* g10/keyedit.c (no_usable_encr_subkeys_warning): New.
(keyedit_menu): Call it after running an expire command.
* g10/import.c (import_one_real): Call it in the show_key case.
--
* kbx/keybox-init.c (ll_buffer_size): New var intialized to 128k
(stream_buffers): New var.
(keybox_set_buffersize): New.
(_keybox_ll_open, _keybox_ll_close): Implement buffering.
* sm/gpgsm.c (oKbxBufferSize): New.
(opts): Add option --kbx-buffer-size.
(main): Call keybox_set_buffersize.
* g10/gpg.c: Include keybox.h.
* (oKbxBufferSize): New.
(opts): Add option --kbx-buffer-size.
(main): Call keybox_set_buffersize.
--
Running a test on Windows using a pubring.kbx with
Total number of blobs: 2098
openpgp: 1294
x509: 803
and a size of 42MiB with
gpgsm -k --with-validation --disable-dirmngr --kbx-buffer-size N >nul
gives these performance figures using procmon
| N(k) | file events | time(s) |
|------+-------------+---------|
| 0 | 4900000 | 86 |
| 16 | 2456000 | 58 |
| 32 | 1233000 | 43 |
| 64 | 622000 | 37 |
| 128 | 317000 | 32 |
| 256 | 164000 | 31 |
| 512 | 88000 | 30 |
Using _open instead of CreateFile give the same number of file events
but increased the time slight by one second for the measured buffer
size of 64k and 128k. Benchmarks for gpg have not been conducted.
* g10/import.c (do_transfer): Force the overridden key import
even when smartcard is available.
--
Fixes-commit: 2c12970550
GnuPG-bug-id: 3456
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* agent/command.c (cmd_keytocard): Add new arg for ECDH params.
* scd/app-openpgp.c (ecc_writekey): Use provided ECDH params to
compute the fingerprint.
* g10/call-agent.c (agent_keytocard): Add arg ecdh_param_str.
* g10/keyid.c (ecdh_param_str_from_pk): New.
* g10/card-util.c (card_store_subkey): Pass ECDH params to writekey.
* g10/keygen.c (card_store_key_with_backup): Ditto.
* scd/app-openpgp.c (store_fpr): Add arg update.
(rsa_read_pubkey, ecc_read_pubkey): Add arg meta_update and avoid
writing the fingerprint back to the card if not set.
(read_public_key): Also add arg meta_update.
(get_public_key): Do not pass it as true here...
(do_genkey): ... but here.
(rsa_write_key, ecc_writekey): Force string the fingerprint.
--
The problem showed up because in 2.4 we changed the standard ECDH
parameter some years ago. Now when trying to write an ECDH key
created by 2.2 with 2.4 to an openpgp card, scdaemon computes a wrong
fingerprint and thus gpg was not able to find the key again by
fingerprint.
The patch also avoids updating the stored fingerprint in certain
situations.
This fix is somewhat related to
GnuPG-bug-id: 6378
* g10/card-util.c (current_card_status): Print a hint for fishy
outputs.
(enum cmdids): Add cmdOPENPGP.
(cmds): Add "openpgp".
(card_edit): Implement that command.
--
If a Yubikey has been used in PIV mode the initial listing does not
look correct. Although we explicitly switch to the OpenPGP app when
needed, we don't want to do this in listing mode. Instead we offer a
new command "openpgp" to force the openpgp mode. The ultimate goal
will be to enhance the gpg-card tool to completely take over the
--card-edit features. But we are not yet there.
GnuPG-bug-id: 6462
* g10/call-agent.c (agent_scd_switchapp): New.
* g10/card-util.c (get_info_for_key_operation): Call it.
--
It may happen that the active card was last used for PIV and in that
case certain commands will fail because they assume the OpenPGP app.
Fortunately we have a pretty central place to assure that the right
app has been selected.
The bug can be easily noticed on Windows.
GnuPG-bug-id: 6378
* g10/import.c (transfer_secret_keys): Only emit a warning when secret
key is not encrypted.
