]> git.ipfire.org Git - thirdparty/krb5.git/log
thirdparty/krb5.git
8 years agoRemove stray exit in t_referral.py 652/head
Greg Hudson [Thu, 4 May 2017 18:11:11 +0000 (14:11 -0400)] 
Remove stray exit in t_referral.py

Commit 1dc619624421002b1e64d3b8c7e270508381b3e6 included a stray
"exit(0)" for debugging.  Remove it.

ticket: 8579

8 years agoUse a random nonce in TGS requests 650/head
Greg Hudson [Tue, 25 Apr 2017 17:31:36 +0000 (13:31 -0400)] 
Use a random nonce in TGS requests

Commit ae0fee058ad883b2e82fa2b34f4e5f059e827a1b (ticket #5425) changed
the AS client code to use a random nonce, but left the TGS client code
using the current timestamp.  Use a random nonce for TGS requests as
well.

ticket: 8582 (new)

8 years agoDon't cache referral and alternate TGT replies 649/head
Greg Hudson [Tue, 25 Apr 2017 21:05:23 +0000 (17:05 -0400)] 
Don't cache referral and alternate TGT replies

During a TGS request, if we get a TGT response that we didn't directly
ask for (a referral TGT or an alternate TGT), don't cache it.  It
would have limited value in the cache as similar operations won't look
for that TGT.  If the overall TGS operation fails and is repeated, we
could wind up caching the same entry multiple times, which doesn't
work well with our current ccache implementations.

ticket: 8579

8 years agoFix uninitialized flags in MSLSA ccache type 646/head
Alexander Karaivanov [Mon, 24 Apr 2017 07:01:12 +0000 (09:01 +0200)] 
Fix uninitialized flags in MSLSA ccache type

The flags field in krb5_lcc_data is not initialized in
krb5_lcc_resolve(), so krb5_lcc_next_cred() can sometimes fail to
include a ticket when retrieving a ccache entry.  This results in a
"Request did not supply a ticket" error from k5_make_tgs_req() when
trying to use the credential.

[ghudson@mit.edu: condensed commit message]

ticket: 8567
target_version: 1.13-next
target_version: 1.14-next
target_version: 1.15-next
tags: pullup

8 years agoFix in_clock_skew() and use it in AS client code 645/head
Greg Hudson [Mon, 24 Apr 2017 06:02:36 +0000 (02:02 -0400)] 
Fix in_clock_skew() and use it in AS client code

Add a context parameter to the in_clock_skew() macro so that it isn't
implicitly relying on a local variable.  Use it in
get_in_tkt.c:verify_as_reply().

8 years agoUse krb5_check_clockskew() in KDC preauth mechs
Greg Hudson [Mon, 24 Apr 2017 05:45:11 +0000 (01:45 -0400)] 
Use krb5_check_clockskew() in KDC preauth mechs

8 years agoAllow clock skew in krb5 gss_context_time() 644/head
Greg Hudson [Sat, 22 Apr 2017 20:51:23 +0000 (16:51 -0400)] 
Allow clock skew in krb5 gss_context_time()

Commit b496ce4095133536e0ace36b74130e4b9ecb5e11 (ticket #8268) adds
the clock skew to krb5 acceptor context lifetimes for
gss_accept_sec_context() and gss_inquire_context(), but not for
gss_context_time().  Add the clock skew in gss_context_time() as well.

ticket: 8581 (new)
target_version: 1.14-next
target_version: 1.15-next
tags: pullup

8 years agoApply TCP timeouts to HTTPS (KKDCP) transport 643/head
Greg Hudson [Thu, 20 Apr 2017 19:33:10 +0000 (15:33 -0400)] 
Apply TCP timeouts to HTTPS (KKDCP) transport

We apply (as of ticket #7604) a ten-second minimum delay after a TCP
connection is accepted before creating new connections or sending UDP
packets.  Apply this timeout to HTTPS connections as well, by removing
the transport check in get_endtime().  As the endtime field is only
set by service_tcp_connect(), it will always have the value 0 for UDP
connection state objects, so there is no need to check the transport
type.

ticket: 8580
target_version: 1.15-next
target_version: 1.14-next
tags: pullup

8 years agoReplace UCS-2 conversions with UTF-16 641/head
Greg Hudson [Tue, 18 Apr 2017 18:01:06 +0000 (14:01 -0400)] 
Replace UCS-2 conversions with UTF-16

Where we convert between UTF-8 and UCS-2 (RC4 string-to-key and PAC
client info), use UTF-16 instead of UCS-2.  Add a test program for
the conversion functions.

ticket: 8577 (new)

8 years agoAdd various bound checks 626/head
Martin Kittel [Thu, 6 Apr 2017 19:03:23 +0000 (21:03 +0200)] 
Add various bound checks

Add bounds checks where Coverity otherwise reports a defect.  Most of
these checks are unlikely to be triggered in practice (Unicode regexps
are unused, and the caller of gss_krb5int_make_seal_token_v3 won't
have a plaintext object larger than half of the address space).  The
checks in dump.c could prevent memory access errors resulting from a
malformed dump file.

[ghudson@mit.edu: rewrote commit message]

ticket: 8578 (new)

8 years agoInitialize some magic fields
Martin Kittel [Thu, 16 Mar 2017 14:20:11 +0000 (15:20 +0100)] 
Initialize some magic fields

Initialize magic fields where Coverity otherwise detects that the
magic value would be used uninitialized.  (The "uses" are always
harmless copying.)

