Re: [PATCH] bpf: Use kmalloc_size_roundup() to match ksize() usage

[Kees Cook <keescook@chromium.org>]

On Tue, Oct 18, 2022 at 11:07:38AM -0700, sdf@google.com wrote:
> On 10/18, Kees Cook wrote:
> > Round up allocations with kmalloc_size_roundup() so that the verifier's
> > use of ksize() is always accurate and no special handling of the memory
> > is needed by KASAN, UBSAN_BOUNDS, nor FORTIFY_SOURCE. Pass the new size
> > information back up to callers so they can use the space immediately,
> > so array resizing to happen less frequently as well. Explicitly zero
> > any trailing bytes in new allocations.
> 
> > Additionally fix a memory allocation leak: if krealloc() fails, "arr"
> > wasn't freed, but NULL was return to the caller of realloc_array() would
> > be writing NULL to the lvalue, losing the reference to the original
> > memory.
> 
> > Cc: Alexei Starovoitov <ast@kernel.org>
> > Cc: Daniel Borkmann <daniel@iogearbox.net>
> > Cc: John Fastabend <john.fastabend@gmail.com>
> > Cc: Andrii Nakryiko <andrii@kernel.org>
> > Cc: Martin KaFai Lau <martin.lau@linux.dev>
> > Cc: Song Liu <song@kernel.org>
> > Cc: Yonghong Song <yhs@fb.com>
> > Cc: KP Singh <kpsingh@kernel.org>
> > Cc: Stanislav Fomichev <sdf@google.com>
> > Cc: Hao Luo <haoluo@google.com>
> > Cc: Jiri Olsa <jolsa@kernel.org>
> > Cc: bpf@vger.kernel.org
> > Signed-off-by: Kees Cook <keescook@chromium.org>
> > ---
> >   kernel/bpf/verifier.c | 49 +++++++++++++++++++++++++++----------------
> >   1 file changed, 31 insertions(+), 18 deletions(-)
> 
> > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> > index 014ee0953dbd..8a0b60207d0e 100644
> > --- a/kernel/bpf/verifier.c
> > +++ b/kernel/bpf/verifier.c
> > @@ -1000,42 +1000,53 @@ static void print_insn_state(struct
> > bpf_verifier_env *env,
> >    */
> >   static void *copy_array(void *dst, const void *src, size_t n, size_t
> > size, gfp_t flags)
> >   {
> > -	size_t bytes;
> > +	size_t src_bytes, dst_bytes;
> 
> >   	if (ZERO_OR_NULL_PTR(src))
> >   		goto out;
> 
> > -	if (unlikely(check_mul_overflow(n, size, &bytes)))
> > +	if (unlikely(check_mul_overflow(n, size, &src_bytes)))
> >   		return NULL;
> 
> > -	if (ksize(dst) < bytes) {
> > +	dst_bytes = kmalloc_size_roundup(src_bytes);
> > +	if (ksize(dst) < dst_bytes) {
> 
> Why not simply do the following here?
> 
> 	if (ksize(dst) < ksize(src)) {
> 
> ?

Yeah, if src always passes through rounding-up allocation path, that
might work. I need to double-check that there isn't a case where "size"
makes this go weird -- e.g. a rounded up "src" may be larger than
"n * size", but I think that's okay because the memcpy/memset does the
right thing.

> It seems like we care about src_bytes/bytes only in this case, so maybe
> move that check_mul_overflow under this branch as well?
> 
> 
> >   		kfree(dst);
> > -		dst = kmalloc_track_caller(bytes, flags);
> > +		dst = kmalloc_track_caller(dst_bytes, flags);
> >   		if (!dst)
> >   			return NULL;
> >   	}
> 
> > -	memcpy(dst, src, bytes);
> > +	memcpy(dst, src, src_bytes);
> > +	memset(dst + src_bytes, 0, dst_bytes - src_bytes);
> >   out:
> >   	return dst ? dst : ZERO_SIZE_PTR;
> >   }
> 
> > -/* resize an array from old_n items to new_n items. the array is
> > reallocated if it's too
> > - * small to hold new_n items. new items are zeroed out if the array
> > grows.
> > +/* Resize an array from old_n items to *new_n items. The array is
> > reallocated if it's too
> > + * small to hold *new_n items. New items are zeroed out if the array
> > grows. Allocation
> > + * is rounded up to next kmalloc bucket size to reduce frequency of
> > resizing. *new_n
> > + * contains the new total number of items that will fit.
> >    *
> > - * Contrary to krealloc_array, does not free arr if new_n is zero.
> > + * Contrary to krealloc, does not free arr if new_n is zero.
> >    */
> > -static void *realloc_array(void *arr, size_t old_n, size_t new_n,
> > size_t size)
> > +static void *realloc_array(void *arr, size_t old_n, size_t *new_n,
> > size_t size)
> >   {
> > -	if (!new_n || old_n == new_n)
> > +	void *old_arr = arr;
> > +	size_t alloc_size;
> > +
> > +	if (!new_n || !*new_n || old_n == *new_n)
> >   		goto out;
> 
> 
> [..]
> 
> > -	arr = krealloc_array(arr, new_n, size, GFP_KERNEL);
> > -	if (!arr)
> > +	alloc_size = kmalloc_size_roundup(size_mul(*new_n, size));
> > +	arr = krealloc(old_arr, alloc_size, GFP_KERNEL);
> > +	if (!arr) {
> > +		kfree(old_arr);
> >   		return NULL;
> > +	}
> 
> Any reason not do hide this complexity behind krealloc_array? Why can't
> it take care of those roundup details?

It might be possible to do this with a macro, yes, but then callers
aren't in a position to take advantage of the new size. Maybe we need
something like:

	arr = krealloc_up(old_arr, alloc_size, &new_size, GFP_KERNEL);

Thanks for looking this over!

-- 
Kees Cook