Roderick Johnstone via FreeIPA-users wrote:
On 25/01/2018 16:56, Roderick Johnstone via FreeIPA-users wrote:
> On 25/01/2018 13:43, Rob Crittenden via FreeIPA-users wrote:
>> Roderick Johnstone via FreeIPA-users wrote:
>>> On 24/01/2018 21:09, Rob Crittenden via FreeIPA-users wrote:
>>>> Roderick Johnstone via FreeIPA-users wrote:
>>>>> On 24/01/2018 15:22, Rob Crittenden via FreeIPA-users wrote:
>>>>>> Roderick Johnstone via FreeIPA-users wrote:
>>>>>>> On 23/01/2018 14:34, Rob Crittenden via FreeIPA-users wrote:
>>>>>>>> Roderick Johnstone via FreeIPA-users wrote:
>>>>>>>>> On 15/01/2018 20:07, Rob Crittenden via FreeIPA-users
wrote:
>>>>>>>>>> Roderick Johnstone via FreeIPA-users wrote:
>>>>>>>>>>> On 15/01/2018 16:06, Rob Crittenden via
FreeIPA-users wrote:
>>>>>>>>>>>> Roderick Johnstone via FreeIPA-users
wrote:
>>>>>>>>>>>>> Hi
>>>>>>>>>>>>>
>>>>>>>>>>>>> Our freeipa certificates need to be
renewed due to passing
>>>>>>>>>>>>> their
>>>>>>>>>>>>> expiry
>>>>>>>>>>>>> dates.
>>>>>>>>>>>>>
>>>>>>>>>>>>> While some certificates have renewed
ok, the ipaCert and
>>>>>>>>>>>>> auditSigningCert are renewing but the
new certificates
>>>>>>>>>>>>> have the
>>>>>>>>>>>>> wrong
>>>>>>>>>>>>> Subject.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Environment is:
>>>>>>>>>>>>> serverA (CRL, first, master) RHEL
7.3, ipa 4.4
>>>>>>>>>>>>> serverB (replica) RHEL 7.3, ipa 4.4
>>>>>>>>>>>>> serverC (replica) RHEL 7.4, ipa 4.5
>>>>>>>>>>>>>
>>>>>>>>>>>>> Once there are renewed certificates
with the wrong Subject
>>>>>>>>>>>>> present,
>>>>>>>>>>>>> there are various problems with
renewing the remaining
>>>>>>>>>>>>> certificates,
>>>>>>>>>>>>> which I think might be related to the
bad Subject:
>>>>>>>>>>>>>
>>>>>>>>>>>>> 1) When just ipaCert has the wrong
subject no further
>>>>>>>>>>>>> renewals
>>>>>>>>>>>>> happen
>>>>>>>>>>>>>
>>>>>>>>>>>>> 2) When auditSigningCert has the
wrong subject the ipa
>>>>>>>>>>>>> pki-tomcatd
>>>>>>>>>>>>> service will not start and no further
renewals happen.
>>>>>>>>>>>>>
>>>>>>>>>>>>> I've been round the following
loop many times on ServerA, our
>>>>>>>>>>>>> first
>>>>>>>>>>>>> master:
>>>>>>>>>>>>>
>>>>>>>>>>>>> 1) Restore good certificates from
backup
>>>>>>>>>>>>> 2) Put the clock back to a time when
certificates are all
>>>>>>>>>>>>> valid
>>>>>>>>>>>>> 3) Resubmit certificates for renewal
>>>>>>>>>>>>>
>>>>>>>>>>>>> Each time the ipaCert renews it has
the same wrong
>>>>>>>>>>>>> Subject. The
>>>>>>>>>>>>> wrong
>>>>>>>>>>>>> Subject includes the host name of one
of our ipa client
>>>>>>>>>>>>> systems.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Each time the auditSigningCert renews
it has the same wrong
>>>>>>>>>>>>> Subject
>>>>>>>>>>>>> but
>>>>>>>>>>>>> a different subject to the ipaCert.
