* agent/command.c (cmd_genkey): Add option --passwd-nonce.
(cmd_passwd): Return a PASSWD_NONCE in verify mode.
* g10/call-agent.c (agent_genkey): Add arg 'passwd_nonce_addr' and do
not send a RESET if given.
(agent_passwd): Add arg 'verify'.
* g10/keygen.c (common_gen): Add optional arg 'passwd_nonce_addr'.
(gen_elg, gen_dsa, gen_ecc, gen_rsa, do_create): Ditto.
(generate_subkeypair): Use sepeare hexgrip var for the to be created
for hexgrip feature. Verify primary key first. Make use of the
passwd nonce. Allow for a static passphrase.
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/command.c (cmd_delete_key): Make the --force option depend on
--disallow-loopback-passphrase.
* g10/call-agent.c (agent_delete_key): Add arg FORCE.
* g10/delkey.c (do_delete_key): Pass opt.answer_yes to
agent_delete_key.
--
Unless the agent has been configured with
--disallow-loopback-passpharse an unattended deletion of a secret key
is now possible with gpg by using --batch _and_ --yes.
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/gpg-agent.c (oNoAllowLoopbackPinentry): New.
(opts): Add --no-allow-loopback-pinentry. Hide
description of --allow-loopback-pinentry.
(parse_rereadable_options): Set opt.allow_loopback_pinentry by
default.
(main): Replace allow-loopback-pinentry by no-allow-loopback-pinentry
in the gpgconf list.
* tools/gpgconf-comp.c (gc_options_gpg_agent): Ditto.
--
Given that a user can anyway change that options in the gpg-agent.conf
file and that gpg needs to be invoked with --pinentry-mode=loopback
the former default does not make much sense - in that option is useful
at all. There was a discussion of this topic on gnupg-devel in April
without a clear result. So we try this new default and just in case
real problems are found for the majority of installations, we can
revert that. The new default is also aligned with GnuPG's policy to
make its use easier and only require users with very high security
standards to tweak certain options (those users have anyway modeled
their threat model and configured their software according to this).
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/findkey.c (write_extended_private_key): New function.
(agent_write_private_key): Detect if an existing file is in extended
format and update the key within if it is.
(read_key_file): Handle the new format.
* agent/keyformat.txt: Document the new format.
* common/Makefile.am: Add the new files.
* common/private-keys.c: New file.
* common/private-keys.h: Likewise.
* common/t-private-keys.c: Likewise.
* common/util.h (alphap, alnump): New macros.
* tests/migrations: Add test demonstrating that we can cope with the
new format.
--
GnuPG 2.3+ will use a new format to store private keys that is both
more flexible and easier to read and edit by human beings. The new
format stores name,value-pairs using the common mail and http header
convention.
This patch adds the parser and support code and prepares GnuPG 2.1 for
the new format.
Signed-off-by: Justus Winter <justus@g10code.com>
* agent/gpg-agent.c (create_private_keys_directory): Set permissions.
* common/sysutils.c (modestr_to_mode): New function.
(gnupg_mkdir): Use new function.
(gnupg_chmod): New function.
* common/sysutils.h (gnupg_chmod): New prototype.
* tests/migrations/from-classic.test: Test migration with existing
directory.
GnuPG-bug-id: 2312
Signed-off-by: Justus Winter <justus@g10code.com>
* agent/protect.c (do_encryption): Fix CBC hashing.
--
The buggy code included an extra closing parenthesis before
the (protected-at) term in the CBC hashing. We now do it by
explicitly hashing the protected stuff and append the rest of the
expression instead of a fixed closing parenthesis. Note that the OCB
hashing only differs that it does no include the protected part.
Fixes-commit: 4159567f7e
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/protect-tool.c (read_and_unprotect): Add arg ctrl and pass to
agent_unprotect.
(main): Allocate a simple CTRL object and pass it to
read_and_unprotect.
(convert_from_openpgp_native): Remove stub.
(agent_key_available, agent_get_cache): New stubs.
(agent_askpin): New emulation for the one in call-pinentry.c.
(agent_write_private_key): New to dump key.
* agent/Makefile.am (gpg_protect_tool_SOURCES): Add cvt-openpgp.c
--
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/protect.c (agent_protect): Add arg use_ocb. Change all caller
to pass -1 for default.
* agent/protect-tool.c: New option --debug-use-ocb.
(oDebugUseOCB): New.
(opt_debug_use_ocb): New.
(main): Set option.
(read_and_protect): Implement option.
* agent/protect.c (OCB_MODE_SUPPORTED): New macro.
(PROT_DEFAULT_TO_OCB): New macro.
(do_encryption): Add args use_ocb, hashbegin, hashlen, timestamp_exp,
and timestamp_exp_len. Implement OCB.
(agent_protect): Change to support OCB.
(do_decryption): Add new args is_ocb, aadhole_begin, and aadhole_len.
Implement OCB.
(merge_lists): Allow NULL for sha1hash.
(agent_unprotect): Change to support OCB.
(agent_private_key_type): Remove debug output.
--
Instead of using the old OpenPGP way of appending a hash of the
plaintext and encrypt that along with the plaintext, the new scheme
uses a proper authenticated encryption mode. See keyformat.txt for a
description. Libgcrypt 1.7 is required.
