[PATCH] ocfs2: fix check if list iterator did find an element

Jakob Koschel posted 1 patch 4 years, 3 months ago
fs/ocfs2/dlm/dlmdebug.c | 12 ++++++------
1 file changed, 6 insertions(+), 6 deletions(-)
[PATCH] ocfs2: fix check if list iterator did find an element
Posted by Jakob Koschel 4 years, 3 months ago
Instead of setting 'res' to NULL, it should only be set if
the suitable element was found.

In the original code 'res' would have been set to an incorrect pointer
if the list is empty.

In preparation to limit the scope of the list iterator to the list
traversal loop, use a dedicated pointer pointing to the found element [1].

Link: https://lore.kernel.org/all/YhdfEIwI4EdtHdym@kroah.com/
Signed-off-by: Jakob Koschel <jakobkoschel@gmail.com>
---
 fs/ocfs2/dlm/dlmdebug.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
index d442cf5dda8a..be5e9ed7da8d 100644
--- a/fs/ocfs2/dlm/dlmdebug.c
+++ b/fs/ocfs2/dlm/dlmdebug.c
@@ -541,7 +541,7 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
 	struct debug_lockres *dl = m->private;
 	struct dlm_ctxt *dlm = dl->dl_ctxt;
 	struct dlm_lock_resource *oldres = dl->dl_res;
-	struct dlm_lock_resource *res = NULL;
+	struct dlm_lock_resource *res = NULL, *iter;
 	struct list_head *track_list;

 	spin_lock(&dlm->track_lock);
@@ -556,11 +556,11 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
 		}
 	}

-	list_for_each_entry(res, track_list, tracking) {
-		if (&res->tracking == &dlm->tracking_list)
-			res = NULL;
-		else
-			dlm_lockres_get(res);
+	list_for_each_entry(iter, track_list, tracking) {
+		if (&iter->tracking != &dlm->tracking_list) {
+			dlm_lockres_get(iter);
+			res = iter;
+		}
 		break;
 	}
 	spin_unlock(&dlm->track_lock);

base-commit: 34e047aa16c0123bbae8e2f6df33e5ecc1f56601
--
2.25.1
Re: [PATCH] ocfs2: fix check if list iterator did find an element
Posted by Joseph Qi 4 years, 3 months ago

On 3/20/22 4:31 AM, Jakob Koschel wrote:
> Instead of setting 'res' to NULL, it should only be set if
> the suitable element was found.
> 
> In the original code 'res' would have been set to an incorrect pointer
> if the list is empty.
> 
The logic before iteration can make sure track_list won't be empty.
Please refer the discussion via:
https://lore.kernel.org/ocfs2-devel/bd0ec87e-b490-83dc-2363-5e5342c59fa4@linux.alibaba.com/T/#m96d4397930201d83d68677c33a9721ae8dbd8f15

Thanks,
Joseph

> In preparation to limit the scope of the list iterator to the list
> traversal loop, use a dedicated pointer pointing to the found element [1].
> 
> Link: https://lore.kernel.org/all/YhdfEIwI4EdtHdym@kroah.com/
> Signed-off-by: Jakob Koschel <jakobkoschel@gmail.com>
> ---
>  fs/ocfs2/dlm/dlmdebug.c | 12 ++++++------
>  1 file changed, 6 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
> index d442cf5dda8a..be5e9ed7da8d 100644
> --- a/fs/ocfs2/dlm/dlmdebug.c
> +++ b/fs/ocfs2/dlm/dlmdebug.c
> @@ -541,7 +541,7 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
>  	struct debug_lockres *dl = m->private;
>  	struct dlm_ctxt *dlm = dl->dl_ctxt;
>  	struct dlm_lock_resource *oldres = dl->dl_res;
> -	struct dlm_lock_resource *res = NULL;
> +	struct dlm_lock_resource *res = NULL, *iter;
>  	struct list_head *track_list;
> 
>  	spin_lock(&dlm->track_lock);
> @@ -556,11 +556,11 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
>  		}
>  	}
> 
> -	list_for_each_entry(res, track_list, tracking) {
> -		if (&res->tracking == &dlm->tracking_list)
> -			res = NULL;
> -		else
> -			dlm_lockres_get(res);
> +	list_for_each_entry(iter, track_list, tracking) {
> +		if (&iter->tracking != &dlm->tracking_list) {
> +			dlm_lockres_get(iter);
> +			res = iter;
> +		}
>  		break;
>  	}
>  	spin_unlock(&dlm->track_lock);
> 
> base-commit: 34e047aa16c0123bbae8e2f6df33e5ecc1f56601
> --
> 2.25.1
Re: [PATCH] ocfs2: fix check if list iterator did find an element
Posted by Jakob Koschel 4 years, 3 months ago
> On 21. Mar 2022, at 02:50, Joseph Qi <joseph.qi@linux.alibaba.com> wrote:
> 
> 
> 
> On 3/20/22 4:31 AM, Jakob Koschel wrote:
>> Instead of setting 'res' to NULL, it should only be set if
>> the suitable element was found.
>> 
>> In the original code 'res' would have been set to an incorrect pointer
>> if the list is empty.
>> 
> The logic before iteration can make sure track_list won't be empty.
> Please refer the discussion via:
> https://lore.kernel.org/ocfs2-devel/bd0ec87e-b490-83dc-2363-5e5342c59fa4@linux.alibaba.com/T/#m96d4397930201d83d68677c33a9721ae8dbd8f15

