Re: [PATCH] scsi: sg: fix memory leak in sg_build_indirect

Re: [PATCH] scsi: sg: fix memory leak in sg_build_indirect

[Markus Elfring]

> Fix a memory leak when there have failed, that we should free the pages
> under the condition rem_sz > 0.

I suggest to improve the change description.

* Please use an imperative wording.

* Will the tag “Fixes” become relevant?

Regards,
Markus

Re: [PATCH] scsi: sg: fix memory leak in sg_build_indirect

[Douglas Gilbert]

On 2020-04-13 3:35 p.m., Markus Elfring wrote:
>> Fix a memory leak when there have failed, that we should free the pages
>> under the condition rem_sz > 0.
> 
> I suggest to improve the change description.
> 
> * Please use an imperative wording.

nominative, vocative, accusative, genitive ...

I have made some comment suggestions in response to the original posts
from the author.

> * Will the tag “Fixes” become relevant?

Yes, both patches from the author are fixes. These leaks would typically
appear on a system under resource pressure (specifically low memory) and
would tend to make the situation worse.

Doug Gilbert

[PATCH] scsi: sg: fix memory leak in sg_build_indirect

[Li Bin]

Fix a memory leak when there have failed, that we should free the pages
under the condition rem_sz > 0.

Signed-off-by: Li Bin <huawei.libin@huawei.com>
---
 drivers/scsi/sg.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c
index 4e6af592..8441ac5 100644
--- a/drivers/scsi/sg.c
+++ b/drivers/scsi/sg.c
@@ -1959,8 +1959,12 @@ static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned lon
 			 k, rem_sz));
 
 	schp->bufflen = blk_size;
-	if (rem_sz > 0)	/* must have failed */
+	if (rem_sz > 0)	{ /* must have failed */
+		for (i = 0; i < k; i++)
+			__free_pages(schp->pages[i], order);
+
 		return -ENOMEM;
+	}
 	return 0;
 out:
 	for (i = 0; i < k; i++)
-- 
1.7.12.4

Re: [PATCH] scsi: sg: fix memory leak in sg_build_indirect

[Douglas Gilbert]

On 2020-04-13 7:32 a.m., Li Bin wrote:
> Fix a memory leak when there have failed, that we should free the pages
> under the condition rem_sz > 0.

May I paraphrase the above:
"Fix a memory leak that occurs when alloc_pages() succeeds several
  times before failing. This condition is noticed when rem_sz > 0."

> 
> Signed-off-by: Li Bin <huawei.libin@huawei.com>
> ---
>   drivers/scsi/sg.c | 6 +++++-
>   1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/scsi/sg.c b/drivers/scsi/sg.c
> index 4e6af592..8441ac5 100644
> --- a/drivers/scsi/sg.c
> +++ b/drivers/scsi/sg.c
> @@ -1959,8 +1959,12 @@ static long sg_compat_ioctl(struct file *filp, unsigned int cmd_in, unsigned lon

It is the sg_build_indirect() function not sg_compat_ioctl() as suggested
above by git. Can be get a replacement for git :-)

>   			 k, rem_sz));
>   
>   	schp->bufflen = blk_size;
> -	if (rem_sz > 0)	/* must have failed */
> +	if (rem_sz > 0)	{ /* must have failed */
> +		for (i = 0; i < k; i++)
> +			__free_pages(schp->pages[i], order);
> +
>   		return -ENOMEM;

It is easier, and less code, to replace 'return -ENOMEM'; with
'goto out'. Or even simpler:

     if (likely(rem_sz == 0))
	return 0;
out:
      ........

Doug Gilbert


BTW I spotted this one during the sg driver rewrite and fixed it.
Note that this bug and several others like it won't be fixed by
me while the sg driver rewrite is pending.

> +	}
>   	return 0;
>   out:
>   	for (i = 0; i < k; i++)
>