jfs: upper bound check of tree index in dbAllocAG

[PATCH] jfs: upper bound check of tree index in dbAllocAG

Posted by Arnaud Lecomte 9 months, 3 weeks ago

When computing the tree index in dbAllocAG, we never check we are not
out of bounds from the size of the stree.
This could happen in a scenario where the filesystem metadata are
corrupted.

Reported-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=cffd18309153948f3c3e
Tested-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
Fixes: 263e55949d89 ("x86/cpu/amd: Fix workaround for erratum 1054")
Signed-off-by: Arnaud Lecomte <contact@arnaud-lcm.com>
---
 fs/jfs/jfs_dmap.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/fs/jfs/jfs_dmap.c b/fs/jfs/jfs_dmap.c
index 26e89d0c69b6..7acebb9a21b0 100644
--- a/fs/jfs/jfs_dmap.c
+++ b/fs/jfs/jfs_dmap.c
@@ -1385,6 +1385,12 @@ dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb, s64 * results)
 	    (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / bmp->db_agwidth;
 	ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
 
+	if (ti < 0 || ti >= le32_to_cpu(dcp->nleafs)) {
+		jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page: ti out of bounds\n");
+		release_metapage(mp);
+		return -EIO;
+	}
+
 	/* dmap control page trees fan-out by 4 and a single allocation
 	 * group may be described by 1 or 2 subtrees within the ag level
 	 * dmap control page, depending upon the ag size. examine the ag's

---
base-commit: 8560697b23dc2f405cb463af2b17256a9888129d
change-id: 20250423-ubsan-jfs-3a58acae5e57

Best regards,
-- 
Arnaud Lecomte <contact@arnaud-lcm.com>

Re: [Jfs-discussion] [PATCH] jfs: upper bound check of tree index in dbAllocAG

Posted by Dave Kleikamp 7 months ago

I'm finally trying to catch up.

On 4/23/25 5:13PM, Arnaud Lecomte via Jfs-discussion wrote:
> When computing the tree index in dbAllocAG, we never check we are not
> out of bounds from the size of the stree.
> This could happen in a scenario where the filesystem metadata are
> corrupted.

Looks good. I'll apply and test this.

> 
> Reported-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
> Closes: https://syzkaller.appspot.com/bug?extid=cffd18309153948f3c3e
> Tested-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
> Fixes: 263e55949d89 ("x86/cpu/amd: Fix workaround for erratum 1054")

Removing the Fixes: line. This doesn't make sense.

> Signed-off-by: Arnaud Lecomte <contact@arnaud-lcm.com>
> ---
>   fs/jfs/jfs_dmap.c | 6 ++++++
>   1 file changed, 6 insertions(+)
> 
> diff --git a/fs/jfs/jfs_dmap.c b/fs/jfs/jfs_dmap.c
> index 26e89d0c69b6..7acebb9a21b0 100644
> --- a/fs/jfs/jfs_dmap.c
> +++ b/fs/jfs/jfs_dmap.c
> @@ -1385,6 +1385,12 @@ dbAllocAG(struct bmap * bmp, int agno, s64 nblocks, int l2nb, s64 * results)
>   	    (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / bmp->db_agwidth;
>   	ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
>   
> +	if (ti < 0 || ti >= le32_to_cpu(dcp->nleafs)) {
> +		jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page: ti out of bounds\n");
> +		release_metapage(mp);
> +		return -EIO;
> +	}
> +
>   	/* dmap control page trees fan-out by 4 and a single allocation
>   	 * group may be described by 1 or 2 subtrees within the ag level
>   	 * dmap control page, depending upon the ag size. examine the ag's
> 
> ---
> base-commit: 8560697b23dc2f405cb463af2b17256a9888129d
> change-id: 20250423-ubsan-jfs-3a58acae5e57
> 
> Best regards,

