* dirmngr/dns-stuff.c (resolve_dns_name): Move up in the file.
(resolve_addr_libdns): New.
(resolve_dns_addr): Divert to resolve_dns_addr.
--
In the old code reverse lookups where disabled in Tor mode. By
implementing the reverse lookups via libdns it is now possible to do
them also in Tor mode.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (resolve_name_libdns): USe flags AI_NUMERICHOST
for literal IP addresses.
(resolve_name_standard): Ditto.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (is_ip_address): Make the return value depend on
the address family.
* dirmngr/ks-engine-hkp.c (map_host): Rename arg R_POOLNAME to
R_HTTPHOST because that is its purpose. Note that the former
behaviour of storing a NULL to indicate that it is not a pool has not
been used.
(make_host_part): Ditto.
(make_host_part): Make sure that literal v6 addresses are correclty
marked in the constructed URL.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns.c (EPROTO) ![EPROTO]: Define to EPROTONOSUPPORT.
--
This is the same replacement we use in Libassuan
(commit 8ab3b9273524bd344bdb90dd5d3bc8e5f53ead6e) to make it work on
OpenBSD and may other BSD based OSes.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/http.c (use_socks): New.
(my_sock_new_for_addr): New.
(connect_server): Replace assuan_sock_new by my_sock_new_for_addr.
--
Libassuan always uses 127.0.0.1 to connect to the local Tor proxy.
https.c used to create a socket for the actual address family and thus
the connect call in Libassuan fails when it tries to connect to a v6
address using a v4 socket.
It would be cleaner to have the my_sock_new_for_addr function as a
public interface in Libassuan; for now we need to duplicate some code.
from Libassuan.
GnuPG-bug-id: 2902
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/http.c (opt_verbose, opt_debug): New vars.
(http_set_verbose): New function.
(_my_socket_new): Add debug output.
(_my_socket_ref, _my_socket_unref, session_unref): Call log_debug if
OPT_DEBUG has ben set to 2 in a debugger.
(http_session_new, http_session_ref): Ditto.
(send_request, http_start_data): Print debug output for the request.
(parse_response): Change to use log_debug_string for the response.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/ks-engine-hkp.c (map_host): Chnage arg NO_SRV to SRVTAG.
(make_host_part): Rewrite.
--
This fixes a regression from 2.0 and 1.4 where these tags have been in
used since 2009. For whatever reason this was not ported to 2.1 and
"hkp" was always used.
GnuPG-bug-id: 2451
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/http.h (parsed_uri_s): Add field EXPLICIT_PORT.
* dirmngr/http.c (do_parse_uri): That it.
* dirmngr/ks-engine-hkp.c (map_host): Add arg NO_SRV.
(make_host_part): Ditto.
(ks_hkp_resolve): Set NO_SRV from EXPLICIT_PORT.
(ks_hkp_search): Ditto.
(ks_hkp_get): Ditto.
(ks_hkp_put): Ditto.
--
This implements the behaviour of the keyserver helpers from 1.4 and
2.0.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/server.c (cmd_wkd_get): Support SRV records.
--
This patch changes the way a WKD query is done. Now we first look for
a SRV record for service "openpgpkey" and port "tcp" under the
to-be-queried domain. If such a record was found and the target host
matches the to-be-queried domain or is a suffix to that domain, that
target host is used instead of the domain name. The SRV record also
allows to change the port and obviously can be used for
load-balancing.
For example a query for the submission address of example.org with the
SRV record specification
_openpgpkey._tcp IN SRV 0 0 0 wkd.foo.org.
IN SRV 0 0 0 wkd.example.net.
IN SRV 0 0 4711 wkd.example.org.
(queried using the name "_openpgpkey._tcp.example.org") would fetch
from this URL:
https://wkd.example.org:4711/.well-known/openpgpkey/submission-address
Note that the first two SRV records won't be used because foo.org and
example.net do not match example.org. We require that the target host
is identical to the domain or be a subdomain of it. This is so that
an attacker modifying the SRV records needs to setup a server in a
sub-domain of the actual domain and can't use an arbitrary domain.