--
Fixing-commit: dbfb7f809b
GnuPG-bug-id: 6322
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* common/homedir.c (gnupg_maybe_make_homedir): Also create a
common.conf.
* g10/keydb.c: Include comopt.h.
(maybe_create_keyring_or_box): Detect the creation of a common.conf.
* g10/gpg.c (main): Avoid adding more resources in this case.
* sm/keydb.c: Include comopt.h.
(maybe_create_keybox): Detect the creation of a common.conf.
* common/comopt.h (comopt): Remove the conditional "extern".
* g10/card-util.c (card_store_subkey): Add arg processed_keys.
* g10/keyedit.c (keyedit_menu): Delete secret key.
--
This used to work using the gpg-agent: learn we called at "save" time.
However, the recent change inhibited the creation of a shadow key by
learn if a regular key still exists. Now we do an explicit delete key
at save time. This syncs the behaviour with the description of the
man page.
GnuPG-bug-id: 6378
* g10/free-packet.c (copy_public_key): Factor some code out to ...
(copy_public_key_basics): new.
* g10/build-packet.c (build_sig_subpkt_from_sig): New arg signhints.
* g10/packet.h (PUBKEY_USAGE_RENC): Fix value.
(SIGNHINT_KEYSIG, SIGNHINT_SELFSIG): Moved from sign.c.
(SIGNHINT_ADSK): New.
(PKT_public_key): Change pubkey_usage from byte to u16.
(PKT_user_id): Cosmetic fix: change help_key_usage from int to u16.
* g10/getkey.c (parse_key_usage): Make public.
* g10/misc.c (openpgp_pk_algo_usage): Take PUBKEY_USAGE_RENC in
account.
* g10/sign.c (update_keysig_packet): Set SIGNHINT_ADSK.
(make_keysig_packet): Ditto.
(do_sign): No time warp check in ADSK mode.
* g10/sig-check.c (check_signature_metadata_validity): Ditto.
* g10/keygen.c (struct opaque_data_usage_and_pk): Remove.
(write_keybinding): Do not use the removed struct.
(do_add_key_flags): Support PUBKEY_USAGE_RENC and others.
(keygen_add_key_flags_and_expire): Rewrite and make public.
* g10/keyedit.c (enum cmdids): Add cmdADDADSK.
(keyedit_menu): Add command "addadsk".
(menu_addadsk): New.
--
This makes use of a new encryption flag:
The "restricted encryption key" (2nd,0x04) does not take part in any
automatic selection of encryption keys. It is only found on a
subkey signature (type 0x18), one that refers to the key the flag
applies to.
Followup patches will add encryption support and a --quick command.
GnuPG-bug-id: 6395
* g10/gpg.c (main): Replace setmode by es_set_binary and use only when
needed.
--
It is better to use our es_set_binary than to use a Windows specific
method which still worked but is fragile because estream might be
changed. We now set binary only when needed. Note that it does not
harm to call es_set_binary more often than needed.
* g10/options.h (LIST_SHOW_UNUSABLE_SIGS): New.
* g10/gpg.c (parse_list_options): Add "show-unusable-sigs".
* g10/keydb.h (keyid_eq): New.
(pk_is_primary): New.
* g10/keylist.c (list_signature_print): Early return for weak key
signatures. Print "self-signature" instead of user-id.
(list_keyblock_print): Simplify and always set self-sig node flag.
--
This patch avoid the printing of often hundreds of "Invalid digest
algorithm" notices during key signature listings if those key
signatures were done with SHA1. The new option can be used to revert
the behaviour.
We now also print "[self-signature]" with --check-sigs or --list-sigs
instead of the primary user id. This makes such listing easier to read.
* common/compliance.c (gnupg_pk_is_allowed): Handle EdDSA.
* g10/gpg.c (oOverrideComplianceCheck): Remove.
(opts): Turn --override-compliance-check into a dummy option.
* g10/options.h (opt): Remove override_compliance_check.
* g10/sig-check.c (check_key_verify_compliance): Remove use of that
option.
