4 years agodoc: Fix name of keep-ownertrust.
Werner Koch [Thu, 26 Feb 2015 14:06:00 +0000 (15:06 +0100)]
doc: Fix name of keep-ownertrust.


Reported-by: Guilhem Moulin <>
4 years agompi: Avoid data-dependent timing variations in mpi_powm.
NIIBE Yutaka [Thu, 26 Feb 2015 12:00:05 +0000 (21:00 +0900)]
mpi: Avoid data-dependent timing variations in mpi_powm.

* include/mpi.h, mpi/mpiutils.c (mpi_set_cond): New.
* mpi/mpi-pow.c (SIZE_PRECOMP): Rename from SIZE_B_2I3.
(mpi_powm): Access all data in the table and use mpi_set_cond.


Access to the precomputed table was indexed by a portion of EXPO,
which could be mounted by a side channel attack.  This change fixes
this particular data-dependent access pattern.

4 years agodoc: Add warning note about not acting as an oracle to --batch.
Werner Koch [Mon, 23 Feb 2015 12:10:57 +0000 (13:10 +0100)]
doc: Add warning note about not acting as an oracle to --batch.


4 years agoProtect against NULL return of mpi_get_opaque.
Werner Koch [Mon, 23 Feb 2015 10:04:35 +0000 (11:04 +0100)]
Protect against NULL return of mpi_get_opaque.

* g10/seckey-cert.c (do_check): Call BUG for NULL return of

This is the suggested addition from commit 6f03218.  We better run
into an fatal error than into a segv.

Signed-off-by: Werner Koch <>
4 years agogpg: Fix segv due to NULL value stored as opaque MPI
Daniel Kahn Gillmor [Sun, 22 Feb 2015 04:10:37 +0000 (23:10 -0500)]
gpg: Fix segv due to NULL value stored as opaque MPI

* g10/build-packet.c (do_secret_key): Check for NULL return from
* g10/keyid.c (hash_public_key): Ditto.

This is a backport of 76c8122adfed0f0f443cce7bda702ba2b39661b3 from
master to the STABLE-BRANCH-1-4

On the STABLE-BRANCH-1-4, we may also want to patch g10/seckey-cert.c,
but that has not been done in this patch.

This fix extends commmit 0835d2f44ef62eab51fce6a927908f544e01cf8f.

  gpg2 --export --no-default-keyring --keyring TESTDATA

With TESTDATA being below after unpacking.



Reported-by: Jodie Cunningham
[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agogpg: Remove an unused variable.
Werner Koch [Sun, 22 Feb 2015 04:10:36 +0000 (23:10 -0500)]
gpg: Remove an unused variable.

* g10/import.c (import): Remove need_armor.

[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agocurl-shim: clean up varargs
Daniel Kahn Gillmor [Sun, 22 Feb 2015 04:10:35 +0000 (23:10 -0500)]
curl-shim: clean up varargs

* keyserver/curl-shim.c (curl_easy_setopt) : ensure that va_end is


stdarg(3) says:
      Each invocation of va_start() must be matched by a
      corresponding invocation of va_end() in the same function.

Observed by Joshua Rogers <>

Debian-Bug-Id: #773475

[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agogpg: Print better diagnostics for keyserver operations.
Werner Koch [Sun, 22 Feb 2015 04:10:34 +0000 (23:10 -0500)]
gpg: Print better diagnostics for keyserver operations.

* g10/armor.c (parse_key_failed_line): New.
(check_input): Watch out for gpgkeys_ error lines.
* g10/filter.h (armor_filter_context_t): Add field key_failed_code.
* g10/import.c (import): Add arg r_gpgkeys_err.
(import_keys_internal): Ditto.
(import_keys_stream): Ditto.
* g10/keyserver.c (keyserver_errstr): New.
(keyserver_spawn): Detect "KEY " lines while sending.  Get gpgkeys_err
while receiving keys.
(keyserver_work): Add kludge for better error messages.

GnuPG-bug-id: 1832

Note that these changes can be backported to 1.4 but they don't make
sense for 2.1 due to the removal of the keyserver helpers.  The error
reporting could be improved even more but given that this is an old
GnuPG branch it is not justified to put too much effort into it.

Signed-off-by: Werner Koch <>
[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agoUse inline functions to convert buffer data to scalars.
Werner Koch [Sun, 22 Feb 2015 04:10:32 +0000 (23:10 -0500)]
Use inline functions to convert buffer data to scalars.

* include/host2net.h (buf16_to_ulong, buf16_to_uint): New.
(buf16_to_ushort, buf16_to_u16): New.
(buf32_to_size_t, buf32_to_ulong, buf32_to_uint, buf32_to_u32): New.

This fixes sign extension on shift problems.  Hanno Böck found a case
with an invalid read due to this problem.  To fix that almost all uses
of "<< 24" and "<< 8" are changed by this patch to use an inline
function from host2net.h.

(back ported from commit 2183683bd633818dd031b090b5530951de76f392)

Signed-off-by: Werner Koch <>
[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agodoc: Change remaining http links to to https
Werner Koch [Sun, 22 Feb 2015 04:10:31 +0000 (23:10 -0500)]
doc: Change remaining http links to to https

GnuPG-bug-id: 1830

[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agogpg: Prevent an invalid memory read using a garbled keyring.
Werner Koch [Sun, 22 Feb 2015 04:10:30 +0000 (23:10 -0500)]
gpg: Prevent an invalid memory read using a garbled keyring.

* g10/keyring.c (keyring_get_keyblock): Whitelist allowed packet

The keyring DB code did not reject packets which don't belong into a
keyring.  If for example the keyblock contains a literal data packet
it is expected that the processing code stops at the data packet and
reads from the input stream which is referenced from the data packets.
Obviously the keyring processing code does not and cannot do that.
However, when exporting this messes up the IOBUF and leads to an
invalid read of sizeof (int).

We now skip all packets which are not allowed in a keyring.

Reported-by: Hanno Böck <>
(back ported from commit f0f71a721ccd7ab9e40b8b6b028b59632c0cc648)

[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agogpg: Fix a NULL-deref in export due to invalid packet lengths.
Werner Koch [Sun, 22 Feb 2015 04:10:29 +0000 (23:10 -0500)]
gpg: Fix a NULL-deref in export due to invalid packet lengths.

