Re: [PATCH] fix detection of non-LUKS CRYPT

[Andrei Borzenkov]

On Mon, Nov 7, 2016 at 9:48 AM, Corey Hickey <address@hidden> wrote:
> On 2016-11-06 02:01, Andrei Borzenkov wrote:
>>
>> 06.11.2016 10:09, Corey Hickey пишет:
>>>
>>> On 2016-11-05 05:31, Andrei Borzenkov wrote:
>>>>>
>>>>> diff --git a/grub-core/osdep/devmapper/getroot.c
>>>>> b/grub-core/osdep/devmapper/getroot.c
>>>>> index 72e5582..a13a39c 100644
>>>>> --- a/grub-core/osdep/devmapper/getroot.c
>>>>> +++ b/grub-core/osdep/devmapper/getroot.c
>>>>> @@ -143,7 +143,7 @@ grub_util_get_dm_abstraction (const char *os_dev)
>>>>>        grub_free (uuid);
>>>>>        return GRUB_DEV_ABSTRACTION_LVM;
>>>>>      }
>>>>> -  if (strncmp (uuid, "CRYPT-LUKS1-", 4) == 0)
>>>>> +  if (strncmp (uuid, "CRYPT-LUKS1-", 12) == 0
>>>>
>>>>
>>>> Committed, thanks! We really need some wrapper around (strncmp (foo,
>>>> "bar", sizeof ("bar") - 1), but for now it is OK as bug fix.
>>>
>>>
>>> Excellent, you're welcome. That seemed like the most simple fix.
>>>
>>> I took a stab at adding such a wrapper, but there are a ton of files
>>> that could use it which I won't have a chance at being able to test. I
>>> can send in an untested patch if you want...
>>>
>>
>> Large scale replacement will have to wait until release, but we sure can
>> discuss (and add) macro itself. I'm leaning towards simple
>>
>> #define GRUB_IS_PREFIX(string,prefix) (strncmp((string), #prefix, sizeof
>> (#prefix) - 1) == 0)
>>
>> Any other idea how to make it constant-safe?
>
>
> I have to admit at this point that I never really got good at C, and most of
> that was years ago. If I understand correctly, though, the stringification
> in your example macro makes "foo" into "\"foo\"", before sizeof() which
> would not be suitable.
>
> My initial idea matched your earlier example:
>
> #define STARTS_WITH(s1, s2) (strncmp((s1), (s2), sizeof(s2) - 1) == 0)
>
> Am I failing to see a use case where that breaks?
>

char *prefix = "CRYPT-LUKS1-";
sizeof (prefix) == 4 or 8 depending on platform. Not 13 in any case.

> The one bad thing I can see is that it can fail poorly if somebody reverses
> the arguments by mistake. In that case, sizeof(s2) returns the pointer size
> rather than the length of the string literal. I don't know how to enforce
> that a macro argument be a string literal. An alternate idea is:
>
> #define STARTS_WITH(s1, s2) (strncmp((s1), (s2), strlen(s2)) == 0)
>

This adds additional overhead of function call instead of constant
expression; unless we are sure gcc/clang are smart enough to reduce it
for constant arguments.

> That should work with arguments in either order (though of course s2 is
> expected to be smaller). I don't know of a drawback other than strlen
> presumably being a bit slower, but I didn't get the impression any of the
> code in question was performance-sensitive.
>

For user space it probably does not matter much; but for boot time
code it also increases code size and better be avoided.

Even for user space it is called often so if we can reduce overhead,
let's do it.

> -Corey
>
>
> _______________________________________________
> Grub-devel mailing list
> address@hidden
> https://lists.gnu.org/mailman/listinfo/grub-devel