* common/iobuf.c (iobuf_seek): Fix for temp streams.
* g10/pubkey-enc.c (get_session_key, get_it): Add some log_clock calls.
* g10/keydb.c (dump_search_desc): New.
(enum_keyblock_states, struct keyblock_cache): New.
(keyblock_cache_clear): New.
(keydb_get_keyblock, keydb_search): Implement a keyblock cache.
(keydb_update_keyblock, keydb_insert_keyblock, keydb_delete_keyblock)
(keydb_rebuild_caches, keydb_search_reset): Clear the cache.
--
Gpg uses the key database at several places without a central
coordination. This leads to several scans of the keybox for the same
key. To improve that we now use a simple cache to store a retrieved
keyblock in certain cases. In theory this caching could also be done
for old keyrings, but it is a bit more work and questionable whether
it is needed; the keybox scheme is anyway much faster than keyrings.
Using a keybox with 20000 384 bit ECDSA/ECHD keypairs and a 252 byte
sample text we get these values for encrypt and decrypt operations on
an Core i5 4*3.33Ghz system. The option --trust-model=always is used.
Times are given in milliseconds wall time.
| | enc | dec | dec,q |
|-----------+-----+-----+-------|
| key 1 | 48 | 96 | 70 |
| key 10000 | 60 | 98 | 80 |
| key 20000 | 69 | 106 | 88 |
| 10 keys | 540 | 290 | 70 |
The 10 keys test uses a mix of keys, the first one is used for
decryption but all keys are looked up so that information about are
printed. The last column gives decryption results w/o information
printing (--quiet).
The keybox is always scanned sequentially without using any index. By
adding an index to the keybox it will be possible to further reduce
the time required for keys stored to the end of the file.
* g10/keydb.c (parse_keyblock_image): Add args PK_NO and UID_NO and
set the note flags accordingly.
(keydb_get_keyblock): Transfer PK_NO and UID_NO to parse_keyblock_image.
* kbx/keybox-search.c (blob_cmp_fpr, blob_cmp_fpr_part)
(blob_cmp_name, blob_cmp_mail): Return the key/user number.
(keybox_search): Set the key and user number into the found struct.
(keybox_get_keyblock): Add args R_PK_NO and R_UID_NO and set them from
the found struct.
--
getkey.c needs to know whether the correct subkey was found. Thus we
need to set the node flags the same way we did it with the keyring
storage.
* common/logging.c (log_clock): New.
* g10/gpg.c (set_debug): Print clock debug flag.
* g10/options.h (DBG_CLOCK_VALUE, DBG_CLOCK): New.
--
To actually use log_clock you need to enable the code in
logginc.c:log_check() and link against librt. --debug 4096 may then
be used to enable it at runtime.
* g10/getkey.c (cache_public_key): Make room in the cache if needed.
--
To create the selfsigs, the key generation code makes use of the key
cache. However, after 4096 the cache is filled up and then disabled.
Thus generating more than 4096 keys in one run was not possible. We
now clear the first half the inserted keys every time the cache gets
full.
* g10/keydb.c (parse_keyblock_image): Add arg SIGSTATUS.
(keydb_get_keyblock): Handle it.
(build_keyblock_image): Add arg SIGSTATUS.
(keydb_insert_keyblock): Handle it.
* kbx/keybox-blob.c (pgp_create_sig_part): Add arg SIGSTATUS.
(_keybox_create_openpgp_blob): Ditto.
* kbx/kbxutil.c (import_openpgp): Adjust for above change.
* kbx/keybox.h (KEYBOX_FLAG_SIG_INFO): New.
* kbx/keybox-search.c (_keybox_get_flag_location): Handle new flag.
(keybox_get_keyblock): Add arg R_SIGSTATUS.
* kbx/keybox-update.c (keybox_insert_keyblock): Add arg SIGSTATUS.
--
With this change a key listing using the keybox format is now double
as fast as using a keyring. The memory use dropped as well. Measured
with about 1500 keys.
