As of 2.2.17, GnuPG will refuse to accept any third-party
certifications from OpenPGP certificates pulled from the keyserver
network.
The SKS keyserver network currently has at least a dozen popular
certificates which are flooded with enough unusable third-party
certifications that they cannot be retrieved in any reasonable amount
of time.
The hkps://keys.openpgp.org keyserver installation offers HKPS,
performs cryptographic validation, and by policy does not distribute
third-party certifications anyway.
It is not distributed or federated yet, unfortunately, but it is
functional, which is more than can be said for the dying SKS pool.
And given that GnuPG is going to reject all the third-party
certifications anyway, there is no clear "web of trust" rationale for
relying on the SKS pool.
One sticking point is that keys.openpgp.org does not distribute user
IDs unless the user has proven control of the associated e-mail
address. This means that on standard upstream GnuPG, retrieving
revocations or subkey updates of those certificates will fail, because
upstream GnuPG ignores any incoming certificate without a user ID,
even if it knows a user ID in the local copy of the certificate (see
https://dev.gnupg.org/T4393).
However, we have three patches in
debian/patches/import-merge-without-userid/ that together fix that
bug.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Name Use-hkps-keys.openpgp.org-as-the-default-keyserver.patch
* dirmngr/http.c (http_session_new): when checking whether the
keyserver is the HKPS pool, check specifically against the pool name,
as ./configure might have been used to select a different default
keyserver. It makes no sense to apply Kristian's certificate
authority to anything other than the literal host
hkps.pool.sks-keyservers.net.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Name dirmngr-Only-use-SKS-pool-CA-for-SKS-pool.patch
* g10/import.c (chk_self_sigs): Set the NODE_GOOD_SELFSIG flag when we
encounter a valid revocation signature. This allows import of subkey
revocation signatures, even in the absence of a corresponding subkey
binding signature.
--
This fixes the remaining test in import-incomplete.scm.
GnuPG-Bug-id: 4393
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic import-merge-without-userid
Gbp-Pq: Name gpg-accept-subkeys-with-a-good-revocation-but-no-self-sig.patch
* g10/import.c (import_one): Accept an incoming OpenPGP certificate that
has no user id, as long as we already have a local variant of the cert
that matches the primary key.
--
This fixes two of the three broken tests in import-incomplete.scm.
GnuPG-Bug-id: 4393
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic import-merge-without-userid
Gbp-Pq: Name gpg-allow-import-of-previously-known-keys-even-without-UI.patch
This commit adds a test case that does the following, in order:
- Import of a primary key plus user id
- Check that import of a subkey works, without a user id present in the
imported key
- Check that import of a subkey revocation works, without a user id or
subkey binding signature present in the imported key
- Check that import of a primary key revocation works, without a user id
present in the imported key
--
Note that this test currently fails. The following changesets will
fix gpg so that the tests pass.
GnuPG-Bug-id: 4393
Signed-Off-By: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic import-merge-without-userid
Gbp-Pq: Name tests-add-test-cases-for-import-without-uid.patch
* g10/keygen.c (keygen_set_std_prefs): prefer SHA-512
and SHA-384 by default.
--
In 8ede3ae29a39641a2f98ad9a4cf61ea99085a892, upstream changed the
defaults for --default-preference-list to advertise a preference for
SHA-512, without touching --personal-digest-preferences. This makes
the same change for --personal-digest-preferences, since every modern
OpenPGP library supports them all.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic update-defaults
Gbp-Pq: Name gpg-Prefer-SHA-512-and-SHA-384-in-personal-digest.patch
* g10/main.h (DEFAULT_DIGEST_ALGO): Use SHA512 instead of SHA256 in
--gnupg mode (leave strict RFC and PGP modes alone).
* configure.ac: Do not allow disabling sha512.
* g10/misc.c (map_md_openpgp_to_gcry): Always support SHA512.
--
SHA512 is more performant on most 64-bit platforms than SHA256, and
offers a better security margin. It is also widely implemented.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic update-defaults
Gbp-Pq: Name gpg-Default-to-SHA-512-for-all-signature-types-on-RS.patch
* g10/main.h (DEFAULT_CIPHER_ALGO): Prefer AES256 by default.