Whether this is a sufficient requirement is not clear and needs
further discussion.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (get_dns_srv): Add args SERVICE and PROTO.
* dirmngr/http.c (connect_server): Simplify SRV lookup.
* dirmngr/ks-engine-hkp.c (map_host): Ditto.
* dirmngr/t-dns-stuff.c (main): Adjust for changed get_dns_srv.
--
This new API is more convenient because it includes commonly used
code. Note that right now http.c's SRV record code is not used.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (enable_dns_tormode): Always succeed.
(reload_dns_stuff): Reset tor port.
* dirmngr/dirmngr.c (set_tor_mode): Also enable Tor mode for DNS.
(main): Remove warning that Tor mode may not fully work.
* dirmngr/server.c (cmd_dns_cert): Remove explicit Tor for DNS
initialization.
* dirmngr/t-dns-stuff.c (main): Remove option --new-circuit and error
checking for enable_dns_tormode.
--
This patch also resets the port on SIGHUP so that after starting Tor
SIGHUP is sufficient to use Tor. Without the SIGHUP and when not
using the Tor browser Dirmngr would keep on trying the Tor browser
port.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (libdns_init): Add debug level diagnostic on
success.
--
This output may help to avoid questions when evaluating an Assuan log.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (DEFAULT_TIMEOUT): New.
(opt_timeout): New var.
(set_dns_timeout): New.
(libdns_res_open): Set the default timeout.
(libdns_res_wait): Use configurable timeout.
(resolve_name_libdns): Ditto.
* dirmngr/dirmngr.c (oResolverTimeout): New const.
(opts): New option --resolver-timeout.
(parse_rereadable_options): Set that option.
(main) <aGPGConfList>: Add --nameserver and --resolver-timeout.
* tools/gpgconf-comp.c (gc_options_dirmngr): Add --resolver-timeout
and --nameserver.
* dirmngr/http.c (connect_server): Fix yesterday introduced bug in
error diagnostic.
--
This timeout is a pretty crude thing because libdns has a few other
internal timeouts as well.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (opt_debug, opt_verbose): New vars.
(set_dns_verbose): New func.
(libdns_switch_port_p): Add debug output.
(resolve_dns_name): Ditto.
(get_dns_cert): Ditto.
(get_dns_cname): Ditto.
(getsrv_libdns, getsrv_standard): Change SRVCOUNT to an unsigend int.
(getsrv): Rename to ...
((get_dns_srv): this. Add arg R_COUNT and return an error. Add debug
output.
* dirmngr/http.c: Adjust for chnaged getsrv().
* dirmngr/ks-engine-hkp.c (map_host): Ditto.
* dirmngr/t-dns-stuff.c (main): Ditto. Call set_dns_verbose.
* dirmngr/dirmngr.c (parse_rereadable_options): Call set_dns_verbose.
--
Due to our switch to Libdns getsrv didn't worked correctly because it
returned -1 for an NXDOMAIN. However, it is perfectly okay to have no
SRV record and thus we change the way this function is called to be
aligned with the other functions and also map NXDOMAIN to a zero SRV
record count.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns.c (FD_SETSIZE): Bump up to 1024.
(dns_poll): Return an error instead of hitting an assertion failure.
--
For unknown reasons socket() return fd with values 244, 252, 268. The
latter is above the FD_SETSIZE of 256. It seems that select has been
build with a highler FD_SETSIZE limit. Bump up to a reasonable large
value.
A better solution would be to grab some code from npth_eselect to
replace select. We could also use npth_eselect direclty in
dns-stuff.c instead of using dns_res_poll.
Signed-off-by: Werner Koch <wk@gnupg.org>
* configure.ac (DNSLIB) {W32]: Add -liphlpapi.
* dirmngr/dns-stuff.c [W32]: Include iphlpapi.h and define
WIN32_LEAN_AND_MEAN.
(libdns_init) [W32]: Use GetNetworkParams to get the nameserver.