8 years agoReplace references to OS X 638/head
Michael Mattioli [Tue, 18 Apr 2017 02:16:37 +0000 (22:16 -0400)] 
Replace references to OS X

Replace references to Apple OS X with the updated OS name "macOS"
where applicable.

8 years agoUpdate Travis CI distribution to Xenial 640/head
Michael Mattioli [Wed, 19 Apr 2017 01:46:59 +0000 (21:46 -0400)] 
Update Travis CI distribution to Xenial

8 years agoFix realm_config.rst typo 642/head
Matt Rogers [Thu, 20 Apr 2017 15:18:50 +0000 (11:18 -0400)] 
Fix realm_config.rst typo

8 years agomake depend 639/head
Greg Hudson [Tue, 18 Apr 2017 16:40:00 +0000 (12:40 -0400)] 
make depend

8 years agoReduce dependencies of k5-input.h
Greg Hudson [Tue, 18 Apr 2017 16:33:48 +0000 (12:33 -0400)] 
Reduce dependencies of k5-input.h

Avoid using the krb5_error_code type (using int32_t instead), and
include k5-platform.h instead of k5-int.h, so that we can use
k5-input.h in libkrb5support.

8 years agoUpdate Travis build for warning-clean clang build 637/head
Greg Hudson [Sun, 16 Apr 2017 01:50:02 +0000 (21:50 -0400)] 
Update Travis build for warning-clean clang build

Create a custom build matrix which passes -Werror to the clang build
via a make variable.  (Using a configure variable does not currently
work, as some of our configure test programs generate warnings.)

Also set the language to C++ (so we use clang++ for the C++ test
programs and not g++ when compiling with clang), and turn on the
maintainer-mode checks for the Travis build.

8 years agoIgnore warnings in libev code
Greg Hudson [Sun, 16 Apr 2017 04:54:36 +0000 (00:54 -0400)] 
Ignore warnings in libev code

Add pragma statements to verto-k5ev.c to ignore warnings generated by
the libev code in gcc and clang.

8 years agoModernize UTF-8/UCS-2 conversion code 635/head
Robbie Harwood [Thu, 6 Apr 2017 16:15:39 +0000 (12:15 -0400)] 
Modernize UTF-8/UCS-2 conversion code

Remove unused entry points as we only need to convert between
little-endian UCS-2 byte buffers and UTF-8.  Rename and simplify the
remaining two function contracts.  Avoid pointer alignment and
endianness issues by operating on byte buffers and using store_16_le()
and load_16_le().  Avoid two-pass operation using k5buf.

[ghudson@mit.edu: simplified code using k5buf; simplified function
names and contracts; rewrote commit message]

8 years agoMake RC4 string-to-key more robust 636/head
Greg Hudson [Sat, 15 Apr 2017 01:41:20 +0000 (21:41 -0400)] 
Make RC4 string-to-key more robust

krb5int_utf8cs_to_ucs2les() can read slightly beyond the end of the
input buffer if the buffer ends with an invalid UTF-8 sequence.  When
computing the RC4 string-to-key result, make a zero-terminated copy of
the input string and use krb5int_utf8s_to_ucs2les() instead.

ticket: 8576 (new)
target_version: 1.15-next
target_version: 1.14-next
tags: pullup

8 years agoAdd FAST encrypted challenge auth indicator 627/head
Matt Rogers [Fri, 31 Mar 2017 02:18:24 +0000 (22:18 -0400)] 
Add FAST encrypted challenge auth indicator

During ec_verify(), look up an authentication indicator string by the
profile realm option "encrypted_challenge_indicator".  If found, add
an indicator to the reply upon succesful creation of the challenge
key.  Add a test to t_authind.py.  Document the option in
kdc_conf.rst.

ticket: 8575 (new)

8 years agoRemove {GETSOCKNAME,GETPEERNAME}_ARG2_TYPE 630/head
Robbie Harwood [Wed, 12 Apr 2017 02:30:25 +0000 (22:30 -0400)] 
Remove {GETSOCKNAME,GETPEERNAME}_ARG2_TYPE

We have had code since at least 1.6 in changepw.c and sendto_kdc.c
which assumes that we can pass a struct sockaddr * as the second
argument to getsockname() and getpeername(), so we can safely get rid
of that configure logic.  Also fix potential alignment issues in
krb5_sendauth() by using a struct sockaddr_storage instead of a
1024-byte character buffer to hold the local and peer addresses.

[ghudson@mit.edu: adjusted style of new code slightly; rewrote commit
message]

8 years agoFix unused finalizer function warning
Robbie Harwood [Thu, 6 Apr 2017 18:43:13 +0000 (14:43 -0400)] 
Fix unused finalizer function warning

When we are building a static object containing a finalizer function
(e.g. for the profile library tests), mark the finalizer as unused to
avoid warnings in gcc and clang.

[ghudson@mit.edu: commented UNUSED definition and moved it so we can
use it elsewhere later; rewrote commit message]

8 years agoFix self-assign of void * in profile_tcl.c
Robbie Harwood [Thu, 6 Apr 2017 18:43:46 +0000 (14:43 -0400)] 
Fix self-assign of void * in profile_tcl.c

Better would be to fix the generation or elminate this altogether.

