverse engineer what is happening.
Perhaps someone else on the list can explain why, given -f is set to 0,
and the kernel has no user space destination for audit, it still prints
(via printk()?)
Regards
On Thu, 2015-08-20 at 13:17 +0300, Alex Beljanski wrote:
 We have custom audit-dispatcher for process events. On some servers
 when auditd fails, all audit messages writes to kernel. 
 We don't want to see all this messages in dmesg and set failure flag
 to "0". This doesn't help. 
 
 
 # cat /etc/audit/auditd.conf
 
 log_file = /var/log/audit/audit.log
 log_format = NOLOG
 log_group = root
 priority_boost = 4
 flush = none
 num_logs = 1
 disp_qos = lossy
 dispatcher = /sbin/audit-dispatcher
 name_format = none
 max_log_file = 1
 max_log_file_action = keep_logs
 space_left = 75
 space_left_action = ignore
 admin_space_left = 50
 admin_space_left_action = ignore
 disk_full_action = ignore
 disk_error_action = ignore
 enable_krb5 = no
 
 cat /etc/audit/rules.d/audit.rules 
 
 -D
 
 -b 8192
 
 -f 0
 -e 1
 
 -a exit,always -F arch=b32 -S 11 -k exec32
 -a exit,always -F arch=b64 -S 59 -k exec64
 
 
 
 
 2015-08-20 12:39 GMT+03:00 Burn Alting <burn(a)swtf.dyndns.org>:
         Alex,
         
         Can you provide a little more detail?
         
         Perhaps your /etc/audit/auditd.conf, /etc/audit/rules.d/*,
         your test
         case, the expected outcome and the outcome you actually get.
         
         Regards
         
         On Thu, 2015-08-20 at 11:09 +0300, Alex Beljanski wrote:
         > Hi!
         >
         >
         > We have problem in CentOS 7 with auditd.
         >
         > For our servers we set failure flag 0, but kernel write
         messages and
         > we see them in dmesg.
         >
         > uname -a
         > Linux 3.10.0-229.11.1.el7.x86_64 #1 SMP Thu Aug 6 01:06:18
         UTC 2015
         > x86_64 x86_64 x86_64 GNU/Linux
         >
         > # rpm -qa | grep audit
         > audit-2.4.1-5.el7.x86_64
         >
         >
         > Why this doesn't work?
         >
         >
         >
         >
         >
         
         > --
         > Linux-audit mailing list
         > Linux-audit(a)redhat.com
         > 
https://www.redhat.com/mailman/listinfo/linux-audit