Performing I18N completely on the server, to leverage the
existing gettext architecture.
Also, the browser does not have access to the Language header.
Added the additional po files for a set of required languages
conflict with install/static/ipa.js was resolved.
Note that the addition of the .po files in this patch is necessary.
In order to get Transifex support, we need to update the LINGUAS
file with the languages for which we want support. If we don't
add the .po files in, they get automatically generated by the rpmbuild
process. Our implementation of gettext has a bug in it (It might
be F13 thing) where the the Plurals line is not getting correctly
transformed, which causes a build failure. However, since the
RPM would have the .po files anyway, we should revision control
the ones we have, even if they are empty.
Fixed the Bug reporting url to the original value.
Corrected the Chartype encoding for UK
Note that this doesn't rely on IPA having a configured DNS server.
It passes the host name to the resolver and doesn't try to do a lookup
within the IPA DNS directly (e.g. no internal LDAP search).
Tries to determine if a domain is included and if not then the IPA
domain is added. This won't do the right thing if there are multiple
configured subdomains.
ticket 106
Quick summary:
- use jQuery UI and jQuery BBQ libraries
- code restructuring
The patch has so many changes they can't be listed here. Many parts
of the code have been rewritten from scrach.
See freeipa-devel mailing list:
webUI code restructuring [wall of text, diagrams, ... you've been warned!]
2010-09-07
Make two krbV imports conditional. These aren't used during a client
install so should cause no problems.
Also fix the client installer to use the new env option in ipautil.run.
We weren't getting the krb5 configuration set in the environment because
we were overriding the environment to set the PATH.
ticket 136
We don't use certmonger to get certificates during installation because
of the chicken-and-egg problem. This means that the IPA web and ldap
certs aren't being tracked for renewal.
This requires some manual changes to the certmonger request files once
tracking has begun because it doesn't store a subject or principal template
when a cert is added via start-tracking.
This also required some changes to the cert command plugin to allow a
host to execute calls against its own service certs.
ticket 67
Fedora 14 introduced the following incompatiblities:
- the kerberos binaries moved from /usr/kerberos/[s]/bin to /usr/[s]bin
- the xmlrpclib in Python 2.7 is not fully backwards compatible to 2.6
Also, when moving the installed host service principals:
- don't assume that krbticketflags is set
- allow multiple values for krbextradata
ticket 155
The problem was that parameters with no values are automatically
set to None by the framework and it wasn't handled properly in
baseldap.py:get_attributes function. Also, there were two logical
bugs in details.js:
1) atttribute callback to update values were called for input elements
instead of dt elements
2) it was always trying to update the primary key
Using the host service principal one should be able to retrieve a keytab
for other services for the host using ipa-getkeytab. This required a number
of changes:
- allow hosts in the service's managedby to write krbPrincipalKey
- automatically add the host to managedby when a service is created
- fix ipa-getkeytab to return the entire prinicpal and not just the
first data element. It was returning "host" from the service tgt
and not host/ipa.example.com
- fix the display of the managedby attribute in the service plugin
This led to a number of changes in the service unit tests. I took the
opportunity to switch to the Declarative scheme and tripled the number
of tests we were doing. This shed some light on a few bugs in the plugin:
- if a service had a bad usercertificate it was impossible to delete the
service. I made it a bit more flexible.
- I added a summary for the mod and find commands
- has_keytab wasn't being set in the find output
ticket 68
This adds a new global option to the ipa command, -f/--no-fallback. If this
is included then just the server configured in /etc/ipa/default.conf is used.
Otherwise that is tried first then all servers in DNS with the ldap SRV record
are tried.
Create a new Local() Command class for local-only commands. The help
command is one of these. It shouldn't need a remote connection to execute.
ticket #15
Date: Tue, 10 Aug 2010 16:41:28 -0400
Subject: [PATCH 2/6] Add a new INTERNAL plugin that exports plugin meta-data into JSON.
This is required for the webUI, since we're dropping Genshi. *ehm* :)
You can't use this command on the CLI. It takes one optional argument:
the name of an IPA object. If not specified, meta-data for all objects
are returned.
setattr and addattr can now be used both to set all values of
ANY attribute. the last setattr always resets the attribute to
the specified value and all addattr append to it.