ah yes, I just read up on the discussion there, sorry for having duplicated it
here.

Was any conclusion reached there which fixes can/should be merged?

This code obviously can always be safe if the list cannot be empty.
That's also not necessarily the reason I'm fixing this. The reason is that
we want to get rid of any use of the list iterator variable after the loop
('res' in this case). This will allow moving the list iterator variable
into the scope of the list iterator macro to forbid any invalid use of it
at compile time. Like this you don't have to rely on assumptions that are
hard to validate (e.g. that a certain list is never empty).

The patch here is the minimal change to simply do that but looking at
Dan Carpenter patch there might be more things in this code that can
be simplified.

[CC'd Dan Carpenter]

See [1] for changes that have already been merged:

[1] https://lore.kernel.org/linux-kernel/20220308171818.384491-3-jakobkoschel@gmail.com/

> 
> Thanks,
> Joseph
> 
>> In preparation to limit the scope of the list iterator to the list
>> traversal loop, use a dedicated pointer pointing to the found element [1].
>> 
>> Link: https://lore.kernel.org/all/YhdfEIwI4EdtHdym@kroah.com/
>> Signed-off-by: Jakob Koschel <jakobkoschel@gmail.com>
>> ---
>> fs/ocfs2/dlm/dlmdebug.c | 12 ++++++------
>> 1 file changed, 6 insertions(+), 6 deletions(-)
>> 
>> diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
>> index d442cf5dda8a..be5e9ed7da8d 100644
>> --- a/fs/ocfs2/dlm/dlmdebug.c
>> +++ b/fs/ocfs2/dlm/dlmdebug.c
>> @@ -541,7 +541,7 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
>> 	struct debug_lockres *dl = m->private;
>> 	struct dlm_ctxt *dlm = dl->dl_ctxt;
>> 	struct dlm_lock_resource *oldres = dl->dl_res;
>> -	struct dlm_lock_resource *res = NULL;
>> +	struct dlm_lock_resource *res = NULL, *iter;
>> 	struct list_head *track_list;
>> 
>> 	spin_lock(&dlm->track_lock);
>> @@ -556,11 +556,11 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
>> 		}
>> 	}
>> 
>> -	list_for_each_entry(res, track_list, tracking) {
>> -		if (&res->tracking == &dlm->tracking_list)
>> -			res = NULL;
>> -		else
>> -			dlm_lockres_get(res);
>> +	list_for_each_entry(iter, track_list, tracking) {
>> +		if (&iter->tracking != &dlm->tracking_list) {
>> +			dlm_lockres_get(iter);
>> +			res = iter;
>> +		}
>> 		break;
>> 	}
>> 	spin_unlock(&dlm->track_lock);
>> 
>> base-commit: 34e047aa16c0123bbae8e2f6df33e5ecc1f56601
>> --
>> 2.25.1