The wrong Subject in this
>>>>>>>>>>>>> case
>>>>>>>>>>>>> includes the host name of a system
which has never been an
>>>>>>>>>>>>> ipa
>>>>>>>>>>>>> client,
>>>>>>>>>>>>> but might have been added and removed
with ipa host-add
>>>>>>>>>>>>> and ipa
>>>>>>>>>>>>> host-del
>>>>>>>>>>>>> for testing something, a while ago.
>>>>>>>>>>>>>
>>>>>>>>>>>>> As far as I can see, the
"cert_subject" is set correctly
>>>>>>>>>>>>> in the
>>>>>>>>>>>>> file
>>>>>>>>>>>>> /var/lib/certmonger/<request
id> until the point at which the
>>>>>>>>>>>>> certificate is actually renewed.
>>>>>>>>>>>>>
>>>>>>>>>>>>> I'd be very grateful for some
pointers as to which
>>>>>>>>>>>>> configuration
>>>>>>>>>>>>> options
>>>>>>>>>>>>> and logs to check through to resolve
this problem on our
>>>>>>>>>>>>> production
>>>>>>>>>>>>> system.
>>>>>>>>>>>>>
>>>>>>>>>>>>> If its of any relevance we did change
which server is the
>>>>>>>>>>>>> first
>>>>>>>>>>>>> master
>>>>>>>>>>>>> some time ago.
>>>>>>>>>>>>
>>>>>>>>>>>> I'd pull the CSR out of dogtag
(CS.cfg) and/or certmonger
>>>>>>>>>>>> to see
>>>>>>>>>>>> what
>>>>>>>>>>>> the subject is.
>>>>>>>>>>>
>>>>>>>>>>> I'm not seeing any obvious CSR fields in
the
>>>>>>>>>>> /etc/pki/pki-tomcat/ca/CS.cfg file.
>>>>>>>>>>
>>>>>>>>>> foo.bar.certreq=
>>>>>>>>>>
>>>>>>>>>>> The CSR in the certmonger requests file for
the
>>>>>>>>>>> auditSigningCert
>>>>>>>>>>> seems
>>>>>>>>>>> to be showing with the correct Subject. This
is different from
>>>>>>>>>>> the bad
>>>>>>>>>>> subject showing in the requests file field:
>>>>>>>>>>> cert_subject=
>>>>>>>>>>
>>>>>>>>>> The value of cert_subject comes from the issued
certificate.
>>>>>>>>>>
>>>>>>>>>>> and the Subject which is showing in the
'getcert list' output
>>>>>>>>>>> (which is
>>>>>>>>>>> the same as that in the cert_subject=
field.>
>>>>>>>>>>> I'm not quite sure what this all means.
>>>>>>>>>>
>>>>>>>>>> It is displayed from the data within the tracked
certmonger
>>>>>>>>>> request.
>>>>>>>>>>
>>>>>>>>>> certmonger logs to syslog so you can check there
or you can stop
>>>>>>>>>> the
>>>>>>>>>> process and run it manually with: certmonger -n
-d 9 2>&1 | tee
>>>>>>>>>> certmonger.log
>>>>>>>>>>
>>>>>>>>>> That will provide a lot of debugging output that
may show
>>>>>>>>>> what is
>>>>>>>>>> going on.
>>>>>>>>>
>>>>>>>>> I've restored certificate databases from backup
and put the clock
>>>>>>>>> back
>>>>>>>>> to a time when certificates are valid and renewed
the
>>>>>>>>> ocspSigining
>>>>>>>>> certificate with:
>>>>>>>>> getcert resubmit -N "CN=OCSP
Subsystem,O=<REALM>" -i
>>>>>>>>> 20161124081302
>>>>>>>>>
>>>>>>>>> (I've previously tried without the -N with
similar results)
>>>>>>>>>
>>>>>>>>> What I am seeing in the certmonger logs is:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> 2017-10-23 00:05:28 [438] Located the key
'ocspSigningCert
>>>>>>>>> cert-pki-ca'.
>>>>>>>>> 2017-10-23 00:05:28 [438] Converted private key
'ocspSigningCert
>>>>>>>>> cert-pki-ca' to public key.