* g10/build-packet.c (write_fake_data): Take care of a NULL stored as
opaque MPI.

Reported-by: Hanno Böck <>
(back ported from commit 0835d2f44ef62eab51fce6a927908f544e01cf8f)

[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agogpg: Fix a NULL-deref due to empty ring trust packets.
Werner Koch [Sun, 22 Feb 2015 04:10:28 +0000 (23:10 -0500)]
gpg: Fix a NULL-deref due to empty ring trust packets.

* g10/parse-packet.c (parse_trust): Always allocate a packet.

Reported-by: Hanno Böck <>
Signed-off-by: Werner Koch <>
(back ported from commit 39978487863066e59bb657f5fe4e8baab510da7e)

[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agogpg: Limit the size of key packets to a sensible value.
Werner Koch [Sun, 22 Feb 2015 04:10:27 +0000 (23:10 -0500)]
gpg: Limit the size of key packets to a sensible value.

* g10/parse-packet.c (MAX_KEY_PACKET_LENGTH): New.
(parse_key): Limit the size of a key packet to 256k.
(parse_user_id): Use macro for the packet size limit.
(parse_attribute): Ditto.
(parse_comment): Ditto.

Without that it is possible to force gpg to allocate large amounts of
memory by using a bad encoded MPI.  This would be an too easy DoS.
Another way to mitigate would be to change the MPI read function to
allocate memory dynamically while reading the MPI.  However, that
complicates and possibly slows down the code.  A too large key packet
is in any case a sign for broken data and thus gpg should not use it.

Reported-by: Hanno Böck
GnuPG-bug-id: 1823
Signed-off-by: Werner Koch <>
(back ported from commit 382ba4b137b42d5f25a7e256bb7c053ee5ac7b64)

[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agogpg: Allow predefined names as answer to the keygen.algo prompt.
Werner Koch [Sun, 22 Feb 2015 04:10:26 +0000 (23:10 -0500)]
gpg: Allow predefined names as answer to the keygen.algo prompt.

* g10/keygen.c (ask_algo): Add list of strings.

Signed-off-by: Werner Koch <>
(backported from commit b1d5ed6ac842469afcb84868d0f6641dc286a6c7)

[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agogpg: Print a warning if the subkey expiration may not be what you want.
Werner Koch [Sun, 22 Feb 2015 04:10:25 +0000 (23:10 -0500)]
gpg: Print a warning if the subkey expiration may not be what you want.

* g10/keyedit.c (subkey_expire_warning): New.
keyedit_menu): Call it when needed.
GnuPG-bug-id: 1715

The heuristic to detect a problem is not very advanced but it should
catch the most common cases.

(backported from commit ae3d1bbb65b65cf3c57bb14886be120f5e31635d)

[dkg: rebased to STABLE-BRANCH-1-4]
Signed-off-by: Daniel Kahn Gillmor <>
4 years agoUse ciphertext blinding for Elgamal decryption.
Werner Koch [Mon, 11 Aug 2014 14:15:40 +0000 (16:15 +0200)]
Use ciphertext blinding for Elgamal decryption.

* cipher/elgamal.c (USE_BLINDING): New.
(decrypt): Rewrite to use ciphertext blinding.

CVE-id: CVE-2014-3591

As a countermeasure to a new side-channel attacks on sliding windows
exponentiation we blind the ciphertext for Elgamal decryption.  This
is similar to what we are doing with RSA.

Unfortunately, the performance impact of Elgamal blinding is quite
noticeable: For a 3072 bit Elgamal key the decryption used to take
13ms; with the blinding it takes 24ms.  This has been measured using
time(1), calling gpg with a 100 byte message, and having gpg modified
to run the pubkey_decrypt function 100 times and finally scale the
result (using an i5-2410M CPU @ 2.30GHz TP 220).

4 years agoUpdate automake helper files.
Werner Koch [Mon, 19 Jan 2015 16:25:57 +0000 (17:25 +0100)]
Update automake helper files.


4 years agoModernize to automake 1.14.
Werner Koch [Mon, 19 Jan 2015 17:59:13 +0000 (18:59 +0100)]
Modernize to automake 1.14.

* (AUTOMAKE_OPTIONS): Move to ...
* (AM_INIT_AUTOMAKE): here and add serial-tests.

* keyserver/ Replace INCLUDES by AM_CPPFLAGS.
* mpi/ Ditto.
* util/ Ditto.
* keyserver/ Ditto.  Adjusted other things.

* m4/intl.m4, m4/po.m4: Use autoconf's AC_PROG_MKDIR_P.

4 years agoFix a problem with select and high fds.
Werner Koch [Mon, 19 Jan 2015 15:46:05 +0000 (16:46 +0100)]
Fix a problem with select and high fds.

* cipher/rndlinux.c (rndlinux_gather_random): Check fd before using

If on systems where the maximum number of fds may be dynamically
configured to a value of FD_MAXSIZE or higher and the RNG is first
used after more than FD_SETSIZE-1 descriptors are in use, we disable
the progress messages from the RNG.  A better solution would be too
use poll but that requires more tests.

The same problem exists in rndunix.c - however this rng is only used
on old Unices and I assume that they don't feature dynamically
configured maximum fd sizes.

(from Libgcrypt commit 9487099071af4478d2882e633a0ade805801d6fa)

This may fix
GnuPG-bug-id: 1818

4 years agodoc: Formatting fixes.
Werner Koch [Thu, 18 Apr 2013 12:40:43 +0000 (14:40 +0200)]
doc: Formatting fixes.

* doc/gpl.texi: Fix enumerate and re-indent examples.


Cherry-pick a part of ff6115227a1ced14e2fb3d160a12181b9dfbc502.

Reported-by: Ian Abbott
Signed-off-by: Werner Koch <>
4 years agoavoid future chance of using uninitialized memory
Daniel Kahn Gillmor [Fri, 19 Dec 2014 22:53:36 +0000 (17:53 -0500)]
avoid future chance of using uninitialized memory

* util/iobuf.c: (iobuf_open): initialize len


Cherry-pick 367b073ab5f439ccf0750461d10c69f36998bd62.