8 years agoAvoid using tmpnam(3) in db2's hash.c
Robbie Harwood [Wed, 29 Mar 2017 22:34:37 +0000 (18:34 -0400)] 
Avoid using tmpnam(3) in db2's hash.c

As we do not rely on anonymous db2 databases, get rid of the code
using tmpnam() for hash databases and reporting EINVAL if a filename
is not specified.

[ghudson@mit.edu: rewrote commit message; condensed conditionals]

8 years agoFix designated initializer check to be defined
Robbie Harwood [Mon, 10 Apr 2017 18:01:40 +0000 (14:01 -0400)] 
Fix designated initializer check to be defined

8 years agoRemove tautological NULL checks in kdc_j_encode.c
Robbie Harwood [Mon, 10 Apr 2017 17:06:25 +0000 (13:06 -0400)] 
Remove tautological NULL checks in kdc_j_encode.c

8 years agoFix types for t_fortuna.c
Robbie Harwood [Thu, 6 Apr 2017 19:45:57 +0000 (15:45 -0400)] 
Fix types for t_fortuna.c

clang warns about abs() being used on an unsigned value, although the
code did work in practice due to the implicit conversion from size_t
to int.

8 years agoRemove use of caddr_t in kpropd.c
Robbie Harwood [Thu, 6 Apr 2017 18:36:05 +0000 (14:36 -0400)] 
Remove use of caddr_t in kpropd.c

8 years agoAvoid increased alignment restriction warnings
Robbie Harwood [Thu, 6 Apr 2017 16:55:36 +0000 (12:55 -0400)] 
Avoid increased alignment restriction warnings

In kdb_log.h, cast through void * after computing the address in the
INDEX macro.

In ipropd_svc.c, use a void * instead of a char * as the generic
handler return value.

In rc4.c, cast through void * when using the cipher state data pointer
as a structure pointer.

In sha256.c and sha512.c, cast through void * when using the save
buffer as a structure pointer.  (This code may not be conformant, but
it should work in practice given the offsets of the save field in the
sha256state and sha512state structures.)

[ghudson@mit.edu: rewrote commit message]

8 years agoMute shift/reduce conflicts in getdate.y
Robbie Harwood [Thu, 6 Apr 2017 15:53:53 +0000 (11:53 -0400)] 
Mute shift/reduce conflicts in getdate.y

These are harmess in a parser of this type.  Also correct header comment on
the number that occur (four, not nine).

8 years agoAvoid deprecated call in kdc5_hammer.c
Robbie Harwood [Thu, 30 Mar 2017 20:59:23 +0000 (16:59 -0400)] 
Avoid deprecated call in kdc5_hammer.c

Use krb5_get_init_creds_password() instead of
krb5_get_in_tkt_with_password().

8 years agoFix unknown pragma with clang on x-deltat.y
Robbie Harwood [Wed, 29 Mar 2017 20:28:56 +0000 (16:28 -0400)] 
Fix unknown pragma with clang on x-deltat.y

Also re-generate deltat.c in order to clear the warning.

8 years agoReplace casts with use of sa2sin and sa2sin6
Robbie Harwood [Fri, 17 Feb 2017 22:16:19 +0000 (17:16 -0500)] 
Replace casts with use of sa2sin and sa2sin6

8 years agoRemove /* in block comment
Robbie Harwood [Thu, 16 Feb 2017 23:09:10 +0000 (18:09 -0500)] 
Remove /* in block comment

8 years agoFix mismatched sign comparisons
Robbie Harwood [Thu, 16 Feb 2017 23:02:09 +0000 (18:02 -0500)] 
Fix mismatched sign comparisons

8 years agoFix unused variables
Robbie Harwood [Thu, 16 Feb 2017 22:49:16 +0000 (17:49 -0500)] 
Fix unused variables

8 years agoRemove tautological comparisons
Robbie Harwood [Thu, 16 Feb 2017 22:20:38 +0000 (17:20 -0500)] 
Remove tautological comparisons

8 years agoRemove unused function system_getnameinfo()
Robbie Harwood [Thu, 16 Feb 2017 22:08:49 +0000 (17:08 -0500)] 
Remove unused function system_getnameinfo()

8 years agoCheck for FAST in encrypted challenge client 634/head
Greg Hudson [Tue, 11 Apr 2017 21:00:01 +0000 (17:00 -0400)] 
Check for FAST in encrypted challenge client

If we reach the encrypted challenge clpreauth process method without
an armor key, error out instead of crashing.  This can happen if (a)
the KDC offers encrypted challenge even though the request doesn't use
FAST (the Heimdal KDC apparently does this), and (b) we fall back to
that preauth method before generating a preauthenticated request,
typically because of a prompter failure in encrypted timestamp.
Reported by Nico Williams.

ticket: 8573 (new)
target_version: 1.15-next
target_version: 1.14-next
tags: pullup

8 years agoPort pkinit debug code to OpenSSL 1.1.0 API 616/head
Jakub Jelen [Thu, 16 Mar 2017 15:39:46 +0000 (16:39 +0100)] 
Port pkinit debug code to OpenSSL 1.1.0 API

[ghudson@mit.edu: added intermediate variables to simplify code]

8 years agoAdd support to query the SSF of a GSS context 625/head
Simo Sorce [Thu, 30 Mar 2017 15:27:09 +0000 (11:27 -0400)] 
Add support to query the SSF of a GSS context

Cyrus SASL provides a Security Strength Factor number to assess the
relative "strength" of the negotiated mechanism, and applications
sometimes make access control decisions based on it.

