Ben Kaduk [Mon, 3 Dec 2012 17:25:07 +0000 (12:25 -0500)]
Leave 'OK' button visible in Leash AboutBox
The AboutBox dialog as specified in the resource file is larger than
the one we display; the dialog init routine marks several things as
non-visible, moves the 'OK' button up to where the now-invisible items
were, and shrinks the dialog's bounding rectangle.
However, the edit boxes containing copyright and version information
seem to always present as being on top of the 'OK' button, and their
background causes the button to appear almost invisible with the current
repositioning.
To keep the 'OK' button visible, reduce the amount that it is moved
(and the amount the dialog is shrunk) so that the button does not overlap
with the edit box.
Greg Hudson [Wed, 5 Dec 2012 16:42:55 +0000 (11:42 -0500)]
Cross-reference account lockout documentation
Link to the database.rst description of policy objects when talking
about them. Briefly mention the "default" policy. Link to the
kadmin_local.rst description of policy fields when referencing them.
Describe policy fields more briefly, and expand the kadmin_local.rst
descriptions where appropriate.
Zhanna Tsitkov [Fri, 30 Nov 2012 21:54:42 +0000 (16:54 -0500)]
Document key usage assigned number conflict
Document the fact that the key usage type 26 is used by both
KBKRB5_KEYUSAGE_PA_S4U_X509_USER_REQUEST and
KRB5_KEYUSAGE_PA_S4U_X509_USER_REQUEST, while 27 - by
KRB5_KEYUSAGE_PA_S4U_X509_USER_REPLY and KRB5_KEYUSAGE_PA_SAM_RESPONSE.
Also, since KRB5_KEYUSAGE_PA_REFERRAL is not actually used in MIT Kerberos
code and is not defined in the latest referrals draft
(http://tools.ietf.org/html/draft-ietf-krb-wg-kerberos-referrals-15)
mark it as "unused".
Zhanna Tsitkov [Thu, 29 Nov 2012 19:01:00 +0000 (14:01 -0500)]
Document param expansion for keytab/ccache names
The DEFCCNAME, DEFCKTNAME and DEFKTNAME configuration options are
subjects to parameter expansion. Also note that this feature
was first introduced in release 1.11.
Greg Hudson [Thu, 29 Nov 2012 01:19:37 +0000 (20:19 -0500)]
Avoid redundant socket cleanup in sendto_kdc.c
Since r25120, kill_conn() has been responsible for closing and
invalidating conn->fd. In the unlikely event that the KDC sends a TCP
response to us before we send a TCP request, we were cleaning up the
socket again, which is useless (though also harmless). Get rid of
that code.
Greg Hudson [Thu, 29 Nov 2012 06:58:13 +0000 (01:58 -0500)]
Fix spin-loop bug in k5_sendto_kdc
In the second part of the first pass over the server list, we passed
the wrong list pointer to service_fds, causing it to see only a subset
of the server entries corresponding to sel_state. This could cause
service_fds to spin if an event is reported on an fd not in the
subset.
Ben Kaduk [Tue, 20 Nov 2012 23:06:22 +0000 (18:06 -0500)]
Remove last "document not in toctree" warnings
There were two of them, for notice.rst and mitK5license.rst.
The former is included in the latter, which is linked to from
the mitK5features document but not listed in a table of contents
otherwise. Includes are processed unconditionally, so we can
add notice.rst to the exclude_patterns array (formerly unused_docs)
and its content will still be included.
If we attempt to do the same for mitK5license.rst, it is not processed
at all, and the attempt to link to it will fail. Instead, put a dummy
table of contents in the mitK5features document (which links to the
license document), with the "hidden" attribute. This satisfies the
Sphinx need to know where all the input documents are without changing
the displayed table of contents.
Ben Kaduk [Sat, 17 Nov 2012 00:48:55 +0000 (19:48 -0500)]
Render macros as literals
Some convenience macros are referring to and dereferencing pointers,
and Sphinx will get a bit confused trying to interpret this as markup.
There should never be any markup intended to be interpreted in the
value of a macro definition, so we can silence this class of
warnings by treating them as literals. (In some sense, they actually
are literals, too.)
This will cause a warning for macros that only cause a
symbol to be defined, that is, a literal "#define MACRO" with no
initializer, due to the lack of body in the inline-literal markup.
Such macros should probably be added to the exclude list for conversion
to reStructuredText in the Doxygen-Sphinx bridge, as was already
done for KRB5_OLD_CRYPTO. Support code to programmatically omit
macros of this sort is deliberately *not* included, so that explicit
action must be taken when a new macro is to be undocumented.
Also, strip leading and trailing whitespace from the macro name,
since this causes problems with the markup.
