On Mon, 2010-11-29 at 05:09 +0300, Pavel Vasilyev wrote:
quoted text
> This patch replace all strncmp(a, b, c) by  memcmp(a, b, c).

But these are not the same. strncmp() will stop when a or b hit a null.
I'm not sure if memcmp() must do so, It may for some reason check
anything within the memory of a+c-1 or b+c-1. What happens if a or b are
right at the end of a vmalloc page, and is just a single character and
null?

	x = vmalloc(32);
	strcpy(x, "some 31 byte string + null");

	call_func(x + 31);

in call_func we have:

	call_func(char *a) {

	strncmp(a, "this is some big string", 23);

With strncmp() when we hit a+1, it will stop comparing because a+1 is
null. With memcmp there's no such guarantee. We can then take a kernel
oops.

That will be a nice thing to try to debug.

Yes the above is contrived, but it demonstrates a possible problem with
this conversion.

-- Steve

quoted text
> 
> I test on x86_64 (AMD Opteron 285).
> 
> #include <string.h>
> char *A = "0000";
> void test_memcmp(void) {
>         memcmp(A, "TEST", 4);
> }
> void test_strn(void) {
>         strncmp(A, "TEST", 4);
> }
> # gcc -c -O2 test.c
> # objdump -d test.o
> ...
> 
> 0000000000000020 <test_strncmp>:
>   20:   f3 c3                   repz retq
>   22:   66 66 66 66 66 2e 0f    data32 data32 data32 data32 nopw
> %cs:0x0(%rax,%rax,1)
>   29:   1f 84 00 00 00 00 00
> 
> 0000000000000030 <test_memcmp>:
>   30:   f3 c3                   repz retq
> 
> Wow, minus  one commad :)
> 


--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/