[v1] ocfs2: Avoid NULL pointer dereference in dx_dir_lookup_rec()

[PATCH] ocfs2: Avoid NULL pointer dereference in dx_dir_lookup_rec()

Posted by Ivan Pravdin 7 months, 2 weeks ago

When a directory entry is not found, ocfs2_dx_dir_lookup_rec() prints an
error message that unconditionally dereferences the 'rec' pointer.
However, if 'rec' is NULL, this leads to a NULL pointer dereference and
a kernel panic.

Add an explicit check for a NULL 'rec' and use an alternate error
message in that case to avoid unsafe access.

Reported-by: syzbot+20282c1b2184a857ac4c@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/67483b75.050a0220.253251.007c.GAE@google.com/T/
Signed-off-by: Ivan Pravdin <ipravdin.official@gmail.com>
---
 fs/ocfs2/dir.c | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
index 7799f4d16ce9..dccf0349e523 100644
--- a/fs/ocfs2/dir.c
+++ b/fs/ocfs2/dir.c
@@ -809,11 +809,17 @@ static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
 	}
 
 	if (!found) {
-		ret = ocfs2_error(inode->i_sb,
-				  "Inode %lu has bad extent record (%u, %u, 0) in btree\n",
-				  inode->i_ino,
-				  le32_to_cpu(rec->e_cpos),
-				  ocfs2_rec_clusters(el, rec));
+		if (rec) {
+			ret = ocfs2_error(inode->i_sb,
+					"Inode %lu has bad extent record (%u, %u, 0) in btree\n",
+					inode->i_ino,
+					le32_to_cpu(rec->e_cpos),
+					ocfs2_rec_clusters(el, rec));
+		} else {
+			ret = ocfs2_error(inode->i_sb,
+					"Inode %lu has no extent records in btree\n",
+					inode->i_ino);
+		}
 		goto out;
 	}
 
-- 
2.45.2

Re: [PATCH] ocfs2: Avoid NULL pointer dereference in dx_dir_lookup_rec()

Posted by Joseph Qi 7 months, 1 week ago

Hi,


On 2025/6/27 10:38, Ivan Pravdin wrote:
> When a directory entry is not found, ocfs2_dx_dir_lookup_rec() prints an
> error message that unconditionally dereferences the 'rec' pointer.
> However, if 'rec' is NULL, this leads to a NULL pointer dereference and
> a kernel panic.
> 

This looks possible, but syzbot reports slab-out-of-bounds Read in
ocfs2_dx_dir_lookup_rec(), not NULL pointer dereference.

So I think it is because it construct a malicious image and set a wrong
l_recs, then access this damaged l_recs.

Thanks,
Joseph


> Add an explicit check for a NULL 'rec' and use an alternate error
> message in that case to avoid unsafe access.
> 
> Reported-by: syzbot+20282c1b2184a857ac4c@syzkaller.appspotmail.com
> Closes: https://lore.kernel.org/all/67483b75.050a0220.253251.007c.GAE@google.com/T/
> Signed-off-by: Ivan Pravdin <ipravdin.official@gmail.com>
> ---
>  fs/ocfs2/dir.c | 16 +++++++++++-----
>  1 file changed, 11 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
> index 7799f4d16ce9..dccf0349e523 100644
> --- a/fs/ocfs2/dir.c
> +++ b/fs/ocfs2/dir.c
> @@ -809,11 +809,17 @@ static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
>  	}
>  
>  	if (!found) {
> -		ret = ocfs2_error(inode->i_sb,
> -				  "Inode %lu has bad extent record (%u, %u, 0) in btree\n",
> -				  inode->i_ino,
> -				  le32_to_cpu(rec->e_cpos),
> -				  ocfs2_rec_clusters(el, rec));
> +		if (rec) {
> +			ret = ocfs2_error(inode->i_sb,
> +					"Inode %lu has bad extent record (%u, %u, 0) in btree\n",
> +					inode->i_ino,
> +					le32_to_cpu(rec->e_cpos),
> +					ocfs2_rec_clusters(el, rec));
> +		} else {
> +			ret = ocfs2_error(inode->i_sb,
> +					"Inode %lu has no extent records in btree\n",
> +					inode->i_ino);
> +		}
>  		goto out;
>  	}
>

Re: [PATCH] ocfs2: Avoid NULL pointer dereference in dx_dir_lookup_rec()

