linux.git/crypto, branch v4.19.59

crypto: cryptd - Fix skcipher instance memory leak

2019-07-10T07:53:41+00:00

commit 1a0fad630e0b7cff38e7691b28b0517cfbb0633f upstream.

cryptd_skcipher_free() fails to free the struct skcipher_instance
allocated in cryptd_create_skcipher(), leading to a memory leak.  This
is detected by kmemleak on bootup on ARM64 platforms:

 unreferenced object 0xffff80003377b180 (size 1024):
   comm "cryptomgr_probe", pid 822, jiffies 4294894830 (age 52.760s)
   backtrace:
     kmem_cache_alloc_trace+0x270/0x2d0
     cryptd_create+0x990/0x124c
     cryptomgr_probe+0x5c/0x1e8
     kthread+0x258/0x318
     ret_from_fork+0x10/0x1c

Fixes: 4e0958d19bd8 ("crypto: cryptd - Add support for skcipher")
Cc: 
Signed-off-by: Vincent Whitchurch 
Signed-off-by: Herbert Xu 
Signed-off-by: Greg Kroah-Hartman

crypto: user - prevent operating on larval algorithms

2019-07-10T07:53:41+00:00

commit 21d4120ec6f5b5992b01b96ac484701163917b63 upstream.

Michal Suchanek reported [1] that running the pcrypt_aead01 test from
LTP [2] in a loop and holding Ctrl-C causes a NULL dereference of
alg->cra_users.next in crypto_remove_spawns(), via crypto_del_alg().
The test repeatedly uses CRYPTO_MSG_NEWALG and CRYPTO_MSG_DELALG.

The crash occurs when the instance that CRYPTO_MSG_DELALG is trying to
unregister isn't a real registered algorithm, but rather is a "test
larval", which is a special "algorithm" added to the algorithms list
while the real algorithm is still being tested.  Larvals don't have
initialized cra_users, so that causes the crash.  Normally pcrypt_aead01
doesn't trigger this because CRYPTO_MSG_NEWALG waits for the algorithm
to be tested; however, CRYPTO_MSG_NEWALG returns early when interrupted.

Everything else in the "crypto user configuration" API has this same bug
too, i.e. it inappropriately allows operating on larval algorithms
(though it doesn't look like the other cases can cause a crash).

Fix this by making crypto_alg_match() exclude larval algorithms.

[1] https://lkml.kernel.org/r/20190625071624.27039-1-msuchanek@suse.de
[2] https://github.com/linux-test-project/ltp/blob/20190517/testcases/kernel/crypto/pcrypt_aead01.c

Reported-by: Michal Suchanek 
Fixes: a38f7907b926 ("crypto: Add userspace configuration API")
Cc:  # v3.2+
Cc: Steffen Klassert 
Signed-off-by: Eric Biggers 
Signed-off-by: Herbert Xu 
Signed-off-by: Greg Kroah-Hartman

crypto: ccm - fix incompatibility between "ccm" and "ccm_base"

2019-05-22T05:37:43+00:00

commit 6a1faa4a43f5fabf9cbeaa742d916e7b5e73120f upstream.

CCM instances can be created by either the "ccm" template, which only
allows choosing the block cipher, e.g. "ccm(aes)"; or by "ccm_base",
which allows choosing the ctr and cbcmac implementations, e.g.
"ccm_base(ctr(aes-generic),cbcmac(aes-generic))".

However, a "ccm_base" instance prevents a "ccm" instance from being
registered using the same implementations.  Nor will the instance be
found by lookups of "ccm".  This can be used as a denial of service.
Moreover, "ccm_base" instances are never tested by the crypto
self-tests, even if there are compatible "ccm" tests.

The root cause of these problems is that instances of the two templates
use different cra_names.  Therefore, fix these problems by making
"ccm_base" instances set the same cra_name as "ccm" instances, e.g.
"ccm(aes)" instead of "ccm_base(ctr(aes-generic),cbcmac(aes-generic))".

This requires extracting the block cipher name from the name of the ctr
and cbcmac algorithms.  It also requires starting to verify that the
algorithms are really ctr and cbcmac using the same block cipher, not
something else entirely.  But it would be bizarre if anyone were
actually using non-ccm-compatible algorithms with ccm_base, so this
shouldn't break anyone in practice.

Fixes: 4a49b499dfa0 ("[CRYPTO] ccm: Added CCM mode")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers 
Signed-off-by: Herbert Xu 
Signed-off-by: Greg Kroah-Hartman

crypto: gcm - fix incompatibility between "gcm" and "gcm_base"

2019-05-22T05:37:37+00:00

commit f699594d436960160f6d5ba84ed4a222f20d11cd upstream.