In iobuf_open, IOBUFCTRL_DESC and IOBUFCTRL_INIT commands are invoked
(via file_filter()) on fcx, passing in a pointer to an uninitialized

With these two commands, file_filter doesn't actually do anything with
the value of len, so there's no actual risk of use of uninitialized
memory in the code as it stands.

However, some static analysis tools might flag this situation with a
warning, and initializing the value doesn't hurt anything, so i think
this trivial cleanup is warranted.

Debian-Bug-Id: 773469

4 years agodoc: Fix memory leak in yat2m.
Werner Koch [Mon, 22 Dec 2014 11:44:13 +0000 (12:44 +0100)]
doc: Fix memory leak in yat2m.

* doc/yat2m.c (write_th): Free NAME.

Reported-by: Joshua Rogers <>
4 years agogpg: Fix possible read of unallocated memory
Werner Koch [Fri, 12 Dec 2014 09:41:25 +0000 (10:41 +0100)]
gpg: Fix possible read of unallocated memory

* g10/parse-packet.c (can_handle_critical): Check content length
before calling can_handle_critical_notation.

The problem was found by Jan Bee and gniibe proposed the used fix.

This bug can't be exploited: Only if the announced length of the
notation is 21 or 32 a memcmp against fixed strings using that length
would be done.  The compared data is followed by the actual signature
and thus it is highly likely that not even read of unallocated memory
will happen.  Nevertheless such a bug needs to be fixed.

Signed-off-by: Werner Koch <>
4 years agoscd: Fix possibly inhibited checkpin of the admin pin.
Werner Koch [Fri, 12 Dec 2014 19:08:45 +0000 (20:08 +0100)]
scd: Fix possibly inhibited checkpin of the admin pin.

* scd/app-openpgp.c (do_check_pin): Do not check a byte of a released

Signed-off-by: Werner Koch <>
4 years agoscd: fix get_public_key for OpenPGPcard v1.0.
Joshua Rogers [Sat, 20 Dec 2014 00:38:53 +0000 (11:38 +1100)]
scd: fix get_public_key for OpenPGPcard v1.0.

* scd/app-openpgp.c (get_public_key): correctly close 'fp' upon use.


Inside the get_public_key function, 'fp' was opened using popen, but
incorrectly closed using fclose.

Debian-Bug-Id: 773474

4 years agogpg: release DEK soon after its use.
NIIBE Yutaka [Fri, 12 Dec 2014 08:41:56 +0000 (17:41 +0900)]
gpg: release DEK soon after its use.

* g10/keygen.c (generate_subkeypair): Release DEK soon.


This fixes the out_of_core error in the test case of adding
RSA-4096 subkey to RSA-4096 primary key with configuration:

    s2k-cipher-algo S10

Debian-bug-id: 772780

5 years agodoc: Do not install gnupg.7
Werner Koch [Mon, 24 Nov 2014 19:01:45 +0000 (20:01 +0100)]
doc: Do not install gnupg.7

GnuPG-bug-id: 1770

5 years agogpg: Fix use of uninit.value in listing sig subpkts.
Werner Koch [Mon, 24 Nov 2014 18:41:46 +0000 (19:41 +0100)]
gpg: Fix use of uninit.value in listing sig subpkts.

* g10/parse-packet.c (dump_sig_subpkt): Print regex subpacket

We may not use "%s" to print an arbitrary buffer.  At least "%.*s"
should have been used.  However, it is in general preferable to escape
control characters while printf user data.

Reported-by: Hanno Böck
Signed-off-by: Werner Koch <>
(backported from commit 596ae9f5433ca3b0e01f7acbe06fd2e424c42ae8)

5 years agogpg: Fix off-by-one read in the attribute subpacket parser.
Werner Koch [Mon, 24 Nov 2014 18:38:04 +0000 (19:38 +0100)]
gpg: Fix off-by-one read in the attribute subpacket parser.

* g10/parse-packet.c (parse_attribute_subpkts): Check that the
attribute packet is large enough for the subpacket type.

Reported-by: Hanno Böck
Signed-off-by: Werner Koch <>
(backported from commit 0988764397f99db4efef1eabcdb8072d6159af76)

5 years agogpg: Fix a NULL-deref for invalid input data.
Werner Koch [Mon, 24 Nov 2014 18:32:47 +0000 (19:32 +0100)]
gpg: Fix a NULL-deref for invalid input data.

* g10/mainproc.c (proc_encrypted): Take care of canceled passpharse

GnuPG-bug-id: 1761
Signed-off-by: Werner Koch <>
(backported from commit 32e85668b82f6fbcb824eea9548970804fb41d9e)

5 years agogpg: Make the use of "--verify FILE" for detached sigs harder.
Werner Koch [Fri, 14 Nov 2014 08:36:19 +0000 (09:36 +0100)]
gpg: Make the use of "--verify FILE" for detached sigs harder.

* g10/openfile.c (open_sigfile): Factor some code out to ...
(get_matching_datafile): new function.
* g10/plaintext.c (hash_datafiles): Do not try to find matching file
in batch mode.
* g10/mainproc.c (check_sig_and_print): Print a warning if a possibly
matching data file is not used by a standard signatures.

Allowing to use the abbreviated form for detached signatures is a long
standing bug which has only been noticed by the public with the
release of 2.1.0.  :-(

What we do is to remove the ability to check detached signature in
--batch using the one file abbreviated mode.  This should exhibit
problems in scripts which use this insecure practice.  We also print a
warning if a matching data file exists but was not considered because
the detached signature was actually a standard signature:

  gpgv: Good signature from "Werner Koch (dist sig)"
  gpgv: WARNING: not a detached signature; \
  file 'gnupg-2.1.0.tar.bz2' was NOT verified!

We can only print a warning because it is possible that a standard
signature is indeed to be verified but by coincidence a file with a
matching name is stored alongside the standard signature.

Reported-by: Simon Nicolussi (to gnupg-users on Nov 7)
Signed-off-by: Werner Koch <>
(backported from commit 69384568f66a48eff3968bb1714aa13925580e9f)

Updated doc/gpg.texi.

5 years agogpg: Add import option "keep-ownertrust".
Werner Koch [Wed, 12 Nov 2014 08:56:40 +0000 (09:56 +0100)]
gpg: Add import option "keep-ownertrust".

