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
* agent/command.c (cmd_getinfo): Return GPG_ERR_FALSE as boolean False.
* g13/server.c (cmd_getinfo): Ditto.
* sm/server.c (cmd_getinfo): Ditto.
--
GPG_ERR_FALSE was introduced with libgpg-error 1.21 and we now require
a later version for gnupg 2. Thus we can switch to this more
descriptive code.
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/command-ssh.c (ssh_key_to_protected_buffer): Update
the length by the second call of gcry_sexp_sprint.
--
GnuPG-bug-id: 4502
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* agent/findkey.c (agent_public_key_from_file): pass size_t as int to
gcry_sexp_build_array's %b.
--
This is only a problem on big-endian systems where size_t is not the
same size as an int. It was causing failures on debian's s390x,
powerpc64, and sparc64 platforms.
There may well be other failures with %b on those platforms in the
codebase, and it probably needs an audit.
Once you have a key in private-keys-v1.d/$KEYGRIP.key with a comment
or a uri of reasonable length associated with it, this fix can be
tested with:
gpg-agent --server <<<"READKEY $KEYGRIP"
On the failing platforms, the printed comment will be of length 0.
Gnupg-bug-id: 4501
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
* agent/divert-scd.c (divert_pkdecrypt): Skip a flags parameter.
--
The libgcrypt docs say that a "flags" parameter should always be used
in the input of pkdecrypt. Thus we should allow that parameter also
when parsing an s-expression to figure out the algorithm for use with
scdaemon.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit a12c3a566e2e4b10bc02976a2819070877ee895c)
* agent/call-scd.c (cancel_inquire): Remove.
(agent_card_pksign, agent_card_pkdecrypt, agent_card_writekey)
(agent_card_scd): Don't call cancel_inquire.
--
Cherry-picked master commit of:
9f5e50e7c85aa8b847d38010241ed570ac114fc3
Since libassuan 2.1.0, cancellation command "CAN" is handled within
the library, by assuan_transact. So, cancel_inquire just caused
spurious "CAN" command to scdaemon which resulted an error.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* agent/agent.h (PINENTRY_STATUS_*): Expose to public.
(struct pin_entry_info_s): Add status.
* agent/call-pinentry.c (agent_askpin): Clearing the ->status
before the loop, let the assuan_transact set ->status. When
failure with PINENTRY_STATUS_PASSWORD_FROM_CACHE, it returns
soon.
* agent/findkey.c (unprotect): Clear the pinentry cache,
when it causes an error.
--
Cherry-picked from master commit of:
02a2633a7f0b7d91aa48ea615fb3a0edfd6ed6bb
Debian-bug-id: 919856
GnuPG-bug-id: 4348
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* agent/command-ssh.c (struct ssh_key_type_spec): Add field
alt_curve_name.
(ssh_key_types): Add some alternate curve names.
(ssh_identifier_from_curve_name): Lookup also bey alternative names
and return the canonical name.
(ssh_key_to_blob): Simplify the ECDSA case by using gcry_pk_get_curve
instead of the explicit mapping.
(ssh_receive_key): Likewise. Use ssh_identifier_from_curve_name to
validate the curve name. Remove the reverse mapping because since
GnuPG-2.2 Libgcrypt 1.7 is required.
(ssh_handler_request_identities): Log an error message.
--
This change will make it easier to support other curves, in particular
those from tokens. Libgcrypt has a large list of alias names which we
now use to to make the mapping more flexible.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit d93797c8a7892fe26672c551017468e9f8099ef6)
* agent/divert-scd.c (getpin_cb): Support --ack option.
--
Cherry-picked master commit of:
827529339a4854886dbb5625238e7e01013efdcd
We are now introducing "acknowledge button" feature to scdaemon,
so that we can support OpenPGPcard User Interaction Flag.
We will (re)use the mechanism of POPUPPINPADPROMPT for this. Perhaps,
we will change the name of POPUPPINPADPROMPT, since it will be no
longer for PINPAD only.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* agent/protect.c (s2k_calibration_time): New file global var.
(calibrate_s2k_count): Use it here.
(get_calibrated_s2k_count): Replace function static var by ...
(s2k_calibrated_count): new file global var.
(set_s2k_calibration_time): New function.
* agent/gpg-agent.c (oS2KCalibration): New const.
(opts): New option --s2k-calibration.
(parse_rereadable_options): Parse that option.
--
Note that using an unrelistic high value (like 60000) takes quite some
time for calibration.
GnuPG-bug-id: 3399
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit cbcc8c19541fe8407f3b6588fce1535c64cf6b25)
* agent/command-ssh.c (ssh_handler_add_identity): Handle other errors
than EOF.
--
GnuPG-bug-id: 4221
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 7385e1babf6eef586c79ad23f8e541aaf608c4e5)
* agent/call-pinentry.c (initialize_module_call_pinentry): It's an
error when npth_mutex_init returns non-zero.
