GitLab

Currently, the ahash API checks the alignment of all key and result
buffers against the algorithm's declared alignmask, and for any
unaligned buffers it falls back to manually aligned temporary buffers.

This is virtually useless, however.  First, since it does not apply to
the message, its effect is much more limited than e.g. is the case for
the alignmask for "skcipher".  Second, the key and result buffers are
given as virtual addresses and cannot (in general) be DMA'ed into, so
drivers end up having to copy to/from them in software anyway.  As a
result it's easy to use memcpy() or the unaligned access helpers.

The crypto_hash_walk_*() helper functions do use the alignmask to align
the message.  But with one exception those are only used for shash
algorithms being exposed via the ahash API, not for native ahashes, and
aligning the message is not required in this case, especially now that
alignmask support has been removed from shash.  The exception is the
n2_core driver, which doesn't set an alignmask.

In any case, no ahash algorithms actually set a nonzero alignmask
anymore.  Therefore, remove support for it from ahash.  The benefit is
that all the code to handle "misaligned" buffers in the ahash API goes
away, reducing the overhead of the ahash API.

This follows the same change that was made to shash.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Currently, the shash API checks the alignment of all message, key, and
digest buffers against the algorithm's declared alignmask, and for any
unaligned buffers it falls back to manually aligned temporary buffers.

This is virtually useless, however.  In the case of the message buffer,
cryptographic hash functions internally operate on fixed-size blocks, so
implementations end up needing to deal with byte-aligned data anyway
because the length(s) passed to ->update might not be divisible by the
block size.  Word-alignment of the message can theoretically be helpful
for CRCs, like what was being done in crc32c-sparc64.  But in practice
it's better for the algorithms to use unaligned accesses or align the
message themselves.  A similar argument applies to the key and digest.

In any case, no shash algorithms actually set a nonzero alignmask
anymore.  Therefore, remove support for it from shash.  The benefit is
that all the code to handle "misaligned" buffers in the shash API goes
away, reducing the overhead of the shash API.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Most shash algorithms don't have custom ->import and ->export functions,
resulting in the memcpy() based default being used.  Yet,
crypto_shash_import() and crypto_shash_export() still make an indirect
call, which is expensive.  Therefore, change how the default import and
export are called to make it so that crypto_shash_import() and
crypto_shash_export() don't do an indirect call in this case.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Fold shash_digest_unaligned() into its only remaining caller.  Also,
avoid a redundant check of CRYPTO_TFM_NEED_KEY by replacing the call to
crypto_shash_init() with shash->init(desc).  Finally, replace
shash_update_unaligned() + shash_final_unaligned() with
shash_finup_unaligned() which does exactly that.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

For an shash algorithm that doesn't implement ->digest, currently
crypto_shash_digest() with aligned input makes 5 indirect calls: 1 to
shash_digest_unaligned(), 1 to ->init, 2 to ->update ('alignmask + 1'
bytes, then the rest), then 1 to ->final.  This is true even if the
algorithm implements ->finup.  This is caused by an unnecessary fallback
to code meant to handle unaligned inputs.  In fact,
crypto_shash_digest() already does the needed alignment check earlier.
Therefore, optimize the number of indirect calls for aligned inputs to 3
when the algorithm implements ->finup.  It remains at 5 when the
algorithm implements neither ->finup nor ->digest.

Similarly, for an shash algorithm that doesn't implement ->finup,
currently crypto_shash_finup() with aligned input makes 4 indirect
calls: 1 to shash_finup_unaligned(), 2 to ->update, and
1 to ->final.  Optimize this to 3 calls.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Some shash algorithms are so simple that they don't have an init_tfm
function.  These can be cloned trivially.  Check this before failing
in crypto_clone_shash.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Checking the config via ifdef incorrectly compiles out the report
functions when CRYPTO_USER is set to =m. Fix it by using IS_ENABLED()
instead.

Fixes: c0f9e01d

 ("crypto: api - Check CRYPTO_USER instead of NET for report")
Signed-off-by: Ondrej Mosnacek <omosnace@redhat.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

This patch adds the helpers crypto_clone_ahash and crypto_clone_shash.
They are the hash-specific counterparts of crypto_clone_tfm.

This allows code paths that cannot otherwise allocate a hash tfm
object to do so.  Once a new tfm has been obtained its key could
then be changed without impacting other users.

