Hi Yann,
On Wed, Sep 12, 2018 at 11:50:00AM +0200, Yann Droneaud wrote:
> Hi,
>
> Le mardi 11 septembre 2018 à 20:05 -0700, Eric Biggers a écrit :
> > From: Eric Biggers <ebiggers@xxxxxxxxxx>
> >
> > In commit 9f480faec58c ("crypto: chacha20 - Fix keystream alignment for
> > chacha20_block()"), I had missed that chacha20_block() can be called
> > directly on the buffer passed to get_random_bytes(), which can have any
> > alignment. So, while my commit didn't break anything, it didn't fully
> > solve the alignment problems.
> >
> > Revert my solution and just update chacha20_block() to use
> > put_unaligned_le32(), so the output buffer need not be aligned.
> > This is simpler, and on many CPUs it's the same speed.
> >
> > But, I kept the 'tmp' buffers in extract_crng_user() and
> > _get_random_bytes() 4-byte aligned, since that alignment is actually
> > needed for _crng_backtrack_protect() too.
> >
> > Reported-by: Stephan Müller <smueller@xxxxxxxxxx>
> > Cc: Theodore Ts'o <tytso@xxxxxxx>
> > Signed-off-by: Eric Biggers <ebiggers@xxxxxxxxxx>
> > ---
> > crypto/chacha20_generic.c | 7 ++++---
> > drivers/char/random.c | 24 ++++++++++++------------
> > include/crypto/chacha20.h | 3 +--
> > lib/chacha20.c | 6 +++---
> > 4 files changed, 20 insertions(+), 20 deletions(-)
> >
> > diff --git a/crypto/chacha20_generic.c b/crypto/chacha20_generic.c
> > index e451c3cb6a56..3ae96587caf9 100644
> > --- a/crypto/chacha20_generic.c
> > +++ b/crypto/chacha20_generic.c
> > @@ -18,20 +18,21 @@
> > static void chacha20_docrypt(u32 *state, u8 *dst, const u8 *src,
> > unsigned int bytes)
> > {
> > - u32 stream[CHACHA20_BLOCK_WORDS];
> > + /* aligned to potentially speed up crypto_xor() */
> > + u8 stream[CHACHA20_BLOCK_SIZE] __aligned(sizeof(long));
> >
> > if (dst != src)
> > memcpy(dst, src, bytes);
> >
> > while (bytes >= CHACHA20_BLOCK_SIZE) {
> > chacha20_block(state, stream);
> > - crypto_xor(dst, (const u8 *)stream, CHACHA20_BLOCK_SIZE);
> > + crypto_xor(dst, stream, CHACHA20_BLOCK_SIZE);
> > bytes -= CHACHA20_BLOCK_SIZE;
> > dst += CHACHA20_BLOCK_SIZE;
> > }
> > if (bytes) {
> > chacha20_block(state, stream);
> > - crypto_xor(dst, (const u8 *)stream, bytes);
> > + crypto_xor(dst, stream, bytes);
> > }
> > }
> >
> > diff --git a/drivers/char/random.c b/drivers/char/random.c
> > index bf5f99fc36f1..d22d967c50f0 100644
> > --- a/drivers/char/random.c
> > +++ b/drivers/char/random.c
> > @@ -1003,7 +1003,7 @@ static void extract_crng(__u32 out[CHACHA20_BLOCK_WORDS])
> > * enough) to mutate the CRNG key to provide backtracking protection.
> > */
> > static void _crng_backtrack_protect(struct crng_state *crng,
> > - __u32 tmp[CHACHA20_BLOCK_WORDS], int used)
> > + __u8 tmp[CHACHA20_BLOCK_SIZE], int used)
> > {
> > unsigned long flags;
> > __u32 *s, *d;
> > @@ -1015,14 +1015,14 @@ static void _crng_backtrack_protect(struct crng_state *crng,
> > used = 0;
> > }
> > spin_lock_irqsave(&crng->lock, flags);
> > - s = &tmp[used / sizeof(__u32)];
> > + s = (__u32 *) &tmp[used];
>
> This introduces a alignment issue: tmp is not aligned for __u32, but is
> dereferenced as such later.
>
> > d = &crng->state[4];
> > for (i=0; i < 8; i++)
> > *d++ ^= *s++;
> > spin_unlock_irqrestore(&crng->lock, flags);
> > }
> >
>
I explained this in the patch; the callers ensure the buffer is aligned.
- Eric
