We derive keysets for virtual host-based service principals, and that
includes the `set_time` field of keys. But applications using the kadm5
API lose that information. Our httpkadmind wants to set a Cache-Control
header with an appropriate max-age so that clients know when to re-fetch
keytabs.
We could extract some of the lib/hdb/common.c functions so that
httpkadmind could re-create an HDB_entry from a kadm5 entry then compute
the desired time, but ultimately we already have an appropriate field in
the HDB_entry and kadm5_principal_ent_rec types: "password expiration".
So let's set the `pw_end` of a virtual host-based service's HDB entry to
the time when a client should next fetch the principal's keys, and we'll
use that in httpkadmind as the `pw_expiration` field of the kadm5 entry
type.
If a virtual host-based service namespace is disabled, then the virtual
services below it cease existing.
This will be useful in a later commit where we'll use virtual host-based
service namespace for providing default attributes for new concrete
host-based service principals created via httpkadmind, whether the
namespace be enabled or disabled.
One user had an entry with duplicate aliases. This happened with an
earlier version of Heimdal.
This commit does not remove the duplicates, but it does tolerate them.
We introduce a notion of soft vs. hard aliases.
Soft aliases are aliases of WELLKNOWN/REFERRALS/TARGET@$some_realm,
where $some_realm is the realm we want the KDC to issue referrals to.
Hard aliases are all other aliases, where if the client requested
canonicalization then the KDC should update the names in the responses,
or else if the client did not request canonicalization, then the KDC
should treat the alias as a distinct principal with the same keys as the
alias' canonical name.
The logic for dealing with these is entirely located in the HDB
backends.
An HDB backend can implement hard aliases by replacing a found
HDB_entry's principal with the name used to look it up.
An HDB backend can implement soft aliases by returning
HDB_ERR_WRONG_REALM to trigger the AS or TGS to return a referral.
Currently only in-tree HDB backends support this feature that use
_hdb_fetch_kvno() as their hdb_fetch_kvno() method implementation.
That's all HDB backends other than SQLite3.
Out-of-tree backends should be unaffected.
We've added a decoration field to HDB_entry: aliased -- an int
(boolean). This is only used internally in libhdb at this time.
Out-of-tree HDB backends could have a use for this decoration, but we
have not decided whether it is a public interface yet.
With HDB_ERR_WRONG_REALM the backend needs to expose the
principal, so we should not free the entry otherwise
the main kdc code will crash.
Signed-off-by: Stefan Metzmacher <metze@samba.org>
Remove hdb_entry_ex and revert to the original design of hdb_entry (except with
an additional context member in hdb_entry which is managed by the free_entry
method in HDB).
Decorate HDB_entry with context and move free_entry callback into HDB structure
itself. Requires updating hdb_free_entry() signature to include HDB parameter.
A follow-up commit will consolidate hdb_entry_ex (which has a single hdb_entry
member) into hdb_entry.
Add a new method for issuing referrals for entire namespaces of hostnames.
An alias of the form WELLKNOWN/HOSTBASED-NAMESPACE/service/namespace-fqdn@REALM
will cause all requests for host-based principals in the given namespace to be
referred to the given realm.
The TGS will issue referrals based on [domain_realm] mappings.
With this change the TGS will also issue referrals based on HDB entry
aliases.
The TGS needed no changes for this, only support in lib/hdb was missing.
All we had to do was return HDB_ERR_WRONG_REALM from hdb_fetch_kvno()
when the given principal is an alias and its canonical name's realm is
different from the alias'.
This feature is important because the KDC currently does not re-read
krb5.conf and must be restarted for changes to e.g., [domain_realm]
mappings to take effect. As well, making krb5.conf changes to all the
KDCs for a realm would need to be arranged. But with aliases in the
HDB, these problems go away.
Relatedly, we should really have an option to store the KDC's entire
configuration in the HDB...
Futures:
- Add support for aliasing of entire namespaces via HDB aliases with
WELLKNOWN namespace name forms. This will round out domain-to-realm
mapping configuration support via HDB.
Deleting an alias causes the HDB_entry_alias entry value encoding to be
written to the iprop log, which later cannot be decoded as an HDB_entry.
Meanwhile, the alias is removed from the HDB but not from the list of
aliases in the canonical principal's HDB entry.
This commit makes deletion of alias names an error.
Status:
- And it works!
- We have an extensive test based on decoding a rich EK certficate.
This test exercises all of:
- decoding
- encoding with and without decoded open types
- copying of decoded values with decoded open types
- freeing of decoded values with decoded open types
Valgrind finds no memory errors.
- Added a manual page for the compiler.
- rfc2459.asn1 now has all three primary PKIX types that we care about
defined as in RFC5912, with IOS constraints and parameterization:
- `Extension` (embeds open type in an `OCTET STRING`)
- `OtherName` (embeds open type in an `ANY`-like type)
- `SingleAttribute` (embeds open type in an `ANY`-like type)
- `AttributeSet` (embeds open type in a `SET OF ANY`-like type)
All of these use OIDs as the open type type ID field, but integer
open type type ID fields are also supported (and needed, for
Kerberos).