Note that only algorithms that implement clone_tfm can be cloned.
However, all keyless hashes can be cloned by simply reusing the
tfm object.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Reviewed-by: Simon Horman <simon.horman@corigine.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Remove the HASH_MAX_STATESIZE limit now that it is unused.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

The report function is currently conditionalised on CONFIG_NET.
As it's only used by CONFIG_CRYPTO_USER, conditionalising on that
instead of CONFIG_NET makes more sense.

This gets rid of a rarely used code-path.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Move all stat code specific to hash into the hash code.

While we're at it, change the stats so that bytes and counts
are always incremented even in case of error.  This allows the
reference counting to be removed as we can now increment the
counters prior to the operation.

After the operation we simply increase the error count if necessary.
This is safe as errors can only occur synchronously (or rather,
the existing code already ignored asynchronous errors which are
only visible to the callback function).
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

kmap_atomic() is used to create short-lived mappings of pages that may
not be accessible via the kernel direct map. This is only needed on
32-bit architectures that implement CONFIG_HIGHMEM, but it can be used
on 64-bit other architectures too, where the returned mapping is simply
the kernel direct address of the page.

However, kmap_atomic() does not support migration on CONFIG_HIGHMEM
configurations, due to the use of per-CPU kmap slots, and so it disables
preemption on all architectures, not just the 32-bit ones. This implies
that all scatterwalk based crypto routines essentially execute with
preemption disabled all the time, which is less than ideal.

So let's switch scatterwalk_map/_unmap and the shash/ahash routines to
kmap_local() instead, which serves a similar purpose, but without the
resulting impact on preemption on architectures that have no need for
CONFIG_HIGHMEM.

Cc: Eric Biggers <ebiggers@kernel.org>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "Elliott, Robert (Servers)" <elliott@hpe.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Previously we limited the maximum alignment mask to 63.  This
is mostly due to stack usage for shash.  This patch introduces
a separate limit for shash algorithms and increases the general
limit to 127 which is the value that we need for DMA allocations
on arm64.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

This reverts commit 22ca9f4a

 because CFI
no longer breaks cross-module function address equality, so
crypto_shash_alg_has_setkey() can now be an inline function like before.

This commit should not be backported to kernels that don't have the new
CFI implementation.
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Add helper function to determine if a given synchronous hash is supported.
Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

crypto_shash_alg_has_setkey() is implemented by testing whether the
.setkey() member of a struct shash_alg points to the default version,
called shash_no_setkey(). As crypto_shash_alg_has_setkey() is a static
inline, this requires shash_no_setkey() to be exported to modules.

Unfortunately, when building with CFI, function pointers are routed
via CFI stubs which are private to each module (or to the kernel proper)
and so this function pointer comparison may fail spuriously.

Let's fix this by turning crypto_shash_alg_has_setkey() into an out of
line function.

Cc: Sami Tolvanen <samitolvanen@google.com>
Cc: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Sami Tolvanen <samitolvanen@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

As said by Linus:

  A symmetric naming is only helpful if it implies symmetries in use.
  Otherwise it's actively misleading.

  In "kzalloc()", the z is meaningful and an important part of what the
  caller wants.

  In "kzfree()", the z is actively detrimental, because maybe in the
  future we really _might_ want to use that "memfill(0xdeadbeef)" or
  something. The "zero" part of the interface isn't even _relevant_.

The main reason that kzfree() exists is to clear sensitive information
that should not be leaked to other future users of the same memory
objects.

Rename kzfree() to kfree_sensitive() to follow the example of the recently
added kvfree_sensitive() and make the intention of the API more explicit.
In addition, memzero_explicit() is used to clear the memory to make sure
that it won't get optimized away by the compiler.

The renaming is done by using the command sequence:

  git grep -w --name-only kzfree |\
  xargs sed -i 's/kzfree/kfree_sensitive/'

followed by some editing of the kfree_sensitive() kerneldoc and adding
a kzfree backward compatibility macro in slab.h.

