[Devel] [PATCH] properly charge and uncharge shmem

[Vladimir Davydov]

On Mon, May 18, 2015 at 10:57:35PM +0300, panda at cloudlinux.com wrote:
> From: Andrew Perepechko <panda at cloudlinux.com>
> 
> Currently, shmem_lock immediately and
> unconditionally uncharges what it has
> just charged for a lock request.
> 
> This, indeed, causes a double uncharge
> with something like the following:
> 
>   shmid = shmget(12345, 8192, IPC_CREAT | 0666);
>   rc = shmctl(shmid, SHM_LOCK, NULL);
>   shmctl(shmid, IPC_RMID, 0);
> 
> with the following in the kernel log:
> 
> [  455.815025] Uncharging too much 2 h 0, res lockedpages ub 0
> 
> Signed-off-by: Andrew Perepechko <panda at cloudlinux.com>
> ---
>  mm/shmem.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/shmem.c b/mm/shmem.c
> index a6b3e30..d09a230 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1359,11 +1359,13 @@ int shmem_lock(struct file *file, int lock, struct user_struct *user)
>  		mapping_set_unevictable(file->f_mapping);
>  	}
>  	if (!lock && (info->flags & VM_LOCKED) && user) {
> +		ub_lockedshm_uncharge(info, inode->i_size);
>  		user_shm_unlock(inode->i_size, user);
>  		info->flags &= ~VM_LOCKED;
>  		mapping_clear_unevictable(file->f_mapping);
>  	}
> -	retval = 0;
> +	spin_unlock(&info->lock);
> +	return 0;

Yeah, I lost this hunk when porting shmem accounting from RH6. Thank you
for catching it!

Acked-by: Vladimir Davydov <vdavydov at parallels.com>

>  
>  out_nomem:
>  	ub_lockedshm_uncharge(info, inode->i_size);