* dirmngr/ks-engine-hkp.c (send_request): Reinitialize HTTP session when
following a HTTP redirection.
--
inspired by patch from Damien Goutte-Gattat <dgouttegattat@incenp.org>
GnuPG-Bug_id: 4566
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
* dirmngr/ks-engine-hkp.c (SEND_REQUEST_EXTRA_RETRIES): New.
(handle_send_request_error): Use it for 503 and 504.
(ks_hkp_search, ks_hkp_get, ks_hkp_put): Pass a new var for
extra_tries.
--
This is a pretty stupid fix but one which works without much risk of
regressions. We could have used the existing TRIES but in that case
the fallback to other host would have been too limited. With the used
value we can have several fallbacks to other hosts. Note that the
TRIES is still cumulative and not per host.
GnuPG-bug-id: 4600
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 8b113bb148f273524682252233b3c65954e1419e)
* dirmngr/http.c (same_host_p): Consider certain subdomains to be the
same.
--
GnuPG-bug-id: 4603
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 37f0c55c7be3fc4912237f2bc72466aef6f8aa36)
* dirmngr/ocsp.c (do_ocsp_request): Remove arg md. Add args r_sigval,
r_produced_at, and r_md. Get the hash algo from the signature and
create the context here.
(check_signature): Allow any hash algo. Print a diagnostic if the
signature does not verify.
--
GnuPG-bug-id: 3966
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/domaininfo.c (struct domaininfo_s): Add field keepmark.
(insert_or_update): Implement new update algorithm.
--
The old algorithm limited the length of a bucket chain by purging the
last 50% or the entries. Thus the first domains entered into the
cache were never purged. The new algorithm is a bit better: It also
limits the chain length on overflow to 50% but tries to keep the
entries indicating that a WKD is available in the cache. If there is
still space to keep more, those which clearly do not support WKD are
also kept.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/ks-engine-hkp.c (send_request): New case for 413.
* dirmngr/ks-engine-http.c (ks_http_fetch): Ditto.
* dirmngr/ocsp.c (do_ocsp_request): Ditto.
--
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns.h (dns_quietinit): Remove.
(dns_hints_i_new): Remove.
--
Even before our change, dns_quietinit was questionable macro; There
was no place in dns.c which requires overrides in initializer list.
Only redundant zero were.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
(cherry picked from commit 371ae25f8f6f2d1ac030bf984bca479393a5ed43)
* dirmngr/dns.h (DNS_OPTS_INIT, dns_opts): Remove.
* dirmngr/dns-stuff.c (libdns_res_open): Use zero-ed, and initialized
automatic variable for opts.
* dirmngr/dns.c (send_query, resolve_query, resolve_addrinfo):
Likewise.
--
In fact, DNS_OPTS_INIT was only needed when args are none. With
partially specified initialization, C99 guarantees zero-ed other
members just like static object.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
(cherry picked from commit d661acd483236d34720a4959fc816d05f89c2cb7)
* dirmngr/dns.h (dns_rr_foreach): Don't use dns_rr_i_new.
Call dns_rr_grep with NULL.
* dirmngr/dns.c (dns_rr_grep): Support NULL for error_.
--
Here we still use C99 feature of struct member initialization in
dns_rr_foreach, for struct dns_rr_i. Note that in C99, it guarantees
non-specified member fields are initialized by zero. So, there's no
need to use dns_rr_i_new at all.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
(cherry picked from commit a1ccfe2b37847cce0db2fb94a7365c9fa501eda4)
* dirmngr/dns.c (dns_inet_pton, dns_so_tcp_keep): Use automatic
variables.
(dns_poll, dns_send_nopipe): Likewise, adding const qualifier.
--
Compound literals is a feature of C99. Because we only use C90 plus
some limited features, in the project, it's better to avoid it.
Besides, we make sure when it's read-only.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
(cherry picked from commit 1c405499388fd5bed0968ab5c6c5d1b3373537b9)
* dirmngr/http.c (same_host_p): Add exception table.
--
Please: Adding entries to this table shall be an exception and not the
rule.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 134c3c16523b1a267ebdd2df6339240fd9e1e3b3)
* dirmngr/dns.c: Include gpgrt.h. Silence -Warray-bounds also gcc.
* tests/gpgscm/scheme.c: Include gpgrt.h.
(Eval_Cycle): Ignore -Wimplicit-fallthrough.
--
The funny use of case and labels in the CASE macro seems confuse the
fallthrough detection.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/ks-engine-hkp.c (handle_send_request_error): Add arg
http_status and handle it.
(ks_hkp_search): Get http_status froms end_request and pass on to
handle_send_request_error.
(ks_hkp_get): Ditto.
(ks_hkp_put): Ditto.
--
GnuPG-bug-id: 4175
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 05ef6282784495a77f4faf76c0de5bc85dfecf06)
* dirmngr/http.c (http_status2string): New.
--
Right now only the standard 5xx codes.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit dc61f4ecea5c9815cb00aeb25439978337c1fd64)
* dirmngr/http.h (parsed_uri_s): Add fields off_host and off_path.
(http_redir_info_t): New.