Examples:
user-mod testuser --setattr=title=msc
title: msc
user-mod testuser --setattr=title=msb
title: msb
user-mod testuser --addattr=title=msc
title: msb, msc
user-mod testuser --setattr=title=
title:
user-mod testuser --setattr=title=msc --addattr=msb
title: msc, msb
user-mod testuser --setattr=title=ing --addattr=bc
title: ing, bc
user-mod testuser --setattr=title=doc
title: doc
It's not very user friendly, but it's going to be used very very
rarely in special conditions in the CLI and we can use it to save
lots of JSON-RPC roundtrips in the webUI.
This version includes calling the validation of Params during the setting of the attrs.
To do this we need to break the link manually on both sides, the user and
the group.
We also have to verify in advance that the user performing this is allowed
to do both. Otherwise the user could be decoupled but not the group
leaving it in a quasi broken state that only ldapmodify could fix.
ticket 75
The remove member function in baseldap was not returning failures at all.
The add member function was only showing them in the group object.
Most of the magic is handled in baseldap. Each plugin just needs to define
object_name and object_name_plural. object_name must be all lower-case
because fake-attributes are created so membership can be broken out
per-object type. I left the plural name lower case as well.
ticket 85
The pattern validator by default displays the pattern that is being
matched against. This isn't helpful, particularly for very hairy patterns.
This adds a new parameter, pattern_errmsg, that is displayed on errors
if set.
ticket #11
This also requires a resolvable hostname on services as well. I want
people to think long and hard about adding things that aren't resolvable.
The cert plugin can automatically create services on the user's behalf when
issuing a cert. It will always set the force flag to True.
We use a lot of made-up host names in the test system, all of which require
the force flag now.
ticket #25
When a service has a certificate and the CA backend doesn't support
revocation (like selfsign) then we simply drop the old certificate in
preparation for adding a new one. We weren't setting the usercertificate
attribute to None so there was nothing to do in ldap_update().
Added a test case for this situation to ensure that re-issuing a certificate
works.
ticket #88
It returns the user prinicpal.
This is required by the webui, as the Kerberos credential mechanism in http
does not expose the cleartext prinicpal to the web browser.
This patch:
- bumps up the minimum version of python-nss
- will initialize NSS with nodb if a CSR is loaded and it isn't already
init'd
- will shutdown NSS if initialized in the RPC subsystem so we use right db
- updated and added a few more tests
Relying more on NSS introduces a bit of a problem. For NSS to work you
need to have initialized a database (either a real one or no_db). But once
you've initialized one and want to use another you have to close down the
first one. I've added some code to nsslib.py to do just that. This could
potentially have some bad side-effects at some point, it works ok now.
When the netgroup plugin was rebased it ended up using the member
attribute for its memberships and not memberuser/memberhost.
I also fixed this same attribute problem in the tests and tried to beef
them up a little. If nis/schema compat are enabled it will try to compare
the generated triplets with a known-good value.
This patch does the following:
- drops our in-tree x509v3 parser to use the python-nss one
- return more information on certificates
- make an API change, renaming cert-get to cert-show
- Drop a lot of duplicated code
I have to do some pretty low-level LDAP work to achieve this. Since
we can't read the key using our modlist generator won't work and lots of
tricks would be needed to use the LDAPUpdate object in any case.
I pulled usercertificate out of the global params and put into each
appropriate function because it makes no sense for service-disable.
This also adds a new variable, has_keytab, to service/host_show output.
This flag tells us whether there is a krbprincipalkey.
Add an optional search_attributes variable in case the attributes you
want to display by default aren't what you want to search on.
Also link in any cn=ipaconfig attributes that contain a comma-separated
list of attributes to search on.
This started as an effort to display a more useful error message in the
Apache error log if retrieving the schema failed. I broadened the scope
a little to include limiting the output in the Apache error log
so errors are easier to find.
This adds a new configuration option, startup_traceback. Outside of
lite-server.py it is False by default so does not display the traceback
that lead to the StandardError being raised. This makes the mod_wsgi
error much easier to follow.
This uses a new 389-ds plugin, Managed Entries, to automatically create
a group entry when a user is created. The DNA plugin ensures that the
group has a gidNumber that matches the users uidNumber. When the user is
removed the group is automatically removed as well.
If the managed entries plugin is not available or if a specific, separate
range for gidNumber is passed in at install time then User-Private Groups
will not be configured.
The code checking for the Managed Entries plugin may be removed at some
point. This is there because this plugin is only available in a 389-ds
alpha release currently (1.2.6-a4).
The problem was trying to operate directly on the ACI itself. I
introduced a new function, _aci_to_kw(), that converts an ACI
into a set of keywords. We can take these keywords, like those passed
in when an ACI is created, to merge in any changes and then re-create the
ACI.