That will cover every typed hole pattern in all our ASN.1 modules.
With this we'll be able to automatically and recursively decode
through all subject DN attributes even when the subject DN is a
directoryName SAN, and subjectDirectoryAttributes, and all
extensions, and all SANs, and all authorization-data elements, and
PA-data, and...
We're not really using `SingleAttribute` and `AttributeSet` yet
because various changes are needed in `lib/hx509` for that.
- `asn1_compile` builds and recognizes the subset of X.681/682/683 that
we need for, and now use in, rfc2459.asn1. It builds the necessary
AST, generates the correct C types, and generates templating for
object sets and open types!
- See READMEs for details.
- Codegen backend not tested; I won't make it implement automatic open
type handling, but it should at least not crash by substituting
`heim_any` for open types not embedded in `OCTET STRING`.
- We're _really_ starting to have problems with the ITU-T ASN.1
grammar and our version of it...
Type names have to start with upper-case, value names with
lower-case, but it's not enough to disambiguate.
The fact the we've allowed value and type names to violate their
respective start-with case rules is causing us trouble now that we're
adding grammar from X.681/682/683, and we're going to have to undo
that.
In preparation for that I'm capitalizing the `heim_any` and
`heim_any_set` types, and doing some additional cleanup, which
requires changes to other parts of Heimdal (all in this same commit
for now).
Problems we have because of this:
- We cannot IMPORT values into modules because we have no idea if a
symbol being imported refers to a value or a type because the only
clue we would have is the symbol's name, so we assume IMPORTed
symbols are for types.
This means we can't import OIDs, for example, which is super
annoying.
One thing we might be able to do here is mark imported symbols as
being of an undetermined-but-not-undefined type, then coerce the
symbol's type the first time it's used in a context where its type
is inferred as type, value, object, object set, or class. (Though
since we don't generate C symbols for objects or classes, we won't
be able to import them, especially since we need to know them at
compile time and cannot defer their handling to link- or
run-time.)
- The `NULL` type name, and the `NULL` value name now cause two
reduce/reduce conflicts via the `FieldSetting` production.
- Various shift/reduce conflicts involving `NULL` values in
non-top-level contexts (in constraints, for example).
- Currently I have a bug where to disambiguate the grammar I have a
CLASS_IDENTIFIER token that is all caps, while TYPE_IDENTIFIER must
start with a capital but not be all caps, but this breaks Kerberos
since all its types are all capitalized -- oof!
To fix this I made it so class names have to be all caps and
start with an underscore (ick).
TBD:
- Check all the XXX comments and address them
- Apply this treatment to Kerberos! Automatic handling of authz-data
sounds useful :)
- Apply this treatment to PKCS#10 (CSRs) and other ASN.1 modules too.
- Replace various bits of code in `lib/hx509/` with uses of this
feature.
- Add JER.
- Enhance `hxtool` and `asn1_print`.
Getting there!
Checking the error code of decoding a blob as an hdb_entry or
hdb_entry_alias to determine which of those the blob is depends on a
detail of the Heimdal ASN.1 compiler and library that shouldn't be
depended on. Using a CHOICE adds no octets to the encoding:
HDB-EntryOrAlias ::= CHOICE {
entry hdb_entry,
alias hdb_entry_alias
}
since we're adding no additional tags and the two arms of the CHOICE
already differ in tag (hdb_entry's tag is a [UNIVERSAL Sequence] tag,
while hdb_entry_alias's is an [APPLICATION 0] tag).
The new [hdb] new_service_key_delay parameter should not apply to
principal entries when used as clients. Otherwise new passwords would
not take effect immediately, and that would be very confusing.
This is a large commit that adds several features:
- Revamps and moves virtual host-based service principal functionality
from kdc/ to lib/hdb/ so that it may be automatically visible to
lib/kadm5/, as well as kadmin(1)/kadmind(8) and ktutil(1).
The changes are backwards-incompatible.
- Completes support for documenting a service principal's supported
enctypes in its HDB entry independently of its long-term keys. This
will reduce HDB bloat by not requiring that service principals have
more long-term keys than they need just to document the service's
supported enctypes.
- Adds support for storing krb5.conf content in principals' HDB
entries. This may eventually be used for causing Heimdal KDC
services to reconfigure primary/secondary roles automatically by
discovering the configured primary in an HDB entry for the realm.
For now this will be used to help reduce the amount of configuration
needed by clients of an upcoming HTTP binding of the kadmin service.
Older databases may lack explicitly stored salts where the salt is the default
one. When fetching a client entry for an AS-REQ, add default salts to keys that
lack one.
Whilst Windows does not canonicalize enterprise principal names if the
canonicalize flag is unset, the original specification in
draft-ietf-krb-wg-kerberos-referrals-03.txt says we should. Non-Windows
deployments of Heimdals are unlikely to understand enterprise principal names
in tickets, and are also unlikely to set the canonicalize flag, so this makes
sense. (It was also the behavior prior to moving the name canonicalization
logic into the KDC.)
This reverts commit 1b7e196e66.