Ben Kaduk [Wed, 21 Nov 2012 16:31:06 +0000 (11:31 -0500)]
Exclude lists for doxygen API docs
Doxygen will pick up every function, macro, and typedef defined
in krb5.h; some of these may not actually be part of the public
API for one reason or another. Provide hardcoded exclude lists
for macro/function/type names for which we do not want to emit
reStructuredText documentation, and check these lists when processing
the Doxygen XML output.
Seed these lists with the macros TRUE, FALSE, KRB5_OLD_CRYPTO,
KRB5_GENERAL__, KRB5_CALLCONV, KRB5_CALLCONV_C, KRB5_CALLCONV_WRONG,
KRB5INT_BEGIN_DECLS, KRB5INT_END_DECLS, and KRB5_ATTR_DEPRECATED,
and typedefs krb5_cc_ops and krb5_responder_context. The booleans
are compatibility cruft that we do not want to advertise, and the other
macros are for internal use for signalling and platform compatibility.
The typedefs are functioning just as forward declarations.
For consistency, remove KRB5_OLD_CRYPTO.rst from the macros index; it
had no content even when we did generate it.
Ben Kaduk [Thu, 15 Nov 2012 21:27:57 +0000 (16:27 -0500)]
Handle adjacent notes from doxygen more correctly
The Doxygen documentation seems to claim that adjacent @note entries
will be collapsed into a single "note" element, as separate paragraphs.
This seems to be reflected in the XML output: multiple <para> entries
in a single <simplesec kind="note"> with a <simplesecsep/> element
between them.
Our XML-to-RST converter gets the entire contents of the simplesec
element, parsed into a unicode string with a single newline between
paragraphs. Paragraphs seem to always start with two spaces, though
I have not tracked down the origin of this behavior. Prior to this
commit, we would just output this entire unicode string directly.
Since our template puts a tab character before the note string, this
means that the first paragraph is indented by a tab and two spaces,
and the second paragraph by just two spaces. Sphinx warns about this,
as "block did not end with blank line; unexpected whitespace", and
the paragraphs are indented differently within the note.
Fix the warning by checking for newlines in the interior of the
parsed unicode string, and introducing the appropriate whitespace
for the Sphinx parser.
Ben Kaduk [Wed, 14 Nov 2012 18:44:08 +0000 (13:44 -0500)]
Fix overflowing cell in HTML table
sphinx-build defaults to the 'tabulary' environment for tables,
but uses regular 'tabular' for those involving literal blocks,
since tabulary does not handle them properly. It seems that this
deficiency of tabulary also applies to inline literals, which
are rendered as \code{} sections by the latex builder.
Fortunately, sphinx provides a workaround to force a particular
column specification for the table.
The requirement for a table:: statement after the tabularcolumns::
statement is very poorly documented.
Ben Kaduk [Tue, 20 Nov 2012 18:13:12 +0000 (13:13 -0500)]
Update doxygen markup in krb5.hin
A few places were using the standard C /* comment */ form, but
this is rendered poorly by doxygen through to our Sphinx bridge.
Use the special /**< comment */ form to get doxygen-specific behavior.
If the standard C comment form is used, the full comment (including
start and end markers) is included in the value of the macro, and
Sphinx then tries to treat the end of the comment as the start of
inline markup with no corresponding end-string, which is a warning.
Using the doxygen form of the comment, the contents of the comment
are put in a separate paragraph block, which is inserted in the
body of the generated RST document.
The markup for krb5_rd_priv() had a line that ended with an @c
markup statement without a symbol following it. This confused
doxygen into not parsing any more of the comment. The beginning
of the next line is a macro identifier with markup to auto-linkify it.
In RST, it is not possible to have a link and a terminal font on the
same text, so removing the @c is the appropriate fix.
There are also eleven deprecated functions which are replaced by
the krb5_c_* family of functions. However, referring to this class
of functions as the "krb5_c_" class of functions results in Sphinx
attempting to interpret this statement as a link to a label elsewhere
in the document, and no such label exists. To avoid this warning, use
"krb5_c_*" to refer to the class of functions, which is arguably
more correct anyways.
Ben Kaduk [Mon, 19 Nov 2012 21:36:56 +0000 (16:36 -0500)]
Make krb5_trace_info a typedef
Our doxygen-to-sphinx documentation bridge only processes typedefs
and not structure definitions, since we almost universally use
typedefs for our data structures. krb5_trace_info is the sole
exception, so bring it into the fold.
Zhanna Tsitkov [Mon, 26 Nov 2012 21:45:08 +0000 (16:45 -0500)]
Move Release tag into footer for Sphinx HTML
The current position of the Release tag is in the body of the main page
of the Sphinx html. Move it to the footer. This way the release number
can be viewed from every page.
