Re: 答复: [PATCH 1/3] crypto: skcipher - fix crash flushing dcache in error path

2018-07-25 Thread Eric Biggers
Hi GaoKui,

On Thu, Jul 26, 2018 at 02:44:30AM +, gaokui (A) wrote:
> Hi, Eric,
>   Thanks for your reply.
> 
>   I have run  your program on an original kernel and it reproduced the 
> crash. And I also run the program on a kernel with our patch, but there was 
> no crash. 
> 
>   I think the reason of the crash is  the parameter buffer is aligned 
> with the page .  So the address of the parameter buffer starts at the 
> beginning of the page, which making "walk->offset = 0" and generating the 
> crash. I add some logs in "scatterwalk_pagedone()" to print the value of 
> walk->offset, and the log before the crash shows that "walk->offset = 0".
> 
>   And I do not understand why "walk->offset = 0" means no data to be 
> processed. In the structure " scatterlist", the member "offset" represents 
> the offset of the buffer in the page, and the member length represents the 
> length of the buffer. In function "af_alg_make_sg()", if a buffer occupies 
> more than one pages, the offset will also be set to 0 in the second and 
> following pages. And In function scatterwalk_done(), walk->offset = 0 will 
> also allow to call "scatterwalk_pagedone()". So I think that when 
> "walk->offset = 0" the page  needs to be flushed  as well. 
> 
> BRs
> GaoKui
> 

Did you test my patches or just yours?  Your patch fixes the crash, but I don't
agree that it's the best fix.  What you're missing is that walk->offset has
already been increased by scatterwalk_advance() to the offset of the *end* of
the data region processed.  Hence, walk->offset = 0 implies that 0 bytes were
processed (as walk->offset must have been 0 initially, then had 0 added to it),
which I think isn't meant to be a valid case.  And in particular it does *not*
make sense to flush any page when 0 bytes were processed.

Note that this could also be a problem for empty scatterlist elements, but
AFAICS the scatterlist walk code doesn't actually support those when the total
length isn't 0.  I think that needs improvement too, but AFAICS other changes
would be needed to properly fix that limitation, and you apparently cannot
generate empty scatterlist elements via AF_ALG anyway so only in-kernel users
would be affected.

- Eric


答复: [PATCH 1/3] crypto: skcipher - fix crash flushing dcache in error path

2018-07-25 Thread gaokui (A)
Hi, Eric,
Thanks for your reply.

I have run  your program on an original kernel and it reproduced the 
crash. And I also run the program on a kernel with our patch, but there was no 
crash. 

I think the reason of the crash is  the parameter buffer is aligned 
with the page .  So the address of the parameter buffer starts at the beginning 
of the page, which making "walk->offset = 0" and generating the crash. I add 
some logs in "scatterwalk_pagedone()" to print the value of walk->offset, and 
the log before the crash shows that "walk->offset = 0".

And I do not understand why "walk->offset = 0" means no data to be 
processed. In the structure " scatterlist", the member "offset" represents the 
offset of the buffer in the page, and the member length represents the length 
of the buffer. In function "af_alg_make_sg()", if a buffer occupies more than 
one pages, the offset will also be set to 0 in the second and following pages. 
And In function scatterwalk_done(), walk->offset = 0 will also allow to call 
"scatterwalk_pagedone()". So I think that when "walk->offset = 0" the page  
needs to be flushed  as well. 