* g10/options.h (IMPORT_KEEP_OWNERTTRUST): New.
* g10/import.c (parse_import_options): Add "keep-ownertrust".
(import_one): Act upon new option.

This option is in particular useful to convert from a pubring.gpg to
the new pubring.kbx in GnuPG 2.1 or vice versa:

gpg1 --export | gpg2 --import-options keep-ownertrust --import

(cherry-picked from commit da95d0d37841b34e2f3d7047f14ab4d98a7c0c56)

5 years agodoc: Minor fix
Werner Koch [Fri, 3 Oct 2014 18:21:28 +0000 (20:21 +0200)]
doc: Minor fix


Thanks to dkg for the reminder.

5 years agompi: Fix compiler warning.
Werner Koch [Fri, 3 Oct 2014 16:28:58 +0000 (18:28 +0200)]
mpi: Fix compiler warning.

* mpi/mpi-inv.c (mpi_invm): Do not return a value.

5 years agogpg: Add build and runtime support for larger RSA keys
Daniel Kahn Gillmor [Fri, 3 Oct 2014 16:01:11 +0000 (12:01 -0400)]
gpg: Add build and runtime support for larger RSA keys

* Added --enable-large-secmem option.
* g10/options.h: Add opt.flags.large_rsa.
* g10/gpg.c: Contingent on configure option: adjust secmem size,
add gpg --enable-large-rsa, bound to opt.flags.large_rsa.
* g10/keygen.c: Adjust max RSA size based on opt.flags.large_rsa
* doc/gpg.texi: Document --enable-large-rsa.


Some older implementations built and used RSA keys up to 16Kib, but
the larger secret keys now fail when used by more recent GnuPG, due to
secure memory limitations.

Building with ./configure --enable-large-secmem will make gpg
capable of working with those secret keys, as well as permitting the
use of a new gpg option --enable-large-rsa, which let gpg generate RSA
keys up to 8Kib when used with --batch --gen-key.

Debian-bug-id: 739424

Minor edits by wk.

GnuPG-bug-id: 1732

5 years agodoc: Cleanup gpg.texi.
Werner Koch [Mon, 29 Sep 2014 09:28:55 +0000 (11:28 +0200)]
doc: Cleanup gpg.texi.


We don't need the gpgone and gpgtwoone macros anymore.

5 years agodoc: Final update from master (gnupg 2.1)
Werner Koch [Mon, 29 Sep 2014 09:11:30 +0000 (11:11 +0200)]
doc: Final update from master (gnupg 2.1)

* doc/ (sources_from_trunk): Remove.
(update-source): Make it a dummy.
* doc/gpg.texi: Update.
* doc/yat2m.c: Update.

Maintaining 3 versions in of the gpg manual in one file is getting
more complicated with 2.1.  Thus we stop this now and keep the manual
for 1.4 separate.

5 years agoAllow use of --debug-level=LEVEL without '='.
Werner Koch [Thu, 25 Sep 2014 07:47:28 +0000 (09:47 +0200)]
Allow use of --debug-level=LEVEL without '='.

* g10/gpg.c (opts): Fix "debug-level".

5 years agompi: Improve mpi_invm to detect bad input.
Werner Koch [Thu, 11 Sep 2014 15:06:16 +0000 (17:06 +0200)]
mpi: Improve mpi_invm to detect bad input.

* mpi/mpi-inv.c (mpi_invm): Return 0 for bad input.

Without this patch the function may enter an endless loop.  This is a
backport from libgcrypt.

GnuPG-bug-id: 1713

5 years agompi: Suppress set-but-unused-variables warnings.
Werner Koch [Wed, 20 Aug 2014 11:25:17 +0000 (13:25 +0200)]
mpi: Suppress set-but-unused-variables warnings.

* include/types.h (GNUPG_GCC_ATTR_UNUSED): Define for gcc >= 3.5.
* mpi/mpih-div.c (mpihelp_divmod_1, mpihelp_mod_1): Mark dummy as
* mpi/mpi-internal.h (UDIV_QRNND_PREINV): Mark _ql as unused.

Due to the use of macros and longlong.h, we use variables which are
only used by some architectures.  At least gcc 4.7.2 prints new
warnings about set but not used variables.  This patch silences them.

5 years agoFix strict-alias warnings for rijndael.c
Werner Koch [Wed, 20 Aug 2014 10:22:35 +0000 (12:22 +0200)]
Fix strict-alias warnings for rijndael.c

* cipher/rijndael.c (do_setkey, prepare_decryption): Use u32_a_t cast.

This extends commit 0ad1458f827c7602ef7f1a4652af05641fd02b62

5 years agogpg: Allow compressed data with algorithm 0.
Werner Koch [Wed, 20 Aug 2014 10:05:16 +0000 (12:05 +0200)]
gpg: Allow compressed data with algorithm 0.

* g10/mainproc.c (proc_compressed): Remove superfluous check for
an algorithm number of 0.

(backport from commit 88633bf3d417aeb5ea0f75508aba8e32adc8acef)

GnuPG-bug-id: 1326, 1684

5 years agogpg: Fix regression due to the keyserver import filter.
Werner Koch [Wed, 6 Aug 2014 15:11:21 +0000 (17:11 +0200)]
gpg: Fix regression due to the keyserver import filter.

* g10/keyserver.c (keyserver_retrieval_filter): Change args.  Rewrite
to take subpakets in account.
* g10/import.c (import_one, import_secret_one): Pass keyblock to

GnuPG-bug-id: 1680

Resolved conflicts:
g10/main.h - s/import_filter/import_filter_t/g

5 years agoAdd kbnode_t for easier backporting.
Werner Koch [Wed, 6 Aug 2014 15:09:15 +0000 (17:09 +0200)]
Add kbnode_t for easier backporting.

* g10/global.h (kbnode_t): New.

5 years agoPost release updates
Werner Koch [Wed, 25 Jun 2014 12:33:34 +0000 (14:33 +0200)]
Post release updates


5 years agoRelease 1.4.18 gnupg-1.4.18
Werner Koch [Wed, 25 Jun 2014 12:33:34 +0000 (14:33 +0200)]
Release 1.4.18

5 years agoLimit keysize for unattended key generation to useful values.
Werner Koch [Wed, 25 Jun 2014 18:25:28 +0000 (20:25 +0200)]
Limit keysize for unattended key generation to useful values.