Posted by Heming Zhao 7 months, 1 week ago

On 6/30/25 09:26, Joseph Qi wrote:
> Hi,
> 
> 
> On 2025/6/27 10:38, Ivan Pravdin wrote:
>> When a directory entry is not found, ocfs2_dx_dir_lookup_rec() prints an
>> error message that unconditionally dereferences the 'rec' pointer.
>> However, if 'rec' is NULL, this leads to a NULL pointer dereference and
>> a kernel panic.
>>
> 
> This looks possible, but syzbot reports slab-out-of-bounds Read in
> ocfs2_dx_dir_lookup_rec(), not NULL pointer dereference.
> 
> So I think it is because it construct a malicious image and set a wrong
> l_recs, then access this damaged l_recs.
> 
> Thanks,
> Joseph

I think this proposed fix (at least the fix method) is acceptable.
the crash occurs at ocfs2_error(), where the pointer 'rec' must be incorrect.
look back at the previous code lines:
         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
                 rec = &el->l_recs[i];

                 if (le32_to_cpu(rec->e_cpos) <= major_hash) {
                         found = 1;
                         break;
                 }
         }

either 'el->l_next_free_rec' or 'el->l_recs[i]' has an incorrect value.
we can do nothing about this kind of error, simply returning errno to caller is sufficient.

btw, ocfs2-tools has a similar function "static errcode_t ocfs2_dx_dir_lookup_rec()"@libocfs2/dir_indexed.c, which sets ret with OCFS2_ET_CORRUPT_EXTENT_BLOCK and return.

- Heming

> 
> 
>> Add an explicit check for a NULL 'rec' and use an alternate error
>> message in that case to avoid unsafe access.
>>
>> Reported-by: syzbot+20282c1b2184a857ac4c@syzkaller.appspotmail.com
>> Closes: https://lore.kernel.org/all/67483b75.050a0220.253251.007c.GAE@google.com/T/
>> Signed-off-by: Ivan Pravdin <ipravdin.official@gmail.com>
>> ---
>>   fs/ocfs2/dir.c | 16 +++++++++++-----
>>   1 file changed, 11 insertions(+), 5 deletions(-)
>>
>> diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
>> index 7799f4d16ce9..dccf0349e523 100644
>> --- a/fs/ocfs2/dir.c
>> +++ b/fs/ocfs2/dir.c
>> @@ -809,11 +809,17 @@ static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
>>   	}
>>   
>>   	if (!found) {
>> -		ret = ocfs2_error(inode->i_sb,
>> -				  "Inode %lu has bad extent record (%u, %u, 0) in btree\n",
>> -				  inode->i_ino,
>> -				  le32_to_cpu(rec->e_cpos),
>> -				  ocfs2_rec_clusters(el, rec));
>> +		if (rec) {
>> +			ret = ocfs2_error(inode->i_sb,
>> +					"Inode %lu has bad extent record (%u, %u, 0) in btree\n",
>> +					inode->i_ino,
>> +					le32_to_cpu(rec->e_cpos),
>> +					ocfs2_rec_clusters(el, rec));
>> +		} else {
>> +			ret = ocfs2_error(inode->i_sb,
>> +					"Inode %lu has no extent records in btree\n",
>> +					inode->i_ino);
>> +		}
>>   		goto out;
>>   	}
>>   
> 
>

Re: [PATCH] ocfs2: Avoid NULL pointer dereference in dx_dir_lookup_rec()

Posted by Joseph Qi 7 months, 1 week ago


On 2025/6/30 10:32, Heming Zhao wrote:
> On 6/30/25 09:26, Joseph Qi wrote:
>> Hi,
>>
>>
>> On 2025/6/27 10:38, Ivan Pravdin wrote:
>>> When a directory entry is not found, ocfs2_dx_dir_lookup_rec() prints an
>>> error message that unconditionally dereferences the 'rec' pointer.
>>> However, if 'rec' is NULL, this leads to a NULL pointer dereference and
>>> a kernel panic.
>>>
>>
>> This looks possible, but syzbot reports slab-out-of-bounds Read in
>> ocfs2_dx_dir_lookup_rec(), not NULL pointer dereference.
>>
>> So I think it is because it construct a malicious image and set a wrong
>> l_recs, then access this damaged l_recs.
>>
>> Thanks,
>> Joseph
> 
> I think this proposed fix (at least the fix method) is acceptable.