>>>>>>>>> 2017-10-23 00:05:28 [439] Located the certificate
>>>>>>>>> "ocspSigningCert
>>>>>>>>> cert-pki-ca".
>>>>>>>>> 2017-10-23 00:05:28 [440] 0x1d Certificate named
"ocspSigningCert
>>>>>>>>> cert-pki-ca" in token "NSS Certificate
DB" in database
>>>>>>>>> "/etc/pki/pki-tomcat/alias" will not be
valid after
>>>>>>>>> 20171025122401.
>>>>>>>>> 2017-10-23 00:05:28 [442] Located the key
'ocspSigningCert
>>>>>>>>> cert-pki-ca'.
>>>>>>>>> 2017-10-23 00:05:28 [442] Converted private key
'ocspSigningCert
>>>>>>>>> cert-pki-ca' to public key.
>>>>>>>>> 2017-10-23 00:05:28 [443] Located the certificate
>>>>>>>>> "ocspSigningCert
>>>>>>>>> cert-pki-ca".
>>>>>>>>> 2017-10-23 00:05:28 [444] Located the key
'ocspSigningCert
>>>>>>>>> cert-pki-ca'.
>>>>>>>>> 2017-10-23 00:05:28 [444] Converted private key
'ocspSigningCert
>>>>>>>>> cert-pki-ca' to public key.
>>>>>>>>> 2017-10-23 00:05:39 [581] Found a certificate with
the same
>>>>>>>>> nickname but
>>>>>>>>> different subject, removing certificate
"ocspSigningCert
>>>>>>>>> cert-pki-ca"
>>>>>>>>> with subject "CN=OCSP
Subsystem,O=<REALM>".
>>>>>>>>> 2017-10-23 00:05:39 [581] Imported certificate
"ocspSigningCert
>>>>>>>>> cert-pki-ca", got nickname "ocspSigningCert
cert-pki-ca".
>>>>>>>>> 2017-10-23 00:05:39 [583] Located the certificate
>>>>>>>>> "ocspSigningCert
>>>>>>>>> cert-pki-ca".
>>>>>>>>> 2017-10-23 00:05:39 [48576] Adding hook
>>>>>>>>> "/usr/libexec/ipa/certmonger/renew_ca_cert
"ocspSigningCert
>>>>>>>>> cert-pki-ca"" (0).
>>>>>>>>> 2017-10-23 00:10:43 [942] 0x1d Certificate named
"ocspSigningCert
>>>>>>>>> cert-pki-ca" in token "NSS Certificate
DB" in database
>>>>>>>>> "/etc/pki/pki-tomcat/alias" issued by CA
and saved.
>>>>>>>>>
>>>>>>>>> I now have a date valid ocspSigningCertificate, but
with the
>>>>>>>>> wrong
>>>>>>>>> subject, and a broken certificate system which will
no longer
>>>>>>>>> start.
>>>>>>>>>
>>>>>>>>> ipactl status
>>>>>>>>> ...
>>>>>>>>> pki-tomcatd Service: STOPPED
>>>>>>>>>
>>>>>>>>> I can't renew other expired certificates.
>>>>>>>>>
>>>>>>>>> I also note that there is now no key for
>>>>>>>>> ocspSigningCertificate as
>>>>>>>>> shown
>>>>>>>>> by:
>>>>>>>>> certutil -K -d /etc/pki/pki-tomcat/alias
>>>>>>>>>
>>>>>>>>> I wonder if this is because the certificate subject
changed?
>>>>>>>>> There
>>>>>>>>> was a
>>>>>>>>> key before the certificate renewed.
>>>>>>>>>
>>>>>>>>> The ca debug logs are showing:
>>>>>>>>>
>>>>>>>>> [23/Oct/2017:00:55:47][localhost-startStop-1]: Found
cert by
>>>>>>>>> nickname:
>>>>>>>>> 'ocspSigningCert cert-pki-ca' with serial
number: 268370108
>>>>>>>>> [23/Oct/2017:00:55:47][localhost-startStop-1]:
converted to
>>>>>>>>> x509CertImpl
>>>>>>>>> [23/Oct/2017:00:55:47][localhost-startStop-1]:
SigningUnit:
>>>>>>>>> Certificate
>>>>>>>>> object not found
>>>>>>>>> Certificate object not found
>>>>>>>>> at
com.netscape.ca.SigningUnit.init(SigningUnit.java:184)
>>>>>>>>>
>>>>>>>>> Any help in repairing my broken ipa system will be
much
>>>>>>>>> appreciated.