* g10/keygen.c (gen_elg): Enforce keysize 1024 to 4096.
(gen_rsa): Enforce keysize 1024 to 4096.
(gen_dsa): Enforce keysize 768 to 3072.

It was possible to create 16k RSA keys in batch mode. In addition to
the silliness of such keys, they have the major drawback that GnuPG,
with its limited amount of specially secured memory areas, the use of
such keys may lead to an "out of secure memory" condition.

5 years agoMake screening of keyserver result work with multi-key commands.
Werner Koch [Wed, 25 Jun 2014 12:33:34 +0000 (14:33 +0200)]
Make screening of keyserver result work with multi-key commands.

* g10/keyserver.c (ks_retrieval_filter_arg_s): new.
(keyserver_retrieval_filter): Use new struct and check all
(keyserver_spawn): Pass filter arg suing the new struct.

This is a fix for commit 52303043.

The old code did only work for a single key.  It failed as soon as
several keys are specified ("gpg --refresh-keys" or "gpg --recv-key A
B C").

5 years agoAdd CVE number
Werner Koch [Tue, 24 Jun 2014 18:12:26 +0000 (20:12 +0200)]
Add CVE number


5 years agoPost release changes.
Werner Koch [Mon, 23 Jun 2014 15:42:21 +0000 (17:42 +0200)]
Post release changes.


5 years agoRelease 1.4.17 gnupg-1.4.17
Werner Koch [Mon, 23 Jun 2014 14:38:09 +0000 (16:38 +0200)]
Release 1.4.17

5 years agopo: Auto-update
Werner Koch [Mon, 23 Jun 2014 14:35:41 +0000 (16:35 +0200)]
po: Auto-update


5 years agodoc: Update from master.
Werner Koch [Mon, 23 Jun 2014 14:09:34 +0000 (16:09 +0200)]
doc: Update from master.

5 years agoFix syntax error introduced with 60bd6488
Werner Koch [Mon, 23 Jun 2014 15:14:15 +0000 (17:14 +0200)]
Fix syntax error introduced with 60bd6488

* g10/apdu.c (pcsc_dword_t): Fix syntax error.

5 years agoScreen keyserver responses.
Stefan Tomanek [Wed, 29 Jan 2014 23:57:43 +0000 (00:57 +0100)]
Screen keyserver responses.

* g10/main.h: Typedef import_filter for filter callbacks.
* g10/import.c (import): Add filter callbacks to param list.
(import_one): Ditto.
(import_secret_one): Ditto.
(import_keys_internal): Ditto.
(import_keys_stream): Ditto.
* g10/keyserver.c (keyserver_retrieval_filter): New.
(keyserver_spawn): Pass filter to import_keys_stream()

These changes introduces import functions that apply a constraining
filter to imported keys. These filters can verify the fingerprints of
the keys returned before importing them into the keyring, ensuring that
the keys fetched from the keyserver are in fact those selected by the
user beforehand.

Signed-off-by: Stefan Tomanek <>
Re-indention and minor changes by wk.

5 years agoPrint hash algorithm in sig records
Werner Koch [Mon, 23 Jun 2014 12:57:32 +0000 (14:57 +0200)]
Print hash algorithm in sig records

* g10/keylist.c (list_keyblock_colon): Print field 16.

We have this info already in gnupg-2 and it is easy to add it to 1.4.

Debian-bug-id: 672658

Patch written and tested by Daniel Leidert.  See above.

5 years agoRemove useless diagnostic in MDC verification.
Werner Koch [Mon, 23 Jun 2014 11:24:43 +0000 (13:24 +0200)]
Remove useless diagnostic in MDC verification.

* g10/encr-data.c (decrypt_data): Do not distinguish between a bad MDC
packet header and a bad MDC.

The separate diagnostic was introduced for debugging a problems.  For
explaining an MDC error a single error message is easier to understand.

5 years agow32: Fix typo in README.W32.
Werner Koch [Mon, 23 Jun 2014 08:43:22 +0000 (10:43 +0200)]
w32: Fix typo in README.W32.

GnuPG-bug-id: 1488

5 years agointl: Fix for uClibc.
Werner Koch [Mon, 23 Jun 2014 08:22:32 +0000 (10:22 +0200)]
intl: Fix for uClibc.

* intl/localename.c (gl_locale_name_thread_unsafe): Take care of

Patch provided by "vriera".

GnuPG-bug-id: 1642

5 years agoPC/SC cleanup.
Werner Koch [Mon, 23 Jun 2014 07:58:34 +0000 (09:58 +0200)]
PC/SC cleanup.

* g10/apdu.c (pcsc_dword_t): New.  It was named as DWORD (double-word)
when a word was 16-bit.
(struct reader_table_s): Fixes for types.
(struct pcsc_readerstate_s) [__APPLE__]: Enable #pragma pack(1).
Throughout: Fixes for types.

GnuPG-bug-id: 1358

This is a backport of commit ae22d629b6028aa994ff09f012e1cb029575eeae.

5 years agogpg: Use more specific reason codes for INV_RECP.
Werner Koch [Tue, 10 Jun 2014 12:54:55 +0000 (14:54 +0200)]
gpg: Use more specific reason codes for INV_RECP.

* g10/pkclist.c (build_pk_list): Use more specific reasons codes for

GnuPG-bug-id: 1650

Note that this patch is a bit more limited than the one in 2.1.

5 years agodoc: Remove outdated Russian man page.
Werner Koch [Mon, 23 Jun 2014 06:52:29 +0000 (08:52 +0200)]
doc: Remove outdated Russian man page.

* (DOCBOOK_TO_MAN): Remove.
* doc/ Remove.
* doc/ Remove all related code.

The man page is outdated and we do not use docbook for a long time
now.  If someone wants to revive such a man page, it would be best to
translate the respective parts of the GnuPG manual in git master.

GnuPG-bug-id: 1652

5 years agogpg: Avoid infinite loop in uncompressing garbled packets.
Werner Koch [Fri, 20 Jun 2014 08:39:26 +0000 (10:39 +0200)]
gpg: Avoid infinite loop in uncompressing garbled packets.

* g10/compress.c (do_uncompress): Limit the number of extra FF bytes.