--
The introduction of --override-compliance-check actually hid the real
cause for the signature verification problem in de-vs mode for the
Ed25519 key. The real fix is to handle the EdDSA algorithm in
gnupg_pk_is_allowed.
Fixes-commit: fb26e144ad
GnuPG-bug-id: 5655
* common/iobuf.c (file_filter_ctx_t): Add fields for the peek feature.
(file_filter): Implement peeking.
(iobuf_ioctl): Add new IOBUF_IOCTL_PEEK.
* common/iobuf.h (IOBUF_IOCTL_PEEK, IOBUFCTRL_PEEK): New.
* common/miscellaneous.c (is_file_compressed): Rewrite. Detect PDF.
* g10/encrypt.c (encrypt_simple): Peek before detecting compression.
(encrypt_crypt): Ditto.
* g10/sign.c (sign_file): Also detect already compressed data.
* g10/options.h (opt): Add explicit_compress_option.
* g10/gpg.c (main): Set opt.explicit_compress_option for -z.
--
Note that this patch also introduces a compression check for signing
which was never done in the past.
GnuPG-bug-id: 6332
* g10/cipher-aead.c (do_flush): Use %llu and a cast.
* g10/decrypt-data.c (aead_underflow): Ditto.
--
Fixes-commit: b2cedc108d
We don't use the system's printf but the one implemented by
us (gpgrt's estream-printf) thus the PRIu64 may or may not be correct.
We can't do much about the -Wformat errors due to our different
implementation.
* agent/command.c (cmd_export_key): Add option --mode1003.
(command_has_option): Ditto.
* g10/build-packet.c (do_key): Implement mode 1003.
* g10/parse-packet.c (parse_key): Ditto.
* g10/options.h (EXPORT_MODE1003): New.o
* g10/call-agent.c (agent_export_key): Add arg mode1003.
* g10/export.c (parse_export_options): Add "mode1003"
(secret_key_to_mode1003): New.
(receive_seckey_from_agent): Add arg mode1003.
(do_export_one_keyblock): Pass option down.
--
This option allows to export a secret key in GnuPG's native format.
Thus no re-encryption is required and further the public key parameters
are also authenticated if a protection passphrase has been used.
Note that --import is not yet able to handle this new mode. Although
old version of GnuPG will bail out with "invalid packet" if a mode1003
exported secret key is seen.
* g10/export.c (receive_seckey_from_agent): Add arg r_key.
(do_export_one_keyblock): Pass NULL for new arg.
(receive_raw_seckey_from_agent): Remove.
(export_secret_ssh_key): Use receive_seckey_from_agent.
* g10/keygen.c (card_store_key_with_backup): Pass NULL for new arg.
* g10/mainproc.c (proc_encrypted): Set complaince_de_vs also if
require-compliance is set.
--
Without this fix require-compliance would fail if no --status-fd was
used.
* g10/gpg.c (oListFilter): New.
(opts): Add --list-filter.
(main): Parse oListFilter.
* g10/keylist.c: Include init.h and recsel.h.
(struct list_filter_s, list_filter): New.
(release_list_filter): New.
(cleanup_keylist_globals): New.
(parse_and_set_list_filter): New.
(list_keyblock): Implement --list-filter type "select".
* g10/import.c (impex_filter_getval): Add scope support and new
property names "key-size", "algostr", "origin", "lastupd", and "url".
--
This option is pretty useful to select keys based on their properties.
The scope thing can be sued to limit a selection to just the primary
key or to subkeys. For example:
gpg -k --list-filter 'select=revoked-f && sub/algostr=ed25519'
Lists all non-revoked keys with an ed25519 (signing)-subkey.
* g10/gpg.c (aQuickUpdatePref): New.
(opts): Add --quick-update-pref.
(main): Implement.
* g10/keyedit.c (keyedit_quick_update_pref): New.
(menu_set_preferences): Add arg 'unattended' and adjust caller.
--
This new quick command is in particular useful to update existing keys
so that they can be used with OCB mode.
* g10/armor.c (is_armored): Add PKT_ENCRYPTED_AEAD.