>>>>>>>>
>>>>>>>> Sorry for the delay. I think my previous reading of the
certmonger
>>>>>>>> csrgen code was wrong.
>>>>>>>>
>>>>>>>> IIRC in your certmonger entry the cert_subject has the
hostname
>>>>>>>> value
>>>>>>>> right? Do you also have cert_subject_der?
>>>>>>>>
>>>>>>>> You can decode that by:
>>>>>>>>
>>>>>>>> 1. Running a hex-to-bin, something hacky like this in
python:
>>>>>>>>
>>>>>>>> import binascii
>>>>>>>>
>>>>>>>> hex_string = "<hex value>"
>>>>>>>>
>>>>>>>> binary_string = binascii.unhexlify(hex_string)
>>>>>>>>
>>>>>>>> fd = open("out", "wb")
>>>>>>>> fd.write(binary_string)
>>>>>>>> fd.close()
>>>>>>>>
>>>>>>>> 2. Run: openssl asn1parse -in out -inform der
>>>>>>>>
>>>>>>>> I'm going to assume this also has the hostname
encoded in it.
>>>>>>>
>>>>>>> Hi Again
>>>>>>>
>>>>>>> Yes, correct. The cert_subject_der does have the bad CN with
the
>>>>>>> hostname encoded in it.
>>>>>>>
>>>>>>>>
>>>>>>>> Can you try this:
>>>>>>>>
>>>>>>>> 1. Make a backup of the request file (just in case) >
2. Remove
>>>>>>>> cert_subject_der
>>>>>>>> 3. Modify cert_subject to be CN=OCSP
Subsystem,O=<YOUR_REALM>
>>>>>>>> 3. Try the renewal again
>>>>>>>
>>>>>>> I ran though all of this, checking that the request file was
still
>>>>>>> what
>>>>>>> we wanted after restarting certmonger with verbose logging,
>>>>>>> restoring
>>>>>>> all the databases, fixing permissions, checking keys etc.,
and
>>>>>>> ran the
>>>>>>> getcert resubmit.
>>>>>>>
>>>>>>> It still generates a certificate with the bad CN including a
>>>>>>> hostname.
>>>>>>>
>>>>>>> Then the pki-tomcat server fails to start again.
>>>>>>>
>>>>>>> Other things I have checked are:
>>>>>>> 1) The csr= field in the (modified) request file has the
correct
>>>>>>> subject.
>>>>>>>
>>>>>>> 2) The dogtag ca debug log file is showing:
>>>>>>> processRenewal: origNotAfter =Wed Oct 25 13:24:01 BST 2017
>>>>>>> processRenewal: orig subj dn =CN=OCSP
Subsystem,O=AST.CAM.AC.UK
>>>>>>> ...
>>>>>>> RenewalSubmitter: renewal original
profileId=caIPAserviceCert
>>>>>>> RenewalSubmitter: for renewal, original authenticator
raCertAuth
>>>>>>> found
>>>>>>> CertProcessor: No authenticator credentials required
>>>>>>> processRenewal: set original Inputs into profile Context
>>>>>>> RenewalSubmitter: setInputsIntoContext() getting input name=
>>>>>>> cert_request_type
>>>>>>> RenewalSubmitter: setInputsIntoContext() setting value in
>>>>>>> ctx:pkcs10
>>>>>>> RenewalSubmitter: setInputsIntoContext() getting input name=
>>>>>>> cert_request
>>>>>>> RenewalSubmitter: setInputsIntoContext() setting value in
ctx:<hex
>>>>>>> encoded csr>
>>>>>>> ...