Re: [Jfs-discussion] [PATCH] jfs: upper bound check of tree index in dbAllocAG

Posted by Arnaud Lecomte 6 months, 2 weeks ago

Lovely, thanks !
Arnaud

On 11/07/2025 20:23, Dave Kleikamp wrote:
> I'm finally trying to catch up.
>
> On 4/23/25 5:13PM, Arnaud Lecomte via Jfs-discussion wrote:
>> When computing the tree index in dbAllocAG, we never check we are not
>> out of bounds from the size of the stree.
>> This could happen in a scenario where the filesystem metadata are
>> corrupted.
>
> Looks good. I'll apply and test this.
>
>>
>> Reported-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
>> Closes: https://syzkaller.appspot.com/bug?extid=cffd18309153948f3c3e
>> Tested-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
>> Fixes: 263e55949d89 ("x86/cpu/amd: Fix workaround for erratum 1054")
>
> Removing the Fixes: line. This doesn't make sense.
>
>> Signed-off-by: Arnaud Lecomte <contact@arnaud-lcm.com>
>> ---
>>   fs/jfs/jfs_dmap.c | 6 ++++++
>>   1 file changed, 6 insertions(+)
>>
>> diff --git a/fs/jfs/jfs_dmap.c b/fs/jfs/jfs_dmap.c
>> index 26e89d0c69b6..7acebb9a21b0 100644
>> --- a/fs/jfs/jfs_dmap.c
>> +++ b/fs/jfs/jfs_dmap.c
>> @@ -1385,6 +1385,12 @@ dbAllocAG(struct bmap * bmp, int agno, s64 
>> nblocks, int l2nb, s64 * results)
>>           (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / 
>> bmp->db_agwidth;
>>       ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
>>   +    if (ti < 0 || ti >= le32_to_cpu(dcp->nleafs)) {
>> +        jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page: ti 
>> out of bounds\n");
>> +        release_metapage(mp);
>> +        return -EIO;
>> +    }
>> +
>>       /* dmap control page trees fan-out by 4 and a single allocation
>>        * group may be described by 1 or 2 subtrees within the ag level
>>        * dmap control page, depending upon the ag size. examine the ag's
>>
>> ---
>> base-commit: 8560697b23dc2f405cb463af2b17256a9888129d
>> change-id: 20250423-ubsan-jfs-3a58acae5e57
>>
>> Best regards,
>
>

Re: [Jfs-discussion] [PATCH] jfs: upper bound check of tree index in dbAllocAG

Posted by Lecomte, Arnaud 6 months, 4 weeks ago

Thanks Dave !

On 11/07/2025 20:23, Dave Kleikamp wrote:
> I'm finally trying to catch up.
>
> On 4/23/25 5:13PM, Arnaud Lecomte via Jfs-discussion wrote:
>> When computing the tree index in dbAllocAG, we never check we are not
>> out of bounds from the size of the stree.
>> This could happen in a scenario where the filesystem metadata are
>> corrupted.
>
> Looks good. I'll apply and test this.
>
>>
>> Reported-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
>> Closes: https://syzkaller.appspot.com/bug?extid=cffd18309153948f3c3e
>> Tested-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
>> Fixes: 263e55949d89 ("x86/cpu/amd: Fix workaround for erratum 1054")
>
> Removing the Fixes: line. This doesn't make sense.
>
>> Signed-off-by: Arnaud Lecomte <contact@arnaud-lcm.com>
>> ---
>>   fs/jfs/jfs_dmap.c | 6 ++++++
>>   1 file changed, 6 insertions(+)
>>
>> diff --git a/fs/jfs/jfs_dmap.c b/fs/jfs/jfs_dmap.c
>> index 26e89d0c69b6..7acebb9a21b0 100644
>> --- a/fs/jfs/jfs_dmap.c
>> +++ b/fs/jfs/jfs_dmap.c
>> @@ -1385,6 +1385,12 @@ dbAllocAG(struct bmap * bmp, int agno, s64 
>> nblocks, int l2nb, s64 * results)
>>           (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / 
>> bmp->db_agwidth;
>>       ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
>>   +    if (ti < 0 || ti >= le32_to_cpu(dcp->nleafs)) {
>> +        jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page: ti 
>> out of bounds\n");
>> +        release_metapage(mp);
>> +        return -EIO;
>> +    }
>> +
>>       /* dmap control page trees fan-out by 4 and a single allocation
>>        * group may be described by 1 or 2 subtrees within the ag level
>>        * dmap control page, depending upon the ag size. examine the ag's
>>
>> ---
>> base-commit: 8560697b23dc2f405cb463af2b17256a9888129d
>> change-id: 20250423-ubsan-jfs-3a58acae5e57
>>
>> Best regards,
>
>

Re: [Jfs-discussion] [PATCH] jfs: upper bound check of tree index in dbAllocAG

Posted by Arnaud Lecomte 6 months, 2 weeks ago

Hi Dave,
I am bumping the thread as I was wondering if you have some updates.
Thanks !