* dirmngr/t-dns-stuff.c (init_sockets): New.
(main): Call it.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (libdns_tor_port): New var.
(set_dns_nameserver): Clear that var.
(libdns_init): Init var to the default port.
(libdns_switch_port_p): New func.
(resolve_dns_name): Use function to switch the port
(get_dns_cert): Ditto.
(getsrv): Ditto.
(get_dns_cname): Ditto.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (libdns_reinit_pending): New var.
(enable_recursive_resolver): Set var.
(set_dns_nameserver): Ditto.
(libdns_init): Avoid double initialization.
(libdns_deinit): New.
(reload_dns_stuff): New.
(libdns_res_open): Act upon LIBDNS_REINIT_PENDING.
* dirmngr/t-dns-stuff.c (main): Call reload_dns_stuff to release
memory.
* dirmngr/dirmngr.c (cleanup): Ditto.
(dirmngr_sighup_action): Call reload_dns_stuff to set
LIBDNS_REINIT_PENDING.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (tor_credentials): Replace by ...
(tor_socks_user, tor_socks_password): new vars.
(enable_dns_tormode): Set these new vars.
(libdns_res_open): Tell libdns the socks credentials.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (libdns_res_open): New. Replace all libdns_init
and dns-res_open by a call to this func.
(libdns_res_submit): New wrapper. Replace all dns_res_sumbit calls.
(libdns_res_wait): New function.
(resolve_name_libdns): Replace loop by libdns_res_wait.
(get_dns_cert_libdns): Ditto.
(getsrv_libdns): Ditto.
Signed-off-by: Werner Koch <wk@gnupg.org>
* configure.ac: Add option --disable-libdns
(USE_LIBDNS): New ac_subst and am_conditional.
(USE_C99_CFLAGS): Set only if libdns is used.
* dirmngr/Makefile.am (dirmngr_SOURCES): Move dns.c and dns.h to ...
(dirmngr_SOURCES) [USE_LIBDNS0: here.
(t_common_src): Ditto.
* dirmngr/dirmngr.c (oRecursiveResolver): New constant.
(opts): New option "--recursive-resolver".
(parse_rereadable_options): Set option.
* dirmngr/t-dns-stuff.c (main): Add option --recursive-resolver.
* dirmngr/server.c (cmd_getinfo): Depend output of "dnsinfo" on the
new variables.
* dirmngr/dns-stuff.c: Include dns.h only if USE_DNSLIB is defined.
Also build and call dnslib functions only if USE_DNSLIB is defined.
(recursive_resolver): New var.
(enable_recursive_resolver): New func.
(recursive_resolver_p): New func.
--
In case users run into problems building GnuPG, the configure option
allows to disable that support and continue w/o Tor support using the
system resolver.
--recursive-resolver was easy enough to implement and may be useful in
some situation. It does not fully work, though.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c (dns_free): New macro.
(libdns): Move var to the top.
(libdns_error_to_gpg_error): Map error codes to the new gpg-error
codes.
(resolve_name_libdns): Restructure code.
(getsrv_libdns): New.
(get_dns_cname_libdns): New.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns-stuff.c: Include dirmngr-err.h to set the correct error
source.
(get_h_errno_as_gpg_error): New.
(get_dns_cert_libdns): Fix error code.
(getsrv_libdns): Add arg R_COUNT and return an error code.
(getsrv_standard): Ditto. Fix handling of res_query errors and
provide the correct size for the return buffer.
(getsrv): Adjust for changed worker functions.
(get_dns_cname_standard): Fix handling of res_query errors and provide
the correct size for the return buffer.
Signed-off-by: Werner Koch <wk@gnupg.org>
* configure.ac (USE_C99_CFLAGS): New ac_subst. Set to -std=gnu99 for
gcc.
* dirmngr/Makefile.am (AM_CFLAGS): Add USE_C99_CFLAGS.
(t_http_CFLAGS): Ditto.
(t_ldap_parse_uri_CFLAGS): Ditto.
