* common/dotlock.c (read_lockfile): Return the file descriptor when
R_FD is available.
(dotlock_take_unix): Check the case the lockfile was already removed.
--
GnuPG-bug-id: 5884
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* common/t-dotlock.c (lock_and_unlock): Use usleep and faster.
Loop at least once. Use getrandom for random time.
(main): Add new option --one-shot to run lock/unlock once.
--
GnuPG-bug-id: 5884
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* 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>
* common/iobuf.c (filter_flush): Use 'iobuf->e_d' if configured.
(iobuf_write): Configure 'iobuf->e_d' for 'filter_flush' if
'iobuf->d.buf' is empty and external buffer is larger than threshold.
--
Zero-copy operation in iobuf_write() and filter_flush() allow bypassing
'iobuf->d.buf' for greater performance. This mainly helps OCB
performance where additional memory copies through iobuf stack
can take significant portion of program time.
GnuPG-bug-id: T5828
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
* common/iobuf.h (iobuf_struct): Add 'e_d' substructure and members.
* common/iobuf.c (IOBUF_ZEROCOPY_THRESHOLD): New.
(iobuf_alloc): Clear 'iobuf->e_d'.
(underflow_target): Use 'iobuf->e_d' when configured to bypass copying
through 'iobuf->d.buf'.
(iobuf_read): Configure 'iobuf->e_d' for 'underflow' if 'iobuf->d.buf'
is empty and external buffer is larger than threshold.
--
Zero-copy operation in iobuf_read() and underflow() allow bypassing
'iobuf->d.buf' for greater performance. This mainly helps OCB
performance where additional memory copies through iobuf stack
can take significant portion of program time.
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>
* tools/gpgtar.c: Replace assert by log_assert.
* tools/gpgtar-extract.c: Ditto.
(extract_regular): Create files with sysopen flag.
* tools/gpgtar-create.c (scan_directory): Use gpgrt_fname_to_wchar.
--
Note that for this change libgpg-error 1.45 is required for Windows.
* scd/app.c (app_auth): It's only APPTYPE_OPENPGP which supports
the challenge response interaction.
* scd/command.c (cmd_pkauth): It only wants if it works or not.
--
GnuPG-bug-id: 5862
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* 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>
* agent/call-daemon.c (wait_child_thread): Print the correct name.
--
This makes sure that the log print the actual used name if for example
--pinentry-program was used.
* agent/trustlist.c (struct trustitem_s): Add flag "qual".
(read_one_trustfile): Rename arg "allow_include" to "systrust" and
change callers. Parse new flag "qual".
(istrusted_internal): Print all flags.
* sm/call-agent.c (istrusted_status_cb): Detect the "qual" flag.
* sm/gpgsm.h (struct rootca_flags_s): Add flag "qualified".
* sm/certchain.c (do_validate_chain): Take care of the qualified flag.
* dirmngr/validate.c (pk_algo_from_sexp): Make public. Support ECC.
* dirmngr/ocsp.c (check_signature): Remove hash preparation out to ...
(check_signature_core): here. This changes the arg s_hash to md.
Support ECDSA.
--
The test was done with my qualified signature certificate from the
Telesec and their responder http://tqrca1.ocsp.telesec.de/ocspr .
See also libksba commit rK24992a4a7a61d93759e1dbd104b845903d4589bf
* 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>
* g10/mainproc.c (mainproc_context): New member
'seen_pkt_encrypted_mdc'.
(release_list): Clear 'seen_pkt_encrypted_mdc'.
(proc_encrypted): Set 'seen_pkt_encrypted_mdc'.
(have_seen_pkt_encrypted_aead): Rename to...
(have_seen_pkt_encrypted_aead_or_mdc): ...this and add check for
'seen_pkt_encrypted_mdc'.
(proc_plaintext): Do not enable extra hash contexts when decrypting
MDC input.
--
Avoiding extra hash contexts speeds up CFB/MDC decryption quite
a lot. For example, decrypting symmetric-key AES-256 encrypted
4 GiB file from RAM to /dev/null sees ~3.4x speed increase on
AMD Ryzen 5800X:
AES256.CFB encryption: 783 MB/s
AES256.CFB decryption: 386 MB/s (before)
AES256.CFB encryption: 1.3 GB/s (after patch)
Note, AEAD is still significantly faster:
AES256.OCB encryption: 2.2 GB/s
AES256.OCB decryption: 3.0 GB/s
GnuPG-bug-id: T5820
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
* g10/cipher-aead.c (do_flush): Use PRIu64.
* g10/decrypt-data.c (aead_underflow): Likewise.
--
Even among LP64 data model machines, uint64_t type may differ;
unsigned long or unsigned long long.
Only portable way is use of PRIu64.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* dirmngr/dirmngr.c (dirmngr_never_use_tor_p): New.
* dirmngr/server.c (ensure_keyserver): Don't even test for the Tor
proxy in never-use-tor Mode.
* tools/gpgtar-create.c: Include unistd.h to avoid a warning on
Windows.
--
This delay of 2 or 3 seconds is in particular annoying on Windows.
This is now suppressed, as it should be, if --no-use-tor is used.
The second patch is unrelated
* agent/command-ssh.c (sexp_key_construct): Do not put an empty string
into an S-expression.
(stream_read_string): Do not not try to a read a zero length block.
--
Actually we could handles this different by not putting a comment tag
into the s-expression, however this requires more code and at other
places we already return "(none)" instead of an empty comment.
The second fix is more or less a cosmetic thing to get better error
messages in case the underlying read system call returns an error.
GnuPG-bug-id: 5794
* g10/filter.h (armor_filter_context_t): New fields dearmor_mode and
dearmor_state.
* g10/dearmor.c (dearmor_file): Set dearmor_mode.
* g10/armor.c (is_armor_header): Magic to switch to generic dearmor
mode.
(parse_header_line): Treat non OpenPGP armor in a special way.
(check_input): Ditto.
(radix64_read): Detect non OpenPGP armor END lines.
* sm/gpgsm.c (oKeyServer_deprecated): New.
(opts): Assign "ldapserver" to the new option and mark it as obsolete.
--
We want to use "ldapserver" in dirmngr but need to keep using
"keyserver" in gpgsm for existant versions of Kleopatra etc.
GnuPG-bug-id: 5801