--
It's 2017, and pretty much everyone has AES-256 available. Symmetric
crypto is also rarely the bottleneck (asymmetric crypto is much more
expensive). AES-256 provides some level of protection against
large-scale decryption efforts, and longer key lengths provide a hedge
against unforseen cryptanalysis.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
(cherry picked from commit 73ff075204df09db5248170a049f06498cdbb7aa)
Gbp-Pq: Topic from-master
Gbp-Pq: Name gpg-default-to-AES-256.patch
* agent/command.c (hlp_genkey): update help text to suggest the use of
3072 bits.
* doc/wks.texi: Make example match default generation.
* g10/keygen.c (DEFAULT_STD_KEY_PARAM): update to
rsa3072/cert,sign+rsa3072/encr, and fix neighboring comment,
(gen_rsa, get_keysize_range): update default from 2048 to 3072).
* g10/keyid.c (pubkey_string): update comment so that first example
is the default 3072-bit RSA.
--
3072-bit RSA is widely considered to be 128-bit-equivalent security.
This is a sensible default in 2017.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
(cherry picked from commit 909fbca19678e6e36968607e8a2348381da39d8c)
Gbp-Pq: Topic from-master
Gbp-Pq: Name gpg-default-to-3072-bit-RSA-keys.patch
* agent/gpg-agent.c (handle_connections): When inotify is working, we
do not need to schedule a timer to evaluate whether we control our own
socket or not.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic gpg-agent-idling
Gbp-Pq: Name agent-Avoid-scheduled-checks-on-socket-when-inotify-.patch
* agent/gpg-agent.c (need_tick): Evaluate whether the short-phase
handle_tick() is needed.
(handle_connections): On each cycle of the select loop, adjust whether
we should call handle_tick() or not.
(start_connection_thread_ssh, do_start_connection_thread): Signal the
main loop when the child terminates.
* agent/call-scd.c (start_scd): Call interrupt_main_thread_loop() once
the scdaemon thread context has started up.
--
With this change, an idle gpg-agent that has no scdaemon running only
wakes up once a minute (to check_own_socket).
Thanks to Ian Jackson and NIIBE Yutaka who helped me improve some of
the blocking and corner cases.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic gpg-agent-idling
Gbp-Pq: Name agent-Avoid-tight-timer-tick-when-possible.patch
* agent/gpg-agent.c (interrupt_main_thread_loop): New function on
non-windows platforms, allows other threads to interrupt the main loop
if there's something that the main loop might be interested in.
--
For example, the main loop might be interested in changes in program
state that affect the timers it expects to see.
I don't know how to do this on Windows platforms, but i welcome any
proposed improvements.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic gpg-agent-idling
Gbp-Pq: Name agent-Allow-threads-to-interrupt-main-select-loop-wi.patch
agent/gpg-agent.c (handle_tick): Remove intermittent call to
check_own_socket.
(tv_is_set): Add inline helper function for readability.
(handle_connections) Create general table of pending scheduled
timeouts.
--
handle_tick() does fine-grained, rapid activity. check_own_socket()
is supposed to happen at a different interval.
Mixing the two of them makes it a requirement that one interval be a
multiple of the other, which isn't ideal if there are different delay
strategies that we might want in the future.
Creating an extensible regular timer framework in handle_connections
should make it possible to have any number of cadenced timers fire
regularly, without requiring that they happen in cadences related to
each other.
It should also make it possible to dynamically change the cadence of
any regularly-scheduled timeout.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic gpg-agent-idling
Gbp-Pq: Name agent-Create-framework-of-scheduled-timers.patch
* dirmngr/dirmngr.c (housekeeping_thread): Avoid automatically
checking upstream's software database. In Debian, software updates
should be handled by the distro mechanism, and additional upstream
checks only confuse the user.