* g10/getkey.c (get_pubkey_fast): Improve the assertion.
* kbx/keybox.h: Include iobuf.h.
* kbx/keybox-blob.c (keyboxblob_uid): Add field OFF.
(KEYBOX_WITH_OPENPGP): Remove use of this macro.
(pgp_create_key_part_single): New.
(pgp_temp_store_kid): Change to use the keybox-openpgp parser.
(pgp_create_key_part): Ditto.
(pgp_create_uid_part): Ditto.
(pgp_create_sig_part): Ditto.
(pgp_create_blob_keyblock): Ditto.
(_keybox_create_openpgp_blob): Ditto.
* kbx/keybox-search.c (keybox_get_keyblock): New.
* kbx/keybox-update.c (keybox_insert_keyblock): New.
* g10/keydb.c (parse_keyblock_image):
(keydb_get_keyblock): Support keybox.
(build_keyblock_image): New.
(keydb_insert_keyblock): Support keybox.
* kbx/kbxutil.c (import_openpgp, main): Add option --dry-run and print
a kbx file to stdout.
* kbx/keybox-file.c (_keybox_read_blob2): Allow keyblocks up to 10^6
bytes.
--
Import and key listing does now work with the keybox format. It is
still quite slow and signature caching is completely missing.
Increasing the maximum allowed length for a keyblock was required due
to a 700k keyblock which inhibited kbxutil to list the file.
kbxutil's option name --import-openpgp is not quite appropriate
because it only creates KBX blobs from OpenPGP data.
* kbx/keybox-defs.h (_keybox_openpgp_key_info): Add field ALGO.
* kbx/keybox-openpgp.c (parse_key): Store algo.
* kbx/kbxutil.c (dump_openpgp_key): Print algo number.
* kbx/keybox-dump.c (_keybox_dump_blob): Print identical Sig-Expire
value lines with a range of indices.
* kbx/keybox-defs.h (_keybox_write_header_blob): Move prototype to ..
* kbx/keybox.h: here.
* kbx/keybox-init.c (keybox_lock): Add dummy function
* g10/keydb.c: Include keybox.h.
(KeydbResourceType): Add KEYDB_RESOURCE_TYPE_KEYBOX.
(struct resource_item): Add field kb.
(maybe_create_keyring_or_box): Add error descriptions to diagnostics.
Add arg IS_BOX. Write a header for a new keybox file.
(keydb_add_resource): No more need for the force flag. Rename the
local variable "force" to "create". Add URL scheme "gnupg-kbx". Add
magic test to detect a keybox file. Add basic support for keybox.
(keydb_new, keydb_get_resource_name, keydb_delete_keyblock)
(keydb_locate_writable, keydb_search_reset, keydb_search2): Add
support for keybox.
(lock_all, unlock_all): Ditto.
* g10/Makefile.am (needed_libs): Add libkeybox.a.
(gpg2_LDADD, gpgv2_LDADD): Add KSBA_LIBS as a workaround.
* g10/keydb.h (KEYDB_RESOURCE_FLAG_PRIMARY)
KEYDB_RESOURCE_FLAG_DEFAULT, KEYDB_RESOURCE_FLAG_READONLY): New.
* g10/gpg.c, g10/gpgv.c (main): Use new constants.
--
I did most of these changes back in 2011 and only cleaned them up
now. More to follow soon.
* g10/import.c (valid_keyblock_packet): New.
(read_block): Store only valid packets.
--
A corrupted key, which for example included a mangled public key
encrypted packet, used to corrupt the keyring. This change skips all
packets which are not allowed in a keyblock.
GnuPG-bug-id: 1455
(cherry-picked from commit f795a0d59e197455f8723c300eebf59e09853efa)
* common/argparse.c (iio_item_def_s, IIO_ITEM_DEF): New.
(initialize): Init field IIO_LIST.
(ignore_invalid_option_p): New.
(ignore_invalid_option_add): New.
