Re: [PATCH 1/7] ext4: avoid overflow when setting values via sysfs

[Baokun Li <libaokun1@huawei.com>]

On 2024/2/23 19:54, Jan Kara wrote:
> On Sat 17-02-24 15:09:06, Baokun Li wrote:
>> On 2024/2/14 0:05, Jan Kara wrote:
>>> On Fri 26-01-24 16:57:10, Baokun Li wrote:
>>>> When setting values of type unsigned int through sysfs, we use kstrtoul()
>>>> to parse it and then truncate part of it as the final set value, when the
>>>> set value is greater than UINT_MAX, the set value will not match what we
>>>> see because of the truncation. As follows:
>>>>
>>>>     $ echo 4294967296 > /sys/fs/ext4/sda/mb_max_linear_groups
>>>>     $ cat /sys/fs/ext4/sda/mb_max_linear_groups
>>>>       0
>>>>
>>>> So when the value set is outside the variable type range, -EINVAL is
>>>> returned to avoid the inconsistency described above. In addition, a
>>>> judgment is added to avoid setting s_resv_clusters less than 0.
>>>>
>>>> Signed-off-by: Baokun Li <libaokun1@huawei.com>
>>>> ---
>>>>    fs/ext4/sysfs.c | 4 +++-
>>>>    1 file changed, 3 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/fs/ext4/sysfs.c b/fs/ext4/sysfs.c
>>>> index 6d332dff79dd..3671a8aaf4af 100644
>>>> --- a/fs/ext4/sysfs.c
>>>> +++ b/fs/ext4/sysfs.c
>>>> @@ -104,7 +104,7 @@ static ssize_t reserved_clusters_store(struct ext4_sb_info *sbi,
>>>>    	int ret;
>>>>    	ret = kstrtoull(skip_spaces(buf), 0, &val);
>>>> -	if (ret || val >= clusters)
>>>> +	if (ret || val >= clusters || (s64)val < 0)
>>>>    		return -EINVAL;
>>> This looks a bit pointless, doesn't it? 'val' is u64, clusters is u64. We
>>> know that val < clusters so how could (s64)val be < 0?
>> When clusters is bigger than LLONG_MAX, (s64)val may be less than 0.
>> Of course we don't have such a large storage device yet, so it's only
>> theoretically possible to overflow here. But the previous patches in this
>> patch set were intended to ensure that the values set via sysfs did not
>> exceed the range of the variable type, so I've modified that here as well.
> Well, my point was that the on disk format is limited to much less than
> 2^63 blocks. But I guess having the additional check does not matter.
OK.
>>>> @@ -463,6 +463,8 @@ static ssize_t ext4_attr_store(struct kobject *kobj,
>>>>    		ret = kstrtoul(skip_spaces(buf), 0, &t);
>>>>    		if (ret)
>>>>    			return ret;
>>>> +		if (t != (unsigned int)t)
>>>> +			return -EINVAL;
>>>>    		if (a->attr_ptr == ptr_ext4_super_block_offset)
>>>>    			*((__le32 *) ptr) = cpu_to_le32(t);
>>>>    		else
>>> I kind of agree with Alexey that using kstrtouint() here instead would look
>>> nicer. And it isn't like you have to define many new variables. You just
>>> need unsigned long for attr_pointer_ul and unsigned int for
>>> attr_pointer_ui.
>> If we use both kstrtouint() and kstrtoul(), then we need to add
>> kstrtouint() or kstrtoul() to each case, which would be a lot of
>> duplicate code as follows:
> Well, it is 5 more lines if I'm counting right :) (3x 3 lines of conversion
> - 2x 2 lines of boundary checks). I kind of find it easier to oversee the
> boundary checks when everything is together at each parameter. But frankly
> this is a bit of nitpicking so if you feel strongly about this I won't
> insist.
Makes sense, there may be some implicit checks that look unintuitive
this way in the original patch. Now keep the string to number conversion
inside the switch does look better. Let me send v2.
>> static ssize_t ext4_generic_attr_store(struct ext4_attr *a,
>>                                         struct ext4_sb_info *sbi,
>>                                         const char *buf, size_t len)
>> {
>>          int ret;
>>          unsigned int t;
>>          unsigned long lt;
>>          void *ptr = calc_ptr(a, sbi);
>>
>>          if (!ptr)
>>                  return 0;
>>
>>          switch (a->attr_id) {
>>          case attr_group_prealloc:
>>                  ret = kstrtouint(skip_spaces(buf), 0, &t);
>>                  if (ret)
>>                          return ret;
>>                  if (t > sbi->s_clusters_per_group)
>>                          return -EINVAL;
>>                  return len;
>>          case attr_pointer_pi:
>>                  ret = kstrtouint(skip_spaces(buf), 0, &t);
>>                  if (ret)
>>                          return ret;
>>                  if ((int)t < 0)
>>                          return -EINVAL;
>>                  return len;
>>          case attr_pointer_ui:
>>                  ret = kstrtouint(skip_spaces(buf), 0, &t);
>>                  if (ret)
>>                          return ret;
>>                  if (t != (unsigned int)t)
>>                          return -EINVAL;
> 		  ^^^ this can go away
I forgot to delete this, thanks!
>>                  if (a->attr_ptr == ptr_ext4_super_block_offset)
>>                          *((__le32 *) ptr) = cpu_to_le32(t);
>>                  else
>>                          *((unsigned int *) ptr) = t;
>>                  return len;
>>          case attr_pointer_ul:
>>                  ret = kstrtoul(skip_spaces(buf), 0, &lt);
>>                  if (ret)
>>                          return ret;
>>                  *((unsigned long *) ptr) = lt;
>>                  return len;
>>          }
>>          return 0;
>>
>> }
>>
>> Also, both kstrtouint() and kstrtoul() are based on the kstrtoull()
>> implementation, so it feels better to opencode kstrtoul() and
>> kstrtouint() to reduce duplicate code.
>> Why is it better to distinguish uint and ulong cases here?
> Hopefully explained above :)
>
>
> 								Honza
Yes, now I understand what you're considering. 😊

Thank you for your explanation!
-- 
With Best Regards,
Baokun Li
.