>>>>>>> Finish parsePKCS10 - CN=<HOSTNAME>,O=<REALM>
>>>>>>>
>>>>>>> The hex encoded csr in the last line decodes to have the bad
>>>>>>> subject
>>>>>>> where the hostname is one of our other ipa servers.
>>>>>>>
>>>>>>> The certificate always getthe same hostname in the bad
subject cn.
>>>>>>>
>>>>>>> Do you have any other ideas of things to check to find out
what's
>>>>>>> generating the bad subject?
>>>>>>
>>>>>> The order certmonger uses for the subject is:
>>>>>>
>>>>>> 1. cert_subject_der
>>>>>> 2. If there is no DER, try to encode cert_subject
>>>>>> 3. If that fails try again a different way
>>>>>> 4. If it fails yet again use CN=localhost
>>>>>>
>>>>>> It is baffling that it would pick a hostname much less one that
>>>>>> certmonger shouldn't even know about.
>>>>>
>>>>> Indeed, and if I try to renew other certificates for some of them it
>>>>> chooses other host names that should not be known about. Each
>>>>> certificate seems to get the same bad hostname each time I try to
>>>>> renew.
>>>>>
>>>>>>
>>>>>> I assume the CSR found in the dogtag logs matches the csr value
>>>>>> in the
>>>>>> certmonger request?
>>>>>
>>>>> No, its different. The issued certificate matches the csr seen in
>>>>> dogtag
>>>>> which makes sense. But the csr in the dogtag logs has the bad
>>>>> subject.
>>>>> The csr in the request directory file has the good subject.
>>>>>
>>>>>>
>>>>>> Can you share the certmonger request file privately?
>>>>>
>>>>> Yes, certainly.
>>>>>
>>>>> Thanks for your continued help on this.
>>>>
>>>> Let's try this. We'll drop the current request and create a new
one.
>>>>
>>>> Stop tomcat, restore the old cert database, start tomcat, then:
>>>>
>>>> # getcert stop-tracking -i <request id>
>>>> # getcert start-tracking -c dogtag-ipa-ca-renew-agent -n
>>>> "ocspSigningCert cert-pki-ca" -d /etc/pki/pki-tomcat/alias -P
<pin> -B
>>>> /usr/libexec/ipa/certmonger/stop_pkicad -C
>>>> '/usr/libexec/ipa/certmonger/renew_ca_cert "ocspSigningCert
>>>> cert-pki-ca"'
>>>>
>>>> Then try resubmitting the request.
>>>>
>>> Hi Rob
>>>
>>> When you say 'stop tomcat', I'm just doing an ipactl stop. That
is
>>> stopiing the ipa pki-tomcat service. Is that good enough or are there
>>> other services that need stopping too?
>>>
>>> I tried the stop-tracking/start-tracking. Exactly the same result as
>>> before. Same hostname appears in the subject CN field of the new
>>> certificate and then the pki-tomcat service won't start.
>>>
>>> I've also been back and redone the manual submits changing the point at
>>> which I copied in the edited request as I wondered if there was some
>>> caching of csr's going on in certmonger and it was remembering a
>>> previous request it had read on startup before I replaced the
>>> request file.
>>>
>>> But nothing seems to stop dogtag issuing a certificate with the Subject
>>> CN being that host name.
>>>
>>> Is it possible that dogtag is somehow overriding the Subject its being
>>> explicitly given?
>>>
>>> FWIW the CSR in the dogtag debug log is always identical, but I guess
>>> thats expected if the CSR information is always the same.
>>>
>>> I'm not sure if its possible to increase the verbosity on the dogtag
>>> side. Can you advise please?
>>>
>>> I've been assuming that its the bad subject that is preventing the
>>> pki-tomcat from starting after the new certificate is issued. Does that
>>> make sense to you?
>>>
>>> I wonder where we can go from here? There must be a good reason for
>>> whats happening!
>>
>> Let's step back and gather some more information.
>>
>> Can you restore the files again and send me the output of:
>>
>> # getcert list
>>
>> # certutil -L -d /etc/pki/pki-tomcat/alias/ -n "ocspSigningCert
>> cert-pki-ca"
>
> Hi Rob
>
> You should have the output you requested by private email now.
