Re: [PATCH] ocfs2: mount fails with buffer overflow in strlen

[Joseph Qi <joseph.qi@linux.alibaba.com>]

On 9/29/21 2:24 PM, Valentin Vidić wrote:
> On Wed, Sep 29, 2021 at 10:38:59AM +0800, Joseph Qi wrote:
>> Okay, you are right, strlen(src) is indeed wrong here.
>>
>> But please note that in strlcpy():
>> size_t ret = strlen(src);
>> if (size) {
>> 	size_t len = (ret >= size) ? size - 1 : ret;
>> 	memcpy(dest, src, len);
>> 	dest[len] = '\0';
>> }
>>
>> Take ci_stack "o2cb" for example, strlen("o2cb") may return wrong if the
>> coming byte is not null, say it is 10.
>> The input size is 5, so len will finally be 4.
>> So dest is still correct ending with null byte. No overflow happens.
>> So the problem here is the wrong return value, but it is discarded in
>> ocfs2_initialize_super().
> 
> strlcpy starts with a call to strlen(src) and this is where the read overflow
> happens. If the kernel is compiled with CONFIG_FORTIFY_SOURCE this gets
> executed instead (include/linux/fortify-string.h):
> 
> __FORTIFY_INLINE __kernel_size_t strlen(const char *p)
> {
>         __kernel_size_t ret;
>         size_t p_size = __builtin_object_size(p, 1);
> 
>         /* Work around gcc excess stack consumption issue */
>         if (p_size == (size_t)-1 ||
>                 (__builtin_constant_p(p[p_size - 1]) && p[p_size - 1] == '\0'))
>                 return __underlying_strlen(p);
>         ret = strnlen(p, p_size);
>         if (p_size <= ret)
>                 fortify_panic(__func__);
>         return ret;
> }
> 
> So while strlcpy did work before this fortify check, it is probably not the
> best option anymore due to the missing null terminator in the source.
> 
Got it, it really triggers panic in strlen().
So could you please update the commit log? I think CONFIG_FORTIFY_SOURCE
is necessary information since it is not default enabled.
And add comments with your changes, e.g.

/*
 * ci_stack and ci_cluster in ocfs2_cluster_info may not null
 * terminated, make sure no overflow happens here.
 */

BTW, since we use kzalloc to alloc osb, so we don't have to manually
set the last null byte.

Thanks,
Joseph