--
With this fix it is now possible to feed a vanilla packet of type 20
without first forcing gpg to assume binary mode.
* g10/gpg.c (oRFC4880bis): Remove.
(opts): Make --rfc4880bis a Noop.
(compliance_options): Make rfc4880bis to gnupg.
(set_compliance_option): Remove rfc4880bis stuff.
(main): Ditto. Note that this now activates the --mimemode option.
* g10/keygen.c (keygen_set_std_prefs): Remove rfc4880bis protection.
(keygen_upd_std_prefs): Always announce support for v5 keys.
(read_parameter_file): Activate the v4 and v5 keywords.
--
* g10/gpg.c (opts): New option--force-ocb as alias for force-aead.
Turn --aead-algo and --personal-aead-preferences into dummy options.
(build_list_md_test_algo, build_list_aead_algo_name): Remove.
(my_strusage): Remove output of AEAD algos.
(main): Remove code from the --aead options.
* g10/encrypt.c (encrypt_seskey): Make file local.
(use_aead): Remove requirement for rfc4880bis. Always return
AEAD_ALGO_OCB.
* g10/main.h (DEFAULT_AEAD_ALGO): Removed unused macro.
* g10/misc.c (default_aead_algo): Remove.
* g10/pkclist.c (select_aead_from_pklist): Return AEAD_ALGO_OCB or 0.
(select_algo_from_prefs): Remove personal AEAD algo setting.
* g10/keygen.c (keygen_set_std_prefs): Remove AEAD preference option
parsing.
* g10/options.h (opt): Remove def_aead_algo and personal_aead_prefs.
--
Due to the meanwhile expired patent on OCB there is no more reason for
using EAX. Thus we forcefully use OCB if the AEAD feature flag is set
on a key.
* g10/gpg.c (oCompatibilityFlags): New.
(opts): Add option.
(compatibility_flags): New list.
(main): Set flags and print help.
* g10/options.h (opt): Add field compatibility_flags.
--
No flags are yet defined but it is good to have the framework.
* g10/kbnode.c (new_kbnode2): New.
* g10/import.c (delete_inv_parts): New arg r_otherrevsigs to store
misplaced revocations.
(import_revoke_cert): Allow to pass an entire list.
(import_one): Import revocations found by delete_inv_parts.
--
It might be useful to distribute revocations of old keys along with
new keys. This is in particicualrr useful for WKD stored keys. This
patch allows to put unrelated standalone revocations into a key. For
example they can simply appended to a keyblock. Right now it is a bit
inaesthetic to see diagnostics about misplaced or bad revocation
signatures.
* g10/ecdh.c (derive_kek): Use GCRY_KDF_ONESTEP_KDF.
--
This change is not yet enabled. We will be able to use the code when
we update NEED_LIBGCRYPT_VERSION to 1.11.0. Before the update, gpg
compiled with libgcrypt 1.11.0 can't work with older libgcrypt
runtime.
GnuPG-bug-id: 5964
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* g10/decrypt.c (decrypt_message_fd): Use INPUT_FD directly.
* g10/encrypt.c (encrypt_crypt): Use FILEFD directly.
--
Before 8402815d, original code was with iobuf_open_fd_or_name, which
used gnupg_fd_t for the file descriptor (FD2INT was relevant at that
time). After the change, because it's not gnupg_fd_t but int, use of
FD2INT is irrelevant.
Fixes-commit: 8402815d8e
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* g10/encrypt.c (encrypt_crypt, encrypt_filter): Factor common code
out to ...
(create_dek_with_warnings): new
(check_encryption_compliance): and new.
* g10/encrypt.c (encrypt_filter): Add the compliance check.
--
GnuPG-bug-id: 6174
Ported-from: f88cb12f8e
* g10/parse-packet.c (enum_sig_subpkt): Show "buffer shorter than
subpacket" only in debug mode.
(parse_signature): Show "signature packet without timestamp / keyid"
only in souble verbose mode.
* g10/sig-check.c (check_signature_metadata_validity): Use ISO
timestamp in UTC for the signature expired note.