Add a call that allows us to query the mechanism that established the
GSS security context to ask what is the current SSF, based on the
enctype of the session key.

ticket: 8569 (new)

8 years agoNotice errors from t_fortuna in make check 632/head
Greg Hudson [Sat, 8 Apr 2017 01:25:28 +0000 (21:25 -0400)] 
Notice errors from t_fortuna in make check

8 years agoUn-deprecate krb5_auth_con_initivector() 631/head
Greg Hudson [Fri, 7 Apr 2017 15:05:16 +0000 (11:05 -0400)] 
Un-deprecate krb5_auth_con_initivector()

The kprop protocol uses cipher state via this call, perhaps along with
other.  As there is no replacement, the call should not be deprecated
in the API.

ticket: 8572 (new)

8 years agoUse the canonical client principal name for OTP 628/head
Matt Rogers [Wed, 5 Apr 2017 20:48:55 +0000 (16:48 -0400)] 
Use the canonical client principal name for OTP

In the OTP module, when constructing the RADIUS request, use the
canonicalized client principal (using the new client_name kdcpreauth
callback) instead of the request client principal.

ticket: 8571 (new)

8 years agoAdd the client_name() kdcpreauth callback
Matt Rogers [Tue, 4 Apr 2017 20:54:56 +0000 (16:54 -0400)] 
Add the client_name() kdcpreauth callback

Add a kdcpreauth callback to returns the canonicalized client principal.

ticket: 8570 (new)

8 years agoExport GSS_KRB5_GET_CRED_IMPERSONATOR on Windows 629/head
Greg Hudson [Wed, 5 Apr 2017 15:42:17 +0000 (11:42 -0400)] 
Export GSS_KRB5_GET_CRED_IMPERSONATOR on Windows

Add the new public data symbol GSS_KRB5_GET_CRED_IMPERSONATOR to the
gssapi DLL export list.

ticket: 8548

8 years agoConvert some pkiDebug messages to TRACE macros 624/head
Matt Rogers [Wed, 29 Mar 2017 14:35:13 +0000 (10:35 -0400)] 
Convert some pkiDebug messages to TRACE macros

ticket: 8568 (new)

8 years agoCorrect formatting error in gss_duplicate_name() 621/head
Martin Kittel [Thu, 16 Mar 2017 15:03:54 +0000 (16:03 +0100)] 
Correct formatting error in gss_duplicate_name()

8 years agoFix krb5int_open_plugin_dirs() error handling
Martin Kittel [Wed, 15 Mar 2017 16:21:28 +0000 (17:21 +0100)] 
Fix krb5int_open_plugin_dirs() error handling

In krb5int_open_plugin_dirs(), if constructing filepath fails,
filepath is set to null but accessed a few lines later.  Add an error
check before calling krb5int_open_plugin().

ticket: 8565 (new)
target_version: 1.15-next
target_version: 1.14-next
tags: pullup

8 years agoAdd null check to placate static analysis tools
Martin Kittel [Wed, 29 Mar 2017 07:22:18 +0000 (09:22 +0200)] 
Add null check to placate static analysis tools

In trace_format() when processing lenstr, if p is NULL and len is 0,
skip the call to buf_add_printable_len() as Coverity considers it an
unconditional dereference of p.

[ghudson@mit.edu: added explanation to commit message]

8 years agoRemove unnecessary null checks
Martin Kittel [Wed, 29 Mar 2017 06:55:57 +0000 (08:55 +0200)] 
Remove unnecessary null checks

Remove some null checks for values that should never be null.  These
checks were performed inconsistently and triggered static code
analysis tools.

8 years agoSimplify null salt handling in string-to-key 622/head
Greg Hudson [Mon, 27 Mar 2017 19:40:08 +0000 (15:40 -0400)] 
Simplify null salt handling in string-to-key

The per-enctype string_to_key implementations are inconsistent about
whether a null salt is treated as empty or results in a null
dereference.  Since the original DES string-to-key allowed a null
salt, substitute an empty salt in krb5_c_string_to_key_with_params().
Eliminate conditionals on accessing salt in the per-enctype
implementations as they are no longer needed.  Based on a patch by
Martin Kittel.

8 years agoCompile less libev code 623/head
Greg Hudson [Tue, 28 Mar 2017 18:12:48 +0000 (14:12 -0400)] 
Compile less libev code

In verto-k5ev.c, turn off optional watchers in ev.c, and enable the
specific watcher types we use.

8 years agoIgnore dotfiles in profile includedir 619/head
Greg Hudson [Fri, 24 Mar 2017 15:07:21 +0000 (11:07 -0400)] 
Ignore dotfiles in profile includedir

Editors and filesystems may create artifacts related to .conf files
which don't change the file suffix; these artifacts generally begin
with "." so that they don't appear in normal directory listings
(e.g. ".#filename" for emacs interlock files).  Make sure to ignore
any such artifacts when processing a profile includedir directive.

ticket: 8563 (new)
target_version: 1.15-next
tags: pullup

8 years agoRemove the NSS PKINIT crypto implementation 620/head
Matt Rogers [Fri, 24 Mar 2017 20:57:42 +0000 (16:57 -0400)] 
Remove the NSS PKINIT crypto implementation

Remove the unused and buggy NSS PKINIT crypto backend.

8 years agoRemove some unnecessary PKINIT code 618/head
Greg Hudson [Thu, 23 Mar 2017 18:26:50 +0000 (14:26 -0400)] 
Remove some unnecessary PKINIT code

In cms_signeddata_create(), alg_buf and digest_buf are allocated but
never used.  (Instead, a combined buffer is allocated and the alg and
digest objects are marshalled into it.)  Remove them.