GCM instances can be created by either the "gcm" template, which only
allows choosing the block cipher, e.g. "gcm(aes)"; or by "gcm_base",
which allows choosing the ctr and ghash implementations, e.g.
"gcm_base(ctr(aes-generic),ghash-generic)".

However, a "gcm_base" instance prevents a "gcm" instance from being
registered using the same implementations.  Nor will the instance be
found by lookups of "gcm".  This can be used as a denial of service.
Moreover, "gcm_base" instances are never tested by the crypto
self-tests, even if there are compatible "gcm" tests.

The root cause of these problems is that instances of the two templates
use different cra_names.  Therefore, fix these problems by making
"gcm_base" instances set the same cra_name as "gcm" instances, e.g.
"gcm(aes)" instead of "gcm_base(ctr(aes-generic),ghash-generic)".

This requires extracting the block cipher name from the name of the ctr
algorithm.  It also requires starting to verify that the algorithms are
really ctr and ghash, not something else entirely.  But it would be
bizarre if anyone were actually using non-gcm-compatible algorithms with
gcm_base, so this shouldn't break anyone in practice.

Fixes: d00aa19b507b ("[CRYPTO] gcm: Allow block cipher parameter")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers 
Signed-off-by: Herbert Xu 
Signed-off-by: Greg Kroah-Hartman

crypto: crct10dif-generic - fix use via crypto_shash_digest()

2019-05-22T05:37:37+00:00

commit 307508d1072979f4435416f87936f87eaeb82054 upstream.

The ->digest() method of crct10dif-generic reads the current CRC value
from the shash_desc context.  But this value is uninitialized, causing
crypto_shash_digest() to compute the wrong result.  Fix it.

Probably this wasn't noticed before because lib/crc-t10dif.c only uses
crypto_shash_update(), not crypto_shash_digest().  Likewise,
crypto_shash_digest() is not yet tested by the crypto self-tests because
those only test the ahash API which only uses shash init/update/final.

This bug was detected by my patches that improve testmgr to fuzz
algorithms against their generic implementation.

Fixes: 2d31e518a428 ("crypto: crct10dif - Wrap crc_t10dif function all to use crypto transform framework")
Cc:  # v3.11+
Cc: Tim Chen 
Signed-off-by: Eric Biggers 
Signed-off-by: Herbert Xu 
Signed-off-by: Greg Kroah-Hartman

crypto: skcipher - don't WARN on unprocessed data after slow walk step

2019-05-22T05:37:37+00:00

commit dcaca01a42cc2c425154a13412b4124293a6e11e upstream.

skcipher_walk_done() assumes it's a bug if, after the "slow" path is
executed where the next chunk of data is processed via a bounce buffer,
the algorithm says it didn't process all bytes.  Thus it WARNs on this.

However, this can happen legitimately when the message needs to be
evenly divisible into "blocks" but isn't, and the algorithm has a
'walksize' greater than the block size.  For example, ecb-aes-neonbs
sets 'walksize' to 128 bytes and only supports messages evenly divisible
into 16-byte blocks.  If, say, 17 message bytes remain but they straddle
scatterlist elements, the skcipher_walk code will take the "slow" path
and pass the algorithm all 17 bytes in the bounce buffer.  But the
algorithm will only be able to process 16 bytes, triggering the WARN.

Fix this by just removing the WARN_ON().  Returning -EINVAL, as the code
already does, is the right behavior.

This bug was detected by my patches that improve testmgr to fuzz
algorithms against their generic implementation.

Fixes: b286d8b1a690 ("crypto: skcipher - Add skcipher walk interface")
Cc:  # v4.10+
Signed-off-by: Eric Biggers 
Signed-off-by: Herbert Xu 
Signed-off-by: Greg Kroah-Hartman

crypto: chacha20poly1305 - set cra_name correctly

2019-05-22T05:37:36+00:00

commit 5e27f38f1f3f45a0c938299c3a34a2d2db77165a upstream.

If the rfc7539 template is instantiated with specific implementations,
e.g. "rfc7539(chacha20-generic,poly1305-generic)" rather than
"rfc7539(chacha20,poly1305)", then the implementation names end up
included in the instance's cra_name.  This is incorrect because it then
prevents all users from allocating "rfc7539(chacha20,poly1305)", if the
highest priority implementations of chacha20 and poly1305 were selected.
Also, the self-tests aren't run on an instance allocated in this way.