On 14/07/2025 21:35, Lecomte, Arnaud wrote:
> Thanks Dave !
>
> On 11/07/2025 20:23, Dave Kleikamp wrote:
>> I'm finally trying to catch up.
>>
>> On 4/23/25 5:13PM, Arnaud Lecomte via Jfs-discussion wrote:
>>> When computing the tree index in dbAllocAG, we never check we are not
>>> out of bounds from the size of the stree.
>>> This could happen in a scenario where the filesystem metadata are
>>> corrupted.
>>
>> Looks good. I'll apply and test this.
>>
>>>
>>> Reported-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
>>> Closes: https://syzkaller.appspot.com/bug?extid=cffd18309153948f3c3e
>>> Tested-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
>>> Fixes: 263e55949d89 ("x86/cpu/amd: Fix workaround for erratum 1054")
>>
>> Removing the Fixes: line. This doesn't make sense.
>>
>>> Signed-off-by: Arnaud Lecomte <contact@arnaud-lcm.com>
>>> ---
>>>   fs/jfs/jfs_dmap.c | 6 ++++++
>>>   1 file changed, 6 insertions(+)
>>>
>>> diff --git a/fs/jfs/jfs_dmap.c b/fs/jfs/jfs_dmap.c
>>> index 26e89d0c69b6..7acebb9a21b0 100644
>>> --- a/fs/jfs/jfs_dmap.c
>>> +++ b/fs/jfs/jfs_dmap.c
>>> @@ -1385,6 +1385,12 @@ dbAllocAG(struct bmap * bmp, int agno, s64 
>>> nblocks, int l2nb, s64 * results)
>>>           (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / 
>>> bmp->db_agwidth;
>>>       ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
>>>   +    if (ti < 0 || ti >= le32_to_cpu(dcp->nleafs)) {
>>> +        jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page: ti 
>>> out of bounds\n");
>>> +        release_metapage(mp);
>>> +        return -EIO;
>>> +    }
>>> +
>>>       /* dmap control page trees fan-out by 4 and a single allocation
>>>        * group may be described by 1 or 2 subtrees within the ag level
>>>        * dmap control page, depending upon the ag size. examine the 
>>> ag's
>>>
>>> ---
>>> base-commit: 8560697b23dc2f405cb463af2b17256a9888129d
>>> change-id: 20250423-ubsan-jfs-3a58acae5e57
>>>
>>> Best regards,
>>
>>

Re: [Jfs-discussion] [PATCH] jfs: upper bound check of tree index in dbAllocAG

Posted by Dave Kleikamp 6 months, 2 weeks ago

On 7/27/25 4:55AM, Arnaud Lecomte wrote:
> Hi Dave,
> I am bumping the thread as I was wondering if you have some updates.
> Thanks !

It's in linux-next. I'll be pushing it to Linus this week.

Thanks,
Shaggy
> 
> On 14/07/2025 21:35, Lecomte, Arnaud wrote:
>> Thanks Dave !
>>
>> On 11/07/2025 20:23, Dave Kleikamp wrote:
>>> I'm finally trying to catch up.
>>>
>>> On 4/23/25 5:13PM, Arnaud Lecomte via Jfs-discussion wrote:
>>>> When computing the tree index in dbAllocAG, we never check we are not
>>>> out of bounds from the size of the stree.
>>>> This could happen in a scenario where the filesystem metadata are
>>>> corrupted.
>>>
>>> Looks good. I'll apply and test this.
>>>
>>>>
>>>> Reported-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
>>>> Closes: https://syzkaller.appspot.com/bug?extid=cffd18309153948f3c3e
>>>> Tested-by: syzbot+cffd18309153948f3c3e@syzkaller.appspotmail.com
>>>> Fixes: 263e55949d89 ("x86/cpu/amd: Fix workaround for erratum 1054")
>>>
>>> Removing the Fixes: line. This doesn't make sense.
>>>
>>>> Signed-off-by: Arnaud Lecomte <contact@arnaud-lcm.com>
>>>> ---
>>>>   fs/jfs/jfs_dmap.c | 6 ++++++
>>>>   1 file changed, 6 insertions(+)
>>>>
>>>> diff --git a/fs/jfs/jfs_dmap.c b/fs/jfs/jfs_dmap.c
>>>> index 26e89d0c69b6..7acebb9a21b0 100644
>>>> --- a/fs/jfs/jfs_dmap.c
>>>> +++ b/fs/jfs/jfs_dmap.c
>>>> @@ -1385,6 +1385,12 @@ dbAllocAG(struct bmap * bmp, int agno, s64 
>>>> nblocks, int l2nb, s64 * results)
>>>>           (1 << (L2LPERCTL - (bmp->db_agheight << 1))) / bmp- 
>>>> >db_agwidth;
>>>>       ti = bmp->db_agstart + bmp->db_agwidth * (agno & (agperlev - 1));
>>>>   +    if (ti < 0 || ti >= le32_to_cpu(dcp->nleafs)) {
>>>> +        jfs_error(bmp->db_ipbmap->i_sb, "Corrupt dmapctl page: ti 
>>>> out of bounds\n");
>>>> +        release_metapage(mp);
>>>> +        return -EIO;
>>>> +    }
>>>> +
>>>>       /* dmap control page trees fan-out by 4 and a single allocation
>>>>        * group may be described by 1 or 2 subtrees within the ag level
>>>>        * dmap control page, depending upon the ag size. examine the 
>>>> ag's
>>>>
>>>> ---
>>>> base-commit: 8560697b23dc2f405cb463af2b17256a9888129d
>>>> change-id: 20250423-ubsan-jfs-3a58acae5e57
>>>>
>>>> Best regards,
>>>
>>>