* dirmngr/http.c (do_parse_uri): Set new fields.
(same_host_p): New.
(http_prepare_redirect): New.
* dirmngr/t-http-basic.c: New test.
* dirmngr/ks-engine-hkp.c (send_request): Use http_prepare_redirect
instead of the open code.
* dirmngr/ks-engine-http.c (ks_http_fetch): Ditto.
--
With this change a http query will not follow a redirect unless the
Location header gives the same host. If the host is different only
the host and port is taken from the Location header and the original
path and query parts are kept.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit fa1b1eaa4241ff3f0634c8bdf8591cbc7c464144)
Summary:
* dirmngr/crlcache.c (crl_cache_flush): Also deinit the cache.
* dirmngr/server.c (hlp_flushcrls, cmd_flushcrls): New.
(register_commands): Add FLUSHCRLS.
--
This allows it to flush the CRL cache of a running dirmngr
server. This can be useful to debug / analyze CRL issues.
GnuPG-Bug-Id: T3967
Differential Revision: https://dev.gnupg.org/D469
Signed-off-by: Andre Heinecke <aheinecke@intevation.de>
* dirmngr/ks-engine-hkp.c (ks_hkp_search): Send SOURCE status also on
NO DATA error.
(ks_hkp_get): Ditto.
* g10/call-dirmngr.c (gpg_dirmngr_ks_search): Print "data source" info
also on error.
(gpg_dirmngr_ks_get): Ditto.
--
If a keyserver does not return any data it can be useful to know which
keyserver out of the pool answered.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit bee65edfbc8cc2c369e5941cc9d1a01a0519b388)
* dirmngr/misc.c (host_and_port_from_url): Fix bad port parsing and a
segv for a missing slash after the host name.
--
Reportted-by: Tomas Mraz
GnuPG-bug-id: 4230
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit a3a5a2451924640588e5ecc03a1d4ba6a6ba94a5)
* dirmngr/ks-engine-ldap.c (extract_attributes): Don't use a variabale
out of scope and cleanup the entire pgpKeySize block.
--
GnuPG-bug-id: 4229
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 2b57a8159cdc3b212a4efc68787b40cafcd91ebe)
* dirmngr/dirmngr.c (parse_ldapserver_file): Only print info
for ENOENT.
--
It is very common that there is no ldapserver file. In that
case users are confused by an error printed at dirmngr startup.
* dirmngr/server.c (proc_wkd_get): Tack the raw local address to the
request.
--
We append the raw non-canonicalized local address part to the hash.
Servers who serve the requests from static files will ignore the
parameters and a test with posteo shows that also services using a
database ignore the parameter. The general idea is that service
providers may use their own canonicalization rules. The problem is
that we currently filter the returned key for the full mail address
and thus we will never see a key if the service did a different
canonicalization than we. So consider this to be an experiment.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 256a280c51f9ea862e4bfb0bb530c2a96f9088f9)
* common/asshelp.c (wait_for_sock): Round SECSLEFT.
* dirmngr/dirmngr.c (main): Take care of --debug-wait also in dameon
mode.
* common/sysutils.c (gnupg_usleep) [HAVE_NANOSLEEP]: Fix nanosleep use.
--
Without the rounding we saw in verbose mose
[...]to come up ... (5s)
[...]to come up ... (4s)
immediately without the expected one second delay. Waiting for the
next seconds did not work if nanosleep was used due to improper passed
parameters in gnupg_usleep.
Adding --debug-wait for dirmngr in daemon mode is required to test
this change.
GnuPG-bug-id: 3490
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 0cfdd3b57d592fb6baa7dafe8fde124e8a6c7798)
Fixes-commit: 1189df2cd7d4b6896ba22aa204c159ff2a425ead
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/server.c (proc_wkd_get): Check the returned SRV record names
to mitigate rogue DNS servers.
--
I am not sure wether this really is very useful because the security
relies on a trustworthy DNS system anyway. However, that check is
easy enough to do.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit ebe727ef596eefebb5eff7d03a98649ffc7ae3ee)
* dirmngr/dns.c (dns_res_exec): If it's DNS_SO_SOCKS_CONN, don't
iterate to other server, but return the error immediately.
--
Cherry picked from master commit:
1aacd12471935a354cfd85ee1805edc7eb16e6c5
In the function libdns_switch_port_p in dns-stuff.c, this patch
allows to fallback using TOR_PORT2 correctly.
Fixes-commit: bcdbf8b8ebe9d61160e0b007dabe1b6462ffbc93
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* dirmngr/dns.c (LEAVE_SELECTION_OF_PORT_TO_KERNEL): New.
(dns_socket): Don't select ephemeral port in user space.
--
Cherry picked from master commit:
861f1da0731bf29dcb9221c4f22c76b40ec15a78
There is no good reason to bind local port aggressively. It might be
some reason to do so, then, a user can specify it in /etc/resolv.conf
by the second argument of "interface" directive.
At least, it causes a problem on Windows. Binding a specified port in
user space can trigger the Firewall dialog on Windows. Since it can
be considered valid question, it is better not to bind with an
ephemeral port which is selected in user space, by default.