8 years agoCorrect error handling bug in prior commit 617/head
Greg Hudson [Thu, 23 Mar 2017 17:42:55 +0000 (13:42 -0400)] 
Correct error handling bug in prior commit

In crypto_encode_der_cert(), if the second i2d_X509() invocation
fails, make sure to free the allocated pointer and not the
possibly-modified alias.

ticket: 8561

8 years agoAdd the certauth dbmatch module 610/head
Matt Rogers [Wed, 15 Mar 2017 23:57:15 +0000 (19:57 -0400)] 
Add the certauth dbmatch module

Add and enable the "dbmatch" builtin module.  Add the
pkinit_client_cert_match() and crypto_req_cert_matching_data() helper
functions.  Add dbmatch tests to t_pkinit.py.  Add documentation to
krb5_conf.rst, pkinit.rst, and kadmin_local.rst.

[ghudson@mit.edu: simplified code, edited docs]

ticket: 8562 (new)

8 years agoSimplify PKINIT cert iteration and selection
Matt Rogers [Wed, 22 Mar 2017 01:24:14 +0000 (21:24 -0400)] 
Simplify PKINIT cert iteration and selection

Remove the pkinit_cert_handle structures and iteration functions used
during certificate matching.  Instead, make pkinit_matching.c obtain a
list of matching data objects from the crypto code, and then select a
cert based on the index into that list.

Also fix a typo in the name of crypto_retrieve_X509_key_usage().

[ghudson@mit.edu: simplified code]

8 years agoAdd certauth pluggable interface
Matt Rogers [Tue, 28 Feb 2017 20:55:24 +0000 (15:55 -0500)] 
Add certauth pluggable interface

Add the header include/krb5/certauth_plugin.h, defining a pluggable
interface to control authorization of PKINIT client certificates.

Add the "pkinit_san" and "pkinit_eku" builtin certauth modules and
related PKINIT crypto X.509 helper functions.  Add authorize_cert() as
the entry function for certauth plugin module checks called in
pkinit_server_verify_padata().  Modify kdcpreauth_moddata to hold the
list of certauth module handles, and load the modules when the PKINIT
kdcpreauth server plugin is initialized.  Change
crypto_retrieve_X509_sans() to return ENOENT when no SAN is found.

Add test modules in plugins/certauth/test.  Create t_certauth.py with
basic certauth tests.  Add plugin interface documentation in
doc/plugindev/certauth.rst and doc/admin/krb5_conf.rst.

[ghudson@mit.edu: simplified code, edited docs]

ticket: 8561 (new)

8 years agoForce autoconf rebuild in maintainer rules 615/head
Greg Hudson [Tue, 14 Mar 2017 23:39:38 +0000 (19:39 -0400)] 
Force autoconf rebuild in maintainer rules

autoconf normally avoids recreating files that it does not consider
obsolete.  Since it knows nothing about patchlevel.h (which we read at
autoconf time using m4's esyscmd()), changes to patchlevel.h won't be
reflected in configure unless another input to configure has changed,
and the maintainer rule will re-run autoconf over and over again.  Fix
this issue by passing the force flag to autoconf when we invoke it
from the maintainer rule.

ticket: 8560 (new)
target_version: 1.15-next
target_verison: 1.14-next
tags: pullup

8 years agoRefactor kvno for simpler memory management
Greg Hudson [Sun, 12 Mar 2017 17:09:38 +0000 (13:09 -0400)] 
Refactor kvno for simpler memory management

Move the kvno loop body into a helper function to reduce indentation
and make freeing temporary values easier.

8 years agoFix minor memory leaks in kvno
Greg Hudson [Sun, 12 Mar 2017 16:42:37 +0000 (12:42 -0400)] 
Fix minor memory leaks in kvno

In do_k5_kvno(), free allocated values on success as well as failure.
In t_kdb.py, run kvno with multiple arguments to manifest this leak in
asan and valgrind.  Reported by Cel Skeggs.

ticket: 8558

8 years agoFix leaks in gss_inquire_cred_by_oid()
Greg Hudson [Sun, 12 Mar 2017 16:30:59 +0000 (12:30 -0400)] 
Fix leaks in gss_inquire_cred_by_oid()

In the mechglue gss_inquire_cred_by_oid(), remove an unnecessary
allocation of ret_set which is overwritten by the first mechanism's
result.

ticket: 8559 (new)
target_version: 1.15-next
target_version: 1.14-next
tags: pullup

8 years agoFix memory leaks in test programs
Greg Hudson [Sun, 12 Mar 2017 16:29:50 +0000 (12:29 -0400)] 
Fix memory leaks in test programs

Eliminate memory leaks detected by asan in test programs (and
introduced since commit 4947c270032691d556140b290e1b10846b692968), to
make it easier to find more serious leaks.

8 years agoAllow null outputs to gss_get_name_attribute() 613/head
Isaac Boukris [Sat, 4 Mar 2017 19:23:32 +0000 (21:23 +0200)] 
Allow null outputs to gss_get_name_attribute()

In krb5_gss_get_name_attribute(), always ask for kvalue and
kdisplay_value when calling krb5_authdata_get_attribute(), as it
currently expect non-null arguments.  This change allows applications
to pass GSS_C_NO_BUFFER for the value and display_value output
parameters.  (Passing NULL for the authenticated and complete output
parameters already works.)

