This driver has been implicitly relying on kmalloc alignment
to be sufficient for DMA. This may no longer be the case with
upcoming arm64 changes.
This patch changes it to explicitly request DMA alignment from
the Crypto API.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
dma_map_sg return 0 on error, fix the error check and return -EIO to
caller.
Cc: Thara Gopinath <thara.gopinath@gmail.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: linux-crypto@vger.kernel.org
Cc: linux-arm-msm@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Jack Wang <jinpu.wang@ionos.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Pointer alg points to sub field of tmpl, it
is dereferenced after tmpl is freed. Fix
this by accessing alg before free tmpl.
Fixes: ec8f5d8f ("crypto: qce - Qualcomm crypto engine driver")
Signed-off-by: Chengfeng Ye <cyeaa@connect.ust.hk>
Acked-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Pointer base points to sub field of tmpl, it
is dereferenced after tmpl is freed. Fix
this by accessing base before free tmpl.
Fixes: ec8f5d8f ("crypto: qce - Qualcomm crypto engine driver")
Signed-off-by: Chengfeng Ye <cyeaa@connect.ust.hk>
Acked-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Pointer alg points to sub field of tmpl, it
is dereferenced after tmpl is freed. Fix
this by accessing alg before free tmpl.
Fixes: 9363efb4 ("crypto: qce - Add support for AEAD algorithms")
Signed-off-by: Chengfeng Ye <cyeaa@connect.ust.hk>
Acked-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Fix to return a negative error code from the error handling
case instead of 0, as done elsewhere in this function.
Fixes: 1339a7c3ba ("crypto: qce: skcipher: Fix incorrect sg count for dma transfers")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Reviewed-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use the sg count returned by dma_map_sg to call into
dmaengine_prep_slave_sg rather than using the original sg count. dma_map_sg
can merge consecutive sglist entries, thus making the original sg count
wrong. This is a fix for memory coruption issues observed while testing
encryption/decryption of large messages using libkcapi framework.
Patch has been tested further by running full suite of tcrypt.ko tests
including fuzz tests.
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Fix to return negative error code from the error handling
cases instead of 0.
Fixes: 9363efb418 ("crypto: qce - Add support for AEAD algorithms")
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Qualcomm crypto engine does not handle the following scenarios and
will issue an abort. In such cases, pass on the transformation to
a fallback algorithm.
- DES3 algorithms with all three keys same.
- AES192 algorithms.
- 0 length messages.
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Remove various redundant checks in qce_auth_cfg. Also allow qce_auth_cfg
to take auth_size as a parameter which is a required setting for ccm(aes)
algorithms
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
rf4309 is the specification that uses aes ccm algorithms with IPsec
security packets. Add a submode to identify rfc4309 ccm(aes) algorithm
in the crypto driver.
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Qualcomm crypto engine allows for IV registers and status register
to be concatenated to the output. This option is enabled by setting the
RESULTS_DUMP field in GOPROC register. This is useful for most of the
algorithms to either retrieve status of operation or in case of
authentication algorithms to retrieve the mac. But for ccm
algorithms, the mac is part of the output stream and not retrieved
from the IV registers, thus needing a separate buffer to retrieve it.
Make enabling RESULTS_DUMP field optional so that algorithms can choose
whether or not to enable the option.
Note that in this patch, the enabled algorithms always choose
RESULTS_DUMP to be enabled. But later with the introduction of ccm
algorithms, this changes.
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
MAC_FAILED gets set in the status register if authenthication fails
for ccm algorithms(during decryption). Add support to catch and flag
this error.
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
totallen is used to get the size of the data to be transformed.
This is also available via nbytes or cryptlen in the qce_sha_reqctx
and qce_cipher_ctx. Similarly offset convey nothing for the supported
encryption and authentication transformations and is always 0.
Remove these two redundant parameters in qce_start.
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
src_table is unused and hence remove it from struct qce_cipher_reqctx
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Set the register REG_ENCR_XTS_DU_SIZE to cryptlen for AES XTS
transformation. Anything else causes the engine to return back
wrong results.
Acked-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The following are the conditions for requesting AES fallback cipher.
- AES-192
- AES-XTS request with len <= 512 byte (Allow messages of length
less than 512 bytes for all other AES encryption algorithms other
than AES XTS)
- AES-XTS request with len > QCE_SECTOR_SIZE and is not a multiple
of it
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
ECB transformations do not have an IV and hence set the ivsize to 0 for
ecb(aes).
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
ECB/CBC encryption/decryption requires the data to be blocksize aligned.
Crypto engine hangs on non-block sized operations for these algorithms.
Return invalid data if data size is not blocksize aligned for these
algorithms.
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Crypto engine BAM dma does not support 0 length data. Return unsupported
if zero length messages are passed for transformation.
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Return unsupported if any three keys are same for DES3 algorithms
since CE does not support this and the operation causes the engine to
hang.
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Crypto engine does not support key1 = key2 for AES XTS algorithm; the
operation hangs the engines. Return -EINVAL in case key1 and key2 are the
same.