--
Cherry-pick from master commit of:
adce73b86fd49d5bbb8884231a26cc7533d400e2
Actually, initialize_module_call_pinentry is only called once from
main. So, this bug had no harm and having the static variable
INITIALIZED is not needed.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* agent/gpg-agent.c (agent_copy_startup_env): Replace explicit list
with the standard list.
--
Although the function agent_copy_startup_env is newer than
session_env_list_stdenvnames the latter was not used. When
DBUS_SESSION_BUS_ADDRESS was added to the latter it was forgotten to
add it to the former as well. Having all stdnames here seems to be
the Right Thing (tm) to do.
GnuPG-bug-id: 3947
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 7ffc1ac7dd95d4cc1897a4c36d5cd628741c12f2)
* agent/command.c (cmd_getinfo): Add sub-command getenv.
* dirmngr/server.c (cmd_getinfo): Ditto.
--
It is sometimes helpful to be able to inspect certain envvars in a
running agent. For example "http_proxy".
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/command-ssh.c (ssh_handler_sign_request): if a flag is passed
during an signature request that we do not know how to apply, return
GPG_ERR_UNKNOWN_OPTION.
--
https://tools.ietf.org/html/draft-miller-ssh-agent-02#section-4.5 says:
If the agent does not support the requested flags, or is otherwise
unable or unwilling to generate the signature (e.g. because it
doesn't have the specified key, or the user refused confirmation of a
constrained key), it must reply with a SSH_AGENT_FAILURE message.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
GnuPG-bug-id: 3880
* agent/command-ssh.c: repoint documentation reference.
--
Damien Miller is now documenting the ssh-agent protocol via the IETF.
Signed-off-by: Daniel Kahn Gillmor <dkg@fifthhorseman.net>
* agent/cache.c (agent_put_cache): Add arg 'ctrl' and change all
callers to pass it.
(agent_get_cache): Ditto.
* agent/cache.c (struct cache_items_s): Add field 'restricted'.
(housekeeping): Adjust debug output.
(agent_flush_cache): Ditto.
(agent_put_cache): Ditto. Take RESTRICTED into account.
(agent_get_cache): Ditto.
--
If requests are coming from different sources they should not share the
same cache. This way we make sure that a Pinentry pops up for a
remote request to a key we have already used locally.
GnuPG-bug-id: 3858
Signed-off-by: Werner Koch <wk@gnupg.org>
* common/shareddefs.h (request_origin_t): New.
* common/agent-opt.c (parse_request_origin): New.
(str_request_origin): New.
* agent/command.c (option_handler): Implement new option.
--
This allows to pretend that a request originated from the extra or
browser socket.
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/cache.c (agent_cache_housekeeping): New func.
* agent/gpg-agent.c (handle_tick): Call it.
--
This change mitigates the risk of having cached items in a post mortem
dump.
GnuPG-bug-id: 3829
Signed-off-by: Werner Koch <wk@gnupg.org>
* common/asshelp2.c (vprint_assuan_status_strings): New.
(print_assuan_status_strings): New.
* agent/command.c (agent_write_status): Replace by call to new
function.
* dirmngr/server.c (dirmngr_status): Ditto.
* g13/server.c (g13_status): Ditto.
* g13/sh-cmd.c (g13_status): Ditto.
* sm/server.c (gpgsm_status2): Ditto.
* scd/command.c (send_status_info): Bump up N.
--
This fixes a potential overflow if LFs are passed to the status
string functions. This is actually not the case and would be wrong
because neither the truncating in libassuan or our escaping is not the
Right Thing. In any case the functions need to be more robust and
comply to the promised interface. Thus the code has been factored out
to a helper function and N has been bumped up correctly and checked in
all cases.
For some uses this changes the behaviour in the error case (i.e. CR or
LF passed): It will now always be C-escaped and not passed to
libassuan which would truncate the line at the first LF.
Reported-by: private_pers
* configure.ac: Check ucred.h as well as sys/ucred.h.
* agent/command-ssh.c: Add inclusion of ucred.h.
--
It was T2981, adding ucred.h for Solaris. We also need sys/ucred.h
for FreeBSD and macOS.
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* agent/gpg-agent.c (oListenBacklog): New const.
(opts): New option --listen-backlog.
(listen_backlog): New var.
(main): Parse new options.
(create_server_socket): Use var instead of 5.
* dirmngr/dirmngr.c: Likewise.
* scd/scdaemon.c: Likewise.
--
GnuPG-bug-id: 3473
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/gpg-agent.c (initialize_modules): Add hook again.
(main): Remove setting of the system houk but add scoket system hook
setting after assuan initialization.
--
Thread initialization is better to be deferred after fork (in case of
UNIX). assuan_sock_init should be earlier. Thus, we need to change
system hooks for assuan_sock_* interface. Or else, on Windows, it may
cause hang on server.
Updates-commit: 1524ba9656f0205d8c6ef504f773b832a7a12ab9
GnuPG-bug-id: 3378
Signed-off-by: Werner Koch <wk@gnupg.org>
* agent/gpg-agent.c (initialize_modules): Move assuan_set_system_hooks.
