Re: [PATCH] Netfilter: Fix integer overflow in net/ipv6/netfilter/ip6_tables.c

From: Xiaotian Feng
Date: Tue Mar 23 2010 - 00:49:57 EST


On Tue, Mar 23, 2010 at 11:48 AM, wzt wzt <wzt.wzt@xxxxxxxxx> wrote:
> 1ã suppose *len = 35, Âsizeof(struct ipt_get_entries) = 36
> 2ã set get.size = 0xffffffff from user space
> 3ã sizeof(struct ipt_get_entries) + get.size = 36 Â+ 0xffffffff = 35;
> 4ã if (*len != sizeof(struct ipt_get_entries) + get.size) was bypassed.
>
> you can test with c code:
>
> #include <stdio.h>
>
> int main(void)
> {
> Â Â Â Âunsigned int arg = 0xffffffff;
>
> Â Â Â Âprintf("%u\n", arg + 36);
> Â Â Â Âif (35 != arg + 36) {
> Â Â Â Â Â Â Â Âprintf("not over flow.\n");
> Â Â Â Â Â Â Â Âreturn -1;
> Â Â Â Â}
> Â Â Â Âprintf("arg over flow.\n");
>
> Â Â Â Âreturn 0;
> }
>

You didn't get his point... The key point is that sizeof() result type
is size_t, slightly modify you code, try the result.
> int main(void)
> {
> Â Â Â Âunsigned int arg = 0xffffffff;
> unsigned int foo;
> Â Â Â Âprintf("%lu\n", arg + sizeof(foo));
> Â Â Â Âif (sizeof(foo) - 1 != arg + sizeof(foo)) {
> Â Â Â Â Â Â Â Âprintf("not over flow.\n");
> Â Â Â Â Â Â Â Âreturn -1;
> Â Â Â Â}
> Â Â Â Âprintf("arg over flow.\n");
>
> Â Â Â Âreturn 0;
> }

> On Tue, Mar 23, 2010 at 11:04 AM, Jan Engelhardt <jengelh@xxxxxxxxxx> wrote:
>>
>> On Tuesday 2010-03-23 03:37, wzt wzt wrote:
>>>> And, for the addition overflow, can it be caught by
>>>>
>>>> "if (*len != sizeof(struct ipt_get_entries) + get.size)" Â???
>>>
>>>sizeof(struct ipt_get_entries) + get.size can be overflow as *len,
>>>get.size is control by user space with copy_from_user().
>>
>> The != should catch it.
>>
>> For 64-bit environments:
>> * + invoked with size_t, unsigned int
>> Â=> right side promoted to size_t, result type is size_t
>> * != invoked with int and size_t
>> Â=> left-side promoted to ssize_t (probably; but something as large as size_t)
>> * get.size is 32-bit bounded, as is *len,
>> Âso no overflow to worry about at all unless you make
>> Âsizeof(X) hilariously big close to 2^64 which is rather unlikely.
>>
>> For 32-bit environments:
>> * Let *len be a number of choice (e.g. 36)
>> * Find a sizeof(X)+get.size that equals 36 mod 2^32.
>> * Since sizeof(X) is const, get.size must be 0 mod 2^32.
>> * So get.size must be a multiple of 2^32 to fool the system.
>> * Since get.size itself is only a 32-bit quantity, you cannot
>> Ârepresent any value larger than 4294967295.
>>
>>
>> What Was What Was Wanted.
>>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/