>
>>
>> And the exact commands you are using to do all this, including
>> re-issuing the cert
>
> Thats a great idea. Maybe I'm just doing something in the wrong order.
>
>>
>> Using ipactl is fine. You just don't want to touch the NSS databases
>> while the service is running. Similarly you probably don't want to mess
>> with certmonger requests while it is running as it won't see the changes
>> until it restarts.
> Here is the procedure I'm using to renew the certificates. One of the
> tricky things is to not have the certificate system working when
> certmonger is started so that it doesn't renew a certificate on
> startup that might not be one I have changed the request file for
> (other certificates have similar problems with bad subject CN and stop
> the certificate system.)
>
> I've found it convenient to keep reinitializing the log files because
> they get very confusing when the clock keeps getting set back.
>
> The request number has changed since we did the stop-tracking /
> start-tracking.
>
> Roderick
>
> ipactl stop
>
> # Set clock back:
> systemctl stop ntpd
> timedatectl set-time "2017-10-23 00:00:00"
> date
>
> #Stop certificate renewals the systemd way
> systemctl stop certmonger
>
> # Sort out certmonger log
> mv /var/log/certmonger.log /var/log/certmonger.log-$(date +%Y%m%d:%H:%M)
>
> # ****** Copy in the fixed certmonger request
> \cp /root/20161124081302.ocsp_backup
> /var/lib/certmonger/requests/20161124081302
>
>
> # In another window
> certmonger -n -d 9 2>&1 | tee /var/log/certmonger.log
>
> # In the original window
> # Sort out dogtag log
> mv /var/log/pki/pki-tomcat/ca/debug
> /var/log/pki/pki-tomcat/ca/debug-$(date +%Y%m%d:%H:%M).log
> touch /var/log/pki/pki-tomcat/ca/debug
> chown pkiuser:pkiuser /var/log/pki/pki-tomcat/ca/debug
>
> mv /var/log/pki/pki-tomcat/ca/selftests.log
> /var/log/pki/pki-tomcat/ca/selftests.log-$(date +%Y%m%d:%H:%M)
> touch /var/log/pki/pki-tomcat/ca/selftests.log
> chown pkiuser:pkiuser /var/log/pki/pki-tomcat/ca/selftests.log
>
> \cp /var/tmp/rmj/etc/httpd/alias/cert8.db /etc/httpd/alias/cert8.db
> \cp /var/tmp/rmj/etc/httpd/alias/key3.db /etc/httpd/alias/key3.db
> \cp /var/tmp/rmj/etc/pki/pki-tomcat/alias/cert8.db
> /etc/pki/pki-tomcat/alias/cert8.db
> \cp /var/tmp/rmj/etc/pki/pki-tomcat/alias/key3.db
> /etc/pki/pki-tomcat/alias/key3.db
>
> \cp /var/tmp/rmj/etc/pki/pki-tomcat/ca/CS.cfg
> /etc/pki/pki-tomcat/ca/CS.cfg
>
> # Fix up certificate permissions and check all is ok
> certutil -M -n "caSigningCert cert-pki-ca" -d
> /etc/pki/pki-tomcat/alias -t CTu,Cu,Cu
> certutil -L -d /etc/pki/pki-tomcat/alias
> certutil -K -d /etc/pki/pki-tomcat/alias
> <pin>
>
> # Fix dogtag so it starts
> pki-server subsystem-enable ca >
> # Start IdM services
>
> ipactl start
> ipactl status
>
> #Resubmit certificate ocsp
> date;getcert resubmit -i 20161124081302
Hi Rob
Does it look like there is anything I should be doing differently in
this command sequence to try to get the certificate renewed correctly?
I'm starting to wonder whether I should look at alternative strategies
for recovering the freeipa servers if the certificates won't renew
correctly. Do you have any ideas on this please?
This looks reasonable. Have you tried stop-tracking and start-tracking I
suggested?
I'm still baffled how the name(s) are getting mixed up.
Something else you might want to try would be to move all the other
requests out of the way in case there is some sort of memory corruption
causing the hostname to get in there somehow.
rob