Re: [PATCH][next] io_uring: Fix incorrect sizeof operator for copy_from_user call

[Pavel Begunkov]

On 6/15/21 12:35 PM, Colin Ian King wrote:
> On 15/06/2021 12:30, Pavel Begunkov wrote:
>> On 6/15/21 11:47 AM, Colin Ian King wrote:
>>> On 15/06/2021 11:45, Colin King wrote:
>>>> From: Colin Ian King <colin.king@xxxxxxxxxxxxx>
>>>>
>>>> Static analysis is warning that the sizeof being used is should be
>>>> of *data->tags[i] and not data->tags[i]. Although these are the same
>>>> size on 64 bit systems it is not a portable assumption to assume
>>>> this is true for all cases.
>>>>
>>>> Addresses-Coverity: ("Sizeof not portable")
>>>> Fixes: d878c81610e1 ("io_uring: hide rsrc tag copy into generic helpers")
>>>> Signed-off-by: Colin Ian King <colin.king@xxxxxxxxxxxxx>
>>>> ---
>>>>  fs/io_uring.c | 2 +-
>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>>>> index d665c9419ad3..6b1a70449749 100644
>>>> --- a/fs/io_uring.c
>>>> +++ b/fs/io_uring.c
>>>> @@ -7231,7 +7231,7 @@ static int io_rsrc_data_alloc(struct io_ring_ctx *ctx, rsrc_put_fn *do_put,
>>>>  		ret = -EFAULT;
>>>>  		for (i = 0; i < nr; i++) {
>>>>  			if (copy_from_user(io_get_tag_slot(data, i), &utags[i],
>>>> -					   sizeof(data->tags[i])))
>>>> +					   sizeof(*data->tags[i])))
>>>>  				goto fail;
>>>>  		}
>>>>  	}
>>>>
>>
> 
> 
>> Yep, thanks Colin. I think `sizeof(io_get_tag_slot(data, i))`
>> would be less confusing. Or
>>
>> u64 *tag_slot = io_get_tag_slot(data, i);
>> copy_from_user(tag_slot, ..., sizeof(*tag_slot));
>>
> BTW, Coverity is complaining about:
> 
> 7220                return -ENOMEM;
> 
> Wrong sizeof argument (SIZEOF_MISMATCH)
> 
> suspicious_sizeof: Passing argument nr * 8UL /* sizeof
> (data->tags[0][0]) */ to function io_alloc_page_table and then casting
> the return value to u64 ** is suspicious.
> 
> 7221        data->tags = (u64 **)io_alloc_page_table(nr *
> sizeof(data->tags[0][0]));

Ah, this one. We want it to be indexed linearly, but can't allocate
as much, so together with io_get_tag_slot() it hides two level
tables from us, providing linear indexing.

> 
> Not sure if that's a false positive or not. This kind of indirection
> makes my brain melt.

So, this one should be a false positive. But agree about the
indirection, it's not the first sizeof bug you found. Any
better ideas how to push it to the type system?

I think something like below would make more sense

#define copy_from_user_typed(from, to) \
    assert(typeof(from) == typeof(to)),
    copy_from_user(from, to, sizeof(*from));

-- 
Pavel Begunkov