Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <307f61a1-b72c-c310-797c-013a8914ec1c@huawei.com>
Date:   Thu, 4 May 2023 10:12:29 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:102.0) Gecko/20100101
 Thunderbird/102.9.0
Subject: Re: [PATCH 1/2] ubi: fix slab-out-of-bounds in
 ubi_eba_get_ldesc+0xfb/0x130
To:     Zhihao Cheng <chengzhihao1@huawei.com>, <richard@nod.at>,
        <miquel.raynal@bootlin.com>, <vigneshr@ti.com>
CC:     <linux-mtd@lists.infradead.org>, <linux-kernel@vger.kernel.org>,
        <yi.zhang@huawei.com>
References: <20230406071331.1247429-1-wangzhaolong1@huawei.com>
 <20230406071331.1247429-2-wangzhaolong1@huawei.com>
 <cdb047f0-c071-3263-0167-e9bce852b5d5@huawei.com>
From:   ZhaoLong Wang <wangzhaolong1@huawei.com>
In-Reply-To: <cdb047f0-c071-3263-0167-e9bce852b5d5@huawei.com>
Content-Type: text/plain; charset="UTF-8"; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

Yes, that could happen. I was able to reproduce the problem despite the
low probability of triggering it.

This race between wear_leveling_work() and ubi_resize_volume() can cause
data corruption in the UBIFS running on the UBI volume.. ubi->volumes_lock
must be added to protect the update of eba_tbl in the ubi_eba_copy_leb().

I'll do a V2 patch later to fix this issue.

With appreciation
ZhaoLong Wang

> HI,
>> From: Guo Xuenan <guoxuenan@huawei.com>
>>
>> When using ioctl interface to resize ubi volume, ubi_resize_volume will
>> resize eba table first, but not change vol->reserved_pebs in the same
>> atomic context which may cause concurrency access eba table.
>>
>> For example, When user do shrink ubi volume A calling ubi_resize_volume,
>> while the other thread is writing (volume B) and triggering 
>> wear-leveling,
>> which may calling ubi_write_fastmap, under these circumstances, KASAN 
>> may
>> report: slab-out-of-bounds in ubi_eba_get_ldesc+0xfb/0x130.
>>
> [...]
>> diff --git a/drivers/mtd/ubi/vmt.c b/drivers/mtd/ubi/vmt.c
>> index 2c867d16f89f..97294def01eb 100644
>> --- a/drivers/mtd/ubi/vmt.c
>> +++ b/drivers/mtd/ubi/vmt.c
>> @@ -408,6 +408,7 @@ int ubi_resize_volume(struct ubi_volume_desc 
>> *desc, int reserved_pebs)
>>       struct ubi_device *ubi = vol->ubi;
>>       struct ubi_vtbl_record vtbl_rec;
>>       struct ubi_eba_table *new_eba_tbl = NULL;
>> +    struct ubi_eba_table *old_eba_tbl = NULL;
>>       int vol_id = vol->vol_id;
>>         if (ubi->ro_mode)
>> @@ -453,10 +454,13 @@ int ubi_resize_volume(struct ubi_volume_desc 
>> *desc, int reserved_pebs)
>>               err = -ENOSPC;
>>               goto out_free;
>>           }
>> +
>>           ubi->avail_pebs -= pebs;
>>           ubi->rsvd_pebs += pebs;
>>           ubi_eba_copy_table(vol, new_eba_tbl, vol->reserved_pebs);
>> -        ubi_eba_replace_table(vol, new_eba_tbl);
>> +        old_eba_tbl = vol->eba_tbl;
>> +        vol->eba_tbl = new_eba_tbl;
>> +        vol->reserved_pebs = reserved_pebs;
>>           spin_unlock(&ubi->volumes_lock);
>>       }
>>   @@ -471,7 +475,9 @@ int ubi_resize_volume(struct ubi_volume_desc 
>> *desc, int reserved_pebs)
>>           ubi->avail_pebs -= pebs;
>>           ubi_update_reserved(ubi);
>>           ubi_eba_copy_table(vol, new_eba_tbl, reserved_pebs);
>> -        ubi_eba_replace_table(vol, new_eba_tbl);
>> +        old_eba_tbl = vol->eba_tbl;
>> +        vol->eba_tbl = new_eba_tbl;
>> +        vol->reserved_pebs = reserved_pebs;
>>           spin_unlock(&ubi->volumes_lock);
>>       }
>>   @@ -493,7 +499,6 @@ int ubi_resize_volume(struct ubi_volume_desc 
>> *desc, int reserved_pebs)
>>       if (err)
>>           goto out_acc;
>>   -    vol->reserved_pebs = reserved_pebs;
>>       if (vol->vol_type == UBI_DYNAMIC_VOLUME) {
>>           vol->used_ebs = reserved_pebs;
>>           vol->last_eb_bytes = vol->usable_leb_size;
>> @@ -501,19 +506,24 @@ int ubi_resize_volume(struct ubi_volume_desc 
>> *desc, int reserved_pebs)
>>               (long long)vol->used_ebs * vol->usable_leb_size;
>>       }
>>   +    /* destroy old table */
>> +    ubi_eba_destroy_table(old_eba_tbl);
>>       ubi_volume_notify(ubi, vol, UBI_VOLUME_RESIZED);
>>       self_check_volumes(ubi);
>>       return err;
>>     out_acc:
>> +    spin_lock(&ubi->volumes_lock);
>> +    vol->reserved_pebs = reserved_pebs - pebs;
>>       if (pebs > 0) {
>> -        spin_lock(&ubi->volumes_lock);
>>           ubi->rsvd_pebs -= pebs;
>>           ubi->avail_pebs += pebs;
>> -        spin_unlock(&ubi->volumes_lock);
>> +        ubi_eba_copy_table(vol, old_eba_tbl, vol->reserved_pebs);
>> +    } else {
>> +        ubi_eba_copy_table(vol, old_eba_tbl, reserved_pebs);
>>       }
>> -    return err;
>> -
>> +    vol->eba_tbl = old_eba_tbl;
>> +    spin_unlock(&ubi->volumes_lock);
>>   out_free:
>>       ubi_eba_destroy_table(new_eba_tbl);
>>       return err;
>>
>
>
> Besides that, it's better to protect 'vol->eba_tbl->entries' 
> assignment like:
> diff --git a/drivers/mtd/ubi/eba.c b/drivers/mtd/ubi/eba.c
> index 403b79d6efd5..5ae0c1bc6f41 100644
> --- a/drivers/mtd/ubi/eba.c
> +++ b/drivers/mtd/ubi/eba.c
> @@ -1450,7 +1450,9 @@ int ubi_eba_copy_leb(struct ubi_device *ubi, int 
> from, int to,
>         }
>
>         ubi_assert(vol->eba_tbl->entries[lnum].pnum == from);
> +       spin_lock(&ubi->volumes_lock);
>         vol->eba_tbl->entries[lnum].pnum = to;
> +       spin_unlock(&ubi->volumes_lock);
>
>  out_unlock_buf:
>         mutex_unlock(&ubi->buf_mutex);
>
> Otherwise, a race between wear_leveling_work and shrinking volume 
> could happen:
>
>  ubi_resize_volume         wear_leveling_worker
>   ubi_eba_copy_table(vol, new_eba_tbl, reserved_pebs);
> vol->eba_tbl->entries[lnum].pnum = to; // update old eba_tbl
>   vol->eba_tbl = new_eba_tbl