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
If the available data to transfer is exactly a multiple of block size, save
the last block to be transferred in qce_ahash_final (with the last block
bit set) if this is indeed the end of data stream. If not this saved block
will be transferred as part of next update. If this block is not held back
and if this is indeed the end of data stream, the digest obtained will be
wrong since qce_ahash_final will see that rctx->buflen is 0 and return
doing nothing which in turn means that a digest will not be copied to the
destination result buffer. qce_ahash_final cannot be made to alter this
behavior and allowed to proceed if rctx->buflen is 0 because the crypto
engine BAM does not allow for zero length transfers.
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Export and import interfaces save and restore partial transformation
states. The partial states were being stored and restored in struct
sha1_state for sha1/hmac(sha1) transformations and sha256_state for
sha256/hmac(sha256) transformations.This led to a bunch of corner cases
where improper state was being stored and restored. A few of the corner
cases that turned up during testing are:
- wrong byte_count restored if export/import is called twice without h/w
transaction in between
- wrong buflen restored back if the pending buffer
length is exactly the block size.
- wrong state restored if buffer length is 0.
To fix these issues, save and restore the partial transformation state
using the newly introduced qce_sha_saved_state struct. This ensures that
all the pieces required to properly restart the transformation is captured
and restored back
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Partial hash was being copied into the final result buffer without the
entire message block processed. Depending on how the end user processes
this result buffer, errors vary from result buffer corruption to result
buffer poisoing. Fix this issue by ensuring that only the final hash value
is copied into the result buffer.
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Currently <crypto/sha.h> contains declarations for both SHA-1 and SHA-2,
and <crypto/sha3.h> contains declarations for SHA-3.
This organization is inconsistent, but more importantly SHA-1 is no
longer considered to be cryptographically secure. So to the extent
possible, SHA-1 shouldn't be grouped together with any of the other SHA
versions, and usage of it should be phased out.
Therefore, split <crypto/sha.h> into two headers <crypto/sha1.h> and
<crypto/sha2.h>, and make everyone explicitly specify whether they want
the declarations for SHA-1, SHA-2, or both.
This avoids making the SHA-1 declarations visible to files that don't
want anything to do with SHA-1. It also prepares for potentially moving
sha1.h into a new insecure/ or dangerous/ directory.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Acked-by: Ard Biesheuvel <ardb@kernel.org>
Acked-by: Jason A. Donenfeld <Jason@zx2c4.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The header file algapi.h includes skbuff.h unnecessarily since
all we need is a forward declaration for struct sk_buff. This
patch removes that inclusion.
Unfortunately skbuff.h pulls in a lot of things and drivers over
the years have come to rely on it so this patch adds a lot of
missing inclusions that result from this.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Even though the qce driver implements asynchronous versions of ecb(aes),
cbc(aes)and xts(aes), the fallbacks it allocates are required to be
synchronous. Given that SIMD based software implementations are usually
asynchronous as well, even though they rarely complete asynchronously
(this typically only happens in cases where the request was made from
softirq context, while SIMD was already in use in the task context that
it interrupted), these implementations are disregarded, and either the
generic C version or another table based version implemented in assembler
is selected instead.
Since falling back to synchronous AES is not only a performance issue, but
potentially a security issue as well (due to the fact that table based AES
is not time invariant), let's fix this, by allocating an ordinary skcipher
as the fallback, and invoke it with the completion routine that was given
to the outer request.
While at it, remove the pointless memset() from qce_skcipher_init(), and
remove the call to it qce_skcipher_init_fallback().
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Crypto test driver's test_ahash_speed calls crypto_ahash_update and
crypto_ahash_final APIs repeatedly for all the available test vector
buffer lengths.
if we mark the end for scatterlist based on the current vector size then
the subsequent vectors might fail if the later buffer lengths are higher.
To avoid this, in qce do not mark the end of scatterlist in update API,
the qce_ahash_async_req_handle API already takes care of this copying
right amount of buffer from the request scatter list.
Signed-off-by: Sivaprakash Murugesan <sivaprak@codeaurora.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
crypto testmgr deliberately corrupts the request context while passing
vectors to the import. This is to make sure that drivers do not rely on
request but they take all the necessary input from io vec passed to it.
qce casts the request context from request parameter, since it is corrupted
the sub squent hash request fails and qce hangs.
To avoid this re-initialize request context on import. The qce import
API alreasy takes care of taking the input vectors from passed io vec.
Signed-off-by: Sivaprakash Murugesan <sivaprak@codeaurora.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
crypto test module passes zero length vectors as test input to sha-1 and
sha-256. To provide correct output for these vectors, hash zero support
has been added as in other crypto drivers.