[akpm@linux-foundation.org: fs/crypto/inline_crypt.c needs linux/slab.h]
[akpm@linux-foundation.org: fix fs/crypto/inline_crypt.c some more]
Suggested-by: Joe Perches <joe@perches.com>
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: David Howells <dhowells@redhat.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Cc: James Morris <jmorris@namei.org>
Cc: "Serge E. Hallyn" <serge@hallyn.com>
Cc: Joe Perches <joe@perches.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: "Jason A . Donenfeld" <Jason@zx2c4.com>
Link: http://lkml.kernel.org/r/20200616154311.12314-3-longman@redhat.com

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Currently the simplest use of the shash API is to use
crypto_shash_digest() to digest a whole buffer.  However, this still
requires allocating a hash descriptor (struct shash_desc).  Many users
don't really want to preallocate one and instead just use a one-off
descriptor on the stack like the following:

	{
		SHASH_DESC_ON_STACK(desc, tfm);
		int err;

		desc->tfm = tfm;

		err = crypto_shash_digest(desc, data, len, out);

		shash_desc_zero(desc);
	}

Wrap this in a new helper function crypto_shash_tfm_digest() that can be
used instead of the above.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

All instances need to have a ->free() method, but people could forget to
set it and then not notice if the instance is never unregistered.  To
help detect this bug earlier, don't allow an instance without a ->free()
method to be registered, and complain loudly if someone tries to do it.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Now that all templates provide a ->create() method which creates an
instance, installs a strongly-typed ->free() method directly to it, and
registers it, the older ->alloc() and ->free() methods in
'struct crypto_template' are no longer used.  Remove them.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Convert shash_free_instance() and its users to the new way of freeing
instances, where a ->free() method is installed to the instance struct
itself.  This replaces the weakly-typed method crypto_template::free().

This will allow removing support for the old way of freeing instances.

Also give shash_free_instance() a more descriptive name to reflect that
it's only for instances with a single spawn, not for any instance.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Add support to shash and ahash for the new way of freeing instances
(already used for skcipher, aead, and akcipher) where a ->free() method
is installed to the instance struct itself.  These methods are more
strongly-typed than crypto_template::free(), which they replace.

This will allow removing support for the old way of freeing instances.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Remove lots of helper functions that were previously used for
instantiating crypto templates, but are now unused:

- crypto_get_attr_alg() and similar functions looked up an inner
  algorithm directly from a template parameter.  These were replaced
  with getting the algorithm's name, then calling crypto_grab_*().

- crypto_init_spawn2() and similar functions initialized a spawn, given
  an algorithm.  Similarly, these were replaced with crypto_grab_*().

- crypto_alloc_instance() and similar functions allocated an instance
  with a single spawn, given the inner algorithm.  These aren't useful
  anymore since crypto_grab_*() need the instance allocated first.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Currently, shash spawns are initialized by using shash_attr_alg() or
crypto_alg_mod_lookup() to look up the shash algorithm, then calling
crypto_init_shash_spawn().

This is different from how skcipher, aead, and akcipher spawns are
initialized (they use crypto_grab_*()), and for no good reason.  This
difference introduces unnecessary complexity.

The crypto_grab_*() functions used to have some problems, like not
holding a reference to the algorithm and requiring the caller to
initialize spawn->base.inst.  But those problems are fixed now.

So, let's introduce crypto_grab_shash() so that we can convert all
templates to the same way of initializing their spawns.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Some of the algorithm unregistration functions return -ENOENT when asked
to unregister a non-registered algorithm, while others always return 0
or always return void.  But no users check the return value, except for
two of the bulk unregistration functions which print a message on error
but still always return 0 to their caller, and crypto_del_alg() which
calls crypto_unregister_instance() which always returns 0.

Since unregistering a non-registered algorithm is always a kernel bug
but there isn't anything callers should do to handle this situation at
runtime, let's simplify things by making all the unregistration
functions return void, and moving the error message into
crypto_unregister_alg() and upgrading it to a WARN().
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

The shash interface supports a dynamic descsize field because of
the presence of fallbacks (it's just padlock-sha actually, perhaps
we can remove it one day).  As it is the API does not verify the
setting of descsize at all.  It is up to the individual algorithms
to ensure that descsize does not exceed the specified maximum value
of HASH_MAX_DESCSIZE (going above would cause stack corruption).

In order to allow the API to impose this limit directly, this patch
adds init_tfm/exit_tfm hooks to the shash_alg structure.  We can
then verify the descsize setting in the API directly.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Reviewed-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

The essiv and hmac templates refuse to use any hash algorithm that has a
->setkey() function, which includes not just algorithms that always need
a key, but also algorithms that optionally take a key.