(t_dns_stuff_CFLAGS): Ditto.
--
C99 and the GCC option is required for use with dns.c which makes
heavy use of C99 features. We should consider to switch GnuPG
entirely to C99 but enable gcc warnings to detect features which are
not supported by not-real-c99 compilers and which makes audits
harder (VLA etc.).
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dns.c: New file.
* dirmngr/dns.h: New file.
* dirmngr/Makefile.am (dirmngr_SOURCES): Add new files.
* dirmngr/dns-stuff.c: Include dns.h.xxx use libdns
(libdns): New global var for the libdns state.
(libdns_error_to_gpg_error): New.
(libdns_init): New.
(resolve_name_libdns): New.
(get_dns_cert_libdns): New stub.
(getsrv_libdns): New stub.
(get_dns_cname_libdns): New stub.
Signed-off-by: Justus Winter <justus@g10code.com>
* autogen.rc: Remove '--with-adns' argument.
* configure.ac: Remove check for ADNS.
* dirmngr/dns-stuff.c: Remove all code that uses ADNS.
* dirmngr/server.c (cmd_getinfo): Update status line.
* doc/dirmngr.texi: Do not mention ADNS.
--
We used ADNS to support queries over Tor. However, our patch to ADNS
that adds Tor support was never accepted upstream. Furthermore, there
are other shortcomings that let us to consider alternatives.
Signed-off-by: Justus Winter <justus@g10code.com>
* dirmngr/loadswdb.c: Set ERR on malloc failure.
* g10/passphrase.c (passphrase_to_dek): Initialize all fields of
HELP_S2K.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dirmngr.c (oStandardResolver): New constant.
(opts): New option --standard-resolver.
(parse_rereadable_options): Set option.
* dirmngr/dns-stuff.c: Refactor all code to support the new option.
(standard_resolver): New var.
(enable_standard_resolver, standard_resolver_p): New func.
* dirmngr/http.c (connect_server): Remove USE_DNS_SRV build
conditional.
* dirmngr/ks-engine-hkp.c (map_host): Ditto.
* dirmngr/server.c (cmd_getinfo) <dnsinfo>: Take care of new option
* configure.ac (HAVE_ADNS_IF_TORMODE): Remove var ADNSLIB. ac_define
USE_ADNS in the adns checking code. Remove options --disable-dns-srv
and --disable-dns-cert. Always look for the system resolver. Print
warning if no system resolver was found.
(USE_DNS_CERT, USE_DNS_SRV): Remove ac_defines.
(HAVE_SYSTEM_RESOLVER): New ac_define.
(USE_DNS_SRV): Remove am_conditional; not used anyway.
--
This option allows for runtime switching to the system's standard
resolver. This is mainly useful to work around possible bugs in the
optional resolver libraries (as of now ADNS). Note that on Windows
there is no code to use systems's resolver and thus for full
functionality dirmngr must be build with a separate resolver.
This patch also does way with configure options to disable the use of
CERT and SRV records.
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/gpg-agent.c (handle_connections): Add a comment.
* dirmngr/dirmngr.c (main): Move assuan_sock_close of the listening
socket to ...
(handle_connections): here. Add a comment why we keep the
listening socket open during a shutdown.
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/gpg-agent.c (handle_connections): Keep on selecting on the
inotify fd even when a shutdown is pending.
* dirmngr/dirmngr.c (handle_connections): Ditto. Also simplifyy the
use of the HAVE_INOTIFY_INIT cpp conditional.
--
Without that patch we won't notice a removed socket when a shutdown is
pending. This is somewhat related to bug report 2849.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dirmngr.c (main): Avoid calling ldap_wrapper_launch_thread()
Before we need it.
* dirmngr/ldap-wrapper.c (ldap_wrapper): Call
ldap_wrapper_launch_thread() just in time (before any attempt to use
an ldap subprocess).