(main): ... here, just before assuan_sock_init.
--
In Assuan, global variable SOCK_CTX is used internally, which is
initialized by assuan_sock_init. When initialized, system hooks
are copied into SOCK_CTX structure. Thus, system hooks should
be set, before the call of assuan_sock_init.
GnuPG-bug-id: 3378
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* agent/gpg-agent.c (oAutoExpandSecmem): New enum value.
(opts): New option --auto-expand-secmem.
(main): Implement that option.
--
Note that this option has an effect only if Libgcrypt >= 1.8.2 is
used.
GnuPG-bug-id: 3530
* agent/gpg-agent.c (handle_connections): Always check inotify fds.
--
I noticed a gpg-agent processed, probably in shutdown_pending state,
which was selecting on only these two inotify fds. The select
returned immediately but because we did not handle the fds in
shutdown_pending state they were not read and the next select call
returned one of them immediately again. Actually that should not
hanppen because the
if (active_connections == 0)
break; /* ready */
should have terminated the loop. For unknown reasons (maybe be just a
connection thread terminated in a gdb session) that did not happen.
By moving the check outside of the shutdown_pending condition and
closing the fd after they have been triggered the code should be more
robust.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 5d83eb9226c0ce608ec284d8c9bc22ce84a00c25)
* agent/protect.c (calibrate_get_time): Use clock or clock_gettime.
--
For calibration, clock(3) is better than times(3) among UNIXen.
Tested on NetBSD 7.1 and FreeBSD 11.1, using QEMU.
Thanks to Damien Goutte-Gattat for the information of use of
CLOCKS_PER_SEC; The old code with times(3) is not 100% correct,
in terms of POSIX. It should have used sysconf (_SC_CLK_TCK) instead
of CLOCKS_PER_SEC. CLOCKS_PER_SEC is specifically for clock(3).
GnuPG-bug-id: 3056, 3276, 3472
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
(cherry picked from commit 380bce13d94ff03c96e39ac1d834f382c5c730a1)
* agent/command.c (cmd_getinfo): New sub-commands.
* agent/protect.c (get_standard_s2k_count): Factor some code out to ...
(get_calibrated_s2k_count): new.
(get_standard_s2k_time): New.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit 52d41c8b0f4af6278d18d8935399ddad16a26856)
* agent/agent.h (opt): New field 's2k_count'.
* agent/gpg-agent.c (oS2KCount): New enum value.
(opts): New option --s2k-count.
(parse_rereadable_options): Set opt.s2k_count.
--
This option is useful to speed up the starting of gpg-agent and in
cases where the auto-calibration runs into problems due to a broken
time measurement facility.
Signed-off-by: Werner Koch <wk@gnupg.org>
(cherry picked from commit f7212f1d11aad5d910d2c77b2e5c6ab31a0e786e)
* agent/agent.h (struct server_control_s): Rename PINENTRY_ACTIVE.
* agent/call-pinentry.c (entry_owner): Remove.
(agent_reset_query): Use thread private object of PINENTRY_ACTIVE.
(unlock_pinentry): Add CTRL to arguments to access thread private.
Check and decrement PINENTRY_ACTIVE for recursive use.
(start_pinentry): Check and increment PINENTRY_ACTIVE for recursion.
(agent_askpin): Follow the change of unlock_pinentry API.
(agent_get_passphrase, agent_get_confirmation): Likewise.
(agent_show_message, agent_popup_message_start): Likewise.
(agent_popup_message_stop, agent_clear_passphrase): Likewise.
--
We use the member PINENTRY_ACTIVE as a thread private object.
It's only valid for a single thread at a time.
It would be possible to have a thread shared object of
PINENTRY_ACTIVE, keeping ENTRY_OWNER for distinguishing its
owner (which is also a thread shared object). But, in this case,
access to ENTRY_OWNER is tricky (only comparison to accessing thread
would be OK with no lock), or we need to introduce another lock for
accessing ENTRY_OWNER, which complicates the code too much.
So, simply have a thread private object for recursive pinentry access.
GnuPG-bug-id: 3190
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
(cherry picked from commit fb7828676cc2c01047498898378711e049f73fee)
* agent/command.c (cmd_scd): Support !BUILD_WITH_SCDAEMON.
* tests/openpgp/defs.scm (create-gpghome): Likewise.
* tests/gpgsm/gpgsm-defs.scm (create-gpgsmhome): Likewise.
--
We could modify gpg-agent to remove all support of scdaemon, with no
inclusion of call-scd.c, divert-scd.c, and learncard.c, but it would
not be worth to do that.
GnuPG-bug-id: 3316
Signed-off-by: NIIBE Yutaka <gniibe@fsij.org>
* agent/agent.h (L_): Define agent_Lunderscore when simple
gettext is used.
--
This fixes a regression introduced by b3286af3 ENABLE_NLS
is not defined if we use simple gettext and not gettext.
GnuPG-Bug-Id: T3364
Signed-off-by: Andre Heinecke <aheinecke@intevation.de>
