Greg Hudson [Sat, 11 Feb 2012 23:25:12 +0000 (23:25 +0000)]
Minimize draft9 PKINIT code by removing dead code
The PKINIT client code doesn't use decode_krb5_pa_pk_as_rep_draft9,
which is fortunate because it doesn't work (see issue #7072).
Instead, it passes both kinds of PKINIT replies through
decode_krb5_pa_pk_as_rep, then decodes the un-enveloped CMS data in
alternative 1 (encKeyPack) as either an RFC or draft9 ReplyKeyPack.
So, remove the unused broken pa_pk_as_rep_draft9 decoder.
For pa_pk_as_req_draft9, we only use two of the fields on encode and
only one of those on decode. So, get rid of the unused fields and
the krb5_trusted_ca structure, and reduce the encoder and decoder
sequences to the minimum necessary fields.
Greg Hudson [Sat, 11 Feb 2012 23:25:08 +0000 (23:25 +0000)]
Eliminate some unused ASN.1 encoding primitives
asn1_make.c contained a variety of utility functions, most of which we
no longer needed. Fold make_tag into asn1_encode.c and get rid of
asn1_make.c and asn1_make.h.
Greg Hudson [Sat, 11 Feb 2012 23:25:04 +0000 (23:25 +0000)]
Fold atype_primitive into atype_fn
atype_primitive is used for only two types (KerberosTime and
KerberosFlags), which doesn't justify the machinery. Turn those types
into atype_fn types and get rid of atype_primitive.
Greg Hudson [Tue, 7 Feb 2012 15:45:59 +0000 (15:45 +0000)]
Fix uninitialized variable warning in trval.c
gcc -O3 is agressive enough to detect that do_cons() doesn't always
set the value of rlen2, so trval2() gets a maybe-uninitialized
warning. Squash it by initializing rlen2.
Greg Hudson [Mon, 6 Feb 2012 23:19:08 +0000 (23:19 +0000)]
Set display_value in krb5_gss_get_name_attribute
A backwards conditional in r25358 caused krb5_gss_get_name_attribute
not to set display_value on success. Fix the sense of the
conditional.
We still don't quite correctly handle the cases where data_to_gss()
fails, but those should be rare and the problem in those cases isn't
severe, so it can be fixed separately.
Also, value and display_value should probably be initialized to null
buffers on failure, as is common with GSS interfaces.
ticket: 7087
status: open
target_version: 1.10.1
tags: pullup
Zhanna Tsitkov [Fri, 3 Feb 2012 19:22:44 +0000 (19:22 +0000)]
Added a new trace logging message TRACE_PROFILE_ERR to improve the diagnostics of the potential misconfiguration.
Added profile_get_(string/integer/boolean)_nodef functions to the profile library to get the typed values from the configuration
files (without setting these values to the defaults).
Used TRACE_PROFILE_ERR for the configuration diagnostics in krb5_init_context_profile API.
Greg Hudson [Fri, 3 Feb 2012 03:19:33 +0000 (03:19 +0000)]
Better short/long descs in gss_display_mech_attr
RFC 5587 defines an "arc name" and a "purpose" for each mechanism
attribute. gss_display_mech_attr was returning the purpose (which
isn't very short) in short_desc and a null buffer in long_desc (which
isn't necessarily conformant to the spec). Instead, output the arc
name as the short_desc and the purpose as the long_desc. This is more
useful and also more consistent with Heimdal, which outputs the arc
name as short_desc and either the purpose string or an empty string
for long_desc.
Greg Hudson [Tue, 31 Jan 2012 21:35:34 +0000 (21:35 +0000)]
Don't check mech in krb5_gss_inquire_cred_by_mech
krb5_gss_inquire_cred_by_mech checks its mech argument against two of
the four mechs a krb5 cred might have (the krb5 mech and the old krb5
mech, but not the wrong Microsoft mech or the IAKERB mech), so would
spuriously fail for the other two mechs. There is no reason to check
the mechanism if we assume a reasonable mechglue is interpreting
application gss_inquire_cred_by_mech calls, so just remove the check.