BRs
GaoKui

> -邮件原件-
> 发件人: linux-crypto-ow...@vger.kernel.org [mailto:linux-crypto-
> ow...@vger.kernel.org] 代表 Eric Biggers
> 发送时间: 2018年7月24日 1:55
> 收件人: linux-crypto@vger.kernel.org; Herbert Xu
> 
> 抄送: Liuchao (H) ; 罗新强
> ; gaokui (A) ; Eric
> Biggers 
> 主题: [PATCH 1/3] crypto: skcipher - fix crash flushing dcache in error path
> 
> From: Eric Biggers 
> 
> scatterwalk_done() is only meant to be called after a nonzero number of
> bytes have been processed, since scatterwalk_pagedone() will flush the
> dcache of the *previous* page.  But in the error case of
> skcipher_walk_done(), e.g. if the input wasn't an integer number of blocks,
> scatterwalk_done() was actually called after advancing 0 bytes.
> This caused a crash ("BUG: unable to handle kernel paging request") during
> '!PageSlab(page)' on architectures like arm and arm64 that define
> ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE, provided that the input was
> page-aligned as in that case walk->offset == 0.
> 
> Fix it by reorganizing skcipher_walk_done() to skip the
> scatterwalk_advance() and scatterwalk_done() if an error has occurred.
> 
> This bug was found by syzkaller fuzzing.
> 
> Reproducer, assuming ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE:
> 
>   #include 
>   #include 
>   #include 
> 
>   int main()
>   {
>   struct sockaddr_alg addr = {
>   .salg_type = "skcipher",
>   .salg_name = "cbc(aes-generic)",
>   };
>   char buffer[4096] __attribute__((aligned(4096))) = { 0 };
>   int fd;
> 
>   fd = socket(AF_ALG, SOCK_SEQPACKET, 0);
>   bind(fd, (void *), sizeof(addr));
>   setsockopt(fd, SOL_ALG, ALG_SET_KEY, buffer, 16);
>   fd = accept(fd, NULL, NULL);
>   write(fd, buffer, 15);
>   read(fd, buffer, 15);
>   }
> 
> Reported-by: Liu Chao 
> Fixes: b286d8b1a690 ("crypto: skcipher - Add skcipher walk interface")
> Cc:  # v4.10+
> Signed-off-by: Eric Biggers 
> ---
>  crypto/skcipher.c | 53 ---
>  1 file changed, 27 insertions(+), 26 deletions(-)
> 
> diff --git a/crypto/skcipher.c b/crypto/skcipher.c index
> 7d6a49fe3047..5f7017b36d75 100644
> --- a/crypto/skcipher.c
> +++ b/crypto/skcipher.c
> @@ -95,7 +95,7 @@ static inline u8 *skcipher_get_spot(u8 *start,
> unsigned int len)
>   return max(start, end_page);
>  }
> 
> -static int skcipher_done_slow(struct skcipher_walk *walk, unsigned int
> bsize)
> +static void skcipher_done_slow(struct skcipher_walk *walk, unsigned int
> +bsize)
>  {
>   u8 *addr;
> 
> @@ -103,23 +103,24 @@ static int skcipher_done_slow(struct
> skcipher_walk *walk, unsigned int bsize)
>   addr = skcipher_get_spot(addr, bsize);
>   scatterwalk_copychunks(addr, >out, bsize,
>  (walk->flags & SKCIPHER_WALK_PHYS) ? 2 : 1);
> - return 0;
>  }
> 
>  int skcipher_walk_done(struct skcipher_walk *walk, int err)  {
> - unsigned int n = walk->nbytes - err;
> - unsigned int nbytes;
> -
> - nbytes = walk->total - n;
> -
> - if (unlikely(err < 0)) {
> - nbytes = 0;
> - n = 0;
> - } else if (likely(!(walk->flags & (SKCIPHER_WALK_PHYS |
> -SKCIPHER_WALK_SLOW |
> -SKCIPHER_WALK_COPY |
> -SKCIPHER_WALK_DIFF {
> + unsigned int n; /* bytes processed */
> + bool more;
> +
> + if (unlikely(err < 0))
> + goto finish;
> +
> + n = walk->nbytes - err;
> + walk->total -= n;
> + more = (walk->total != 0);
> +
> + if (likely(!(walk->flags & (SKCIPHER_WALK_PHYS |
> + SKCIPHER_WALK_SLOW |
> +