* doc/dirmngr.texi: document that --allow-version-check does nothing.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic dirmngr-idling
Gbp-Pq: Name dirmngr-Avoid-automatically-checking-upstream-swdb.patch
* dirmngr/ks-engine-hkp.c (host_is_alive): New function. Test whether
host is alive and resurrects it if it has been dead long enough.
(select_random_host, map_host, ks_hkp_mark_host): Use host_is_alive
instead of testing hostinfo_t->dead directly.
(ks_hkp_housekeeping): Remove function, no longer needed.
* dirmngr/dirmngr.c (housekeeping_thread): Remove call to
ks_hkp_housekeeping.
--
Rather than resurrecting hosts upon scheduled resurrection times, test
whether hosts should be resurrected as they're inspected for being
dead. This removes the need for explicit housekeeping, and makes host
resurrections happen "just in time", rather than being clustered on
HOUSEKEEPING_INTERVAL seconds.
According to 392e068e9f143d41f6350345619543cbcd47380f,
dns_stuff_housekeeping only works on Windows, so it also isn't
necessary in debian, but it remains in place for now.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic dirmngr-idling
Gbp-Pq: Name dirmngr-Avoid-need-for-hkp-housekeeping.patch
* dirmngr/ks-engine-hkp.c (select_random_host): Use atomic pass
through the host table instead of risking out-of-bounds write.
--
Multiple threads may write to hosttable[x]->dead while
select_random_host() is running. For example, a housekeeping thread
might clear the ->dead bit on some entries, or another connection to
dirmngr might manually mark a host as alive.
If one or more hosts are resurrected between the two loops over a
given table in select_random_host(), then the allocation of tbl might
not be large enough, resulting in a write past the end of tbl on the
second loop.
This change collapses the two loops into a single loop to avoid this
discrepancy: each host's "dead" bit is now only checked once.
As Werner points out, this isn't currently strictly necessary, since
npth will not switch threads unless a blocking system call is made,
and no blocking system call is made in these two loops.
However, in a subsequent change in this series, we will call a
function in this loop, and that function may sometimes write(2), or
call other functions, which may themselves block. Keeping this as a
single-pass loop avoids the need to keep track of what might block and
what might not.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Gbp-Pq: Topic dirmngr-idling
Gbp-Pq: Name dirmngr-hkp-Avoid-potential-race-condition-when-some.patch
This avoids needing to setgid gpg-agent. It probably doesn't defend
against all possible attacks, but it defends against one specific (and
easy) one. If there are other protections we should do them too.
This will make it slightly harder to debug the agent because the
normal user won't be able to attach gdb to it directly while it runs.
The remaining options for debugging are:
* launch the agent from gdb directly
* connect gdb to a running agent as the superuser
Upstream bug: https://dev.gnupg.org/T1211
Gbp-Pq: Topic block-ptrace-on-secret-daemons
Gbp-Pq: Name Avoid-simple-memory-dumps-via-ptrace.patch
upstream ships doc/defsincdate in its tarballs. but doc/Makefile.am
tries to rewrite doc/defsincdate if it notices that any of the files
have been modified more recently, and it does so assuming that we're
running from a git repo.
However, we'd rather ship the documents cleanly without regenerating
defsincdate -- we don't have a git repo available (debian builds from
upstream tarballs) and any changes to the texinfo files (e.g. from
debian/patches/) might result in different dates on the files than we
expect after they're applied by dpkg or quilt or whatever, which makes
the datestamp unreproducible.
Gbp-Pq: Topic debian-packaging
Gbp-Pq: Name avoid-regenerating-defsincdate-use-shipped-file.patch
avoid self-describing as a beta
Using autoreconf against the source as distributed in tarball form
invariably results in a package that thinks it's a "beta" package,
which produces the "THIS IS A DEVELOPMENT VERSION" warning string.
since we use dh_autoreconf, i need this patch to avoid producing
builds that announce themselves as DEVELOPMENT VERSIONs.
See discussion at:
http://lists.gnupg.org/pipermail/gnupg-devel/2014-November/029065.html
Gbp-Pq: Topic debian-packaging
Gbp-Pq: Name avoid-beta-warning.patch
Daniel Kahn Gillmor