[ghudson@mit.edu: initialized kvalue and kdisplay_value for safety]

ticket: 8557 (new)
target_version: 1.15-next
target_version: 1.14-next
tags: pullup

8 years agoFix udp_preference_limit with SRV records 612/head
Greg Hudson [Tue, 28 Feb 2017 03:35:07 +0000 (22:35 -0500)] 
Fix udp_preference_limit with SRV records

In sendto_kdc:resolve_server() when resolving a server entry with a
specified transport, defer the resulting addresses if the strategy
dictates that the specified transport is not preferred.  Reported by
Jochen Hein.

ticket: 8554
target_version: 1.15-next
target_version: 1.14-next
tags: pullup

8 years agoFix PKINIT two-component matching rule parsing 611/head
Greg Hudson [Fri, 24 Feb 2017 18:41:53 +0000 (13:41 -0500)] 
Fix PKINIT two-component matching rule parsing

In pkinit_matching.c:parse_rule_set(), apply the default relation when
parsing the second component of a rule, not the third.  Otherwise we
apply no default relation to two-component matching rules, effectively
reducing such rules to their second components.  Reported by Sumit
Bose.

ticket: 8553 (new)
target_version: 1.15-next
target_version: 1.14-next
tags: pullup

8 years agoAdd test cases for preauth fallback behavior 604/head
Greg Hudson [Fri, 20 Jan 2017 17:44:12 +0000 (12:44 -0500)] 
Add test cases for preauth fallback behavior

Add options to icred for performing optimistic preauth and setting
preauth options, and for choosing between the normal and stepwise
interfaces.  Add options to the test preauth module to allow induced
failures at several points in processing, factoring out some padata
manipulation functions into a new file to avoid repeating too much
code.  Add test cases to t_preauth.py using the new facilities to
exercise and verify several preauth fallback scenarios.  Amend the
tryagain test case in t_pkinit.py to look for more trace log messages.

ticket: 8537

8 years agoContinue after KDC_ERR_PREAUTH_FAILED
Greg Hudson [Mon, 16 Jan 2017 20:09:32 +0000 (15:09 -0500)] 
Continue after KDC_ERR_PREAUTH_FAILED

If the KDC sends KDC_ERR_PREAUTH_FAILED, try another mechanism, or
send an unauthenticated request if optimistic preauth failed.

ticket: 8537

8 years agoContinue preauth after client-side failures
Greg Hudson [Sat, 14 Jan 2017 18:55:22 +0000 (13:55 -0500)] 
Continue preauth after client-side failures

If the module for the selected preauth mechanism fails when processing
a KDC_ERR_MORE_PREAUTH_DATA_REQUIRED error, or fails a tryagain
operation, try again with a different preauth mech using the cached
method data.

If optimistic preauth fails on the client side, send an
unauthenticated request, allowing the mechanisms we tried
optimistically to be tried again.

ticket: 8537

8 years agoPreserve method data in get_in_tkt.c
Greg Hudson [Fri, 13 Jan 2017 20:35:48 +0000 (15:35 -0500)] 
Preserve method data in get_in_tkt.c

To continue after preauth failures, we need a persistent field in
krb5_init_creds_context containing the METHOD-DATA from a
KDC_PREAUTH_REQUIRED or KDC_PREAUTH_FAILED error.  If we overwrite
this field with the padata in a KDC_MORE_PREAUTH_DATA_REQUIRED error,
or conflate it with an optimistic padata list, we won't be able to
correctly continue after a preauth failure.

In krb5_init_creds_context, split the preauth_to_use field into
optimistic_padata, method_padata, and more_padata.  Separately handle
KDC_ERR_MORE_PREAUTH_DATA_REQUIRED in init_creds_step_request() and
init_creds_step_reply(), and separately handle optimistic preauth in
init_creds_step_request().  Do not call k5_preauth() if none of the
padata lists are set.

Also stop clearing ctx->err_reply when processing a
KDC_ERR_PREAUTH_REQUIRED response.  Instead look for that error code
in init_creds_step_request().  Eliminate the preauth_required field of
krb5_init_creds_context as it can be inferred from whether we are
performing optimistic preauth.

ticket: 8537

8 years agoAdd OID to inquire GSS cred impersonator name 601/head
Simo Sorce [Thu, 26 Jan 2017 10:45:17 +0000 (05:45 -0500)] 
Add OID to inquire GSS cred impersonator name

In the krb5 GSS mechanism, add support in gss_inquire_cred_by_oid()
for inquiring the impersonator name of a credential object, using OID
1.2.840.113554.1.2.2.5.14.

[ghudson@mit.edu: edited code slightly; added documentation; expanded
commit message]

ticket: 8548 (new)

8 years agoAdd GSSAPI S4U documentation 609/head
Greg Hudson [Fri, 17 Feb 2017 18:38:19 +0000 (13:38 -0500)] 
Add GSSAPI S4U documentation

Describe how a GSS application can perform S4U2Self and S4U2Proxy
requests using the MIT krb5 GSS library.  Also add a reference to RFC
7546 at the top, and fix a reference to gssapi_krb5.h.

ticket: 8552 (new)
target_version: 1.15-next
tags: pullup

8 years agoUse fallback realm for GSSAPI ccache selection 606/head
Matt Rogers [Fri, 10 Feb 2017 17:53:42 +0000 (12:53 -0500)] 
Use fallback realm for GSSAPI ccache selection