(ignore_invalid_option_clear): New.
(optfile_parse): Implement meta option.
--
This option is currently of no use. However, as soon as it has been
deployed in all stable versions of GnuPG, it will allow the use of the
same configuration file with an old and a new version of GnuPG. For
example: If a new version implements the option "foobar", and a user
uses it in gpg.conf, an old version of gpg would bail out with the
error "invalid option". To avoid that the following line can be put
above that option in gpg.conf
ignore-invalid-option foobar
This meta option may be given several times or several option names
may be given as arguments (space delimited). Note that this option is
not available on the command line.
* common/utf8conv.c [HAVE_ANDROID_SYSTEM]: Do not include iconv.h.
(iconv_open, iconv_close, load_libiconv) [HAVE_ANDROID_SYSTEM]: New
dummy functions.
(set_native_charset) [HAVE_ANDROID_SYSTEM]: Force use of "utf-8".
(jnlib_iconv_open) [HAVE_ANDROID_SYSTEM]: Act the same as under W32.
(jnlib_iconv) [HAVE_ANDROID_SYSTEM]: Ditto.
(jnlib_iconv_close) [HAVE_ANDROID_SYSTEM]: Ditto.
--
Co-authored-by: Hans of Guardian <hans@guardianproject.info>
* agent/command-ssh.c (SPEC_FLAG_IS_ECDSA): New.
(struct ssh_key_type_spec): Add fields CURVE_NAME and HASH_ALGO.
(ssh_key_types): Add types ecdsa-sha2-nistp{256,384,521}.
(ssh_signature_encoder_t): Add arg spec and adjust all callers.
(ssh_signature_encoder_ecdsa): New.
(sexp_key_construct, sexp_key_extract, ssh_receive_key)
(ssh_convert_key_to_blob): Support ecdsa.
(ssh_identifier_from_curve_name): New.
(ssh_send_key_public): Retrieve and pass the curve_name.
(key_secret_to_public): Ditto.
(data_sign): Add arg SPEC and change callers to pass it.
(ssh_handler_sign_request): Get the hash algo from SPEC.
* common/ssh-utils.c (get_fingerprint): Support ecdsa.
* agent/protect.c (protect_info): Add flag ECC_HACK.
(agent_protect): Allow the use of the "curve" parameter.
* agent/t-protect.c (test_agent_protect): Add a test case for ecdsa.
* agent/command-ssh.c (ssh_key_grip): Print a better error code.
--
The 3 standard curves are now supported in gpg-agent's ssh-agent
protocol implementation. I tested this with all 3 curves and keys
generated by OpenSSH 5.9p1.
Using existing non-ssh generated keys will likely fail for now. To fix
this, the code should first undergo some more cleanup; then the fixes
are pretty straightforward. And yes, the data structures are way too
complicated.
* agent/command-ssh.c: Remove dirent.h.
(control_file_s): Add struct item.
(rewind_control_file): New.
(search_control_file): Factor code out to ...
(read_control_file_item): New.
(ssh_handler_request_identities): Change to iterate over entries in
sshcontrol.
--
Formerly we scanned the private key directory for matches of entries
in sshcontrol. This patch changes it to scan the sshcontrol file and
thus considers only keys configured there. The rationale for this is
that it is common to have only a few ssh keys but many private keys.
Even if that assumption does not hold true, the scanning of the
sshcontrol file is faster than reading the directory and only then
scanning the ssh control for each directory entry.
* agent/command-ssh.c (SSH_CONTROL_FILE_NAME): New macro to replace
the direct use of the string.
(struct control_file_s, control_file_t): New.
(open_control_file, close_control_file): New. Use them instead of
using fopen/fclose directly.
* agent/gpg-agent.c (main): Pass new envar gnupg_SSH_AUTH_SOCK_by to
an invoked process.
--
This environment variable is useful for debugging if
--use-standard-socket is used (which is the default since 2.1).