A packet like (a3 01 5b ff) leads to an infinite loop.  Using
--max-output won't help if it is a partial packet.  This patch
actually fixes a regression introduced on 1999-05-31 (c34c6769).
Actually it would be sufficient to stuff just one extra 0xff byte.
Given that this problem popped up only after 15 years, I feel safer to
allow for a very few FF bytes.

Thanks to Olivier Levillain and Florian Maury for their detailed

5 years agogpg: Need to init the trustdb for import.
Werner Koch [Thu, 6 Mar 2014 15:11:34 +0000 (16:11 +0100)]
gpg: Need to init the trustdb for import.

* g10/trustdb.c (clear_ownertrusts): Init trustdb.

This is actually a hack to fix a bug introduced with commit 2528178.
Debian uses it and thus we should do too.

GnuPG-bug-id: 1622

5 years agoSupport building using the latest mingw-w64 toolchain.
Werner Koch [Thu, 23 Jan 2014 14:19:35 +0000 (15:19 +0100)]
Support building using the latest mingw-w64 toolchain.

* acinclude.m4 (GNUPG_SYS_SYMBOL_UNDERSCORE): Change mingw detection.

  From: Stephen Kitt <>

  All MinGW targets require underscores when linking. This patch fixes
  acinclude.m4 and the resulting configure so they don't limit the use
  of underscores to the old mingw32msvc targets.

Debian-bug-id: 730271
Signed-off-by: Werner Koch <>
5 years agoPost release version number bump.
Werner Koch [Fri, 13 Dec 2013 09:03:19 +0000 (10:03 +0100)]
Post release version number bump.


5 years agoRelease 1.4.16 gnupg-1.4.16
Werner Koch [Fri, 13 Dec 2013 08:07:11 +0000 (09:07 +0100)]
Release 1.4.16

5 years agoChange --show-session-key to print the session key earlier.
Werner Koch [Wed, 11 Dec 2013 09:20:15 +0000 (10:20 +0100)]
Change --show-session-key to print the session key earlier.

* g10/mainproc.c (proc_encrypted): Move show_session_key code to ...
* g10/decrypt-data.c (decrypt_data): here.

This feature can be used to return the session key for just a part of
a file.  For example to downloading just the first 32k of a huge file,
decrypting that incomplete part and while ignoring all the errors
break out the session key.  The session key may then be used on the
server to decrypt the entire file without the need to have the private
key on the server.

This is the same feature as
commit 101a54add351ff62793cbfbf3877787c4791f833 for 2.1 and
commit 3ae90ff28c500967cb90b1176299d2ca01ef450f for 2.0.

GnuPG-bug-id: 1389
Signed-off-by: Werner Koch <>
5 years agoUpdate config.{guess,sub} and some copyright notices.
Werner Koch [Tue, 10 Dec 2013 19:33:48 +0000 (20:33 +0100)]
Update config.{guess,sub} and some copyright notices.

* scripts/config.guess, scripts/config.sub: Update to version

Signed-off-by: Werner Koch <>
6 years agoPrepare for newer automakes which default to parallel tests.
Werner Koch [Tue, 22 Oct 2013 12:26:53 +0000 (14:26 +0200)]
Prepare for newer automakes which default to parallel tests.

* checks/ Add a list of test dependencies.

We want to keep the tests in a specific order because that helps to
compare tests and some tests rely on others anyway.

Signed-off-by: Werner Koch <>
6 years agoNormalize the MPIs used as input to secret key functions.
Werner Koch [Wed, 27 Nov 2013 13:22:10 +0000 (14:22 +0100)]
Normalize the MPIs used as input to secret key functions.

* cipher/rsa.c (secret): Normalize the INPUT.
(rsa_decrypt): Pass reduced data to secret.
* cipher/elgamal.c (decrypt): Normalize A and B.
* cipher/dsa.c (sign): Normalize HASH.

mpi_normalize is in general not required because extra leading zeroes
do not harm the computation.  However, adding extra all zero limbs or
padding with multiples of N may be useful in side-channel attacks. In
particular they are used by the acoustic crypt-analysis.  This is an
extra pre-caution which alone would not be sufficient to mitigate the
described attack.

CVE-id: CVE-2013-4576

Signed-off-by: Werner Koch <>
6 years agoUse blinding for the RSA secret operation.
Werner Koch [Tue, 22 Oct 2013 12:26:53 +0000 (14:26 +0200)]
Use blinding for the RSA secret operation.

* cipher/random.c (randomize_mpi): New.
* g10/gpgv.c (randomize_mpi): New stub.
* cipher/rsa.c (USE_BLINDING): Define macro.
(secret): Implement blinding.

GPG 1.x has never used any protection against timing attacks on the
RSA secret operation.  The rationale for this has been that there was
no way to mount a remote timing attack on GnuPG.  With the turning up
of Acoustic Cryptanalysis ( this
assumption no longer holds true and thus we need to do do something
about it.  Blinding seems to be a suitable mitigation to the threat of
key extraction.  It does not help against distinguishing used keys,

Note that GPG 2.x uses Libgcrypt which does blinding by default.

The performance penalty is negligible: Modifying the core pubkey_sign
or pubkey_decrypt function to run 100 times in a loop, the entire
execution times for signing or decrypting a small message using a 4K
RSA key on a Thinkpad X220 are

  Without blinding:  5.2s  (8.9s)
  With blinding:     5.6s  (9.3s)

The numbers in parentheses give the values without the recently
implemented k-ary exponentiation code.  Thus for the next release the
user will actually experience faster signing and decryption.  A
drawback of blinding is that we need random numbers even for
decryption (albeit at low quality).

Signed-off-by: Werner Koch <>
CVE-id: CVE-2013-4576

6 years agogpg: Change armor Version header to emit only the major version.
Werner Koch [Wed, 27 Nov 2013 08:20:02 +0000 (09:20 +0100)]
gpg: Change armor Version header to emit only the major version.

