Date: Thu, 14 Mar 2024 21:47:28 +0800
From: Baokun Li <libaokun1@...wei.com>
To: Jan Kara <jack@...e.cz>
CC: <linux-ext4@...r.kernel.org>, <tytso@....edu>, <adilger.kernel@...ger.ca>,
	<ritesh.list@...il.com>, <ojaswin@...ux.ibm.com>, <adobriyan@...il.com>,
	<linux-kernel@...r.kernel.org>, <yi.zhang@...wei.com>,
	<yangerkun@...wei.com>, <stable@...r.kernel.org>, Baokun Li
	<libaokun1@...wei.com>
Subject: Re: [PATCH v2 4/9] ext4: fix slab-out-of-bounds in
 ext4_mb_find_good_group_avg_frag_lists()

On 2024/3/14 20:50, Jan Kara wrote:
> On Thu 14-03-24 20:37:38, Baokun Li wrote:
>> On 2024/3/14 20:00, Jan Kara wrote:
>>> On Thu 14-03-24 19:24:56, Baokun Li wrote:
>>>> Hi Jan,
>>>>
>>>> On 2024/3/14 18:30, Jan Kara wrote:
>>>>> On Tue 27-02-24 17:11:43, Baokun Li wrote:
>>>>>
>>>>>
>>>>> At 4k block size, the length of the s_mb_avg_fragment_size list is 14,
>>>>> but an oversized s_mb_group_prealloc is set, causing slab-out-of-bounds
>>>>> to be triggered by an attempt to access an element at index 29.
>>>>>
>>>>> Add a new attr_id attr_clusters_in_group with values in the range
>>>>> [0, sbi->s_clusters_per_group] and declare mb_group_prealloc as
>>>>> that type to fix the issue. In addition avoid returning an order
>>>>> from mb_avg_fragment_size_order() greater than MB_NUM_ORDERS(sb)
>>>>> and reduce some useless loops.
>>>>>
>>>>> Fixes: 7e170922f06b ("ext4: Add allocation criteria 1.5 (CR1_5)")
>>>>> CC: stable@...r.kernel.org
>>>>> Signed-off-by: Baokun Li <libaokun1@...wei.com>
>>>>> Looks good. Just one nit below. Otherwise feel free to add:
>>>>>
>>>>> Reviewed-by: Jan Kara <jack@...e.cz>
>>>>>
>>>>>> ---
>>>>>>     fs/ext4/mballoc.c |  6 ++++++
>>>>>>     fs/ext4/sysfs.c   | 13 ++++++++++++-
>>>>>>     2 files changed, 18 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
>>>>>> index 85a91a61b761..7ad089df2408 100644
>>>>>> --- a/fs/ext4/mballoc.c
>>>>>> +++ b/fs/ext4/mballoc.c
>>>>>> @@ -831,6 +831,8 @@ static int mb_avg_fragment_size_order(struct super_block *sb, ext4_grpblk_t len)
>>>>>>     		return 0;
>>>>>>     	if (order == MB_NUM_ORDERS(sb))
>>>>>>     		order--;
>>>>>> +	if (WARN_ON_ONCE(order > MB_NUM_ORDERS(sb)))
>>>>>> +		order = MB_NUM_ORDERS(sb) - 1;
>>>>>>     	return order;
>>>>>>     }
>>>>>> @@ -1057,6 +1059,10 @@ static void ext4_mb_choose_next_group_best_avail(struct ext4_allocation_context
>>>>>>     			ac->ac_flags |= EXT4_MB_CR_BEST_AVAIL_LEN_OPTIMIZED;
>>>>>>     			return;
>>>>>>     		}
>>>>>> +
>>>>>> +		/* Skip some unnecessary loops. */
>>>>>> +		if (unlikely(i > MB_NUM_ORDERS(ac->ac_sb)))
>>>>>> +			i = MB_NUM_ORDERS(ac->ac_sb);
>>>>> How can this possibly trigger now? MB_NUM_ORDERS is sb->s_blocksize_bits +
>>>>> 2. 'i' is starting at fls(ac->ac_g_ex.fe_len) and ac_g_ex.fe_len shouldn't
>>>>> be larger than clusters per group, hence fls() should be less than
>>>>> sb->s_blocksize_bits? Am I missing something? And if yes, we should rather
>>>>> make sure 'order' is never absurdly big?
>>>>>
>>>>> I suspect this code is defensive upto a point of being confusing :)
>>>>>
>>>>> Honza
>>>> Yes, this is indeed defensive code! Only walk into this branch when
>>>> WARN_ON_ONCE(order > MB_NUM_ORDERS(sb)) is triggered.
>>>> As previously mentioned by ojaswin in the following link:
>>>>
>>>> "The reason for this is that otherwise when order is large eg 29,
>>>> we would unnecessarily loop from i=29 to i=13 while always
>>>> looking at the same avg_fragment_list[13]."
>>>>
>>>> Link：https://lore.kernel.org/lkml/ZdQ7FEA7KC4eAMpg@li-bb2b2a4c-3307-11b2-a85c-8fa5c3a69313.ibm.com/
>>>>
>>>> Thank you so much for the review! ღ( ´･ᴗ･` )
>>> Thanks for the link. So what Ojaswin has suggested has been slightly
>>> different though. He suggested to trim the order before the for loop, not
>>> after the first iteration as you do which is what was confusing me. I'd
>>> even suggest to replace your check with:
>>>
>>>           /*
>>>            * mb_avg_fragment_size_order() returns order in a way that makes
>>>            * retrieving back the length using (1 << order) inaccurate. Hence, use
>>>            * fls() instead since we need to know the actual length while modifying
>>>            * goal length.
>>>            */
>>> -       order = fls(ac->ac_g_ex.fe_len) - 1;
>>> +	order = min(fls(ac->ac_g_ex.fe_len), MB_NUM_ORDERS(ac->ac_sb)) - 1;
>>>           min_order = order - sbi->s_mb_best_avail_max_trim_order;
>>>           if (min_order < 0)
>>>                   min_order = 0;
>>>
>>> 								Honza
>> Yes, I changed it that way because it only happens when an exception
>> somewhere causes fe_len to be a huge value. I think in this case we
>> should report the exception via WARN_ON_ONCE(), and trimming the
>> order before the for loop will bypass WARN_ON_ONCE and not report
>> any errors.
> Fair enough. Then:
>           /*
>            * mb_avg_fragment_size_order() returns order in a way that makes
>            * retrieving back the length using (1 << order) inaccurate. Hence, use
>            * fls() instead since we need to know the actual length while modifying
>            * goal length.
>            */
> 	order = fls(ac->ac_g_ex.fe_len) - 1;
> +	if (WARN_ON_ONCE(order > MB_NUM_ORDERS(ac->ac_sb) - 1))
> +		order = MB_NUM_ORDERS(ac->ac_sb) - 1;
>          min_order = order - sbi->s_mb_best_avail_max_trim_order;
>          if (min_order < 0)
>                  min_order = 0;
>
> Still much less confusing...
>
> 								Honza
Yes this does look much better!
Let me send v3!

Thanks for the suggestion!
-- 
With Best Regards,
Baokun Li
.

Powered by blists - more mailing lists