--
I have seen to many of these diagnostics and in particular the first
one seems to be a connected to the others. Thus it does not make
sense to show them in standard verbose mode.
The ISO timestamp is much easier to read than than the localized
timestamp and switching from localtime to UTC should not harm.
* g10/mainproc.c (proc_encrypted): Change compliance logic.
--
For the description of the proplem see
https://dev.gnupg.org/T6205#163306
GnuPG-bug-id: 6205
* g10/packet.h (PUBKEY_USAGE_RENC): New.
(PUBKEY_USAGE_TIME): New.
(PUBKEY_USAGE_GROUP): New.
* g10/getkey.c (parse_key_usage): Set the new key flags.
* g10/keyedit.c (show_key_with_all_names_colon): Show the new key
flags.
* g10/keyid.c (usagestr_from_pk): Ditto
* g10/keylist.c (print_capabilities): Ditto.
* g10/keygen.c (parse_usagestr): Parse line and set new flags.
(quickgen_set_para): Show flags.
--
See draft-koch-openpgp-2015-rfc4880bis-00 for the current version.
Actually these flags have been in the draft for years now. This patch
is a first step to make use of them.
* common/comopt.c (opts): Add "no-autostart".
(parse_comopt): Set it.
* common/comopt.h (comopt): Add no_autostart.
* g10/gpg.c (main): Take care of the new option.
* sm/gpgsm.c (main): Ditto.
* tools/gpg-connect-agent.c (INCLUDED_BY_MAIN_MODULE): Add.
(main): Parse common options and handle new option.
* tools/gpg-card.c (main): Ditto.
(cmd_yubikey): Fix minor error reporting issue.
* common/util.h (GNUPG_MODULE_NAME_CARD): New const.
* common/homedir.c (gnupg_module_name): Support it.
--
Having a global option makes it easier to use disable autostart on a
server which is required to use a remote gpg-agent reliable.
* g10/keygen.c (keygen_set_std_prefs): Allow extra spaces before
preference elements. Detect the bracketed versions of the strings.
Ignore "aead".
--
This allows to c+p the list shown by pref with out remove the
brackets.
* g10/keyedit.c (keyedit_quick_set_primary): Issue a status error.
--
This allows GpgME to detect and report a failure if setting the primary
user ID of a key failed.
GnuPG-bug-id: 6126
* g10/keyedit.c (find_userid_by_namehash, find_userid): Add argument
want_valid. Skip invalid user IDs if valid is wanted.
(keyedit_quick_revuid): Ask find_userid() for any matching user ID.
(keyedit_quick_set_primary): Use find_userid() to find the user ID to
mark as primary.
* tests/openpgp/quick-key-manipulation.scm: Change second call of the
quick-set-primary-uid test to specify the user ID by its hash.
--
This makes it possible to specify the user ID to mark as primary via its
UID hash when calling --quick-set-primary-uid.
GnuPG-bug-id: 6126
* g10/mainproc.c (proc_encrypted): Make symmetric + pubkey encryption
de-vs compliant.
* g10/mainproc.c (struct symlist_item): New.
(struct mainproc_context): Add field symenc_list.
(release_list): Free that list.
(proc_symkey_enc): Record infos from symmetric session packet.
(proc_encrypted): Check symkey packet algos
--
The original check was too strong because it is in fact compliant to
encrypt with a symmetric key and and public key. Thus decryption
should issue a compliance status.
In addition we now check that the cipher algorithms used to
symmetrically encrypt the session key are all compliant. This is
similar to our check for all public key encrypted session key packets.
GnuPG-bug-id: 6119
Fixes-commit: b03fab09e1
Backported from 2.2
Signed-off-by: Werner Koch <wk@gnupg.org>
* g10/cpr.c (write_status_text_and_buffer): Fix off-by-one
--
Depending on the escaping and line wrapping the computed remaining
buffer length could be wrong. Fixed by always using a break to
terminate the escape detection loop. Might have happened for all
status lines which may wrap.