Greg Hudson [Fri, 27 Jan 2012 21:27:47 +0000 (21:27 +0000)]
Suppress maybe-uninitialized warning in x-deltat.y
Recent versions of gcc can generate a maybe-uninitialized warning from
bison output instead of a regular uninitialized warning. Suppress
both. Fix from nalin@redhat.com.
Greg Hudson [Thu, 26 Jan 2012 22:05:25 +0000 (22:05 +0000)]
Clean up dispatch lookaside code slightly
Always log when we get a lookaside cache hit, eliminating a confusing
conditional. This is a slight behavior change because we never used
to log a lookaside cache hit when we can't deliver the response via
UDP, but that was never really deliberate or important--we log all
sorts of stuff about responses which might turn out to be too big.
Also eliminate a signed/unsigned comparison warning in
finish_dispatch.
Greg Hudson [Thu, 26 Jan 2012 21:56:16 +0000 (21:56 +0000)]
Various lookaside cache fixes
Don't touch the lookaside cache if we're responding with a lookaside
cache entry. Also, leave the null entry behind if we're deliberately
dropping a request (a rare case) so that we don't have to process it
again. Fixes several lookaside problems in 1.10:
* When dropping a request because it was already being processed, we
were erroneously removing the null entry, causing us to process the
request again upon a second retransmit.
* When responding to a finished request with a lookaside entry, we
were removing and re-adding the entry to the cache, resetting its
time and performing unnecessary work.
* We were not caching responses we couldn't deliver because they were
too big for UDP, causing us to re-process the request when it came
in again via TCP instead of simply delivering the cached response.
Greg Hudson [Thu, 26 Jan 2012 17:01:05 +0000 (17:01 +0000)]
Don't use stack variable address in as_req state
The KDC's process_as_req was storing the address of a stack variable
(client_princ), which fails if the request is not immediately
serviced. Move that variable to the state structure so its address
remains valid for the lifetime of the request.
Greg Hudson [Tue, 24 Jan 2012 16:12:41 +0000 (16:12 +0000)]
Avoid sizeof(void) in ASN.1 encoder
We were computing sizeof(void) when referencing atype_int_immediate
types in other types (primarily atype_tagged_thing). gcc gives this a
pass but the Solaris compiler does not. Use "int" as the dummy type
instead. (The type is unimportant since int-immediates aren't used by
offset or sequence-of types.)
Replace the "field" half of the ASN.1 encoder machinery with several
new struct atype_info types (offset, counted, and int_immediate) and a
new counted_type machinery used exclusively for object/integer pairs.
This change eliminates the two-way entanglement between atypes and
fields, and makes the encoder more independent of RFC 4120 ASN.1
conventions.
Defining sequences with the preprocessor macros is now slightly more
cumbersome; we need to define a type for each field and then string
them together with an array of struct atype_info *. But it's not so
bad, and hopefully we can auto-generate some of this soon.
*_optional functions now return a bitmask of sequence field offsets
not present in the C type. This will need some more attention before
we add decoding support to the engine, since it means sequence fields
no longer know whether they are optional or not.
Tom Yu [Fri, 13 Jan 2012 20:47:26 +0000 (20:47 +0000)]
Add shadow manpages for k5login.5 and k5identity.5
Add shadow manpages dot.k5login and dot.k5identity for k5login.5 and
k5identity.5. Stop generating .k5login.5 and .k5identity.5 from
sphinx (these will be taken care of by make install in src/man). Add
generated k5identity.5.
Add SYNOPSIS sections to k5login.5 and k5identity.5 to make it more
clear that the filenames start with a dot.
Greg Hudson [Fri, 13 Jan 2012 15:04:35 +0000 (15:04 +0000)]
Add PKINIT decoder tests
Add tests to krb5_decode_test.c for PKINIT types. Where the decoders
do not match the encoders, include comments and hand-generated
encodings which work with the decoder.
Add a comment to asn1_k_encode.c documenting inconsistencies between
the draft 9 PA-PK-AS-REQ spec and our encoder results.