* g10/options.h (opt): Rename field no_version to emit_version.
* g10/gpg.c (main): Init opt.emit_vesion to 1.  Change --emit-version
to bump up opt.emit_version.
* g10/armor.c (armor_filter): Implement different --emit-version

GnuPG-bug-id: 1572
Signed-off-by: Werner Koch <>
(cherry picked from commit e951782e937ce290be0d89d83e84b3daea997587)

Resolved conflicts:

6 years agompi: mpi-pow improvements
Werner Koch [Fri, 18 Oct 2013 08:54:55 +0000 (10:54 +0200)]
mpi: mpi-pow improvements

(mpi_powm) [!USE_ALGORITHM_SIMPLE_EXPONENTIATION]: New implementation
of left-to-right k-ary exponentiation.

This is a backport from Libgcrypt commit

Signed-off-by: NIIBE Yutaka <>
    For the Yarom/Falkner flush+reload cache side-channel attack, we
    changed the code so that it always calls the multiplication
    routine (even if we can skip it to get result).  This results some
    performance regression.

    This change is for recovering performance with efficient

Signed-off-by: Werner Koch <>
6 years agoPrint the keyid for key packets with --list-packets.
Werner Koch [Fri, 18 Oct 2013 08:24:32 +0000 (10:24 +0200)]
Print the keyid for key packets with --list-packets.

* g10/parse-packet.c (parse_key): Add keyid printing.

This is backport from GnuPG-2.  Note that the --list-packets command
is for debugging only and not part iof the stable API.

Signed-off-by: Werner Koch <>
6 years agompi: Fix syntax error for mips64 and gcc < 4.4
Werner Koch [Fri, 11 Oct 2013 12:11:43 +0000 (14:11 +0200)]
mpi: Fix syntax error for mips64 and gcc < 4.4

* mpi/longlong.h [__mips && gcc < 4.4]: Fix cpp syntax error.

GnuPG-bug-id: 1465

6 years agogpg: Do not require a trustdb with --always-trust.
Werner Koch [Fri, 11 Oct 2013 07:25:58 +0000 (09:25 +0200)]
gpg: Do not require a trustdb with --always-trust.

* g10/tdbio.c (tdbio_set_dbname): Add arg R_NOFILE.
* g10/trustdb.c (trustdb_args): Add field no_trustdb.
(init_trustdb): Set that field.
(revalidation_mark):  Take care of a nonexistent trustdb file.
(read_trust_options): Ditto.
(get_ownertrust): Ditto.
(get_min_ownertrust): Ditto.
(update_ownertrust): Ditto.
(update_min_ownertrust): Ditto.
(clear_ownertrusts): Ditto.
(cache_disabled_value): Ditto.
(check_trustdb_stale): Ditto.
(get_validity): Ditto.
* g10/gpg.c (main): Do not create a trustdb with most commands for
trust-model always.

This slightly changes the semantics of most commands in that they
won't create a trustdb if --trust-model=always is used.  It just does
not make sense to create a trustdb if there is no need for it.

Signed-off-by: Werner Koch <>
(cherry picked from commit 1a0eeaacd1bf09fe5125dbc3f56016bc20f3512e)

Resolved conflicts:
 (indentation fixes)

6 years agoPost release updates.
Werner Koch [Fri, 4 Oct 2013 19:29:50 +0000 (21:29 +0200)]
Post release updates.


6 years agoRelease 1.4.15 gnupg-1.4.15
Werner Koch [Fri, 4 Oct 2013 19:10:52 +0000 (21:10 +0200)]
Release 1.4.15

6 years agopo: Autoupdate due to changed order of strings.
Werner Koch [Fri, 4 Oct 2013 19:03:40 +0000 (21:03 +0200)]
po: Autoupdate due to changed order of strings.


6 years agodoc: Update from master.
Werner Koch [Fri, 4 Oct 2013 19:01:16 +0000 (21:01 +0200)]
doc: Update from master.

6 years agogpg: Print a "not found" message for an unknown key in --key-edit.
Werner Koch [Fri, 4 Oct 2013 16:34:56 +0000 (18:34 +0200)]
gpg: Print a "not found" message for an unknown key in --key-edit.

* g10/keyedit.c (keyedit_menu): Print message.

GnuPG-bug-id: 1420
Signed-off-by: Werner Koch <>
(cherry picked from commit 0bf54e60d31389812d05c3fd29bece876204561d)

6 years agogpg: Protect against rogue keyservers sending secret keys.
Werner Koch [Fri, 4 Oct 2013 11:44:39 +0000 (13:44 +0200)]
gpg: Protect against rogue keyservers sending secret keys.

* g10/options.h (IMPORT_NO_SECKEY): New.
* g10/keyserver.c (keyserver_spawn, keyserver_import_cert): Set new
* g10/import.c (import_secret_one): Deny import if flag is set.

By modifying a keyserver or a DNS record to send a secret key, an
attacker could trick a user into signing using a different key and
user id.  The trust model should protect against such rogue keys but
we better make sure that secret keys are never received from remote

Suggested-by: Stefan Tomanek
Signed-off-by: Werner Koch <>
(cherry picked from commit e7abed3448c1c1a4e756c12f95b665b517d22ebe)

Resolved conflicts:

6 years agogpg: Allow setting of all zero key flags
Daniel Kahn Gillmor [Tue, 19 Mar 2013 15:25:25 +0000 (11:25 -0400)]
gpg: Allow setting of all zero key flags

* g10/keygen.c (do_add_key_flags): Do not check for empty key flags.
(cherry picked from commit b693ec02c467696bf9d7324dd081e279f9965151)
(cherry picked from commit dd868acb0d13a9f119c0536777350a6c237a66a1)

6 years agogpg: Distinguish between missing and cleared key flags.
Werner Koch [Fri, 15 Mar 2013 14:46:03 +0000 (15:46 +0100)]
gpg: Distinguish between missing and cleared key flags.

* include/cipher.h (PUBKEY_USAGE_NONE): New.
* g10/getkey.c (parse_key_usage): Set new flag.

We do not want to use the default capabilities (derived from the
algorithm) if any key flags are given in a signature.  Thus if key
flags are used in any way, the default key capabilities are never

This allows to create a key with key flags set to all zero so it can't
be used.  This better reflects common sense.
(cherry picked from commit 4bde12206c5bf199dc6e12a74af8da4558ba41bf)
(cherry picked from commit 0a805ed1604ef3e9b27f3e22a936a2d439300e9f)

Resolved conflicts:

6 years agokeyserver: Allow use of cURL's default CA store.
Werner Koch [Fri, 4 Oct 2013 06:28:12 +0000 (08:28 +0200)]
keyserver: Allow use of cURL's default CA store.

