| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
New authority monitor code requires the USN plugin to be
enabled in the database to ensure that the entryUSN attribute
is added to authority entries.
In the case where this plugin was disabled, accessing this
attribute resulted in a null pointer exception whch prevented server
startup.
The code has been changed so as not to throw a null pointer exception
on startup if the entryusn is not present, and also to call an LDIF
to enable the plugin when a subsystem is configured through pkispawn.
|
|
|
|
|
|
|
|
|
|
| |
When a lightweight CA is created, clones will initialise a local
object when the LDAP replication takes place, however, the signing
keys will not yet have been replicated. Therefore, indicate CA
readiness in authority data and respond appropriately (HTTP 503)
when signing operations are attempted.
Part of: https://fedorahosted.org/pki/ticket/1625
|
|
|
|
|
|
|
|
|
|
|
|
| |
When initialising a lightweight CA, if we do not have the signing
cert and key in the NSSDB yet, we do not initialise the DN. This
causes NPE in other code that expects getX500Name() to return a
value, e.g. REST API to list or show CA.
To work around this, when loading lightweight CAs set the DN based
on the 'authorityDN' value stored in its LDAP entry.
Part of: https://fedorahosted.org/pki/ticket/1625
|
|
|
|
|
|
|
|
|
|
|
| |
Implement a thread that performs an LDAP persistent search to keep a
running CA's view of lightweight CAs in sync with the database.
Signing key replication is not yet supported; this will be
implemented in a later patch and will not use the database to
propagate keys.
Part of: https://fedorahosted.org/pki/ticket/1625
|
|
|
|
|
|
|
|
|
| |
LDAP code to add, modify and delete authority entries exists in
multiple places. Extract these methods to remove this duplication
and provide a cleaner basis for upcoming implementation of
replication handling.
Part of: https://fedorahosted.org/pki/ticket/1625
|
|
|
|
|
|
|
|
| |
To reduce the amount of code that would be run in the persistent
search thread, extract the host authority entry creation out of the
'loadLightweightCAs' method, into 'CertificateAuthority.init'.
Part of: https://fedorahosted.org/pki/ticket/1625
|
|
|
|
| |
Part of: https://fedorahosted.org/pki/ticket/1625
|
|
|
|
|
|
|
| |
Use a static database connection factory that is initialised by the
host authority and used by all CertificateAuthority instances.
Part of: https://fedorahosted.org/pki/ticket/1625
|
|
|
|
|
|
|
|
|
|
| |
Add the CAMissingCertException and CAMissingKeyException classes and
throw when signing unit initialisation fails due to a missing
object. In CertificateAuthority, store the exception if it occurs
for possible re-throwing later. Also add the private 'hasKeys'
field for internal use.
Part of: https://fedorahosted.org/pki/ticket/1625
|
|
|
|
|
|
|
|
| |
The CertificateAuthority.getCACert() has been modified to re-throw
the exception instead of ignoring it. All callers have been
modified to bubble up the exception.
https://fedorahosted.org/pki/ticket/1654
|
|
|
|
|
|
| |
If an OCSP request includes CertIDs for certificates issued by
multiple CAs, return 'unknown' CertStatus for all certificates not
issued by the "signing" CA.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch makes a low risk attempt to slow down the loop that could be
caused by an unexpected exception caused by the unavailability of a
dependant component (e.g. HSM, LDAP) in the middle of CRL generation/update.
New configuration parameters are:
ca.crl.MasterCRL.unexpectedExceptionWaitTime
- the wait time in minutes; default is 30
- normally you want it to be less than ca.crl.MasterCRL.autoUpdateInterval
and ca.crl.MasterCRL.cacheUpdateInterval
ca.crl.MasterCRL.unexpectedExceptionLoopMax
- the max number of tries allowed before the slow down mechanism kicks in;
default is 10
When such unexpected failure happens, a loop counter is kept and checked
against the unexpectedExceptionLoopMax. If the loop counter exceeds the
unexpectedExceptionLoopMax, then the current time is checked against the
time of the failure, where the time lapse must exceed the
unexpectedExceptionWaitTime to trigger a delay. This delay is the
counter measure to mitigate the amount of log messages that could flood
the log(s).
The delay is calcuated like this:
waitTime = mUnexpectedExceptionWaitTime - (now - timeOfUnexpectedFailure);
|
|
|
|
|
|
|
|
|
|
|
|
| |
Enrol new CA certs via the profile subsystem to ensure that the
usual audit events are logged and to avoid the nasty ConfigStore
hack used to generate the cert via CertUtil.
This commit also fixes an issue where the new CA certificate does
not have the correct Authority Key Identifier extension.