Tom Yu [Mon, 26 Nov 2012 21:01:23 +0000 (16:01 -0500)]
Remove .doctrees when cleaning src/doc
Sphinx produces .doctree pickles that can be over 17MB with the
current documentation. Remove them when running "make clean" in
src/doc so that they don't pollute distribution tar files.
Zhanna Tsitkov [Wed, 21 Nov 2012 19:45:44 +0000 (14:45 -0500)]
Update feature list in the documentation
In Quick facts section:
- restructure the Supported platforms section;
- do not tie KfW to 1.11 release;
- move references to GSS-API extensions to Feature list table.
In Feature list section:
- reformat the table;
- move PRNG and Pre-auth mechanisms into their own tables;
- clarify GS2 feature description;
- reference rfc6680 for GSS-API naming extensions.
Lowercase the words in the title of the document.
Greg Hudson [Sun, 25 Nov 2012 22:53:02 +0000 (17:53 -0500)]
Remove broken clean_hostname trace messages
The trace messages in krb5int_clean_hostname were outputting the
entire contents of the output buffer (mostly uninitialized garbage)
into the trace log. Since these messages were essentially redundant
with messages in the callers, and were arguably at too low of a level
to begin with, simply remove them.
Greg Hudson [Tue, 20 Nov 2012 17:33:14 +0000 (12:33 -0500)]
Reword krb5_unparse_name_ext doxygen markup
Avoid using asterix characters in the documentation for
krb5_unparse_ext_name, since they get intepreted as markdown
punctuation when translated to RST.
Greg Hudson [Sat, 17 Nov 2012 20:30:32 +0000 (15:30 -0500)]
Fix quoting issues in LDAP KDB module
Modify ldap_filter_correct() to quote special characters for DN
strings as well as filters, since it is already used to quote a DN
string in krb5_ldap_name_to_policydn() and there's no harm in
over-quoting. In krb5_ldap_put_principal(), quote the unparsed
principal name for use in DNs we choose. In
krb5_ldap_create_password_policy(), use the policy name for the CN of
the policy entry instead of the (possibly quoted) first element of the
DN.
Adapted from a patch by Jim Shi <hanmao_shi@apple.com>.
Ben Kaduk [Thu, 27 Sep 2012 16:46:26 +0000 (12:46 -0400)]
Make kinit smarter about using keytabs
Previously, we would happily accept -i or -t name and do nothing
with it, if -k was not given. If the user is passing -i or -t, they
clearly want to use a keytab, so do so (but print a warning).
While here, enforce that only one of -t and -i is given.
Greg Hudson [Sun, 11 Nov 2012 03:08:39 +0000 (22:08 -0500)]
Add automated tests for LDAP KDB module
Add new tests kdbtest.c and t_kdb.py. Together these exercise most of
the code in the LDAP back end. kdbtest is also run against the DB2
module, which is mostly redundant with other tests, but does exercise
the lockout logic a little more thoroughly than t_lockout.py can.
To test the LDAP back end, we look for slapd and ldapadd binaries in
the path. The system slapd is sometimes constrained by AppArmor or
the like, which we can typically work around by making a copy of the
binary. slapd detaches before listening on its server socket (this
got better in 2.4.27 but still isn't perfect), so we unfortunately
have to use a one-second sleep in the slapd setup.
Ben Kaduk [Wed, 14 Nov 2012 22:09:23 +0000 (17:09 -0500)]
Rebuild krb5.conf.man for default enctypes
Now that the Camellia enctypes are in the default lists for
permitted_enctypes, default_tkt_enctypes, and default_tgs_enctypes,
regenerate the man page to reflect that change.
Greg Hudson [Wed, 14 Nov 2012 21:49:33 +0000 (16:49 -0500)]
Add Camellia enctypes to default enctype lists
Add camellia256-cts-cmas and camellia128-cts-cmac to the default
permitted_enctypes, default_tkt_enctypes, and default_tgs_enctypes
lists, to simplify deployment of Camellia. The new enctypes still
aren't on supported_enctypes, so won't be in the set of long-term keys
for principals without administrator intervention.
Ben Kaduk [Wed, 17 Oct 2012 00:11:14 +0000 (20:11 -0400)]
Make glue for building PDFs
sphinx-build's latex output engine creates a subdirectory with
various latex files, and a Makefile. The generated Makefile assumes
gmake, which we do not. The logic needed in this makefile is rather
simple, so we just include it in src/doc/Makefile.in, even if we
do need a rather complicated shell expression to work in the subdirectory.
Greg Hudson [Tue, 13 Nov 2012 03:08:05 +0000 (22:08 -0500)]
Make kdb5_util dump work with LDAP again
The LDAP module doesn't support locking. There's code to ignore this
in load but not in dump. dump used to only lock for iprop dumps, but
now locks all the time after e65a16d898f3a686525e83661f4fd86c76e27bbf
(#7384), causing it to fail with LDAP.