But this seems different from the issue syzbot reports, right?
https://lore.kernel.org/all/67483b75.050a0220.253251.007c.GAE@google.com/T/

Thanks,
Joseph

> the crash occurs at ocfs2_error(), where the pointer 'rec' must be incorrect.
> look back at the previous code lines:
>         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
>                 rec = &el->l_recs[i];
> 
>                 if (le32_to_cpu(rec->e_cpos) <= major_hash) {
>                         found = 1;
>                         break;
>                 }
>         }
> 
> either 'el->l_next_free_rec' or 'el->l_recs[i]' has an incorrect value.
> we can do nothing about this kind of error, simply returning errno to caller is sufficient.
> 
> btw, ocfs2-tools has a similar function "static errcode_t ocfs2_dx_dir_lookup_rec()"@libocfs2/dir_indexed.c, which sets ret with OCFS2_ET_CORRUPT_EXTENT_BLOCK and return.
>

Re: [PATCH] ocfs2: Avoid NULL pointer dereference in dx_dir_lookup_rec()

Posted by Heming Zhao 7 months, 1 week ago

On 6/30/25 14:58, Joseph Qi wrote:
> 
> 
> On 2025/6/30 10:32, Heming Zhao wrote:
>> On 6/30/25 09:26, Joseph Qi wrote:
>>> Hi,
>>>
>>>
>>> On 2025/6/27 10:38, Ivan Pravdin wrote:
>>>> When a directory entry is not found, ocfs2_dx_dir_lookup_rec() prints an
>>>> error message that unconditionally dereferences the 'rec' pointer.
>>>> However, if 'rec' is NULL, this leads to a NULL pointer dereference and
>>>> a kernel panic.
>>>>
>>>
>>> This looks possible, but syzbot reports slab-out-of-bounds Read in
>>> ocfs2_dx_dir_lookup_rec(), not NULL pointer dereference.
>>>
>>> So I think it is because it construct a malicious image and set a wrong
>>> l_recs, then access this damaged l_recs.
>>>
>>> Thanks,
>>> Joseph
>>
>> I think this proposed fix (at least the fix method) is acceptable.
> 
> But this seems different from the issue syzbot reports, right?
> https://lore.kernel.org/all/67483b75.050a0220.253251.007c.GAE@google.com/T/
> 
> Thanks,
> Joseph

OK, I think I may have found the problem. In the URL above,
there are two issues/reports:
1> From: syzbot @ 2024-11-28  9:44 UTC
2> From: syzbot @ 2025-03-09 11:40 UTC

Your review comments about report <1>; Ivan's patch is about report <2>.

And the dashboard shows the <2>:
- https://syzkaller.appspot.com/bug?extid=20282c1b2184a857ac4c

It seems syzbot changed the fuzz code and won't trigger issue <1>.

For issue <1>, the rec was an illegal value, and made KASAN report a bug.

> 
>> the crash occurs at ocfs2_error(), where the pointer 'rec' must be incorrect.
>> look back at the previous code lines:
>>          for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
>>                  rec = &el->l_recs[i];
>>
>>                  if (le32_to_cpu(rec->e_cpos) <= major_hash) {
>>                          found = 1;
>>                          break;
>>                  }
>>          }
>>
>> either 'el->l_next_free_rec' or 'el->l_recs[i]' has an incorrect value.
>> we can do nothing about this kind of error, simply returning errno to caller is sufficient.
>>
>> btw, ocfs2-tools has a similar function "static errcode_t ocfs2_dx_dir_lookup_rec()"@libocfs2/dir_indexed.c, which sets ret with OCFS2_ET_CORRUPT_EXTENT_BLOCK and return.
>>
> 
> 
>

Re: [PATCH] ocfs2: Avoid NULL pointer dereference in dx_dir_lookup_rec()

Posted by Ivan Pravdin 7 months, 1 week ago

On Mon, Jun 30, 2025 at 10:32:35AM GMT, Heming Zhao wrote:
> On 6/30/25 09:26, Joseph Qi wrote:
> > Hi,
> > 
> > 
> > On 2025/6/27 10:38, Ivan Pravdin wrote:
> > > When a directory entry is not found, ocfs2_dx_dir_lookup_rec() prints an
> > > error message that unconditionally dereferences the 'rec' pointer.
> > > However, if 'rec' is NULL, this leads to a NULL pointer dereference and
> > > a kernel panic.
> > > 
> > 
> > This looks possible, but syzbot reports slab-out-of-bounds Read in
> > ocfs2_dx_dir_lookup_rec(), not NULL pointer dereference.
> > 
> > So I think it is because it construct a malicious image and set a wrong
> > l_recs, then access this damaged l_recs.
> > 
> > Thanks,
> > Joseph
> 
> I think this proposed fix (at least the fix method) is acceptable.
> the crash occurs at ocfs2_error(), where the pointer 'rec' must be incorrect.
> look back at the previous code lines:
>         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
>                 rec = &el->l_recs[i];
> 
>                 if (le32_to_cpu(rec->e_cpos) <= major_hash) {
>                         found = 1;
>                         break;
>                 }
>         }
> 
> either 'el->l_next_free_rec' or 'el->l_recs[i]' has an incorrect value.
> we can do nothing about this kind of error, simply returning errno to caller is sufficient.
> 
> btw, ocfs2-tools has a similar function "static errcode_t ocfs2_dx_dir_lookup_rec()"@libocfs2/dir_indexed.c, which sets ret with OCFS2_ET_CORRUPT_EXTENT_BLOCK and return.

