why I have lost messages on boot even with very big backlog while I hunting only 2 syscalls?
Lev Olshvang
levonshe at yandex.com
Sat Sep 30 12:48:23 UTC 2017
28.09.2017, 17:02, "Steve Grubb" <sgrubb at redhat.com>:
> Hello,
>
> On Thursday, September 28, 2017 4:51:38 AM EDT Lev Olshvang wrote:
>> 28.09.2017, 00:32, "Steve Grubb" <sgrubb at redhat.com>:
>> > On Wednesday, September 27, 2017 4:41:29 PM EDT Lev Olshvang wrote:
>> >> Hello list !
>> >>
>> >> A very technical question
>> >> I have Ubuntu 16.10 Virtual Box , auditd 2.7.8
>> >> I have audit=1 parameter in grub.cfg
>> >> I see that /proc/cmdline indeed sees it
>> >>
>> >> I see that auditd is started with PID 564
>> >>
>> >> root 312 2 0 23:12 ? 00:00:00 [kauditd]
>> >> root 564 1 0 23:12 ? 00:00:00 /sbin/auditd
>> >>
>> >> And I have 15 lost messages ???
>> >> auditctl -s
>> >> enabled 1
>> >> failure 1
>> >> pid 564
>> >> rate_limit 0
>> >> backlog_limit 16384
>> >> lost 15
>> >> backlog 0
>> >> backlog_wait_time 30
>> >> loginuid_immutable 0 unlocked
>> >>
>> >> auditctl -l
>> >> -a always,exit -F arch=b64 -S execve,execveat -F key=exec
>> >>
>> >> Do I understand correctly that auiditd is indeed started by systemd
>> >> before
>> >> other services, except 2 that is listed in auditd.service dependencuies
>> >> -
>> >> local-fs and some temp setup of systemd ?
>> >
>> > Yes, it is started before most services. However. systemd-journal for some
>> > reason feels obligated to enable auditing. And sometimes people put
>> > audit=1 on the kernel command line. Either way, auditing is on way before
>> > auditd starts. The audit logs have a 64 entry buffer by default. So, as
>> > the system boots events pile up and eventually overflows the 64 entry
>> > limit.
>> >
>> > The fix is to add another boot command option audit_backlog_limit=8192 or
>> > some other suitable number. The test to check for this is to boot your
>> > system, login and run auditctl -s. If you have just booted and lost
>> > events during boot, this should fix it.
>> >
>> > -Steve
>>
>> Hi Steve
>>
>> Thank you for your answer.
>> I added backlog parameter as you advised, but it did not solve the problem
>>
>> cat /proc/cmdline
>> BOOT_IMAGE=/vmlinuz-4.8.0-59-generic root=/dev/mapper/kubuntu--vg-root ro
>> net.ifnames=0 biosdevname=0 audit=1 audit_backlog_limit=8192 debug splash
>> auditctl -s
>> enabled 1
>> failure 1
>> pid 672
>> rate_limit 0
>> backlog_limit 16384
>> lost 16
>> backlog 10
>> backlog_wait_time 30
>> loginuid_immutable 0 unlocked
>>
>> Perhaps something else in configuration ?
>
> You have a backlog of 10. That should normally be 0 unless the system is very
> busy. What do you have for the flush and freq settings in auditd.conf?
>
> -Steve
Hi Steve,
I overloked your mail yesterday, sorry for delay.
Here the auditd.conf
local_events = yes
write_logs = yes
log_format = RAW
log_file = /var/log/audit/audit.log
log_group = root
priority_boost = 16
flush = INCREMENTAL_ASYNC
freq = 20
num_logs = 5
disp_qos = lossy
I increased priority_boost from 4 to 16 in a hope to solve lost messages problem.
I observed other values of backlog, it was sometimes 6, sometimes 7.
Today I made very big backlog, here are results
enabled 1
failure 1
pid 663
rate_limit 0
backlog_limit 32768
lost 15
backlog 0
backlog_wait_time 15000
Still 15 losts, now events in backlog
Perhaps I need to add some tracer to lost messages code in kernel to debug it.
Regards,
Lev
More information about the Linux-audit
mailing list