* src/engine-gpgsm.c (gpgsm_assuan_simple_command): Do not terminate
on a status lines.
--
This bug has been with us since the support for gpgsm: If there is no
status line handler but a status line is received anyway the command
handling loop terminates and thus the command/answer order gets out of
sync. In the case of the bug report this is triggered by sending an
option which starts the agent and that starting emits a "PROGRESS"
status line.
The solution is not to stop reading after a status line but record a
possible error code and return that only after OK or ERR.
GnuPG-bug-id: 1795
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/debug.h: Change macros to not have a literal 0 as last
expression of the comma operator.
* src/debug.c (_gpgme_debug_frame_end): Return 0.
(_gpgme_debug): Return 0.
--
Instead of using
foo(), 0
for the trace macros we let foo() return 0 instead.
Signed-off-by: Werner Koch <wk@gnupg.org>
* tests/gpgsm/final.test: New.
* tests/gpgsm/initial.test: New.
* tests/gpg/start-stop-agent: Move to ../.
* tests/gpgsm/Makefile.am (TESTS_ENVIRONMENT): Export top_srcdir.
(TESTS): Add intial.test and final.test.
(AM_LDFLAGS): Add -no-install.
(clean-local): Use start-stop-agent
(initial.test): Add dependency.
* tests/gpg/Makefile.am (top_srcdir): Export top_srcdir.
(AM_LDFLAGS): Add -no-install.
(check-local): Depend on pubring-stamp instead of pubring.gpg.
(initial.test): Depend on check-local.
(./pubring-gpg): Replace by rule for ./pubring-stamp.
--
There are also a couple of other changes which should make the tests a
bit more robust and the gpg and gpgsm tests more similar.
The -no-install avoids creating wrappers for test programs, which make
debugging easier.
The dependency on check-local guarantees that its rules are run before
the first test. This is important because conf files are setup by
this rule. Earlier automake versions seem to have run check-local
always before the tests but today the order of execution is not
defined.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/verify.c (calc_sig_summary): Handle GPG_ERR_CERT_REVOKED.
--
parse_new_sig() handles a revoked key by setting sig->status to
GPG_ERR_CERT_REVOKED, but then later calc_sig_summary() expects that
code in sig->validity_reason.
Additional comments added by wk.
* src/engine-gpg.c (gpg_keylist_preprocess): Increment SRC for a
backslash.
--
This bug is not exploitable because this bug fills up .data with
backslashes and thus causes the segv.
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/Makefile.am (extra_ltoptions): New.
(libgpgme_la_LDFLAGS): Use it.
(libgpgme_pthread_la_LDFLAGS): Ditto.
(libgpgme_glib_la_LDFLAGS): Ditto.
--
Since gcc 4.8 there is a regression in Mingw64 in that plain C
programs may link to libgcc_s.a which has a dependency on
libgcc_s_sjlj.dll. This is for example triggered by using long long
arithmetic on a 32 bit Windows (e.g symbol __udivdi3).
Note that we don't use this patch for the Qt version which, as C++
programs, actually requires that DLL,
Signed-off-by: Werner Koch <wk@gnupg.org>
* src/engine-spawn.c (add_data): Fix malloc
--
Bummer. Why did I subtracted one from the size? Did I assume a
dynamically allocated structure with a string field which was not
going to be used? Very strange.
Not a real problem though because malloc will anyway round up the
allocation to at least the next word size.
Detected by Stack 0.3.
--
Somehow the doc/gpl.texi from gpgme and gnupg drifted out of sync.
This patch to gpgme's file brings it in line with gnupg's master
branch, and avoids the following errors during make:
./gpl.texi:667: @section seen before @end enumerate
./gpl.texi:724: unmatched `@end enumerate'
./gpl.texi:1: warning: node next `Copying' in menu `Concept Index'
and in sectioning `Function and Data Index' differ
* src/context.h (OPDATA_EXPORT): New.