Greg Hudson [Wed, 11 Jan 2012 21:20:08 +0000 (21:20 +0000)]
Fix spurious clock skew caused by gak_fct delay
In get_in_tkt.c, a time offset is computed between the KDC's auth_time
and the current system time after the reply is decrypted. Time may
have elapsed between these events because of a gak_fct invocation
which blocks on user input. The resulting spurious time offset can
cause subsequent TGS-REQs to fail and can also cause the end time of
the next AS request to be in the past (issue #889) in cases where the
old ccache is opened to find the default principal.
Use the system time, without offset, for the request time of an AS
request, for more predictable kinit behavior. Use this request time,
rather than the current time, when computing the clock skew after the
reply is decrypted.
Greg Hudson [Tue, 10 Jan 2012 17:27:38 +0000 (17:27 +0000)]
Use DEFTAGGEDTYPE for FAST choice types
pa_fx_fast_request and pa_fx_fast_reply are defined in RFC 6113 as
extensible choice types with only one choice, which means they encode
as another type with a [0] tag wrapper. Use DEFTAGGEDTYPE to add
this wrapper instead of DEFFIELDTYPE/FIELDOF_ENCODEAS, and add a
comment describing why.
After this change, all uses of DEFFIELDYPE are for dataptr/lenptr
types or just simple offset transformations on the base pointer.
Greg Hudson [Tue, 10 Jan 2012 07:33:41 +0000 (07:33 +0000)]
Simplify asn1_decode_kdc_dh_key_info
Due to some lingering confusion, the last commit changing
asn1_decode_kdc_dh_key_info was correct but overly complicated (and
contained an incorrect comment). Change it to just use get_lenfield
for subjectPublicKey.
Greg Hudson [Mon, 9 Jan 2012 21:03:58 +0000 (21:03 +0000)]
Make dh_key_info encoder and decoder symmetric
The dh_key_info encoder expects subjectPublicKey to contain the
contents of a bit string, but the decoder outputs the DER encoding of
the bit string including tag. The PKINIT client code expects this, so
everything works, but the encoder and decoder should be symmetric.
Change the decoder to process the bit string (adding a bit string
decoding primitive) and modify the PKINIT client code to expect only
the bit string contents.
Tom Yu [Mon, 9 Jan 2012 20:13:10 +0000 (20:13 +0000)]
install sphinx-generated manpages
Install sphinx-generated manpages. Original nroff manpages remain for
reference until proofreading is complete. Modify
doc/rst_source/conf.py to better deal with shadow manpages -- sphinx
will now build k5login.5 instead of .k5login.5, and kadmin.1 instead
of both kadmin.1 and kadmin.local.8.
Proofreaders should ensure that the original nroff manpages (and
associated Makefile rules) are deleted once their reST format
equivalents have been proofread.
Greg Hudson [Sun, 8 Jan 2012 21:54:29 +0000 (21:54 +0000)]
Remove unneeded kdcRealm field in PKINIT structure
krb5_pk_authenticator_draft9 had a kdcRealm field which was set by the
client code but never encoded or decoded. Remove it. Eliminating this
field exposed a bug in auth_pack_draft9_optional; fix that.
Greg Hudson [Sun, 8 Jan 2012 21:27:59 +0000 (21:27 +0000)]
Fix PKINIT serverDHNonce encoding
Use an explicit tag for serverDHNonce, as specified in RFC 4556,
rather than the implicit tag we historically used. This bug had no
practical effect (and creates no interoperability issues) because we
never generate a serverDHNonce.
Greg Hudson [Sun, 8 Jan 2012 02:10:47 +0000 (02:10 +0000)]
Clean up the asn1 encoder design
Now that the PKINIT types have been converted and atype_fn has only
one use, we can more easily modify the encoder so that any object can
be encoded without its tag, which makes for a cleaner design. The
basic building block is now krb5int_asn1_encode_type, which encodes
the contents of a function and returns its tag information to the
caller.
atype_fn now has its own structure, and the encoder function it
references follows the semantics of krb5int_asn1_encode_type.
atype_opaque is now atype_der and goes with a new corresponding field
type (field_der); stored DER encodings are parsed to separate the tag
from the content.