GnuPG-bug-id: T6027
* g10/keygen.c (keygen_set_std_prefs): Use the right variable when
reading AEAD preference string
--
GnuPG-bug-id: 6019
Signed-off-by: Jakub Jelen <jjelen@redhat.com>
* g10/export.c (do_export_stream): Merge the key to get the properties
ready.
--
This makes
gpg --export --export-filter 'drop-subkey=usage=~a'
(Export all subkeys but those with the auth usage)
work without using the workaound of adding
--export-options export-clean
* g10/keyedit.c (find_userid_by_namehash, find_userid): New.
(keyedit_quick_revuid): Use find_userid() instead of iterating over the
nodes of the keyblock.
* tests/openpgp/quick-key-manipulation.scm: Add test for revoking a
user ID specified by its hash.
--
This makes it possible to specify the user ID to revoke as UID hash when
calling --quick-revoke-uid.
GnuPG-bug-id: 5936
* g10/mainproc.c (proc_symkey_enc): Issue new error code.
(proc_encrypted): Ditto.
--
This allows GPGME to return a better error message than "bad session
key" to the user. Technically we could get run into these errors also
in other cases but this more unlikley. For the command line use we
don't do anything to not change the expected output of the command
line interface.
GnuPG-bug-id: 5943
* g10/keyedit.c (menu_adduid): Move linefeed character to the format
string.
--
This fixes a literal '\n' in the error message and a missing line feed
after the error message.
* g10/gpg.c (main): Set LOG_NO_REGISTRY.
* sm/gpgsm.c (main): Ditto.
* tools/gpg-connect-agent.c (main): Ditto.
* tools/gpgconf.c (main): Ditto.
(show_other_registry_entries): Print "DefaultLogFile".
--
The intention of this mostly forgotten registry entry was to allow for
easy debugging of the tools. However, with the global config file
common.conf things are anyway better. We disable the use for the
commonly used tools so that it does not look like calling gpg on the
command line seems to block with no output if the log
server (e.g. tcp://1.2.3.4:11111) is not reachable.
* g10/keygen.c (ask_user_id): Allow for the name to start with a
digit. Allow names shorter than 5.
--
The reason for this change is that we don't enforce these constraints
in the --quick-gen-key interface. I added the constraints right in the
beginning of gnupg to make sure that we have a uniform style for
user-ids. However, this is all problematic with non-Latin names
and we prefer to use mail addresses anyway.
* g10/gpg.c (main): Remove note about rfc4880bis.
* g10/keygen.c (keygen_set_std_prefs): Use only OCB in the AEAD
preference list.
--
It is more than unlikely that EAX will ever be used in practice and
thus we remove it from the preference list.
* g10/misc.c (openpgp_cipher_algo_mode_name): New.
* g10/decrypt-data.c (decrypt_data): Use function here.
--
Note that openpgp_cipher_algo_mode_name is different from the version
2.2 becuase we append ".CFB" here.
Without this change we would see
gpg: cipher algorithm 'AES256' may not be used in
--compliance=de-vs mode
This is confusing because AES256 is compliant. Now we see
gpg: cipher algorithm 'AES256.OCB' may not be used in
--compliance=de-vs mode
which gives a hint on the problem.
* g10/decrypt-data.c (decrypt_data): Add arg compliance_error. Adjust
all callers. Fail on compliance error only in --require-compliance
mode. Make sure to return an error if the buffer is missing; actually
that should be an assert.
* g10/mainproc.c (proc_encrypted): Delay printing of the compliance
mode status. Consult the compliance error now returned by
decrypt_data.
--
The actual case here is that we fail hard if a message has been AEAD
encrypted with one AEAD capable key and also with one passphrase. In
general the preference system takes care of not using AEAD if one
recipient's key does not support it. However, if the sender uses her
own AEAD-capable key _and_ a passphrase the message will be AEAD
encrypted. This change allows to decrypt that anyway along with a
warning message.
Note that this does currently not work in 2.3 due to a non-compliant
libgcrypt. We will however, backport this to 2.2.
