Hit a warning when building QAT, indicating that sz_out might be uninitalized
before use. Looks like if you hit an error path and jump to err: you might find
yourself trying to unmap an arbirarily long dma region. Its safe on intel since
intel defines the invalid dma address as zero, but other arches don't, and if
qat makes its way to one of those, that can cause all sorts of corruption.
Fix is pretty easy, just init sz_out to zero, and gate the unmapping on sz_out
being non-zero
Signed-off-by: Neil Horman <[email protected]>
CC: Herbert Xu <[email protected]>
CC: "David S. Miller" <[email protected]>
CC: Tadeusz Struk <[email protected]>
CC: [email protected] (open list:QAT DRIVER)
---
drivers/crypto/qat/qat_common/qat_algs.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
diff --git a/drivers/crypto/qat/qat_common/qat_algs.c b/drivers/crypto/qat/qat_common/qat_algs.c
index 067402c..35ab752 100644
--- a/drivers/crypto/qat/qat_common/qat_algs.c
+++ b/drivers/crypto/qat/qat_common/qat_algs.c
@@ -667,8 +667,9 @@ static int qat_alg_sgl_to_bufl(struct qat_crypto_instance *inst,
dma_addr_t blp;
dma_addr_t bloutp = 0;
struct scatterlist *sg;
- size_t sz_out, sz = sizeof(struct qat_alg_buf_list) +
- ((1 + n + assoc_n) * sizeof(struct qat_alg_buf));
+ size_t sz_out = 0;
+ size_t sz = sizeof(struct qat_alg_buf_list) +
+ ((1 + n + assoc_n) * sizeof(struct qat_alg_buf));
if (unlikely(!n))
return -EINVAL;
@@ -793,7 +794,7 @@ err:
dma_unmap_single(dev, buflout->bufers[i].addr,
buflout->bufers[i].len,
DMA_BIDIRECTIONAL);
- if (!dma_mapping_error(dev, bloutp))
+ if (sz_out && !dma_mapping_error(dev, bloutp))
dma_unmap_single(dev, bloutp, sz_out, DMA_TO_DEVICE);
kfree(buflout);
}
--
2.1.0
On 06/26/2015 10:56 AM, Neil Horman wrote:
> Hit a warning when building QAT, indicating that sz_out might be uninitalized
> before use. Looks like if you hit an error path and jump to err: you might find
> yourself trying to unmap an arbirarily long dma region. Its safe on intel since
> intel defines the invalid dma address as zero, but other arches don't, and if
> qat makes its way to one of those, that can cause all sorts of corruption.
Hi Neil,
This is a false positive. The sz_out is always initialized before used because
the same condition i.e. if (sgl != sglout && buflout) is in the error path as
well as on the path where is is initialized.
This warning is printed by an old gcc version. If you'll use gcc 4.9 or later
it wont print it.
It didn't make it's way to linux-crypto for whatever reason so resending again.
regards,
T