[PATCH v3] md/raid5: reject raid4/5 arrays with too few disks

[PATCH v3] md/raid5: reject raid4/5 arrays with too few disks

[Chen Cheng]

From: Chen Cheng <chencheng@fnnas.com>

raid4 and raid5 require at least two disks, while raid6 requires at least
four disks. setup_conf() only rejects raid6 arrays with fewer than four
disks. As a result, an invalid raid4 or raid5 array can be assembled with
no data disks and later reach raid5_set_limits().

Reject such arrays before the r5conf is created. This prevents the invalid
geometry from reaching queue limit setup, where roundup_pow_of_two() can be
called with a zero stripe size.

Reported-by: syzbot+de94ddbfff0c9e6fe030@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/r/6a778c82.01d0871a.3a0d52.006a.GAE@google.com/
Fixes: f63f17350e53 ("md/raid5: use the atomic queue limit update APIs")
Signed-off-by: Chen Cheng <chencheng@fnnas.com>
---
v1->v2:
- Fix wrong tags
- Check raid-4 also

v2->v3:
- Keep two-disk raid5 arrays valid by rejecting raid4/5 only below two disks
---
 drivers/md/raid5.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index b91545ce090d..e1f20035703b 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -7655,13 +7655,15 @@ static struct r5conf *setup_conf(struct mddev *mddev)
 	     && !algorithm_valid_raid6(mddev->new_layout))) {
 		pr_warn("md/raid:%s: layout %d not supported\n",
 			mdname(mddev), mddev->new_layout);
 		return ERR_PTR(-EIO);
 	}
-	if (mddev->new_level == 6 && mddev->raid_disks < 4) {
-		pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n",
-			mdname(mddev), mddev->raid_disks);
+	if ((mddev->new_level != 6 && mddev->raid_disks < 2) ||
+	    (mddev->new_level == 6 && mddev->raid_disks < 4)) {
+		pr_warn("md/raid:%s: not enough configured devices (%d, minimum %d)\n",
+			mdname(mddev), mddev->raid_disks,
+			mddev->new_level == 6 ? 4 : 2);
 		return ERR_PTR(-EINVAL);
 	}
 
 	if (!mddev->new_chunk_sectors ||
 	    (mddev->new_chunk_sectors << 9) % PAGE_SIZE ||
-- 
2.55.0

Re: [PATCH v3] md/raid5: reject raid4/5 arrays with too few disks

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Division by zero in `raid5_run()` when assembling an array with an active reshape and insufficient data disks.
--

commit e2bb0701d8414cbf17eed6287965f8022ffe5eab
Author: Chen Cheng <chencheng@fnnas.com>

md/raid5: reject raid4/5 arrays with too few disks

raid4 and raid5 require at least two disks, while raid6 requires at least
four disks. setup_conf() only rejects raid6 arrays with fewer than four
disks. As a result, an invalid raid4 or raid5 array can be assembled with
no data disks and later reach raid5_set_limits().

Reported-by: syzbot+de94ddbfff0c9e6fe030@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/r/6a778c82.01d0871a.3a0d52.006a.GAE@google.com/
Fixes: f63f17350e53 ("md/raid5: use the atomic queue limit update APIs")

> diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
> index b91545ce090de..e1f20035703ba 100644
> --- a/drivers/md/raid5.c
> +++ b/drivers/md/raid5.c
> @@ -7657,9 +7657,11 @@ static struct r5conf *setup_conf(struct mddev *mddev)
>  			mdname(mddev), mddev->new_layout);
>  		return ERR_PTR(-EIO);
>  	}
> -	if (mddev->new_level == 6 && mddev->raid_disks < 4) {
> -		pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n",
> -			mdname(mddev), mddev->raid_disks);
> +	if ((mddev->new_level != 6 && mddev->raid_disks < 2) ||
> +	    (mddev->new_level == 6 && mddev->raid_disks < 4)) {
> +		pr_warn("md/raid:%s: not enough configured devices (%d, minimum %d)\n",
> +			mdname(mddev), mddev->raid_disks,
> +			mddev->new_level == 6 ? 4 : 2);

[Severity: High]
This isn't a bug introduced by this patch, but does this fully prevent
invalid array geometries from causing panics during array assembly?

In raid5_run(), if there is an active reshape and the array is assembled with
insufficient data disks, we calculate new_data_disks and divide by it before
setup_conf() is called to validate the array geometry:

drivers/md/raid5.c:raid5_run() {
    if (mddev->reshape_position != MaxSector) {
        ...
        new_data_disks = mddev->raid_disks - max_degraded;
        if (sector_div(here_new, chunk_sectors * new_data_disks)) {
            ...
        }
        ...
        sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
    }
    ...
    if (mddev->private == NULL)
        conf = setup_conf(mddev);
}

If a crafted superblock sets raid_disks <= max_degraded and reshape_position
is not MaxSector, these divisions by zero are reached before the new checks
in setup_conf() can reject the array.

Could these checks be moved earlier in the assembly path to prevent the
divide-by-zero panics?

