The problem is that a user may unintentionally use the suggested filename without checking that it does not harm to write to this file. It is better not to present a default name at all.

The revision was actually applied (rG0947a20c28cf: gpgsm: Fix output of card serial number in colon listing.), but Phabricator doesn't allow me to set it as applied because not all reviewers have approved it.

Fix applied. Thanks.

This is not a composite key specific thing despite that this is an extra challenge. The creation date is used to reconstruct a key if the public key has been lost and only the fingerprint is still available. A solution might be to test the all combinations of stored creation dates to match the fingerprint.

I can confirm that the patch fixes the issue. Thanks!

I am currently working on backup/restore of Kyber keys. The error message will go away.

There are a lot of other ways to confuse the user. We can't fix them all because the whole purpose of a cleartext signature is to make it easy to use in legacy environments like an BBS. Modern systems use MIME to handle this in a more stringent specified way. For any use it is stongly suggested to check the actual signed data which is avaialable with the --output options. At least a sanitizing viewer should be used which filters out all escape characters (something like cat -v |less).

We have seen wrong encodings quite often in the past and thus we won't apply the patch. After all the armor header is a different layer and could also be applied or removed by other software or tools. The integrity of an OpenPGP message does not depend on its concrete outer encoding.

For someone who considers the rejection is the requirement of the spec., here is a change:

Here is a revised patch (hopefully the last one):

This is caused by a bug in gpgsm. The card serial numbers are missing in the with-colon key listing.

That's my mistake with KEM API.

Alright, I change it from for notation data (and name).

[GNUPG:] NOTATION_NAME foo@foo.org
[GNUPG:] NOTATION_FLAGS 0 1
[GNUPG:] NOTATION_DATA bla%20bla%20��%20blub

with change:

[GNUPG:] NOTATION_NAME foo@foo.org
[GNUPG:] NOTATION_FLAGS 0 1
[GNUPG:] NOTATION_DATA bla%20bla%20%81%82%20blub

Since rfc2440 the PGP specs say:

Here is a patch (revised since the email of mine):

For t-stringhelp.c, here is a fix.

For

openpgp_oid_from_str has a wrong memory-safety assumption

For Integer Overflow in literals_seen, I think that the claim is wrong.
When literals_seen becomes two, GnuPG stops processing to return an error.
It's not possible for GnuPG to keep processing.

For Integer Truncation of Arguments report, since the int value is limited to one-octet value, we could add check for the length of input when we use strtol.

We have fixed it but the commit also states:

Added a compatibility flag in: rGad0c6c33c3d6: gpg: Do not use a default when asking for another output filename.

Fixed in rG115d138ba599: gpg: Fix possible memory corruption in the armor parser.

Fixed in rGdb9705ef594d: gpg: Avoid potential downgrade to SHA1 in 3rd party key signatures.

Fixed in rG8abc320f2a75: gpg: Error out on unverified output for non-detached signatures.

Mitigation would be: adding context validation in add_onepass_sig function, which checks WHAT of armored input.

That's a good question. Looking at https://datatracker.ietf.org/doc/draft-koch-librepgp/, it doesn't really specify what encoding is used for "human-readable" notation, so I'd personally lean towards encoding it to stay on the safe side. Unless I'm mistaken, status-fd will only be used locally, so escaping overhead should not be a problem.

The question is who shall correct the wrong encoding of notation data (assuming it is flagged as human readable). Escaping is a solution but needs a lot of extra bytes.

It is not an ADSK issue. The problem is that the new subkey has not been entered into the fingerprint table and can thus not be found.

That's what gpg-card url --clear does

if (!strcmp (argstr, "--clear"))
  url = xstrdup (" "); /* No real way to clear; set to space instead. */

Fixed in 2.5.13.

So we need to find out what gpg-card url --clear does to avoid the card error for the ZeitControl cards.

In gpg4win-4.4.1 it works too.

Note: In the current vsd beta (29) it works (pinentry for the next key is opened):

@werner Proposed patch for gpg:

diff --git a/g10/export.c b/g10/export.c
index 5dcb9c665..908a6b6a0 100644
--- a/g10/export.c
+++ b/g10/export.c
@@ -1961,7 +1961,9 @@ do_export_one_keyblock (ctrl_t ctrl, kbnode_t keyblock, u32 *keyid,
           if (strchr (hexgrip, ','))
             {
               log_error ("exporting a secret dual key is not yet supported\n");
-              return gpg_error (GPG_ERR_NOT_IMPLEMENTED);
+              err = gpg_error (GPG_ERR_NOT_IMPLEMENTED);
+              write_status_error ("export_keys.secret", err);
+              return err;
             }

Note: It works with gpg-card url --clear.

I could reproduce this with a ZeitControl OpenPGP v3.4 card, but (as Tobias) not with an (old) Yubikey. Looks like a bug in the card firmware.

This issue should be fixed in 2.6, too.

Pushed the change to gnupg master: rG61ff3759e827: common,dirmngr:w32: Fix for semi-hosted environment.

In libgpg-error, I pushed thread-safe version : rE0313b660f8bd: w32: Don't convert slash->backslash when it's under Wine.
I'm going to push similar code to gnupg master.

For completeness, that's https://gitlab.freedesktop.org/poppler/poppler/-/issues/1595. dkg obviously filed that but it may be useful for others finding themselves here.

I'm fixing this issue under T7855. So, I move this ticket as a child of T7855.

@timegrid Thank you for your confirmation.

I can't reproduce this in vsd-3.3.90.19 @ win10 anymore.
Probably the fixes in https://dev.gnupg.org/T7827 or https://dev.gnupg.org/T7855 solved this, too.

The problem here is that iobuf_readbyte returns -1 on error and on EOF. parse_packet is not able to distinguish that because for histroic reasons we do not return a gpg-error code (GPG_ERR_EOF). To fix this we need to change all callers of parse_packet to not act upon -1 but only on an error code.