Thanks, I will include it in v2.

> 
> - Heming
> 
> > 
> > 
> > > Add an explicit check for a NULL 'rec' and use an alternate error
> > > message in that case to avoid unsafe access.
> > > 
> > > Reported-by: syzbot+20282c1b2184a857ac4c@syzkaller.appspotmail.com
> > > Closes: https://lore.kernel.org/all/67483b75.050a0220.253251.007c.GAE@google.com/T/
> > > Signed-off-by: Ivan Pravdin <ipravdin.official@gmail.com>
> > > ---
> > >   fs/ocfs2/dir.c | 16 +++++++++++-----
> > >   1 file changed, 11 insertions(+), 5 deletions(-)
> > > 
> > > diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
> > > index 7799f4d16ce9..dccf0349e523 100644
> > > --- a/fs/ocfs2/dir.c
> > > +++ b/fs/ocfs2/dir.c
> > > @@ -809,11 +809,17 @@ static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
> > >   	}
> > >   	if (!found) {
> > > -		ret = ocfs2_error(inode->i_sb,
> > > -				  "Inode %lu has bad extent record (%u, %u, 0) in btree\n",
> > > -				  inode->i_ino,
> > > -				  le32_to_cpu(rec->e_cpos),
> > > -				  ocfs2_rec_clusters(el, rec));
> > > +		if (rec) {
> > > +			ret = ocfs2_error(inode->i_sb,
> > > +					"Inode %lu has bad extent record (%u, %u, 0) in btree\n",
> > > +					inode->i_ino,
> > > +					le32_to_cpu(rec->e_cpos),
> > > +					ocfs2_rec_clusters(el, rec));
> > > +		} else {
> > > +			ret = ocfs2_error(inode->i_sb,
> > > +					"Inode %lu has no extent records in btree\n",
> > > +					inode->i_ino);
> > > +		}
> > >   		goto out;
> > >   	}
> > 
> > 
> 

	Ivan Pravdin

Re: [PATCH] ocfs2: Avoid NULL pointer dereference in dx_dir_lookup_rec()

Posted by Ivan Pravdin 7 months, 1 week ago

On Sun, Jun 29, 2025 at 10:45:15PM GMT, Ivan Pravdin wrote:
> On Mon, Jun 30, 2025 at 10:32:35AM GMT, Heming Zhao wrote:
> > On 6/30/25 09:26, Joseph Qi wrote:
> > > Hi,
> > > 
> > > 
> > > On 2025/6/27 10:38, Ivan Pravdin wrote:
> > > > When a directory entry is not found, ocfs2_dx_dir_lookup_rec() prints an
> > > > error message that unconditionally dereferences the 'rec' pointer.
> > > > However, if 'rec' is NULL, this leads to a NULL pointer dereference and
> > > > a kernel panic.
> > > > 
> > > 
> > > This looks possible, but syzbot reports slab-out-of-bounds Read in
> > > ocfs2_dx_dir_lookup_rec(), not NULL pointer dereference.
> > > 
> > > So I think it is because it construct a malicious image and set a wrong
> > > l_recs, then access this damaged l_recs.
> > > 
> > > Thanks,
> > > Joseph
> > 
> > I think this proposed fix (at least the fix method) is acceptable.
> > the crash occurs at ocfs2_error(), where the pointer 'rec' must be incorrect.
> > look back at the previous code lines:
> >         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
> >                 rec = &el->l_recs[i];
> > 
> >                 if (le32_to_cpu(rec->e_cpos) <= major_hash) {
> >                         found = 1;
> >                         break;
> >                 }
> >         }
> > 
> > either 'el->l_next_free_rec' or 'el->l_recs[i]' has an incorrect value.
> > we can do nothing about this kind of error, simply returning errno to caller is sufficient.
> > 
> > btw, ocfs2-tools has a similar function "static errcode_t ocfs2_dx_dir_lookup_rec()"@libocfs2/dir_indexed.c, which sets ret with OCFS2_ET_CORRUPT_EXTENT_BLOCK and return.
> 
> Thanks, I will include it in v2.

I have realized that ret is already set to EROFS that acts as a general
FS errno. Unless there are any other objectives this is the latest
version.

	Ivan Pravdin