Signed-off-by: Sivaprakash Murugesan <sivaprak@codeaurora.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The CONFIG_CRYPTO_DEV_QCE_SOFT_THRESHOLD symbol was renamed during
development, but the stringify reference in the parameter description
sneaked by unnoticed.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
QCE hangs when presented with an AES-XTS request whose length is larger
than QCE_SECTOR_SIZE (512-bytes), and is not a multiple of it. Let the
fallback cipher handle them.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Process small blocks using the fallback cipher, as a workaround for an
observed failure (DMA-related, apparently) when computing the GCM ghash
key. This brings a speed gain as well, since it avoids the latency of
using the hardware engine to process small blocks.
Using software for all 16-byte requests would be enough to make GCM
work, but to increase performance, a larger threshold would be better.
Measuring the performance of supported ciphers with openssl speed,
software matches hardware at around 768-1024 bytes.
Considering the 256-bit ciphers, software is 2-3 times faster than qce
at 256-bytes, 30% faster at 512, and about even at 768-bytes. With
128-bit keys, the break-even point would be around 1024-bytes.
This adds the 'aes_sw_max_len' parameter, to set the largest request
length processed by the software fallback. Its default is being set to
512 bytes, a little lower than the break-even point, to balance the cost
in CPU usage.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The qce crypto driver appends an extra entry to the dst sgl, to maintain
private state information.
When the gcm driver sends requests to the ctr skcipher, it passes the
authentication tag after the actual crypto payload, but it must not be
touched.
Commit 1336c2221bee ("crypto: qce - save a sg table slot for result
buf") limited the destination sgl to avoid overwriting the
authentication tag but it assumed the tag would be in a separate sgl
entry.
This is not always the case, so it is better to limit the length of the
destination buffer to req->cryptlen before appending the result buf.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The CRYPTO_TFM_RES_BAD_KEY_LEN flag was apparently meant as a way to
make the ->setkey() functions provide more information about errors.
However, no one actually checks for this flag, which makes it pointless.
Also, many algorithms fail to set this flag when given a bad length key.
Reviewing just the generic implementations, this is the case for
aes-fixed-time, cbcmac, echainiv, nhpoly1305, pcrypt, rfc3686, rfc4309,
rfc7539, rfc7539esp, salsa20, seqiv, and xcbc. But there are probably
many more in arch/*/crypto/ and drivers/crypto/.
Some algorithms can even set this flag when the key is the correct
length. For example, authenc and authencesn set it when the key payload
is malformed in any way (not just a bad length), the atmel-sha and ccree
drivers can set it if a memory allocation fails, and the chelsio driver
sets it for bad auth tag lengths, not just bad key lengths.
So even if someone actually wanted to start checking this flag (which
seems unlikely, since it's been unused for a long time), there would be
a lot of work needed to get it working correctly. But it would probably
be much better to go back to the drawing board and just define different
return values, like -EINVAL if the key is invalid for the algorithm vs.
-EKEYREJECTED if the key was rejected by a policy like "no weak keys".
That would be much simpler, less error-prone, and easier to test.
So just remove this flag.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Reviewed-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Allow the user to choose whether to build support for all algorithms
(default), hashes-only, or skciphers-only.
The QCE engine does not appear to scale as well as the CPU to handle
multiple crypto requests. While the ipq40xx chips have 4-core CPUs, the
QCE handles only 2 requests in parallel.
Ipsec throughput seems to improve when disabling either family of
algorithms, sharing the load with the CPU. Enabling skciphers-only
appears to work best.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Adjust cra_flags to add CRYPTO_NEED_FALLBACK only for AES ciphers, where
AES-192 is not handled by the qce hardware, and don't allocate & free
the fallback skcipher for other algorithms.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Update the IV after the completion of each cipher operation.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
When ctr-aes-qce is used for gcm-mode, an extra sg entry for the
authentication tag is present, causing trouble when the qce driver
prepares the dst-results sg table for dma.
It computes the number of entries needed with sg_nents_for_len, leaving
out the tag entry. Then it creates a sg table with that number plus
one, used to store a result buffer.
When copying the sg table, there's no limit to the number of entries
copied, so the extra slot is filled with the authentication tag sg.
When the driver tries to add the result sg, the list is full, and it
returns EINVAL.
By limiting the number of sg entries copied to the dest table, the slot
for the result buffer is guaranteed to be unused.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
XTS-mode uses two keys, so the keysizes should be doubled in
skcipher_def, and halved when checking if it is AES-128/192/256.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Set blocksize of ctr-aes-qce to 1, so it can operate as a stream cipher,
adding the definition for chucksize instead, where the underlying block
size belongs.
Signed-off-by: Eneas U de Queiroz <cotequeiroz@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Commit 7a7ffe65c8 ("crypto: skcipher - Add top-level skcipher interface")
dated 20 august 2015 introduced the new skcipher API which is supposed to
replace both blkcipher and ablkcipher. While all consumers of the API have
been converted long ago, some producers of the ablkcipher remain, forcing
us to keep the ablkcipher support routines alive, along with the matching
code to expose [a]blkciphers via the skcipher API.
So switch this driver to the skcipher API, allowing us to finally drop the
ablkcipher code in the near future.
Reviewed-by: Stanimir Varbanov <stanimir.varbanov@linaro.org>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Herbert Xu