Greg Hudson [Sat, 7 Jan 2012 20:57:36 +0000 (20:57 +0000)]
Convert securid module edata method
r25348 made modified the edata method of the kdcpreauth interface to
be async-capable, but neglected to convert the securid_sam2 module's
edata function. Do that now.
Greg Hudson [Sat, 7 Jan 2012 16:50:06 +0000 (16:50 +0000)]
Remove some unused structures and encoders
krb5_alt_method was added in r6604 but never supported. krb5_pwd_data
became unused when the Sandia kadmin system was replaced.
krb5_pa_server_referral_data and krb5_pa_svr_referral_data were added
in r21690 with internally-visible encoders which nothing uses.
Leave behind structure declarations in krb5.hin for API compatibility.
Greg Hudson [Sat, 7 Jan 2012 15:50:14 +0000 (15:50 +0000)]
Remove SAM encoders and structures
r24403 removed the old SAM support, but left behind the structures,
free functions, and ASN.1 encoders/decoders. Remove those now.
(SAM-2 support is still present.)
Greg Hudson [Fri, 6 Jan 2012 21:17:14 +0000 (21:17 +0000)]
Convert all remaining macro-coded ASN.1 encoders
Use data-driven encoders for all of the remaining types which still
used macros (primarily PKINIT types), and get rid of the macros. Do
not change any encoding behavior, but add some comments where behavior
differs from the spec.
DEFFNTYPE is now unused except for the kdc_req_body hack.
Greg Hudson [Fri, 6 Jan 2012 21:13:59 +0000 (21:13 +0000)]
Add support for CHOICE in ASN.1 encoder
Add a new field type where the length offset indicates a distinguisher
and the data offset indicates a union address. The field's type is an
atype_choice containing a seq_info indexed by the distinguisher.
Greg Hudson [Fri, 6 Jan 2012 21:08:54 +0000 (21:08 +0000)]
Support implicit context tags in ASN.1 fields
Add a field_info bit (the 32nd bit of the bitfields) indicating whether
the context tag is implicit, and support it in encode_a_field. Adjust
all field-generating macros and invocations to include the new bit
(always 0 for the moment).
For atype_tagged_thing, narrow the construction field to six bits and
add an implicit bit. We could remove the construction field if it
weren't for DEFOCTETWRAPPEDTYPE abusing atype_tagged_thing a little
bit, since (normal) explicit tags are always constructed and implicit
tag construction is computed from the base type.
Given how rarely implicit tagging is used, it might be nice to have
separate _IMPLICIT macros rather than an extra argument to every
field. But we already have separate _OPT macros for optional fields
and FIELDOF_STRING vs. FIELDOF_STRINGL, so we start to get a
combinatoric explosion in the number of macros.
Greg Hudson [Fri, 6 Jan 2012 21:06:10 +0000 (21:06 +0000)]
Support ASN.1 encoding without the outer tag
In order to support implicit tagging, make it possible to ASN.1-encode
a value without its outer tag, instead remembering the construction
bit of the omitted tag.
A cleaner design would be to have separate functions for encoding a
value's contents and its tag. However, we can't do that for atype_fn
or atype_opaque, and the possible indirections between types and
fields mean we want to stay at the "encode everything" level for as
long as possible to allow implicit tagging of the largest possible
subset of types. If we can get rid of atype_fn, we may be able to
switch to the cleaner design with some adjustments to atype_opaque.
Greg Hudson [Fri, 6 Jan 2012 20:52:12 +0000 (20:52 +0000)]
Use content-only ASN.1 primitives
As part of implicit tag support, rework ASN.1 encoding primitives so
that they encode only content, not tags. Combine primitives which
become identical with this change. The new atype_primitive type
invokes a primitive encoder and adds a tag. atype_fn_len is split
into atype_string and atype_opaque, both of which are hardcoded to
use asn1_encode_bytestring.
For the encoders still using macros, create asn1_addprimitive,
asn1_addinteger, and asn1_addstring macros which call the primitive
encoder function and add a tag.