>  		return ERR_PTR(-EINVAL);
>  	}
>

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260911041521.2433222-1-chencheng@fnnas.com?part=1

Re: [PATCH v3] md/raid5: reject raid4/5 arrays with too few disks

[yu kuai]

Hi,

在 2026/9/11 12:27, sashiko-bot@kernel.org 写道:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
>
> Pre-existing issues:
> - [High] Division by zero in `raid5_run()` when assembling an array with an active reshape and insufficient data disks.
> --
>
> commit e2bb0701d8414cbf17eed6287965f8022ffe5eab
> Author: Chen Cheng <chencheng@fnnas.com>
>
> md/raid5: reject raid4/5 arrays with too few disks
>
> raid4 and raid5 require at least two disks, while raid6 requires at least
> four disks. setup_conf() only rejects raid6 arrays with fewer than four
> disks. As a result, an invalid raid4 or raid5 array can be assembled with
> no data disks and later reach raid5_set_limits().
>
> Reported-by: syzbot+de94ddbfff0c9e6fe030@syzkaller.appspotmail.com
> Closes: https://lore.kernel.org/r/6a778c82.01d0871a.3a0d52.006a.GAE@google.com/
> Fixes: f63f17350e53 ("md/raid5: use the atomic queue limit update APIs")
>
>> diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
>> index b91545ce090de..e1f20035703ba 100644
>> --- a/drivers/md/raid5.c
>> +++ b/drivers/md/raid5.c
>> @@ -7657,9 +7657,11 @@ static struct r5conf *setup_conf(struct mddev *mddev)
>>   			mdname(mddev), mddev->new_layout);
>>   		return ERR_PTR(-EIO);
>>   	}
>> -	if (mddev->new_level == 6 && mddev->raid_disks < 4) {
>> -		pr_warn("md/raid:%s: not enough configured devices (%d, minimum 4)\n",
>> -			mdname(mddev), mddev->raid_disks);
>> +	if ((mddev->new_level != 6 && mddev->raid_disks < 2) ||
>> +	    (mddev->new_level == 6 && mddev->raid_disks < 4)) {
>> +		pr_warn("md/raid:%s: not enough configured devices (%d, minimum %d)\n",
>> +			mdname(mddev), mddev->raid_disks,
>> +			mddev->new_level == 6 ? 4 : 2);
> [Severity: High]
> This isn't a bug introduced by this patch, but does this fully prevent
> invalid array geometries from causing panics during array assembly?
>
> In raid5_run(), if there is an active reshape and the array is assembled with
> insufficient data disks, we calculate new_data_disks and divide by it before
> setup_conf() is called to validate the array geometry:
>
> drivers/md/raid5.c:raid5_run() {
>      if (mddev->reshape_position != MaxSector) {
>          ...
>          new_data_disks = mddev->raid_disks - max_degraded;
>          if (sector_div(here_new, chunk_sectors * new_data_disks)) {
>              ...
>          }
>          ...
>          sector_div(here_old, chunk_sectors * (old_disks-max_degraded));
>      }
>      ...
>      if (mddev->private == NULL)
>          conf = setup_conf(mddev);
> }
This looks like a valid problem as well.
>
> If a crafted superblock sets raid_disks <= max_degraded and reshape_position
> is not MaxSector, these divisions by zero are reached before the new checks
> in setup_conf() can reject the array.
>
> Could these checks be moved earlier in the assembly path to prevent the
> divide-by-zero panics?
>
>>   		return ERR_PTR(-EINVAL);
>>   	}
>>
-- 
Thanks,
Kuai

Re: [PATCH v3] md/raid5: reject raid4/5 arrays with too few disks

[yu kuai]

在 2026/9/11 12:15, Chen Cheng 写道:

> From: Chen Cheng<chencheng@fnnas.com>
>
> raid4 and raid5 require at least two disks, while raid6 requires at least
> four disks. setup_conf() only rejects raid6 arrays with fewer than four
> disks. As a result, an invalid raid4 or raid5 array can be assembled with
> no data disks and later reach raid5_set_limits().
>
> Reject such arrays before the r5conf is created. This prevents the invalid
> geometry from reaching queue limit setup, where roundup_pow_of_two() can be
> called with a zero stripe size.
>
> Reported-by:syzbot+de94ddbfff0c9e6fe030@syzkaller.appspotmail.com
> Closes:https://lore.kernel.org/r/6a778c82.01d0871a.3a0d52.006a.GAE@google.com/
> Fixes: f63f17350e53 ("md/raid5: use the atomic queue limit update APIs")
> Signed-off-by: Chen Cheng<chencheng@fnnas.com>
> ---
> v1->v2:
> - Fix wrong tags
> - Check raid-4 also
>
> v2->v3:
> - Keep two-disk raid5 arrays valid by rejecting raid4/5 only below two disks
> ---
>   drivers/md/raid5.c | 8 +++++---
>   1 file changed, 5 insertions(+), 3 deletions(-)
Applied to md-7.3

-- 
Thanks,
Kuai