GnuPG-bug-id: 3610
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* dirmngr/dns.c (dns_so_check): When EINVAL, release the association
by connect with AF_UNSPEC and try again. Also try again for
ECONNREFUSED.
(dns_res_exec): Try next nameserver when ECONNREFUSED.
--
Cherry pick from master commit:
bcdbf8b8ebe9d61160e0b007dabe1b6462ffbc93
GnuPG-bug-id: T3374
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* dirmngr/dns.c (dns_resconf_pton): Clear SS.
(dns_resconf_setiface): Clear ->IFACE.
(dns_hints_root, send_query): Clear SS.
--
Cherry pick from master commit:
1c0b6681e4f322b88ac35d1f21c03d3cfc35fc23
POSIX requires clear the structure of struct sockaddr_in6. On macOS,
in some case like bind, it is better to clear even for struct
sockaddr_in.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* dirmngr/dns-stuff.c (libdns_init): Initialize options.recurse.
--
Cherry pick from master commit:
5b40338f12762cd74238c2d2b3101c33dd2d0ed3
To reproduce an error, run:
./t-dns-stuff --debug --recursive-resolver www.gnupg.org
Then, it returns "No name" error. That's because there was only setup
for root servers, and no setup for recursive query in fact.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* dirmngr/ldap-wrapper.c (wrapper_list): Rename to reaper_list.
(ldap_reaper_thread): Protect all list modification with a mutex. Use
a condition var to wake up the reaper thread.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/ldap-wrapper.c: Revamp module to make use of es_poll for
portability.
* configure.ac: Always use the ldap wrapper.
--
Since the migration from GNU Pth to nPth the ldap wrapper never worked
reliable on Windows. Our long term use of the old Window CE wrapper
thing didn't fixed this either. The new code uses the portable
es_poll function and thus code which is tested at several other
places. It Should(tm) fix the Windows issues.
GnuPG-bug-id: 3937
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dirmngr_ldap.c: Remove assert.h.
(main): Replace assert by log_assert.
* dirmngr/ldap.c (run_ldap_wrapper): Use debug options to pass
verbose options to dirmngr_ldap.
(start_cert_fetch_ldap): Ditto.
--
verbose is a pretty common option in dirmngr.conf and it would clutter
the logs with output from dirmngr_ldap. Now we require DBG_EXTPROG
or DBG_LOOKUP to make dirmngr_ldap more verbose.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/ks-engine-hkp.c (RESURRECT_INTERVAL): Decrease.
(INITIAL_HOSTTABLE_SIZE): Increase because the old values was likely
for development.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/ks-engine-hkp.c (map_host): Don't use the cname for HTTPHOST.
* dirmngr/server.c (make_keyserver_item): Map keys.gnupg.net.
--
For a description of the problem see the comment in
make_keyserver_item.
GnuPG-bug-id: 3755
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/http.c (send_request): Print the used TLS library in debug
mode.
--
We allow two different TLS libararies and thus it is useful to see
that in the debug output of bug reports.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/ks-engine.h (KS_HTTP_FETCH_NOCACHE): New flag.
(KS_HTTP_FETCH_TRUST_CFG): Ditto.
(KS_HTTP_FETCH_NO_CRL): Ditto.
(KS_HTTP_FETCH_ALLOW_DOWNGRADE): Ditto.
* dirmngr/ks-engine-http.c (ks_http_fetch): Replace args send_no_cache
and extra_http_trust_flags by a new flags arg. Allow redirectiong
from https to http it KS_HTTP_FETCH_ALLOW_DOWNGRADE is set.
* dirmngr/loadswdb.c (fetch_file): Call with KS_HTTP_FETCH_NOCACHE.
* dirmngr/ks-action.c (ks_action_get): Ditto.
(ks_action_fetch): Ditto.
* dirmngr/crlfetch.c (crl_fetch): Call with the appropriate flags.
--
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/http.h (HTTP_FLAG_TRUST_CFG): New flag.
* dirmngr/http.c (http_register_cfg_ca): New.
(http_session_new) [HTTP_USE_GNUTLS]: Implement new trust flag.
* dirmngr/certcache.c (load_certs_from_dir): Call new function.
(cert_cache_deinit): Ditto.
* dirmngr/http-ntbtls.c (gnupg_http_tls_verify_cb): Ditto.
* dirmngr/ks-engine-http.c (ks_http_fetch): Add new args
'send_no_cache' and 'extra_http_trust_flags'. Change all callers to
provide the default value.
* dirmngr/crlfetch.c (crl_fetch): Rewrite to make use of
ks_http_fetch.
--
The old code simply did not use https for downloading of CRLS.
Instead it rewrote https to http under the assumption that the CRL
service was also available without encryption. Note that a CRL is
self-standing and thus it does not need to have extra authenticity as
provided by TLS. These days we should not use any unencrypted content
and thus this patch.
Be aware that cacert.org give a https CRL DP but that currently
redirects to to http! This is a downgrade attack which we detect and
don't allow. The outcome is that it is right now not possible to use
CAcert certificates.
Signed-off-by: Werner Koch <wk@gnupg.org>