	Jakob
Re: [PATCH] ocfs2: fix check if list iterator did find an element
Posted by Dan Carpenter 4 years, 3 months ago
On Mon, Mar 21, 2022 at 02:34:34PM +0100, Jakob Koschel wrote:
> >> @@ -556,11 +556,11 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
> >> 		}
> >> 	}
> >> 
> >> -	list_for_each_entry(res, track_list, tracking) {
> >> -		if (&res->tracking == &dlm->tracking_list)
> >> -			res = NULL;
> >> -		else
> >> -			dlm_lockres_get(res);
> >> +	list_for_each_entry(iter, track_list, tracking) {
> >> +		if (&iter->tracking != &dlm->tracking_list) {

This is an open coded version of:

	if (!list_entry_is_head(iter, &dlm->tracking_list, tracking)) {

Ideally someone would come through with enough confidence to just delete
it but the second best option is to just make it readable...

regards,
dan carpenter

> >> +			dlm_lockres_get(iter);
> >> +			res = iter;
> >> +		}
> >> 		break;
> >> 	}
RE: [PATCH] ocfs2: fix check if list iterator did find an element
Posted by David Laight 4 years, 3 months ago
From: Dan Carpenter
> Sent: 21 March 2022 13:55
> On Mon, Mar 21, 2022 at 02:34:34PM +0100, Jakob Koschel wrote:
> > >> @@ -556,11 +556,11 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
> > >> 		}
> > >> 	}
> > >>
> > >> -	list_for_each_entry(res, track_list, tracking) {
> > >> -		if (&res->tracking == &dlm->tracking_list)
> > >> -			res = NULL;
> > >> -		else
> > >> -			dlm_lockres_get(res);
> > >> +	list_for_each_entry(iter, track_list, tracking) {
> > >> +		if (&iter->tracking != &dlm->tracking_list) {
> 
> This is an open coded version of:
> 
> 	if (!list_entry_is_head(iter, &dlm->tracking_list, tracking)) {

Doesn't list_for_each_entry() terminate before that happens?
So this code is probably still horribly broken.

My worry about bugs with these lists isn't really the code
that uses list_for_each_entry() - they are fairly easy to
locate, read and fix.
The problem is all the other code that is scanning these
list in other more obscure ways.

It really isn't a list structure that is easy to use at all.
It's only slight advantage is that you can unlink an item
without knowing the list head.
(Which can stop buggy code generating cross-linked lists.)

Unless you actually need to traverse backwards the 'back
pointer points to forwards pointer' list is much better.
Even if you have to maintain a 'pointer to last' to get
FIFO operation.

The other option is to double-link the items into a loop and
have a 'head' that points to the first item.

Both these loops have 'pointer to items' to don't need
container_of() and get better type checking from the compiler.

	David

> 
> Ideally someone would come through with enough confidence to just delete
> it but the second best option is to just make it readable...
> 
> regards,
> dan carpenter
> 
> > >> +			dlm_lockres_get(iter);
> > >> +			res = iter;
> > >> +		}
> > >> 		break;
> > >> 	}

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
Re: [PATCH] ocfs2: fix check if list iterator did find an element
Posted by Dan Carpenter 4 years, 3 months ago
On Mon, Mar 21, 2022 at 04:00:10PM +0000, David Laight wrote:
> From: Dan Carpenter
> > Sent: 21 March 2022 13:55
> > On Mon, Mar 21, 2022 at 02:34:34PM +0100, Jakob Koschel wrote:
> > > >> @@ -556,11 +556,11 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
> > > >> 		}
> > > >> 	}
> > > >>
> > > >> -	list_for_each_entry(res, track_list, tracking) {
> > > >> -		if (&res->tracking == &dlm->tracking_list)
> > > >> -			res = NULL;
> > > >> -		else
> > > >> -			dlm_lockres_get(res);
> > > >> +	list_for_each_entry(iter, track_list, tracking) {
> > > >> +		if (&iter->tracking != &dlm->tracking_list) {
> > 
> > This is an open coded version of:
> > 
> > 	if (!list_entry_is_head(iter, &dlm->tracking_list, tracking)) {
> 
> Doesn't list_for_each_entry() terminate before that happens?
> So this code is probably still horribly broken.