* g10/sign.c (sign_symencrypt_file): Insert correct version and AEAD
information into symkey packet.
--
GnuPG-bug-id: 5856
Signed-off-by: Jakub Jelen <jjelen@redhat.com>
* g10/options.h (opt): Add field flags.require_compliance.
* g10/gpg.c (oRequireCompliance): New.
(opts): Add --require-compliance.
(main): Set option.
* g10/mainproc.c (proc_encrypted): Emit error if non de-vs compliant.
(check_sig_and_print): Ditto.
* g10/encrypt.c (encrypt_crypt): Ditto.
--
Note that in the --encrypt and --verify cased other checks may kick in
earlier than this new --require-compliance controlled one.
* common/iobuf.c (filter_flush): Remove "src_len == 0" check.
* g10/compress-bz2.c (do_compress): Exit early if flush not
forced and input length is zero.
* g10/compress.c (do_compress): Likewise.
--
Remove "(src_len == 0)" check in filter_flush which was
introduced to fix compress failure caused by zero length
flush from iobuf_close. However this check broke enarmoring
file with length of zero. Patch instead fixes zero length
flush problem in compress filters.
GnuPG-bug-id: T5828
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
* g10/cipher-aead.c (do_hash): Add faster path for encrypting directly
from input buffer instead of memcpying then encrypting inplace.
--
When encrypting with AES256.OCB on AMD Ryzen 5800X, memcpy shows as
taking largest portion of computation time. Patch adds fast path for
AEAD encryption which largely eliminates need for memcpying when
doing AEAD encryption. AES256.OCB throughput increases from 2.2GB/s
to 4.2GB/s on said system (1.9x faster).
GnuPG-bug-id: T5828
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
* g10/decrypt-data.c (decrypt_data): Disable estream buffering for
output file.
--
Here estream is filled with iobuf_copy which already uses large buffers
so additional buffering in estream was just adding memory copy overhead.
GnuPG-bug-id: T5828
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
* g10/plaintext.c (handle_plaintext): Disable estream buffering in
binary modes.
--
Since in binary mode, large buffers are passed from source iobuf to
target estream, extra buffering in estream only causes extra memory
copying and overhead.
GnuPG-bug-id: T5828
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
* common/iobuf.c (iobuf_copy): Use iobuf buffer size for temporary
buffers.
* g10/plaintext.c (handle_plaintext, do_hash): Likewise.
* g10/sign.c (sign_file): Likewise.
--
As iobuf will have zerocopy operation for read/write, it is better to
use same size buffers as iobuf for temporary copy buffers.
GnuPG-bug-id: T5828
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
* g10/sign.c (sign_file): Use iobuf_read instead of iobuf_get for
reading data from detached file.
--
This patch reduces iobuf_read per byte processing overhead and speeds
up detached signing.
Detached signing speed on AMD Ryzen 5800X (4.3GiB file, SHA256):
gpg process
user time
before: 3.951s
after: 1.898s (2.0x faster)
GnuPG-bug-id: T5826
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
* g10/plaintext.c (do_hash): Use iobuf_read instead of iobuf_get for
reading data; Use gcry_md_write instead of gcry_md_putc for hash data.
--
This patch reduces iobuf_read per byte processing overhead and speeds
up detached signature verifying.
Detached verifying speed on AMD Ryzen 5800X (4.3GiB file, SHA256):
gpg process
user time
before: 9.410s
after: 1.913s (4.9x faster)
GnuPG-bug-id: T5826
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
* g10/mainproc.c (proc_tree): Call ask_for_detached_datafile
with MD2=NULL.
--
Here, c->mfx.md2 is always NULL, in fact. But, text-wise
(when searching the use of "mfx.md2"), before the change, it used
mfx.md2, which is irrelevant in the handling of PKT_ONEPASS_SIG.
Note that: One-Pass Signature is not available in PGP2.
This fix removes (text-wise) unmatch of the calls of functions
hash_datafile_by_fd hash_datafiles, and ask_for_detached_datafile.
Fixes-commit: 88a916cdd4
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
