* scd/app.c (new_card_lock): New.
(select_application): Scanning is serialized by NEW_CARD_LOCK.
For app_new_register, we hold the W-lock.
(initialize_module): Initialize NEW_CARD_LOCK.
--
GnuPG-bug-id: 7402
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* tools/gpg-mail-tube.c (mail_tube_encrypt): Fix content type for an
attached message.
--
We can't use message/rfc822 if we encrypt this message as a simple PGP
file.
* scd/app.c (send_card_and_app_list): Only handle the case with
WANTCARD=NULL.
(app_send_card_list): Follow the change.
(app_send_active_apps): Factor out the case with WANTCARD!=NULL.
--
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* tools/gpgtar-extract.c (extract_directory): Factor parent directory
creation out to ..
(try_mkdir_p): new.
(extract_regular): Create directory on ENOENT.
* g10/pubkey-enc.c (get_it): Use log_info instead of log_error if the
public key was not found for preference checking.
--
If tarball was created with
tar cf tarball file1.txt foo/file2.txt
the tarball has no entry for foo/ and thus the extraction fails. This
patch fixes this.
GnuPG-bug-id: 7380
The second patch avoid a wrong exist status status line due to the use
of log_error. But the actual cause needs stuill needs tobe
investigated.
* tools/gpg-mail-tube.c (mail_tube_encrypt): Rename var ct_text for
clarity. Replace debug diagnostic by log_info. Assume text/plain for
missing content-type.
--
Without this fix we would create message/rfc822 attachment instead of
a text/plain attachment with the encrypted body.
* g10/packet.h (PKT_public_key): Increased size of req_usage to 16.
* g10/getkey.c (key_byname): Set allow_adsk in the context if ir was
requested via req_usage.
(finish_lookup): Allow RENC usage matching.
* g10/keyedit.c (append_adsk_to_key): Adjust the assert.
* g10/keygen.c (prepare_adsk): Also allow to find an RENC subkey.
--
If an ADSK is to be added it may happen that an ADSK subkey is found
first and this should then be used even that it does not have the E
usage. However, it used to have that E usage when it was added.
While testing this I found another pecularity: If you do
gpg -k ADSK_SUBKEY_FPR
without the '!' suffix and no corresponding encryption subkey is dound,
you will get an unusabe key error. I hesitate to fix that due to
possible side-effects.
GnuPG-bug-id: 6882
* agent/trustlist.c (istrusted_internal): When LISTMODE is enabled,
TRUSTLISTFPR status output should be done.
--
GnuPG-bug-id: 7363
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
Fixes-commit: 4275d5fa7a51731544d243ba16628a9958ffe3ce
* g10/trustdb.c (validate_keys): Use log_info instead of log_error for
not found or expired UTKs.
--
Actually the not-found case used log_error for decades. The
semantically simialr expired case did thus the same. The actual
problem is for example in the import case where gpg exits with a
failure despite that a key validation was requested.
GnuPG-bug-id: 7351
* build-aux/speedo.mk (speedo_w32_cflags): Remove -mms-bitfields
because it is for a long time the gcc default. Enable control flow
protection.
--
Note that due to mingw static linking problems with libssp the stack
protector is not yet enabled.
(cherry picked from commit afe87ffc08e14317f4ef5bbe2940d07203a43808)
* sm/keylist.c (list_internal_keys): Detect write errors to the output
stream.
* sm/server.c (any_failure_printed): New var.
(gpgsm_status2): Handle new var. Move statusfp init to ...
(gpgsm_init_statusfp): new function.
(gpgsm_exit_failure_status): New.
* sm/gpgsm.c (main): Explicit statusfp init.
(gpgsm_exit): Print failure status on error.
--
Test by using
gpgsm -k >/dev/full
gpgsm -k --wit-colons >/dev/full
and also by redirecting to a file on a small partition.
GnuPG-bug-id: 6185
* common/compliance.c (gnupg_status_compliance_flag): Fix test.
--
In general the cache is used to query this flag but in this function
it is used directly and we need to adjust the test.
Thanks to Ingo for reporting this.
* common/homedir.c (_gnupg_socketdir_internal): Check return code
of gnupg_mkdir and handle the case of GPG_ERR_EEXIST.
--
GnuPG-bug-id: 7332
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* dirmngr/crlcache.c (crl_cache_list): Print a summary of URLs.
* sm/call-dirmngr.c (gpgsm_dirmngr_run_command): Print a notice to
stdout if the dirmngr has been disabled.
--
GnuPG-bug-id: 7337
* acinclude.m4 (GNUPG_CHECK_ENDIAN): Remove.
* configure.ac (BIG_ENDIAN_HOST): Use AC_C_BIGENDIAN
to detect endian and set BIG_ENDIAN_HOST.
--
Reported-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* common/compliance.c (assumed_de_vs_compliance): New.
(get_compliance_cache): Check envvar and fake compliance.
(gnupg_status_compliance_flag): Return 2023 for de-vs if in faked
mode.