Tom Yu [Thu, 1 Nov 2012 21:47:06 +0000 (17:47 -0400)]
Update Camellia feature description
Remove the potentially misleading "experimental" annotation on the
description of the Camellia encryption feature. Also update the I-D
version to match the IESG-approved version.
Greg Hudson [Thu, 25 Oct 2012 16:51:17 +0000 (12:51 -0400)]
Always rebuild rst_composite in src/doc
Avoid using "rst_composite" as the target name for building the
rst_composite directory, since we can't give it proper dependencies.
Instead use the target name "composite", which (like "html", "clean",
etc.) doesn't correspond to the name of a file or directory created by
the build rules.
Greg Hudson [Wed, 24 Oct 2012 16:06:01 +0000 (12:06 -0400)]
Don't save empty cc_config_out in ccache
Add an internal json function to make it easier to detect if an object
is empty, and use it to avoid creating a ccache config entry for
preauth module config data if there isn't any to save.
Greg Hudson [Tue, 23 Oct 2012 17:16:44 +0000 (13:16 -0400)]
Alter responder function signature for consistency
For the responder callback signature, put the closure argument just
after the context, and use KRB5_CALLCONV. These changes make the
signature consistent with most other libkrb5 callbacks.
Greg Hudson [Sun, 21 Oct 2012 23:46:16 +0000 (19:46 -0400)]
Add dependencies for some test programs
Some recently added test programs under lib/krb5 didn't have their
source files added to the appropriate Makefile.in variables, and
weren't getting dependencies as a result.
Greg Hudson [Fri, 19 Oct 2012 16:39:04 +0000 (12:39 -0400)]
Improve C style checking
* Avoid space-before-paren false positives on some function pointer
declarations by checking the identifier for simple type names.
* Check for space before close parenthesis.
* Check (carefully) for asymmetric spaces around binary operators.
* Handle nesting when checking for 2+ line flow control bodies.
* Check for asymmetric bracing around else statements.
* Save the vendor name of the token we used to create the challenge.
* If we saved the name of a token vendor previously, prune out any
tokeninfos which contain different vendor names.
Greg Hudson [Thu, 18 Oct 2012 17:11:28 +0000 (13:11 -0400)]
Fix sam2 client preauth after salt changes
Commit bc096a77ffdab283d77c2e0fc1fdd15b9f77eb41 altered the internal
contracts relating to salts, but neglected to adjust the sam2 preauth
code to match. Do that now.
* Keep track of the names of client identities when we load them.
* Store the client identity we just used when we create or retry a
client request.
* If we read a client identity from the configuration, treat it like the
KDC does: pick the "this is it, there is no other" logic branch.
* Read a "pa_config_data" item from an in_ccache, if provided, and add a
callback which client preauth plugins can use to retrieve a string
value from it that's keyed by a string.
* Add a callback which client preauth plugins can use to provide string
key/value pairs to be stored in the ccache.
* Moves the definition of (struct krb5_clpreauth_rock_st) from k5-int.h
to init_creds_ctx.h to try to reduce the number of files that will
need to include k5-json.h to understand k5_json_value.
Ben Kaduk [Wed, 17 Oct 2012 15:37:01 +0000 (11:37 -0400)]
Remove install-oldman
The old man pages are gone, so we can't install them anymore.
Also clean up install and install-unix targets that were installing
the old man pages by hand.
Nalin Dahyabhai [Mon, 1 Oct 2012 22:51:06 +0000 (18:51 -0400)]
Test in_ccache and pa_types functionality
* Add a krb5int_build_conf_principals() function to allow our get/set
code to directly prune out duplicate config entries.
* Verify that when we specify a pa_type, it affects whether or not we
will use a particular preauth plugin.
* Verify that we correctly save the KDC's preauth type number, that we
tried to answer, to the out_ccache.
* When producing preauth data, keep track of the type of padata in the
KDC's list of acceptable types which prompted the module to produce
padata.
* After obtaining credentials, store that value as a "pa_type"
configuration item in the out_ccache.
* Read that allowed preauth type from an in_ccache, if possible.
* If we have an allowed preauth type, only call "real" modules that
handle that value when filling in responder items and producing a
client request.
Add a krb5_get_init_creds_opt_set_in_ccache() function. An input
ccache may hold configuration data which the client libraries can
use to influence their decisions.
Greg Hudson [Tue, 16 Oct 2012 20:23:49 +0000 (16:23 -0400)]
Avoid bit shifting in krb5.h constants
Our traditional practice is to use hex integer literals for flag bits.
Bit shifting can be dangerous because shifting into the sign bit is
undefined. Convert existing bit shift expressions to hex literals.
projects / thirdparty / krb5.git / log