Fixes: https://fedorahosted.org/pki/ticket/1624
Fixes: https://fedorahosted.org/pki/ticket/1632
|
|
|
|
| |
Fixes: https://fedorahosted.org/pki/ticket/1628
|
|
|
|
|
|
|
|
|
|
|
|
| |
Lightweight CAs mean that a single database can include certificates
from many issuers. Update CRLIssuingPoint to only include
certificates issued by its associated CA.
For backwards compatibility, if the associated CA is the host CA,
certificate records with missing 'issuerName' attribute are also
included.
Fixes: https://fedorahosted.org/pki/ticket/1626
|
|
|
|
| |
client and server This patch provides subsystem->subsystem cipher configuration when acting as a client
|
|
|
|
|
|
|
| |
Implement lightweight authority deletion including CLI command. To
be deleted an authority must be disabled and have no sub-CAs.
Fixes: https://fedorahosted.org/pki/ticket/1324
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is an interim solution for supporting HSM failover by automatically
shutting down the server when signing key becomes inaccessible.
At auto-shutdown, a crumb fiile will be left in the instance directory
for an external daemon to detect and restart, if necessary.
Due to limitation of the watch dog (nuxwdog) at present time,
the restart option currently only works if started with watch dog (nuxwdog),
and it will prompt for passwords on the terminals.
The restart counter is to prevent the server from going into an infinite restart
loop. Administrator will have to reset autoShutdown.restart.count to 0 when max
is reached.
(cherry picked from commit 5a9ecad9172f76ca1b94b40aedcdd49d009aceb1)
|
|
|
|
|
|
|
|
|
| |
Some access to caMap was not correctly synchronized, with
authorities (of which there could be many) acquiring their own
intrinsic lock rather than the shared caMap.
Use 'Collections.synchronizedSortedMap' to fix this. As a bonus,
locking is now more fine-grained.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit adds initial support for "lightweight CAs" - CAs that
inhabit an existing CA instance and share the request queue and
certificate database of the "top-level CA".
We initially support only sub-CAs under the top-level CA - either
direct sub-CAs or nested. The general design will support hosting
unrelated CAs but creation or import of unrelated CAs is not yet
implemented.
Part of: https://fedorahosted.org/pki/ticket/1213
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We currently disable the cert status maintenance thread on
clone CAs because CRL processing should only be done on the
master CA. Currently, the maintenance thread also performs
other checks on serial number ranges and settings. By disabling
the maintenance thread, we disable these checks too.
To fix this, we have separated the serial number checks into a
different maintenance thread, so that these tasks will occur
even if the cert status thread is disabled.
Bugzilla # 1251606
|
|
|
|
|
|
|
|
|
|
|
| |
The HttpConnection class has been modified to support fail-over
and timeout more consistently. The targets are parsed into a list
during initialization. All direct calls to HttpClient.connect()
are replaced with a method that will connect to the first available
target. All connections are now created with a timeout (which by
default is 0).
https://fedorahosted.org/pki/ticket/891
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
schedules.
Addresses the complaint of this ticket. Tested to work in a few basic cases. The minor code change
was designed to only affect the specific scenario when we have a daily scedule that spans only one day.
More Info:
How to duplicate and test:
Perform a manual crl generate from the agent interface because the code to be tested relies heavily upon the "lastUpdate" which will appear in the logs. Do this to have a nice launching off point.
Go to the ca's pkiconsole and select : Certificate Manager -> CRL Issuing Points -> MasterCRL.
Check "updateCRL at: " and give a schedule such as : 15:03, 15:10 .. This gives us a chance to watch the two regularly scheduled updates happen.
When the first event triggers, have a look at the CA's "debug" log and note the following or similar entry:
[CRLIssuingPoint-MasterCRL]: findNextUpdate: Wed May 06 15:10:00 PDT 2015 delay: 86301873
Wait for the 15:00 even to happen. When that triggers at the end of that cycle, we should see one more similar entry.
[CRLIssuingPoint-MasterCRL]: findNextUpdate: Wed May 06 15:03 PDT 2015 delay: 86301873
That is the correct behavior after the fix. We want the next update to be at the first entry of the daily schedule , but tomorrow. The current bug would print out this value as something like:
Wed May 06 00:00:00 or similar to indicate midnight. This is not what we want.
|
| |
|
|
|
|
|
|
|
|
| |
The REST service classes have been moved into org.dogtagpki.server
namespace. A new upgrade script has been added to update existing
instances.
Ticket #114
|
|
|
|
|
|
|
|
|
|
| |
New ACL has been added to allow only the administrators to access
TPS authenticators.
The set of interceptors in each application has been modified to
preserve the order.
Ticket #652
|
|
|
|
|
|
|
| |
Due to a regression RESTEasy is unable to find some sub-resources properly.
As a workaround some resources need to be merged into the parent resource.