In krb5_cc_select(), if the server principal has an empty realm, use
krb5_get_fallback_host_realm() and set the server realm to the first
fallback found.  This helps with the selection of a non-default ccache
when there is no [domain_realms] configuration for the server domain.
Modify t_ccselect.py tests to account for fallback behavior.

ticket: 8549 (new)

8 years agoUse LDIF config and add mdb support in t_kdb.py 585/head
Greg Hudson [Fri, 23 Dec 2016 02:49:09 +0000 (21:49 -0500)] 
Use LDIF config and add mdb support in t_kdb.py

When setting up slapd, use slapadd with cn=config LDIF directives
instead of the deprecated config file.  By adding one cn=config object
at a time, we can detect specific specific configuration failures,
including unsupported database types.  Try the mdb and bdb database
types, to work with older and newer OpenLDAP versions.

8 years agoAdd OpenLDAP LDIF file for Kerberos schema
Greg Hudson [Fri, 23 Dec 2016 02:16:39 +0000 (21:16 -0500)] 
Add OpenLDAP LDIF file for Kerberos schema

Add an LDIF version of the Kerberos schema suitable for use with
OpenLDAP's cn=config.

ticket: 8529 (new)

8 years agoAvoid draft 9 fallback after PKINIT failure 605/head
Greg Hudson [Tue, 7 Feb 2017 18:12:24 +0000 (13:12 -0500)] 
Avoid draft 9 fallback after PKINIT failure

If a KDC offers both RFC 4556 and draft 9 PKINIT, and we experience a
client-side failure trying RFC 4556 PKINIT (e.g. due to the user
entering the wrong PKCS #11 PIN), do not try to use draft 9 PKINIT.

ticket: 8544

8 years agoTrack preauth failures instead of tries 602/head
Greg Hudson [Fri, 13 Jan 2017 17:16:04 +0000 (12:16 -0500)] 
Track preauth failures instead of tries

In preauth2.c, instead of noting whenever we try a real preauth mech,
note when a mechanism fails on our side.  Tracking only failures
eliminates the need to reset the list for multi-step preauth exchanges
or for processing padata in the AS-REP, but we will need the function
later for continuing after optimistic preauth failures.

ticket: 8537

8 years agoRemove sent_nontrivial_preauth field
Greg Hudson [Mon, 16 Jan 2017 18:42:18 +0000 (13:42 -0500)] 
Remove sent_nontrivial_preauth field

In krb5_init_creds_context, the selected_preauth_type field subsumes
the need for sent_nontrivial_preauth.  Use it instead.

8 years agoSimplify k5_preauth_tryagain()
Greg Hudson [Sat, 14 Jan 2017 01:45:48 +0000 (20:45 -0500)] 
Simplify k5_preauth_tryagain()

When retrying pre-authentication for an error, try only the module for
the selected preauth type, not all preauth types in the original
method data.  Pass the error and its padata to k5_preauth_tryagain()
explicitly, so that those fields of krb5_init_creds_context are only
referenced in get_in_tkt.c.  Handle a degenerate case in
init_creds_step_reply() to simplify the code in
init_creds_step_request().

ticket: 8537

8 years agoAdjust processing of pa_type ccache config
Greg Hudson [Fri, 13 Jan 2017 15:14:36 +0000 (10:14 -0500)] 
Adjust processing of pa_type ccache config

Read the allowed preauth type from the input ccache in
restart_init_creds_loop(); there is no need to reread it each time we
produce a request.  Move read_allowed_preauth_type() earlier in the
file to allow it to be called from restart_init_creds_loop() without a
prototype.

Clear the selected preauth type in restart_init_creds_loop(), not in
init_creds_step_request().  We want to make sure that it doesn't
survive a restart due to a realm referral or expiry, but we don't want
to forget about it when retrying after an error.

8 years agoDocument multi-component PKINIT client certs 603/head
Greg Hudson [Mon, 30 Jan 2017 17:30:51 +0000 (12:30 -0500)] 
Document multi-component PKINIT client certs

In pkinit.rst, note that the extensions.client file only works for
single-component client principals, and describe how to modify it for
multi-component principals.

ticket: 7940
target_version: 1.15-next
tags: pullup

8 years agoEcho KDC cookies in preauth tryagain 599/head
Greg Hudson [Sat, 21 Jan 2017 18:20:38 +0000 (13:20 -0500)] 
Echo KDC cookies in preauth tryagain

When trying again after a mechanism-specific error, we should send the
KDC cookie for conformance with RFC 6113.

ticket: 8539

8 years agoDocument default realm and login authorization 600/head
Greg Hudson [Wed, 25 Jan 2017 18:07:42 +0000 (13:07 -0500)] 
Document default realm and login authorization

Add documentation to host_config.rst describing what the default realm
does.  Also add documentation discussing login authorization
configuration, and give an example showing how to give login access to
principals from a realm other than the default realm.

ticket: 8540 (new)
target_version: 1.15-next
tags: pullup

8 years agoDocument and check init_creds context requirement 593/head
Greg Hudson [Mon, 9 Jan 2017 16:44:29 +0000 (11:44 -0500)] 
Document and check init_creds context requirement

To ensure that the same clpreauth plugin modules and moddata pointers
are used for each step of an initial creds operation, the caller must
use the same library context for krb5_init_creds_init(),
krb5_init_creds_step(), and krb5_init_creds_free().  Document and
enforce this requirement.