I also switched the ACI tests to be declarative and added a lot more
cases around the modify operation.
EmptyModlist exception was generated by pwpolicy2-mod when modifying
policy priority only. It was because the priority attribute is stored
outside of the policy entry (in a CoS entry) and there was nothing
left to be changed in the policy entry.
This patch uses the new exception callbacks in baseldap.py classes
to catch the EmptyModlist exception and checks if there was really
nothing to be modified before reraising the exception.
It enables plugin authors to supply their own handlers for
ExecutionError exceptions generated by calls to ldap2 made from
the execute method of baseldap.py classes that extend CallbackInterface.
Summaries were appearing as "Gettext(...")
Embedded dictionaries, such as group membership failures, didn't have
labels so were basically just being dumped.
The DNS plugin is getting old, tired and already looking forward to his
pension in the Carribean. It will be replaced soon by a younger, faster,
safer, shorter (in terms of code) and more maintainable version.
Until that happens, here's some medicine for the old guy:
- proper output definitions: the DNS plugin was created before we
had the has_output attribute in place
- --all: this is related to the output definitions as
Command.get_options() adds the --all and --raw options automatically
if has_output contains entries
- dns-add-rr overwritting: missing .lower() caused records to be
overwritten everytime a new one was added from the CLI
I also changed the default value of the print_all argument in
textui.print_entry from False to True. It think it makes more sense this
way, because:
1) if order is None, it will still print something
2) if order is not None, it will print what's in order first and then the
rest
3) commands that care about the print_all argument have to set it in any
case, those that don't care usually want to print everything
This will alert the user that nothing was done and is handy when used
with --attr=''. This can be used to delete a non-required attribute but
can be set to any valid attribute, present or not. We should alert the
user if they attempt to delete a non-existant value.
This fixes:
- Consistent usage of priority vs cospriority in options
- Fixes bug introduced with recent patch where global policy couldn't be
updated
- Doesn't allow cospriority to be removed for groups (#570536)
- returns the priority with group policy so it can be displayed
- Properly unicode encode group names for display
Method overrides the Command get_output_params() method and only returns
the object params, not anything defined within the method itself. Return
those as well so they are displayed in output. Some care needs to be taken
to avoid returning duplicate values. In the case of duplicates the
value in obj.params wins.
A number of doc strings were not localized, wrap them in _().
Some messages were not localized, wrap them in _()
Fix a couple of failing tests:
The method name in RPC should not be unicode.
The doc attribute must use the .msg attribute for comparison.
Also clean up imports of _() The import should come from
ipalib or ipalib.text, not ugettext from request.
None is passed if the option is set with --minlife=''. This is a valid
use case to delete a non-required attribute. In this case we simply
don't do the math on None and things work as expected.
569847
As a consequence of using doc=_('some message') the _()
method was returning a Gettext instance, thus when optparse
was handed the help text it received a Gettext instance instead
of a basestring. optparse tried to operate on the Gettext instance
as if it were a basestring and since it wasn't threw an exception.
The fix is to promote (e.g. cast) the option.doc to unicode.
If the option.doc was a str it becomes unicode, if it was unicode
nothing happens, if it was Gettext (or any other object implementing
the __unicode__() method) object is converted to unicode via the
objects rules.
By the way, while debugging this I discovered strings which were not
localized, sometimes option.doc would be a str and sometimes a Gettext
object. In a subsequent patch I'll fix all those unlocalized doc
strings, but I don't want to bury this fix along with a load of
string fixes.
This is a temporary fix until we either use Params to determine
output type or treat integers differently from other binary values
internally (as unicode instead of str, for example).
The attributes displayed is now dependant upon their definition in
a Param. This enhances that, giving some level of control over how
the result is displayed to the user.
This also fixes displaying group membership, including failures of
adding/removing entries.
All tests pass now though there is still one problem. We need to
return the dn as well. Once that is fixed we just need to comment
out all the dn entries in the tests and they should once again
pass.
This primarily affects the installer. We want to log to the install/
uninstall file in DEBUG. This was getting reset to INFO causing lots of
details to not show in the logs.
Let the user, upon installation, set the certificate subject base
for the dogtag CA. Certificate requests will automatically be given
this subject base, regardless of what is in the CSR.
The selfsign plugin does not currently support this dynamic name
re-assignment and will reject any incoming requests that don't
conform to the subject base.
The certificate subject base is stored in cn=ipaconfig but it does
NOT dynamically update the configuration, for dogtag at least. The
file /var/lib/pki-ca/profiles/ca/caIPAserviceCert.cfg would need to
be updated and pki-cad restarted.
