Re: [PATCH] staging: lustre: replace kzalloc with copy_from_user with memdup_user

["Drokin, Oleg" <oleg.drokin@intel.com>]

On Mar 31, 2015, at 11:57 AM, gregkh@linuxfoundation.org wrote:

> On Tue, Mar 31, 2015 at 05:15:23PM +0200, Julia Lawall wrote:
>> On Tue, 31 Mar 2015, Dhere, Chaitanya (C.) wrote:
>> 
>>> This patch replaces kzalloc and copy_from_user with memdup_user call
>>> This change was detected with coccinelle tool
>>> 
>>> Signed-off-by: Chaitanya Dhere <cvijaydh@visteon.com>
>>> ---
>>> drivers/staging/lustre/lustre/llite/file.c |   11 +++--------
>>> 1 file changed, 3 insertions(+), 8 deletions(-)
>>> 
>>> diff --git a/drivers/staging/lustre/lustre/llite/file.c b/drivers/staging/lustre/lustre/llite/file.c
>>> index 85e74d1..85b5567 100644
>>> --- a/drivers/staging/lustre/lustre/llite/file.c
>>> +++ b/drivers/staging/lustre/lustre/llite/file.c
>>> @@ -2368,14 +2368,9 @@ ll_file_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
>>> 		struct hsm_state_set	*hss;
>>> 		int			 rc;
>>> 
>>> -		hss = kzalloc(sizeof(*hss), GFP_NOFS);
>>> -		if (!hss)
>>> -			return -ENOMEM;
>>> -
>>> -		if (copy_from_user(hss, (char *)arg, sizeof(*hss))) {
>>> -			OBD_FREE_PTR(hss);
>>> -			return -EFAULT;
>>> -		}
>>> +		hss = memdup_user((char *)arg, sizeof(*hss));
>> 
>> memdup_user will use the flag GFP_KERNEL, ie (__GFP_WAIT | __GFP_IO |
>> __GFP_FS), rather than the flag GFP_NOFS, ie (__GFP_WAIT | __GFP_IO), that
>> is specified.  I don't know if this is a problem here.
> 
> Yes, this is a filesystem, so this can't be changed, as we can't have
> the allocation go out and ask for more filesystem accesses in the middle
> of trying to do a filesystem access :)

Technically in this place we are not really holding any locks or anything else of value to cause a deadlock,
so we might be fine here.
More importantly, I totally missed this OBD_ALLOC replacement with kzalloc when it happened.
In theory all OBD_ALLOC() calls add up all allocated memory in a counter and then OBD_FREE() calls
subtract freed memory (for a poor man's memory leak detection and tracing).
Now since it's out of match, there should have been tons of very loud warnings about it, but I don't see
any in my logs and I wonder why.

Julia, I wonder if you happen to have a bunch of other patches to get rid of the rest of OBD_ALLOC and OBD_FREE stuff by any chance?

Bye,
    Oleg