* src/export.c (op_data_t): New.
(release_op_data): New.
(parse_error): New.
(export_status_handler): New.
(export_start, export_ext_start): Prepare op_data.
(gpgme_op_export_ext, gpgme_op_export_keys): Return an error from the
status handler.
--
To support an error return also for the async functions we need to
extend the API. Until we have done that this new features helps at
least in some cases; in particular for --send-keys.
* src/sign.c (gpgme_op_sign_result): Reformat and take care of failed
malloc.
--
Although _gpgme_debug_trace() is current always true, the code should
be run always and not just in trace mode. Also added error checking
to malloc and strdup. And while at replace some while by for loop for
easier readability.
* src/sign.c (gpgme_op_sign_result): Test that invalid and valid
signatures add up to gpgme_signers_count().
--
When invalid and valid signatures do not equal gpgme_signers_count() it
means that there was a bad passphrase during signing after the first
signer. This leaves the result.signatures from previous signers intact
which isn't correct since gpg will report:
gpg: number of one-pass packets does not match number of signature
packets
gpg: can't handle this ambiguous signature data
during verify. So when this happens append the valid signatures to the
.invalid_signers list with .reason set to GPG_ERR_GENERAL.
--
The --no-sk-comments flag is (or should be) a no-op in modern versions
of gnupg, but gpgme should still use its full form rather than the
(slightly) abbreviated --no-sk-comment
* src/engine-gpgsm.c (status_handler):
* src/engine-uiserver.c (status_handler):
--
After a realloc (realloc is also used for initial alloc) the allocated
size if the buffer is not correctly recorded. Thus an overflow can be
introduced by receiving data with different line lengths in a specific
order. This is not easy exploitable because libassuan constructs the
line. However a crash has been reported and thus it might be possible
to constructs an exploit.
CVE-id: CVE-2014-3564
Reported-by: Tomáš Trnka
* src/gpgme.h.in (GPGME_KEYLIST_MODE_WITH_SECRET): New.
* src/engine-gpg.c (gpg_keylist_build_options): Handle new mode.
* src/engine-gpgsm.c (gpgsm_keylist, gpgsm_keylist_ext): Ditto.
* src/keylist.c (parse_sec_field15): Add arg key and take care of
--with-secret output.
* src/gpgme-tool.c (gt_get_keylist_mode, cmd_keylist_mode): Add
"with_secret". Print card info and and secret flag for subkeys.
--
Note: This mode may only be used with GnuPG >= 2.1.
* src/gpgme.h.in (struct _gpgme_subkey): Add field CURVE.
* src/key.c (gpgme_key_unref): Free CURVE.
* src/keylist.c (keylist_colon_handler): Set CURVE.
* src/gpgme.c (gpgme_release): For failsafe reasons reset engine and
engine info after freeing.
--
The engine hack is useful in case the other release functions
accidently call engine release.
* src/engine-spawn.c (engspawn_start): Allocate space for list
terminator.
* src/posix-util.c (walk_path): Fix trailing slash detection.
--
Kudos to Valgrind for pointing out these two problems.
The first is a plain allocation bug in a code pattern I have written
thousands of times - this time it went wrong. The allocation is not
user controlled thus not directly exploitable.
The second is missed to do what it intended to do. Found due to the
access of malloced but not initialized memory. Not using calloc
again proved to be helpful to detect logical error.
* src/w32-io.c (create_writer): Free CTX in cased of bad FD.
* src/w32-util.c (_gpgme_mkstemp): Free TMPNAME in case of a failed
mkstemp.
--
Found by Hans-Christoph Steiner with cppcheck.
* src/spawn.c (gpgme_op_spawn_start, gpgme_op_spawn): Pass FLAGS dow
to spawn_start and add FLAGS args along the call path.
* src/engine-spawn.c (engspawn_start): Hack to automagically provide
argv[0].