ticket: 7877

8 years agoAdd tests for per-request preauth data scoping
Greg Hudson [Wed, 4 Jan 2017 23:31:15 +0000 (18:31 -0500)] 
Add tests for per-request preauth data scoping

Add a test harness which interleaves calls for multiple initial creds
contexts using the same library context.  Add a test case to
t_preauth.py using the new harness and the test preauth module to
verify that modreq pointers are correctly tracked.

ticket: 7877

8 years agoProperly scope per-request preauth data
Greg Hudson [Tue, 20 Dec 2016 21:06:24 +0000 (16:06 -0500)] 
Properly scope per-request preauth data

It should be possible to successfully use multiple initial credentials
contexts with the same library context.  Create a new internal type
krb5_preauth_req_context containing per-request preauth state,
including the clpreauth modreq handles and the list of preauth types
already tried.  Remove this state from clpreauth_handle and
krb5_preauth_context.

ticket: 7877

8 years agoMake krb5_preauth_context a pointer type
Greg Hudson [Tue, 20 Dec 2016 20:25:29 +0000 (15:25 -0500)] 
Make krb5_preauth_context a pointer type

For consistency with krb5_context and krb5_init_creds_context, make
krb5_preauth_context a pointer type.  In preauth2.c, use the typedef
name rather than the structure tag except when defining the structure.

8 years agoAdd free_principal_e_data KDB method 596/head
Andreas Schneider [Wed, 18 Jan 2017 10:52:48 +0000 (11:52 +0100)] 
Add free_principal_e_data KDB method

Add an optional method to kdb_vftabl to free e_data pointer in a
principal entry, in case it was populated by a module using a more
complex structure than a single memory region.

[ghudson@mit.edu: handled minor version bump; simplified code; rewrote
commit message]

ticket: 8538
target_version: 1.15-next
tags: pullup

8 years agoExplicitly copy KDB vtable fields
Greg Hudson [Wed, 18 Jan 2017 16:40:49 +0000 (11:40 -0500)] 
Explicitly copy KDB vtable fields

In preparation for bumping the kdb_vftabl minor version, use explicit
field assignments when copying the module vtable to the internal copy,
so that we can conditionalize assignments for minor versions greater
than 0.

ticket: 8538

8 years agoUse expected_msg in test scripts 598/head
Greg Hudson [Wed, 18 Jan 2017 16:22:58 +0000 (11:22 -0500)] 
Use expected_msg in test scripts

8 years agoUse expected_trace in test scripts
Greg Hudson [Tue, 17 Jan 2017 16:25:22 +0000 (11:25 -0500)] 
Use expected_trace in test scripts

8 years agoAdd k5test expected_msg, expected_trace
Greg Hudson [Tue, 17 Jan 2017 16:24:41 +0000 (11:24 -0500)] 
Add k5test expected_msg, expected_trace

In k5test.py, add the optional keyword argument "expected_msg" to
methods that run commands, to make it easier to look for substrings in
the command output.  Add the optional keyword "expected_trace" to run
the command with KRB5_TRACE enabled and look for an ordered series of
substrings in the trace output.

8 years agoAdd test case for PKINIT DH renegotiation 595/head
Greg Hudson [Wed, 11 Jan 2017 15:49:30 +0000 (10:49 -0500)] 
Add test case for PKINIT DH renegotiation

In t_pkinit.py, add a PKINIT test case where the KDC sends
KDC_ERR_DH_KEY_PARAMETERS_NOT_ACCEPTED and the client retries with the
KDC's TD_DH_PARAMETERS value, using the clpreauth tryagain method.
Use the trace log to verify that the renegotiation actually takes
place.

8 years agoClean up krb5_db2_free_policy() 594/head
Greg Hudson [Wed, 11 Jan 2017 14:46:46 +0000 (09:46 -0500)] 
Clean up krb5_db2_free_policy()

Commit 03d34fcfa329fbc2f686a0b34e2731e37f483a34 (ticket 8414) removed
the prototype and all uses of krb5_db2_free_policy(), but neglected to
remove the function definition, resulting in a warning.  Remove the
definition now.

8 years agoFix detection of supported warnings in clang 592/head
Ken Raeburn [Sat, 7 Jan 2017 03:39:39 +0000 (22:39 -0500)] 
Fix detection of supported warnings in clang

Without -Werror=unknown-warning-option, clang will warn for
unrecognized -W options like "-Werror=discarded-qualifiers" but won't
return a nonzero exit status, leading configure to think the options
are supported and thus include them during the build, leading to a
rather noisy log.

This option isn't needed during the build, though it won't hurt
anything either.  It is desirable during the testing of other -W
options for cleaner results, but the existing code tests each option
independently, requiring different handling for this option than for
other -W options.

8 years agoRemove addevent flag in net-server.c 588/head
Greg Hudson [Mon, 2 Jan 2017 19:20:29 +0000 (14:20 -0500)] 
Remove addevent flag in net-server.c

The addevent flag to make_event() was always set to true except when
setting up the routing socket.  Since we no longer set up the routing
socket (ticket 8348), we can remove the flag.

8 years agoRemove struct socksetup from net-server.c
Greg Hudson [Mon, 2 Jan 2017 19:10:12 +0000 (14:10 -0500)] 
Remove struct socksetup from net-server.c

struct socksetup was required when we iterated over local addresses
using foreach_localaddr.  Since we no longer do that (ticket 8348),
the functions which use it can simply accept the parameters they
require and return error codes.