Fix it by setting the instance's cra_name from the underlying
algorithms' actual cra_names, rather than from the requested names.
This matches what other templates do.

Fixes: 71ebc4d1b27d ("crypto: chacha20poly1305 - Add a ChaCha20-Poly1305 AEAD construction, RFC7539")
Cc:  # v4.2+
Cc: Martin Willi 
Signed-off-by: Eric Biggers 
Reviewed-by: Martin Willi 
Signed-off-by: Herbert Xu 
Signed-off-by: Greg Kroah-Hartman

crypto: salsa20 - don't access already-freed walk.iv

2019-05-22T05:37:36+00:00

commit edaf28e996af69222b2cb40455dbb5459c2b875a upstream.

If the user-provided IV needs to be aligned to the algorithm's
alignmask, then skcipher_walk_virt() copies the IV into a new aligned
buffer walk.iv.  But skcipher_walk_virt() can fail afterwards, and then
if the caller unconditionally accesses walk.iv, it's a use-after-free.

salsa20-generic doesn't set an alignmask, so currently it isn't affected
by this despite unconditionally accessing walk.iv.  However this is more
subtle than desired, and it was actually broken prior to the alignmask
being removed by commit b62b3db76f73 ("crypto: salsa20-generic - cleanup
and convert to skcipher API").

Since salsa20-generic does not update the IV and does not need any IV
alignment, update it to use req->iv instead of walk.iv.

Fixes: 2407d60872dd ("[CRYPTO] salsa20: Salsa20 stream cipher")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers 
Signed-off-by: Herbert Xu 
Signed-off-by: Greg Kroah-Hartman

crypto: x86/poly1305 - fix overflow during partial reduction

2019-04-27T07:36:37+00:00

commit 678cce4019d746da6c680c48ba9e6d417803e127 upstream.

The x86_64 implementation of Poly1305 produces the wrong result on some
inputs because poly1305_4block_avx2() incorrectly assumes that when
partially reducing the accumulator, the bits carried from limb 'd4' to
limb 'h0' fit in a 32-bit integer.  This is true for poly1305-generic
which processes only one block at a time.  However, it's not true for
the AVX2 implementation, which processes 4 blocks at a time and
therefore can produce intermediate limbs about 4x larger.

Fix it by making the relevant calculations use 64-bit arithmetic rather
than 32-bit.  Note that most of the carries already used 64-bit
arithmetic, but the d4 -> h0 carry was different for some reason.

To be safe I also made the same change to the corresponding SSE2 code,
though that only operates on 1 or 2 blocks at a time.  I don't think
it's really needed for poly1305_block_sse2(), but it doesn't hurt
because it's already x86_64 code.  It *might* be needed for
poly1305_2block_sse2(), but overflows aren't easy to reproduce there.

This bug was originally detected by my patches that improve testmgr to
fuzz algorithms against their generic implementation.  But also add a
test vector which reproduces it directly (in the AVX2 case).

Fixes: b1ccc8f4b631 ("crypto: poly1305 - Add a four block AVX2 variant for x86_64")
Fixes: c70f4abef07a ("crypto: poly1305 - Add a SSE2 SIMD variant for x86_64")
Cc:  # v4.3+
Cc: Martin Willi 
Cc: Jason A. Donenfeld 
Signed-off-by: Eric Biggers 
Reviewed-by: Martin Willi 
Signed-off-by: Herbert Xu 
Signed-off-by: Greg Kroah-Hartman

crypto: testmgr - skip crc32c context test for ahash algorithms

2019-03-23T19:09:55+00:00

commit eb5e6730db98fcc4b51148b4a819fa4bf864ae54 upstream.

Instantiating "cryptd(crc32c)" causes a crypto self-test failure because
the crypto_alloc_shash() in alg_test_crc32c() fails.  This is because
cryptd(crc32c) is an ahash algorithm, not a shash algorithm; so it can
only be accessed through the ahash API, unlike shash algorithms which
can be accessed through both the ahash and shash APIs.

As the test is testing the shash descriptor format which is only
applicable to shash algorithms, skip it for ahash algorithms.

(Note that it's still important to fix crypto self-test failures even
 for weird algorithm instantiations like cryptd(crc32c) that no one
 would really use; in fips_enabled mode unprivileged users can use them
 to panic the kernel, and also they prevent treating a crypto self-test
 failure as a bug when fuzzing the kernel.)

Fixes: 8e3ee85e68c5 ("crypto: crc32c - Test descriptor context format")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers 
Signed-off-by: Herbert Xu 
Signed-off-by: Greg Kroah-Hartman