Add a new get_subject() helper and return the subject when retrieving
certificates.
Add a normalizer so that everything before and after the BEGIN/END
block is removed.
Need to add a few more places where the DN will not be automatically
normalized. The krb5 server expects a very specific format and normalizing
causes it to not work.
This profile enables subject validation and ensures that the subject
that the CA issues is uniform. The client can only request a specific
CN, the rest of the subject is fixed.
This is the first step of allowing the subject to be set at
installation time.
Also fix 2 more issues related to the return results migration.
The idnsUpdatePolicy takes a list of BIND dynamic update policies, each
of which must be terminated by ";". Also fix a minor error in the
documentation string.
Ignore NotImplementedError when revoking a certificate as this isn't
implemented in the selfsign plugin.
Also use the new type argument in x509.load_certificate(). Certificates
are coming out of LDAP as binary instead of base64-encoding.
The pyOpenSSL PKCS#10 parser doesn't support attributes so we can't identify
requests with subject alt names.
Subject alt names are only allowed if:
- the host for the alt name exists in IPA
- if binding as host principal, the host is in the services managedBy attr
The parsing bug was looking for the string 'version' expecting to find
the ACI version. This blew up with the attribute nsosversion. Use
the string 'version 3.0' instead.
The comparison bug appeared if neither ACI had a targetattr attribute.
It was trying to create a set out of a None which is illegal. If an
ACI doesn't have any targetattrs then return () instead.
This modifies the original patch by including a unit test, handling floats
when passed as unicode, and handling large magnitude values beyond maxint.
The INT parameter class was not respecting any radix prefix (e.g. 0x) the user
may have supplied. This patch implements _convert_scalar method for the Int
class so that we can pass the special radix base of zero to the int constructor
telling it to determine the radix from the prefix (if present).
Signed-off-by: John Dennis <jdennis@redhat.com>
If plugin fails to load log the traceback
If a plugin fails to load due to some kind of error it would be nice
if the error log contained the traceback so you can examine what went
wrong rather than being left blind as to why it failed to load.
If an exception is not handled here then the context isn't destroyed
leaving at least an LDAP connection dangling. This means the next time
this thread/process tries to handle a connection it will fail because
a context already exists.
This introduces 2 new params: --setattr and --addattr
Both take a name/value pair, ala:
ipa user-mod --setattr=postalcode=20601 jsmith
--setattr replaces or sets the current attribute to the value
--addattr adds the value to an attribute (or sets a new attribute)
OptionsParser allows multiple versions of this, so you can have multiple
setattr and addattr, either for the same attribute or for different
attributes.
ipa user-mod --addattr=postalcode=20601 --addattr=postalcode=30330 jsmith
Values are silent dropped if either of these on an existing param:
ipa user-mod --setattr=givenname=Jerry jsmith
Is a no-op.
Using the client IP address was a rather poor mechanism for controlling
who could request certificates for whom. Instead the client machine will
bind using the host service principal and request the certificate.
In order to do this:
* the service will need to exist
* the machine needs to be in the certadmin rolegroup
* the host needs to be in the managedBy attribute of the service
It might look something like:
admin
ipa host-add client.example.com --password=secret123
ipa service-add HTTP/client.example.com
ipa service-add-host --hosts=client.example.com HTTP/client.example.com
ipa rolegroup-add-member --hosts=client.example.com certadmin
client
ipa-client-install
ipa-join -w secret123
kinit -kt /etc/krb5.keytab host/client.example.com
ipa -d cert-request file://web.csr --principal=HTTP/client.example.com
We want to only allow a machine to request a certificate for itself, not for
other machines. I've added a new taksgroup which will allow this.
The requesting IP is resolved and compared to the subject of the CSR to
determine if they are the same host. The same is done with the service
principal. Subject alt names are not queried yet.
This does not yet grant machines actual permission to request certificates
yet, that is still limited to the taskgroup request_certs.
Use a Class of Service template to do per-group password policy. The
design calls for non-overlapping groups but with cospriority we can
still make sense of things.
The password policy entries stored under the REALM are keyed only on
the group name because the MIT ldap plugin can't handle quotes in the
DN. It also can't handle spaces between elements in the DN.
- The aci plugin didn't quite work with the new ldap2 backend.