It turns out that, contrary to the referrals draft, Windows does not
canonicalize enterprise principal names if the canonicalize KDC option is
unset.
Enterprise principal client names in AS-REQs should always be canonicalized
irrespective of the setting the canonicalize KDC option. Perform this check in
the KDC rather than HDB.
Do not set the HDB_F_GET_KRBTGT flag unless the client actually requested a TGS
principal.
Mirroring the logic recently introduced in the TGS, this patch modifies the KDC
to perform client and server canonicalization itself rather than relying on the
backend to do so. Per RFC 6806, the behavior is slightly different for the AS
in that the setting of the canonicalize flag in the AS-REQ does impact the
returned names in the ticket. In order to support realm canonicalization or
other custom behavior, we allow the backend to force the KDC to canonicalize by
setting the force-canonicalize flag in the returned client or server entries.
e11abf41 added support in libhdb for always dereferencing principal aliases
during an AS-REQ (where dereferencing refers to enabling alias lookups, and
rewriting the returned entry with the alias name unless canonicalization was
enabled).
Due to the KDC setting HDB_F_FOR_AS_REQ for all lookups from the AS, this
allowed aliases on the TGS itself to be dereferenced during an AS-REQ; however,
on presenting the TGT, the TGS would fail to resolve. Creating an explicit TGS
principal for the aliased realm would work (at least prior to c555ed6a), but
this could be confusing to deploy.
This commit changes enables alias dereferencing when HDB_F_GET_ANY is set,
which essentially means dereference whenever the request is coming from the KDC
(as opposed to, say, kadmin).
We also backout c555ed6a, which changed the TGS to always canonicalize the
server realm, as this breaks serving multiple realms from a single KDC, where
server principals in different realms share a single canonical entry.
HDB_F_CANON is now passed to the backend as a hint only, and per RFC 6806 the
principal name is never changed in TGS replies. (However, for Samba interop,
backends can override this by setting the force-canonicalize HDB flag.)
Adds support for "hard" aliases when initially authenticating, that is,
allowing a client or server principal to be known by many names without
requiring that the client support name canonicalization.
In order to avoid changing the behavior for other backends such as Samba, this
is implemented in the HDB backend rather than the KDC.
To use, add an alias for both the client and TGS ("krbtgt") principals using
kadmin. This behavior is unchanged if name canonicalization is enabled.
We used to update the iprop log and HDB in different orders depending on
the kadm5 operation, which then led to various race conditions.
The iprop log now functions as a two-phase commit (with roll forward)
log for HDB changes. The log is auto-truncated, keeping the latest
entries that fit in a configurable maximum number of bytes (defaults to
50MB). See the log-max-size parameter description in krb5.conf(5).
The iprop log format and the protocol remain backwards-compatible with
earlier versions of Heimdal. This is NOT a flag-day; there is NO need
to update all the slaves at once with the master, though it is advisable
in general. Rolling upgrades and downgrades should work.
The sequence of updates is now (with HDB and log open and locked):
a) check that the HDB operation will succeed if attempted,
b) append to iprop log and fsync() it,
c) write to HDB (which should fsync()),
d) mark last log record committed (no fsync in this case).
Every kadm5 write operation recover transactions not yet confirmed as
committed, thus there can be at most one unconfirmed commit on a master
KDC.
Reads via kadm5_get_principal() also attempt to lock the log, and if
successful, recover unconfirmed transactions; readers must have write
access and must win any race to lock the iprop log.
The ipropd-master daemon also attempts to recover unconfirmed
transactions when idle.
The log now starts with a nop record whose payload records the offset of
the logical end of the log: the end of the last confirmed committed
transaction. This is kown as the "uber record". Its purpose is
two-fold: act as the confirmation of committed transactions, and provide
an O(1) method of finding the end of the log (i.e., without having to
traverse the entire log front to back).
Two-phase commit makes all kadm5 writes single-operation atomic
transactions (though some kadm5 operations, such as renames of
principals, and changes to principals' aliases, use multiple low-level
HDB write operations, but still all in one transaction). One can still
hold a lock on the HDB across many operations (e.g., by using the lock
command in a kadmin -l or calling kadm5_lock()) in order to push
multiple transactions in sequence, but this sequence will not be atomic
if the process or host crashes in the middle.
As before, HDB writes which do not go through the kadm5 API are excluded
from all of this, but there should be no such writes.
Lastly, the iprop-log(1) command is enhanced as follows:
- The dump, last-version, truncate, and replay sub-commands now have an
option to not lock the log. This is useful for inspecting a running
system's log file, especially on slave KDCs.
- The dump, last-version, truncate, and replay sub-commands now take an
optional iprop log file positional argument, so that they may be used
to inspect log files other than the running system's
configured/default log file.
Extensive code review and some re-writing for clarity by Viktor Dukhovni.
most of these warnings are not problems because of ample
use of abort() calls. However, the large number of warnings
makes it difficult to identify real problems. Initialize
the variables to shut up the compilers.
Change-Id: I8477c11b17c7b6a7d9074c721fdd2d7303b186a8