This mode is not yet enabled because there would be no way to return
to an older GnuPG version. To test the new scheme use
gpg-protect-tool:
./gpg-protect-tool -av -P abc -p --debug-use-ocb <plain.key >prot.key
./gpg-protect-tool -av -P abc -u <prot.key
Any key from the private key storage should work.
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/command-ssh.c (SPEC_FLAG_WITH_CERT): New.
(ssh_key_types): Add OpenSSH cert types.
(stream_read_string): Allow a dummy read.
(ssh_receive_mpint_list): Pass SPEC by reference.
(ssh_receive_mpint_list): New arg CERT and use it.
(ssh_receive_key): Read certificate into an estream object and modify
parser to make use of that object.
--
This is a first step to support certificate via the agent. The only
effect of this change is the removal of an error message parsing the
certificate. Note that ssh-add sends the private key anyway first and
only then follows with the certificate+private key.
What we need to implement next is a way to store the certificate in
the agent and return it on request.
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/pksign.c (agent_pksign_do): Use S-exp of public key, instead
of shadowed key.
--
Reported-by: Justus Winter
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* agent/cvt-openpgp.c (convert_from_openpgp_main): Add an option not
to check existing key.
(convert_from_openpgp): Ditto.
(convert_from_openpgp_native): Call convert_from_openpgp_main with
dontcare_exist=0.
* agent/command.c (cmd_import_key): Call with dontcare_exist=force.
* common/stringhelp.c (utf8_charcount): Take additional parameter,
len. Process at most LEN bytes.
--
Signed-off-by: Neal H. Walfield <neal@g10code.com>
* agent/trustlist.c (istrusted_internal): Initialize 'err'.
--
There is a plausible path of execution so that a branch condition uses
the uninitialized value.
Found using the Clang Static Analyzer.
Signed-off-by: Justus Winter <justus@g10code.com>
* agent/pksign.c (agent_pksign_do): Use public key smartcard.
--
Since gcry_pk_verify can't handle shadowed private key, public
key SEXP should be prepared for smartcard.
* agent/genkey.c (agent_ask_new_passphrase): Use a separate malloc for
PI2. Check return value of the malloc function.
* agent/command-ssh.c (ssh_identity_register): Use a separate malloc
for PI2. Wipe PI2.
--
For whatever stupid reasons I once allocated only one memory area and
split that into PI and PI2. This is actually a common pattern with
malloc but here we used a made up object size and do not take the
extra alignment required into account. One of these not yet hit by
a (sig)bus PC/VAX hacker bugs.
Instead of trying to fix the alignment, it is better to use a second
calloc for the second struct.
GnuPG-bug-id: 2112
Signed-off-by: Werner Koch <wk@gnupg.org>
* g10/sign.c (do_sign): Let verify signature by gpg-agent.
* agent/pksign.c (agent_pksign_do): Call gcry_pk_verify for RSA.
--
RSA signature verification should be done to prevent attacks against
RSA CRT implementations and not to return invalid signature to
adversary. Newer libgcrypt does so. For older libgcrypt and
smartcards, gpg-agent does signature verification.
* agent/cvt-openpgp.c (convert_from_openpgp_main): Only encrypted
parameters are stored as opaque.
(apply_protection): ARRAY members are all normal, non-opaque MPI.
(extract_private_key): Get public key as normal, non-opaque MPI.
Remove support of ECC key with '(flags param)'.
Remove support of "ecdsa" and "ecdh" keys of our experiment.
* agent/command.c (cmd_get_passphrase): Don't repeat passphrase for
pinentry loopback mode.
* agent/genkey.c (check_passphrase_constraints): Immediately return when
pinentry mode is loopback.
--
Fixes endless loop when inquiring a passphrase with
pinentry-mode=loopback that may not satisfy passphrase contraints.
* agent/cvt-openpgp.c (get_keygrip): Handle Curve25519.
(convert_secret_key, convert_transfer_key): Ditto.
* common/openpgp-oid.c (oidtable): Add Curve25519.
(oid_crv25519, openpgp_oid_is_crv25519): New.
* common/util.h (openpgp_oid_is_crv25519): New.
* g10/ecdh.c (pk_ecdh_encrypt_with_shared_point): Handle the case
with Montgomery curve which uses x-only coordinate.
* g10/keygen.c (gen_ecc): Handle Curve25519.
(ask_curve): Change the API and second arg is to return subkey algo.
(generate_keypair, generate_subkeypair): Follow chage of ask_curve.
* g10/keyid.c (keygrip_from_pk): Handle Curve25519.
* g10/pkglue.c (pk_encrypt): Handle Curve25519.
* g10/pubkey-enc.c (get_it): Handle the case with Montgomery curve.
* scd/app-openpgp.c (ECC_FLAG_DJB_TWEAK): New.
(send_key_attr): Work with general ECC, Ed25519, and Curve25519.
(get_public_key): Likewise.
(ecc_writekey): Handle flag_djb_tweak.
--
When libgcrypt has Curve25519, GnuPG now supports Curve25519.
* common/util.h: Provide replacement for GPGRT_ATTR_ macros when using
libgpg-error < 1.20.
* common/mischelp.h: Ditto.
* common/types.h: Ditto.
--
Given that libgpg-error is a dependency of all GnuPG related libraries
it is better to define such macros at only one place instead of having
similar macros at a lot of places. For now we need repalcement
macros, though.