There is ! in the condition.  It's just a complicated way of writing
if (!0) { so code works fine, it's just messy.

regards,
dan carpenter
Re: [PATCH] ocfs2: fix check if list iterator did find an element
Posted by Joseph Qi 4 years, 3 months ago

On 3/21/22 9:34 PM, Jakob Koschel wrote:
> 
>> On 21. Mar 2022, at 02:50, Joseph Qi <joseph.qi@linux.alibaba.com> wrote:
>>
>>
>>
>> On 3/20/22 4:31 AM, Jakob Koschel wrote:
>>> Instead of setting 'res' to NULL, it should only be set if
>>> the suitable element was found.
>>>
>>> In the original code 'res' would have been set to an incorrect pointer
>>> if the list is empty.
>>>
>> The logic before iteration can make sure track_list won't be empty.
>> Please refer the discussion via:
>> https://lore.kernel.org/ocfs2-devel/bd0ec87e-b490-83dc-2363-5e5342c59fa4@linux.alibaba.com/T/#m96d4397930201d83d68677c33a9721ae8dbd8f15
> 
> ah yes, I just read up on the discussion there, sorry for having duplicated it
> here.
> 
> Was any conclusion reached there which fixes can/should be merged?
> 
> This code obviously can always be safe if the list cannot be empty.
> That's also not necessarily the reason I'm fixing this. The reason is that
> we want to get rid of any use of the list iterator variable after the loop
> ('res' in this case). This will allow moving the list iterator variable
> into the scope of the list iterator macro to forbid any invalid use of it
> at compile time. Like this you don't have to rely on assumptions that are
> hard to validate (e.g. that a certain list is never empty).
> 
> The patch here is the minimal change to simply do that but looking at
> Dan Carpenter patch there might be more things in this code that can
> be simplified.
> 
Agree, so I'm fine with this change.
So could you please update the description and send v2?

Thanks,
Joseph

> [CC'd Dan Carpenter]
> 
> See [1] for changes that have already been merged:
> 
> [1] https://lore.kernel.org/linux-kernel/20220308171818.384491-3-jakobkoschel@gmail.com/
> 
>>
>> Thanks,
>> Joseph
>>
>>> In preparation to limit the scope of the list iterator to the list
>>> traversal loop, use a dedicated pointer pointing to the found element [1].
>>>
>>> Link: https://lore.kernel.org/all/YhdfEIwI4EdtHdym@kroah.com/
>>> Signed-off-by: Jakob Koschel <jakobkoschel@gmail.com>
>>> ---
>>> fs/ocfs2/dlm/dlmdebug.c | 12 ++++++------
>>> 1 file changed, 6 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
>>> index d442cf5dda8a..be5e9ed7da8d 100644
>>> --- a/fs/ocfs2/dlm/dlmdebug.c
>>> +++ b/fs/ocfs2/dlm/dlmdebug.c
>>> @@ -541,7 +541,7 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
>>> 	struct debug_lockres *dl = m->private;
>>> 	struct dlm_ctxt *dlm = dl->dl_ctxt;
>>> 	struct dlm_lock_resource *oldres = dl->dl_res;
>>> -	struct dlm_lock_resource *res = NULL;
>>> +	struct dlm_lock_resource *res = NULL, *iter;
>>> 	struct list_head *track_list;
>>>
>>> 	spin_lock(&dlm->track_lock);
>>> @@ -556,11 +556,11 @@ static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
>>> 		}
>>> 	}
>>>
>>> -	list_for_each_entry(res, track_list, tracking) {
>>> -		if (&res->tracking == &dlm->tracking_list)
>>> -			res = NULL;
>>> -		else
>>> -			dlm_lockres_get(res);
>>> +	list_for_each_entry(iter, track_list, tracking) {
>>> +		if (&iter->tracking != &dlm->tracking_list) {
>>> +			dlm_lockres_get(iter);
>>> +			res = iter;
>>> +		}
>>> 		break;
>>> 	}
>>> 	spin_unlock(&dlm->track_lock);
>>>
>>> base-commit: 34e047aa16c0123bbae8e2f6df33e5ecc1f56601
>>> --
>>> 2.25.1
> 
> 	Jakob