The UserCertResource and UserMembershipResource have been merged into
UserResource. The GroupMemberResource has been merged into GroupResource.
|
|
|
|
|
|
|
| |
A new REST service and clients have been added to manage the audit
configuration in all subsystems.
Ticket #652
|
|
|
|
|
|
|
| |
New REST service and clients have been added for managing selftests
in all subsystems.
Ticket #652
|
|
|
|
|
|
| |
The ACLInterceptor and AuthMethodInterceptor interceptors only run
on the server, so they have been moved from the base package into
the server package.
|
|
|
|
|
|
| |
This patch provides an option to generate CRLs with nextUpdate calculated as sum of thisUpdate and an offset.
Ticket #571
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
A new mechanism has been added to specify the authentication methods that
can be used to invoke the REST methods. The AuthMethodMapping annotation
maps each REST method to a list of allowed authentication methods. When a
client calls a REST method, the AuthMethodInterceptor will intercept the
call and verify that the client uses an allowed authentication method.
Most REST methods that require authentication have been configured to
require client certificate authentication. Authentication using username
and password will only be used to get the installation token from security
domain.
Ticket #477
|
|
|
|
|
|
|
|
| |
New CLI's have been added to search, add, and remove user membership.
The group member management code has been refactored into a processor
to allow reuse.
Ticket #190
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously nonces were stored in a global map which might not scale
well due to some issues:
1. The map uses the nonces as map keys. There were possible nonce
collisions which required special handling.
2. The collision handling code was not thread safe. There were
possible race conditions during concurrent modifications.
3. The map was shared and size limited. If there were a lot of
users using the system, valid nonces could get pruned.
4. The map maps the nonces to client certificates. This limits
the possible authentication methods that can be supported.
Now the code has been modified such that each user has a private map
in the user's session to store the nonces. Additional locking has been
implemented to protect against concurrent modifications. The map now
uses the target of the operation as the map key, eliminating possible
collisions and allowing the use of other authentication methods. Since
this is a private map, it's not affected by the number of users using
the system.
Ticket #474
|
|
|
|
|
|
|
|
|
|
|
| |
* Added RESTful servlet to add/remove a KRA connector from the CA.
* Modified ACL to allow KRA subsystem user to remove connector.
* Modified connector code to allow the connector to be replaced without a server restart.
* Added functionality to pki CLI to add/remove connector
* Added code to pkidestroy to remove the connector (using both pki CLI and sslget)
When the issues with pki connection are resolved, we will use that method instead.
* Modified sslget to accept HTTP return codes != 200. In this case, we were returning
204 - which is perfectly legitimate.
|
|
|
|
|
|
|
|
|
| |
Previously ACL checking was done in PKIRealm by matching the URL.
This code has been replaced by ACLInterceptor which will intercept
RESTEasy method invocations. This allows more precise mapping of
REST methods to ACL entries in acl.ldif.
Ticket #287
|
|
|
|
|
|
|
|
| |
Some synchronized methods in CertificateRepository may block
modifyCeritifcateRecord() too long, so they have been moved
into CRLIssuingPoint and CertStatusUpdateThread.
Ticket #313
|
|
|
|
|
|
|
|
| |
The CertificateAuthorityApplication has been modified to deploy
the REST service for security domain only if the server has been
configured with a new security domain.
Ticket #309
|
|
|
|
|
|
|
|
|
| |
A REST account service has been added to allow client to login
to establish a session and to logout to destroy the session. This
way multiple operations can be executed using the same session
without having to re-authenticate.
Ticket #357
|
|
|
|
|
|
|
|
| |
The REST interface for security domain has been refactored and
configured such that it requires authentication. A CLI has been
added to get an installation token.
Ticket #309
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The shutdown() methods in several classes have been fixed to allow
more graceful shutdown and clean restart. There are two types of
object attributes that need to be handled differently.
Attributes that are initialized by the constructor should not be
nulled during shutdown because they won't be reinitialized during
restart. If they require a cleanup (e.g. emptying collections,
closing LDAP connections) it's not necessary to check for null
before calling the cleanup method because they're never null.
For attributes that are initialized during init(), it may not be
necessary to do a cleanup or null the attribute since they might
still be used by other threads and they will be reinitialized
during restart so the old objects will be garbage collected. If
they do need a cleanup they should be checked for null because
they might still be null due to init() failure or initialization
conditionals.
If the attributes are initialized conditionally, the logic has been
modified to ensure the attributes are either initialized or set to
null.
Ticket #247
|
|
|
|
|
|
|
| |
The common classes used by REST client and services have been moved
into the com.netscape.certsrv.<component> packages.
Ticket #215
|
|
|
|
|
|
|
| |
The REST common classes have been renamed for better clarity
and consistency.
Ticket #259
|
|
|
|
|
|
|
| |
The REST server classes have been renamed for better clarity
and consistency.
Ticket #259
|
| |
|
| |
|