Previously the only optionally-keyed hash algorithms in the crypto API
were non-cryptographic algorithms like crc32, so this didn't really
matter.  But that's changed with BLAKE2 support being added.  BLAKE2
should work with essiv and hmac, just like any other cryptographic hash.

Fix this by allowing the use of both algorithms without a ->setkey()
function and algorithms that have the OPTIONAL_KEY flag set.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Based on 1 normalized pattern(s):

  this program is free software you can redistribute it and or modify
  it under the terms of the gnu general public license as published by
  the free software foundation either version 2 of the license or at
  your option any later version

extracted by the scancode license scanner the SPDX license identifier

  GPL-2.0-or-later

has been chosen to replace the boilerplate/reference in 3029 file(s).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Allison Randal <allison@lohutok.net>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190527070032.746973796@linutronix.de

Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

The flags field in 'struct shash_desc' never actually does anything.
The only ostensibly supported flag is CRYPTO_TFM_REQ_MAY_SLEEP.
However, no shash algorithm ever sleeps, making this flag a no-op.

With this being the case, inevitably some users who can't sleep wrongly
pass MAY_SLEEP.  These would all need to be fixed if any shash algorithm
actually started sleeping.  For example, the shash_ahash_*() functions,
which wrap a shash algorithm with the ahash API, pass through MAY_SLEEP
from the ahash API to the shash API.  However, the shash functions are
called under kmap_atomic(), so actually they're assumed to never sleep.

Even if it turns out that some users do need preemption points while
hashing large buffers, we could easily provide a helper function
crypto_shash_update_large() which divides the data into smaller chunks
and calls crypto_shash_update() and cond_resched() for each chunk.  It's
not necessary to have a flag in 'struct shash_desc', nor is it necessary
to make individual shash algorithms aware of this at all.

Therefore, remove shash_desc::flags, and document that the
crypto_shash_*() functions can be called from any context.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

The crypto_yield() in shash_ahash_digest() occurs after the entire
digest operation already happened, so there's no real point.  Remove it.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

shash_ahash_digest(), which is the ->digest() method for ahash tfms that
use an shash algorithm, has an optimization where crypto_shash_digest()
is called if the data is in a single page.  But an off-by-one error
prevented this path from being taken unless the user happened to provide
extra data in the scatterlist.  Fix it.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

crypto_init_shash_ops_async() only gives the ahash tfm non-NULL
->export() and ->import() if the underlying shash alg has these
non-NULL.  This doesn't make sense because when an shash algorithm is
registered, shash_prepare_alg() sets a default ->export() and ->import()
if the implementor didn't provide them.  And elsewhere it's assumed that
all shash algs and ahash tfms have non-NULL ->export() and ->import().

Therefore, remove these unnecessary, always-true conditions.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Prevent registering shash algorithms that implement ->export() but not
->import(), or ->import() but not ->export().  Such cases don't make
sense and could confuse the check that shash_prepare_alg() does for just
->export().

I don't believe this affects any existing algorithms; this is just
preventing future mistakes.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Some algorithms have a ->setkey() method that is not atomic, in the
sense that setting a key can fail after changes were already made to the
tfm context.  In this case, if a key was already set the tfm can end up
in a state that corresponds to neither the old key nor the new key.

It's not feasible to make all ->setkey() methods atomic, especially ones
that have to key multiple sub-tfms.  Therefore, make the crypto API set
CRYPTO_TFM_NEED_KEY if ->setkey() fails and the algorithm requires a
key, to prevent the tfm from being used until a new key is set.

Note: we can't set CRYPTO_TFM_NEED_KEY for OPTIONAL_KEY algorithms, so
->setkey() for those must nevertheless be atomic.  That's fine for now
since only the crc32 and crc32c algorithms set OPTIONAL_KEY, and it's
not intended that OPTIONAL_KEY be used much.

[Cc stable mainly because when introducing the NEED_KEY flag I changed
 AF_ALG to rely on it; and unlike in-kernel crypto API users, AF_ALG
 previously didn't have this problem.  So these "incompletely keyed"
 states became theoretically accessible via AF_ALG -- though, the
 opportunities for causing real mischief seem pretty limited.]