Commonly you should have this in your init script (e.g. ~/.bashrc):
unset GPG_AGENT_INFO
unset SSH_AGENT_PID
SSH_AUTH_SOCK="${HOME}/.gnupg/S.gpg-agent.ssh"
export SSH_AUTH_SOCK
The problem is that gpg-agent won't be able to override the
SSH_AUTH_SOCK envvar if gpg-agent has been invoked as
gpg-agent --enable-ssh-support --daemon /bin/bash
To fix this you should instead use this code in the init script:
unset GPG_AGENT_INFO
unset SSH_AGENT_PID
if [ ${gnupg_SSH_AUTH_SOCK_by:-0} -ne $$ ]; then
export SSH_AUTH_SOCK="${HOME}/.gnupg/S.gpg-agent.ssh"
fi
This will work in all cases and thus allows to start gpg-agent for
testing purposes with a different homedir and use this gpg-agent as an
ssh-agent. Example:
GNUPGHOME=$(pwd) gpg-agent --enable-ssh-support --daemon /bin/bash
gnupg_SSH_AUTH_SOCK_by is set to the PID of the exec-ed process and
thus will work safely if called recursively.
* scd/apdu.c (pcsc_no_service): Remove.
(open_pcsc_reader_direct, open_pcsc_reader_wrapped): Remove
pcsc_no_service support.
(apdu_open_reader): Remove R_NO_SERVICE.
* scd/apdu.h (apdu_open_reader): Remove R_NO_SERVICE.
* scd/command.c (reader_disabled): Remove.
(get_current_reader): Follow the change of R_NO_SERVICE.
(open_card, cmd_serialno, scd_command_handler): Remove reader_disabled
support.
* scd/sc-copykeys.c (main): Follow the change of R_NO_SERVICE.
--
Daemon should handle all possible cases. Even if such a difficult
case like reader_disabled, it should not exit.
* scd/command.c (update_reader_status_file): Don't call
get_current_reader.
--
This fix has a impact that the insertion of a card reader will not be
detected upon the insertion, but will be deferred until user tries to
access his card.
* keyserver.c (print_keyrec): Honor --keyid-format when getting back
full fingerprints from the keyserver (the comment in the code was
correct, the code was not).
* configure.ac (HAVE_BROKEN_TTYNAME): New ac_define set for Android
systems.
* common/util.h (gnupg_ttyname): New macro. Change all callers of
ttyname to use this macro instead.
(ttyname) [W32]: Rename to _gnupg_ttyname and use also if
HAVE_BROKEN_TTYNAME is defined.
* common/simple-pwquery.c (agent_send_all_options): Keep on using
ttyname unless HAVE_BROKEN_TTYNAME is set. This is because this file
may be used standalone.
* scd/command.c (MAXLEN_SETDATA): New.
(cmd_setdata): Add option --append.
* agent/call-scd.c (agent_card_pkdecrypt): Use new option for long
data.
* scd/app-openpgp.c (struct app_local_s): Add field manufacturer.
(app_select_openpgp): Store manufacturer.
(do_decipher): Print a note for broken cards.
--
Please note that I was not able to run a full test because I only have
broken cards (S/N < 346) available.
* findkey.c (agent_public_key_from_file): Fix use of
gcry_sexp_build_array.
--
A test case leading to a segv in Libgcrypt is
gpg-connect-agent \
"READKEY 9277C5875C8AFFCB727661C18BE4E0A0DEED9260" /bye
The keygrip was created by "monkeysphere s", which has a comment.
gcry_sexp_build_array expects pointers to the arguments which is quite
surprising. Probably ARG_NEXT was accidentally implemented wrongly.
Anyway, we can't do anything about it and thus need to fix the check
the users of this function.
Some-comments-by: Werner Koch <wk@gnupg.org>
* configure.ac (HAVE_ANDROID_SYSTEM, RUN_TESTS): New.
(AH_BOTTOM) [__ANDROID__]: Do not re-define ttyname.
* Makefile.am: Depend tests on new RUN_TESTS conditional.