Greg Hudson [Fri, 6 Jan 2012 20:52:09 +0000 (20:52 +0000)]
Make ASN.1 struct atype_info more extensible
Instead of including all of the possible type fields in struct
atype_info, use a pointer to a type-specific structure. This might
save a little space, but more importantly, if we get to the point of
exposing this stuff across plugin APIs, it allows ASN.1 type
information to be extensible via defining new atype_type values.
Greg Hudson [Fri, 6 Jan 2012 20:52:02 +0000 (20:52 +0000)]
Add test cases for PKINIT ASN.1 encoders
Do not add decode tests, because those would trip some bugs in the
decoders, and we can't safely fix some of those bugs without interop
testing. Encode tests are sufficient to detect when we
unintentionally change the output of the encoders.
Fix trval2() not to use the context shortcut on primitive context
tags.
Greg Hudson [Fri, 6 Jan 2012 20:46:17 +0000 (20:46 +0000)]
Fix asn1_encode_subject_pk_info without params
r20923 inadvertently broke asn1_encode_subject_pk_info in the case
where algorithm.parameters.length == 0. Fortunately this case never
happens, but fix it anyway.
Greg Hudson [Wed, 21 Dec 2011 22:52:52 +0000 (22:52 +0000)]
Stop using krb5_typed_data structure type
Use the krb5_pa_data structure type when encoding or decoding
TYPED-DATA. Leave the krb5_typed_data structure definition in krb5.h
with a comment saying not to use it. Remove krb5_free_typed_data
(which was never declared in krb5.h). Remove some vestigial accessor
stuff related to PKINIT encoding and decoding TYPED-DATA, which was
unneeded since r25483. Bump the accessor structure version to 19
accordingly.
Greg Hudson [Wed, 21 Dec 2011 22:52:43 +0000 (22:52 +0000)]
Stop using krb5_octet_data
For consistency with the rest of the code base, make PKINIT use
krb5_data as a pointer/length container. Leave krb5_octet_data and
krb5_free_octet_data behind for API compatibility.
Greg Hudson [Fri, 16 Dec 2011 23:19:01 +0000 (23:19 +0000)]
Do mech fallback for first SPNEGO context token
When producing the first SPNEGO security context token, if the first
mechanism's init_sec_context fails, fall back to a later mechanism.
This fixes a regression in 1.10 for SPNEGO initiators using non-krb5
credentials. The identity selection work causes errors to be deferred
from krb5's acquire_cred in some cases, which means SPNEGO doesn't see
an error until it tries the krb5 init_sec_context.
Greg Hudson [Fri, 16 Dec 2011 23:18:54 +0000 (23:18 +0000)]
Verify acceptor's mech in SPNEGO initiator
In spnego_gss_ctx_id_rec, store the set of negotiable mechanisms as
well as the currently selected internal_mech, which becomes an alias
into mech_set. In init_ctx_reselect, locate the acceptor's counter-
proposal in sc->mech_set and consider the token defective if it is not
found.
If krb5_server_decrypt_ticket_keytab doesn't find a key of the
appropriate enctype in an iterable keytab, it returns 0 (without
decrypting the ticket) due to a misplaced initialization of retval.
This bug causes kinit -k to claim "keytab entry valid" when it
shouldn't. Reported by mark@mproehl.net.
Tom Yu [Mon, 12 Dec 2011 20:46:20 +0000 (20:46 +0000)]
Split cci_thread_init into per-process and per-thread portions
Call the per-thread code on thread attach and per-process once per
process. Previously, while the function was named 'thread', it was
only actually called once per process. Currently, the per-thread
code does nothing on non-windows platforms and is not even actually
invoked.
Fixes a windows bug when multiple non-main threads try to use ccapi
at the same time.
Signed-off-by: Kevin Wasserman <kevin.wasserman@painless-security.com>
ticket: 7050
mfc bug causes assertions when dialog is generated from
within PreTranslateMessages() (MSG input param points to a global
variable which is corrupted in the dialog message loop). So we need
to instead PostMessage() to cause the popup later.
Also fixed logic to cause warning dialog to actually be modal as intended
when the leash window is not minimized.
Signed-off-by: Kevin Wasserman <kevin.wasserman@painless-security.com>
ticket: 7050
projects / thirdparty / krb5.git / log