* keyserver/gpgkeys_curl.c (main): Set CURLOPT_CAINFO only if a file
has been given.
* keyserver/gpgkeys_hkp.c (main): Ditto.

GnuPG-bug-id: 1542
Signed-off-by: Werner Koch <>
(cherry picked from commit e957b9b3f408491f36660499b215aebcf2633a95)

6 years agogpg: Limit the nesting level of I/O filters.
Werner Koch [Fri, 4 Oct 2013 06:20:49 +0000 (08:20 +0200)]
gpg: Limit the nesting level of I/O filters.

* until/iobuf.c (MAX_NESTING_FILTER): New.
(iobuf_push_filter2): Limit the nesting level.

This is a more general fix for the nested compression packet bug.  In
particular this helps g10/import.c:read_block to stop pushing
compression filters onto an iobuf stream.

Signed-off-by: Werner Koch <>
6 years agogpg: Fix bug with deeply nested compressed packets.
Werner Koch [Wed, 2 Oct 2013 07:11:43 +0000 (09:11 +0200)]
gpg: Fix bug with deeply nested compressed packets.

* g10/mainproc.c (MAX_NESTING_DEPTH): New.
(proc_compressed): Return an error code.
(check_nesting): New.
(do_proc_packets): Check packet nesting depth.  Handle errors from

Signed-off-by: Werner Koch <>
6 years agoFix bug in mpi_tdiv_q_2exp.
Werner Koch [Mon, 16 Sep 2013 13:57:01 +0000 (15:57 +0200)]
Fix bug in mpi_tdiv_q_2exp.

* mpi/mpi-internal.h (MPN_COPY_INCR): Make it work.

This bug has been with us since the version 0.0.0 of GnuPG.
Fortunately it only affects an optimized code path which is rarely
used in practice: If the shift size matches the size of a
limb (i.e.. 32 or 64); this is is_prime in primegen.c.  Over there the
Rabin-Miller test may fail with a probability of 2^-31 (that is if the
to be tested prime - 1 has the low 32 bits cleared).  In practice the
probability is even much less because we first do a Fermat test on the
randomly generated candidates which sorts out the majority of
composite numbers.

The bug in MPN_COPY_INCR was found by Sven Bjorn.

Signed-off-by: Werner Koch <>
(back ported from Libgcrypt
 commit 7f7a5ef59962ae1a819b5060f9b781469bfe27d5)

6 years agompicalc: Change copyright notice.
Werner Koch [Wed, 4 Sep 2013 13:25:18 +0000 (15:25 +0200)]
mpicalc: Change copyright notice.


Except for a minor change by David Shaw, all code has been written by
me.  After termination of my copyright assignment, I change that
notice to reflect the new legal state.

Signed-off-by: Werner Koch <>
6 years agogpg: Use 2048 as the default keysize in batch mode.
Werner Koch [Fri, 30 Aug 2013 08:19:14 +0000 (10:19 +0200)]
gpg: Use 2048 as the default keysize in batch mode.

* g10/keygen.c (gen_elg, gen_dsa, gen_rsa): Set default keysize to

Signed-off-by: Werner Koch <>
6 years agogpg: No need to create a trustdb when encrypting with --always-trust.
Werner Koch [Fri, 2 Aug 2013 07:17:52 +0000 (09:17 +0200)]
gpg: No need to create a trustdb when encrypting with --always-trust.

* g10/gpg.c (main): Special case setup_trustdb for --encrypt.

(back ported from commit 498b9a95dc65c43240835d64cc92d8fb43014d53)

Signed-off-by: Werner Koch <>
6 years agoPost release updates.
Werner Koch [Thu, 25 Jul 2013 09:00:04 +0000 (11:00 +0200)]
Post release updates.


6 years agoRelease 1.4.14. gnupg-1.4.14
Werner Koch [Thu, 25 Jul 2013 07:33:23 +0000 (09:33 +0200)]
Release 1.4.14.

6 years agoAutoupdate a translation.
Werner Koch [Thu, 25 Jul 2013 08:44:01 +0000 (10:44 +0200)]
Autoupdate a translation.


6 years agoUpdate Chinese translation.
Jedi Lin [Thu, 25 Jul 2013 07:30:05 +0000 (09:30 +0200)]
Update Chinese translation.

6 years agoUpdate to modern beta release numbering scheme.
Werner Koch [Thu, 25 Jul 2013 07:21:46 +0000 (09:21 +0200)]
Update to modern beta release numbering scheme.

* s/my_/mym4_/.  Add new release building code.

6 years agoPrepare for a forthcoming new algorithm id.
Werner Koch [Thu, 25 Jul 2013 07:11:08 +0000 (09:11 +0200)]
Prepare for a forthcoming new algorithm id.

* include/cipher.h (PUBKEY_ALGO_ECC): New.
* g10/keyid.c (pubkey_letter): Add letter 'C'.

ID 22 will be used for generic ECC, i.e. one which can be used for
ECDSA and ECDH.  The only support in 1.4 will pretty printing the
algorithm id.

6 years agoMitigate a flush+reload cache attack on RSA secret exponents.
Werner Koch [Fri, 19 Jul 2013 11:49:23 +0000 (13:49 +0200)]
Mitigate a flush+reload cache attack on RSA secret exponents.

* mpi/mpi-pow.c (mpi_powm): Always perform the mpi_mul for exponents
hold in secure memory.

The attack is described in a paper to be pusblished at

Flush+Reload: a High Resolution, Low Noise, L3 Cache Side-Channel
Attack by Yuval Yarom and Katrina Falkner. 18 July 2013.

  Flush+Reload is a cache side-channel attack that monitors access to
  data in shared pages. In this paper we demonstrate how to use the
  attack to extract private encryption keys from GnuPG.  The high
  resolution and low noise of the Flush+Reload attack enables a spy
  program to recover over 98% of the bits of the private key in a
  single decryption or signing round. Unlike previous attacks, the
  attack targets the last level L3 cache. Consequently, the spy
  program and the victim do not need to share the execution core of
  the CPU. The attack is not limited to a traditional OS and can be
  used in a virtualised environment, where it can attack programs
  executing in a different VM.

Signed-off-by: Werner Koch <>