- We already walk through the target part of the ACI syntax so skip that
in the regex altogether. This now lets us handle all current ACIs in IPA
(some used to be ignored/skipped)
- Add support for user groups so one can do v1-style delegation (group A
can write attributes x,y,z in group B). It is actually quite a lot more
flexible than that but you get the idea)
- Improve error messages in the aci library
- Add a bit of documentation to the aci plugin
This will create a host service principal and may create a host entry (for
admins). A keytab will be generated, by default in /etc/krb5.keytab
If no kerberos credentails are available then enrollment over LDAPS is used
if a password is provided.
This change requires that openldap be used as our C LDAP client. It is much
easier to do SSL using openldap than mozldap (no certdb required). Otherwise
we'd have to write a slew of extra code to create a temporary cert database,
import the CA cert, ...
External CA signing is a 2-step process. You first have to run the IPA
installer which will generate a CSR. You pass this CSR to your external
CA and get back a cert. You then pass this cert and the CA cert and
re-run the installer. The CSR is always written to /root/ipa.csr.
A run would look like:
# ipa-server-install --ca --external-ca -p password -a password -r EXAMPLE.COM -u dirsrv -n example.com --hostname=ipa.example.com -U
[ sign cert request ]
# ipa-server-install --ca --external-ca -p password -a password --external_cert_file=/tmp/rob.crt --external_ca_file=/tmp/cacert.crt -U -p password -a password -r EXAMPLE.COM -u dirsrv -n example.com --hostname=ipa.example.com
This also abstracts out the RA backend plugin so the self-signed CA we
create can be used in a running server. This means that the cert plugin
can request certs (and nothing else). This should let us do online replica
creation.
To handle the self-signed CA the simple ca_serialno file now contains
additional data so we don't have overlapping serial numbers in replicas.
This isn't used yet. Currently the cert plugin will not work on self-signed
replicas.
One very important change for self-signed CAs is that the CA is no longer
held in the DS database. It is now in the Apache database.
Lots of general fixes were also made in ipaserver.install.certs including:
- better handling when multiple CA certificates are in a single file
- A temporary directory for request certs is not always created when the
class is instantiated (you have to call setup_cert_request())
Also, member attributes are now mapped to 'member user', 'member group',
etc. instead of 'member users', 'member groups'. In other words,
the second word is now taken from LDAPObject.object_name instead of
LDAPObject.object_name_plural.
ldapi: grants httpd and krb5kdc to access the DS ldapi socket
ctypes: the Python uuid module includes ctypes which makes httpd segfault
due to SELinux problems.
dogtag: remove the CRL publishing permissions. This only worked if you
had dogtag installed. In the near future will publish elsewhere so for
the time being CRL file publishing will be broken with SELinux enabled.
- remove obsolete code related to PluginProxy
- remove parent_key attribute, for the purpose of nested objects
the parent's primary key is retrieved automatically
- added support for auto-generating of UUIDs
- make use of the improved attribute printing in CLI
- make LDAPDelete delete all sub-entries, not just one-level
- minor bug fixes
If you don't want to use ldapi then you can remove the ldap_uri setting
in /etc/ipa/default.conf. The default for the framework is to use
ldap://localhost:389/
ipaObject is defined as an auxiliary objectclass so it is up to the
plugin author to ensure that the objectclass is included an a UUID generated.
ipaUniqueId is a MUST attribute so if you include the objectclass you must
ensure that the uuid is generated.
This also fixes up some unrelated unit test failures.
- attribute re-mapping, ordering and hiding
(Enables plugins to completely hide LDAP internals from users
and full localisation of command output.)
- translation of member DNs into object names
(No more DNs when listing group members etc.)
- support for "singleton" LDAP objects
(Objects like "pwpolicy"; not accessed by primary key.)
- new base classes for commands: LDAPModMember, LDAPAddMember
and LDAPRemoveMember
(Providing support for objects with 'member'-like attributes.)
- LDAPSearch implicit exit code changed to 1 when nothing is found
This involves creating a new CA instance on the replica and using pkisilent
to create a clone of the master CA.
Also generally fixes IPA to work with the latest dogtag SVN tip. A lot of
changes to ports and configuration have been done recently.
Returning the exception value doesn't work because a shell return value
is in the range of 0-255.
The default return value is 1 which means "something went wrong." The only
specific return value implemented so far is 2 which is "not found".
There are some operations, like those for the certificate system, that
don't need to write to the directory server. So instead we have an entry
that we test against to determine whether the operation is allowed or not.
This is done by attempting a write on the entry. If it would succeed then
permission is granted. If not then denied. The write we attempt is actually
invalid so the write itself will fail but the attempt will fail first if
access is not permitted, so we can distinguish between the two without
polluting the entry.