Fixes: 9fa68f62

 ("crypto: hash - prevent using keyed hashes without setting key")
Cc: stable@vger.kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

There have been a pretty ridiculous number of issues with initializing
the report structures that are copied to userspace by NETLINK_CRYPTO.
Commit 4473710d ("crypto: user - Prepare for CRYPTO_MAX_ALG_NAME
expansion") replaced some strncpy()s with strlcpy()s, thereby
introducing information leaks.  Later two other people tried to replace
other strncpy()s with strlcpy() too, which would have introduced even
more information leaks:

    - https://lore.kernel.org/patchwork/patch/954991/
    - https://patchwork.kernel.org/patch/10434351/

Commit cac5818c

 ("crypto: user - Implement a generic crypto
statistics") also uses the buggy strlcpy() approach and therefore leaks
uninitialized memory to userspace.  A fix was proposed, but it was
originally incomplete.

Seeing as how apparently no one can get this right with the current
approach, change all the reporting functions to:

- Start by memsetting the report structure to 0.  This guarantees it's
  always initialized, regardless of what happens later.
- Initialize all strings using strscpy().  This is safe after the
  memset, ensures null termination of long strings, avoids unnecessary
  work, and avoids the -Wstringop-truncation warnings from gcc.
- Use sizeof(var) instead of sizeof(type).  This is more robust against
  copy+paste errors.

For simplicity, also reuse the -EMSGSIZE return value from nla_put().
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

In the quest to remove all stack VLA usage from the kernel[1], this uses
the newly defined max alignment to perform unaligned hashing to avoid
VLAs, and drops the helper function while adding sanity checks on the
resulting buffer sizes. Additionally, the __aligned_largest macro is
removed since this helper was the only user.

[1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com

Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

In the quest to remove all stack VLA usage from the kernel[1], this
removes the VLAs in SHASH_DESC_ON_STACK (via crypto_shash_descsize())
by using the maximum allowable size (which is now more clearly captured
in a macro), along with a few other cases. Similar limits are turned into
macros as well.

A review of existing sizes shows that SHA512_DIGEST_SIZE (64) is the
largest digest size and that sizeof(struct sha3_state) (360) is the
largest descriptor size. The corresponding maximums are reduced.

[1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com

Signed-off-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Currently, almost none of the keyed hash algorithms check whether a key
has been set before proceeding.  Some algorithms are okay with this and
will effectively just use a key of all 0's or some other bogus default.
However, others will severely break, as demonstrated using
"hmac(sha3-512-generic)", the unkeyed use of which causes a kernel crash
via a (potentially exploitable) stack buffer overflow.

A while ago, this problem was solved for AF_ALG by pairing each hash
transform with a 'has_key' bool.  However, there are still other places
in the kernel where userspace can specify an arbitrary hash algorithm by
name, and the kernel uses it as unkeyed hash without checking whether it
is really unkeyed.  Examples of this include:

    - KEYCTL_DH_COMPUTE, via the KDF extension
    - dm-verity
    - dm-crypt, via the ESSIV support
    - dm-integrity, via the "internal hash" mode with no key given
    - drbd (Distributed Replicated Block Device)

Th...

Because the HMAC template didn't check that its underlying hash
algorithm is unkeyed, trying to use "hmac(hmac(sha3-512-generic))"
through AF_ALG or through KEYCTL_DH_COMPUTE resulted in the inner HMAC
being used without having been keyed, resulting in sha3_update() being
called without sha3_init(), causing a stack buffer overflow.

This is a very old bug, but it seems to have only started causing real
problems when SHA-3 support was added (requires CONFIG_CRYPTO_SHA3)
because the innermost hash's state is ->import()ed from a zeroed buffer,
and it just so happens that other hash algorithms are fine with that,
but SHA-3 is not.  However, there could be arch or hardware-dependent
hash algorithms also affected; I couldn't test everything.

Fix the bug by introducing a function crypto_shash_alg_has_setkey()
which tests whether a shash algorithm is keyed.  Then update the HMAC
template to require that its underlying hash algorithm is unkeyed....

The shash ahash digest adaptor function may crash if given a
zero-length input together with a null SG list.  This is because
it tries to read the SG list before looking at the length.

This patch fixes it by checking the length first.

Cc: <stable@vger.kernel.org>
Reported-by: Stephan Müller<smueller@chronox.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Tested-by: Stephan Müller <smueller@chronox.de>