--
A dirmngr process that never looks anything up in LDAP has no need for
a reaper thread, but one was started automatically. This thread wakes
up every two seconds to look for ldap processes that might never have
been running. We won't start more than one reaper thread for any
given dirmngr due to the static int "done" in
ldap_wrapper_launch_thread(), so it's safe to call this every time
there is a use of ldap_wrapper.
If someone wants to do further dirmngr optimizations for ldap users,
the reaper thread itself could use dynamically-calculated timeouts
(and probably needs to be alerted dynamically when a new ldap
subprocess is available so it can re-calculate those timeouts).
Note: It's not clear to me how to test ldap access effectively; i know
of no public ldap services that i can verify against, and i do not run
my own ldap servers. If someone has a publicly-available ldap server
that developers can run tests against, i would be happy to hear about
it.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
* dirmngr/dirmngr.c (TIMERTICK_INTERVAL): Always use 60 seconds like
we did for WindowsCE.
--
Given that the timer tick is only used for housekeeping tasks and
these are done every 10 minutes, it makes no sense to use 2 seconds.
The minor drawback is tha the housekeeping may be delayed by one
minute.
NB: For the purpose of power saving, we already make sure that the
process wakes up at the full second so that it is synchronized to the
wakeup time of other processes.
Signed-off-by: Werner Koch <wk@gnupg.org>
* dirmngr/dirmngr.c (handle_tick): Remove w32 tests for
shutdown_pending; no longer needed.
--
In d83ba4897b, we removed the
Windows-specific system daemon features, where shutdown_pending was
set from w32_service_control(). shutdown_pending is now never
assigned outside of handle_signal() or within an inotify test, neither
of which are available on w32.
As a result, this stanza in handle_tick() should be dead code, and can
be removed to keep things simple.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
s/win32/w32/ to please RMS ;-) -wk
* configure.ac: Define DIRMNGR_DEFAULT_KEYSERVER.
* dirmngr/server.c (ensure_keyserver): Use it if no keyservers are set.
* doc/dirmngr.texi: Document this behavior.
--
A user who doesn't specify a keyserver, but asks gnupg to fetch a key
currently just gets a simple error messages "No keyserver available".
If the user is asking to contact a keyserver, we should have a
reasonable default, and not require them to fiddle with settings when
they might not know what settings to choose. This patch makes the
default hkps://hkps.pool.sks-keyservers.net.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
* dirmngr/dirmngr.c (http_session_new): If the user isn't talking to
the HKPS pool, and they have not specified any hkp-cacert, then we
should default to the system CAs, rather than nothing.
* doc/dirmngr.texi: Document choice of CAs.
--
Consider three possible classes of dirmngr configuration:
a) no hkps:// keyserver URLs at all (communication with keyservers is
entirely in the clear)
b) hkps:// keyserver URLs, but no hkp-cacert directives
c) hkps:// keyserver URLs, and at least one hkp-cacert directive
class (a) provides no confidentiality of requests.
class (b) currently will never work because the server certificate
cannot be validated.
class (c) is currently supported as intended.
This patch allows users with configurations in class (b) to work as
most users expect (relying on the system certificate authorities),
without affecting users in classes (a) or (c).
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
o minor indentation fix
- wk
* dirmngr/dirmngr.c (parse_readable_options): If we're unable to turn
an argument for hkp-cacert into an absolute filename, terminate
completely.
* dirmngr/http.c (http_register_tls_ca): Show a warning if file is not
immediately accessible, but register it anyway.
--
Without this changeset, the condition of the filesystem when dirmngr
is initialized will have an effect on later activities of dirmngr.
For example, if a file identified by a hkp-cacert directive doesn't
exist when dirmngr starts, dirmngr will behave as though it simply
didn't have the hkp-cacert directive set at all, even if the file
should appear later.
dirmngr currently behaves differently if no hkp-cacert directives have
been set then it does when at least one hkp-cacert directive has been
set. For example, its choice of CA cert for
hkps://hkps.pool.sks-keyservers.net depends on whether a TLS CA file
has been registered. That behavior shouldn't additionally depend on
the state of the filesystem at the time of dirmngr launch.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
Werner Koch