* g10/gpg.c (gpgconf_list): For compliance_de_vs return 23 or 2023.
--
The user visible changes are that
GNUPG_ASSUME_COMPLIANCE=de-vs gpgconf --list-options gpg \
| awk -F: '$1=="compliance_de_vs" {print $8}'
returns 2023 if "compliance de-vs" is found in gpg.conf. If
eventually the software is arpproved the returned value will be 23 and
not 1 as it was before. Consumers should check whether they see value
of true (Kleopatra does this right now) and also check whether the
value is > 2000 and in this case print a beta/non-approved warning.
The envvar is currently used to assume that the underlying libgcrypt
is compliant and approved. This is not yet the case but eventually
libgcrypt will announce this itself and from then on the envvar is not
anymore required for testing.
* sm/gpgsm.h (struct keyinfo_cache_item_s): New.
(struct server_control_s): Add keyinfo_cache and keyinfo_cache_valid.
* sm/call-agent.c (keyinfo_cache_disabled): New flag.
(release_a_keyinfo_cache): New.
(gpgsm_flush_keyinfo_cache): New.
(struct keyinfo_status_parm_s): New.
(keyinfo_status_cb): Implement a fill mode.
(gpgsm_agent_keyinfo): Implement a cache.
* sm/server.c (reset_notify): Flush the cache.
* sm/gpgsm.c (gpgsm_deinit_default_ctrl): Ditto.
--
In almost all cases we have just a few private keys in the agent and
thus it is better to fetch them early. This does not work in a
restricted connection but we take care and disable the cache in this
case.
This cache gives a a minor speed up.
GnuPG-bug-id: 7308
* sm/call-agent.c (struct istrusted_cache_s): New.
(istrusted_cache, istrusted_cache_valid): New.
(istrusted_cache_disabled): New.
(flush_istrusted_cache): New.
(struct istrusted_status_parm_s): New.
(istrusted_status_cb): Fill the cache.
(gpgsm_agent_istrusted): Implement a cache.
--
Not a really measurable performance improvements on Linux but maybe
somewhat on Windows (not yet tested). However, it does not clutter
the log files with IPC calls returning NOT_TRUSTED.
GnuPG-bug-id: 7308
* agent/trustlist.c (istrusted_internal): Add arg listmode and print
new status line in this mode. Adjust callers.
(agent_listtrusted): Add new args ctrl and status_mode. Get all
trusted keys and then call is_trusted_internal for all of them.
* agent/command.c (cmd_listtrusted): Add new option --status.
--
This allows in a non-restricted connection to list all trusted keys in
one go.
* sm/minip12.c (parse_shrouded_key_bag): Increase size of salt buffer.
--
Reported on the mailing list. The change does not seem to have a big
regression risk, thus applied. See below for the mail
# ------------------------ >8 ------------------------
https://lists.gnupg.org/pipermail/gnupg-users/2024-September/067312.html
* sm/certchain.c (ask_marktrusted): Avoid fingerprint printing in
quiet mode
--
And also don't print it anymore after the agent told us that the
feature has been disabled.
* sm/gpgsm.h (COMPAT_NO_CHAIN_CACHE): New.
(struct cert_cache_item_s, cert_cache_item_t): New.
(struct server_control_s): Add parent_cert_cache.
* sm/gpgsm.c (compatibility_flags): Add "no-chain-cache".
(parent_cache_stats): New.
(gpgsm_exit): Print the stats with --debug=memstat.
(gpgsm_deinit_default_ctrl): Release the cache.
* sm/certchain.c (gpgsm_walk_cert_chain): Cache the certificates.
(do_validate_chain): Ditto.
--
This gives another boost of 30% (from 6.5 to 4.0 seconds in the test
environment with ~1000 certs). do_validate_chain actually brings us
the speedup becuase the gpgsm_walk_cert_chain is not used during a key
listing. For the latter we actually cache all certificates because
that was easier.
GnuPG-bug-id: 7308
* kbx/keybox-search.c (keybox_get_cert): Store the blob clags in the
cert object.
* sm/certchain.c (do_validate_chain): Skip clearing of the ephemeral
flag if we know that it is not set.
--
GnuPG-bug-id: 7308
* g10/trustdb.c (copy_key_item): New.
(validate_keys): Use a stripped down UTK list w/o expired keys.
--
This patch makes sure that an expired trusted key is not used for
trust computation. The test case is to delete a trusted key from the
keyring, import a copy of that key which has already expired, check
that a signed key is not anymore fully trusted and finally import a
prolonged version of the trusted key and check that the signed key is
now again fully trusted.
GnuPG-bug-id: 7200
* g10/import.c (import_one_real): Rename non_self to non_self_or_utk.
If not set after chk_self_sigs check whether the imported key is an
ultimately trusted key.
--
The revalidation mark was only set if the imported key had a new key
signature. This is in general correct but not if the imported key is
a trusted key.
GnuPG-bug-id: 7200
