Re: [PATCH] mm/mempolicy: do not duplicate policy if it is not applicable for set_mempolicy_home_node

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

On 2022-12-15 09:49, Michal Hocko wrote:
> On Thu 15-12-22 09:33:54, Mathieu Desnoyers wrote:
>> On 2022-12-15 02:51, Michal Hocko wrote:
> [...]
>>> Btw. looking at the code again it seems rather pointless to duplicate
>>> the policy just to throw it away anyway. A slightly bigger diff but this
>>> looks more reasonable to me. What do you think? I can also send it as a
>>> clean up on top of your fix.
>>
>> I think it would be best if this comes as a cleanup on top of my fix. The
>> diff is larger than the minimal change needed to fix the leak in stable
>> branches.
>>
>> Your approach looks fine, except for the vma_policy(vma) -> old change
>> already spotted by Aneesh.
> 
> This shouldn't have any real effect on the functionality. Anyway, here
> is a follow up cleanup:
> ---
>  From f3fdb6f65fa3977aab13378b8e299b168719577c Mon Sep 17 00:00:00 2001
> From: Michal Hocko <mhocko@suse.com>
> Date: Thu, 15 Dec 2022 15:41:27 +0100
> Subject: [PATCH] mm/mempolicy: do not duplicate policy if it is not applicable
>   for set_mempolicy_home_node
> 
> set_mempolicy_home_node tries to duplicate a memory policy before
> checking it whether it is applicable for the operation. There is
> no real reason for doing that and it might actually be a pointless
> memory allocation and deallocation exercise for MPOL_INTERLEAVE.
> 
> Not a big problem but we can do better. Simply check the policy before
> acting on it.
> 
> Signed-off-by: Michal Hocko <mhocko@suse.com>

Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

> ---
>   mm/mempolicy.c | 28 ++++++++++++----------------
>   1 file changed, 12 insertions(+), 16 deletions(-)
> 
> diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> index 02c8a712282f..becf41e10076 100644
> --- a/mm/mempolicy.c
> +++ b/mm/mempolicy.c
> @@ -1489,7 +1489,7 @@ SYSCALL_DEFINE4(set_mempolicy_home_node, unsigned long, start, unsigned long, le
>   {
>   	struct mm_struct *mm = current->mm;
>   	struct vm_area_struct *vma;
> -	struct mempolicy *new;
> +	struct mempolicy *new, *old;
>   	unsigned long vmstart;
>   	unsigned long vmend;
>   	unsigned long end;
> @@ -1521,31 +1521,27 @@ SYSCALL_DEFINE4(set_mempolicy_home_node, unsigned long, start, unsigned long, le
>   		return 0;
>   	mmap_write_lock(mm);
>   	for_each_vma_range(vmi, vma, end) {
> -		vmstart = max(start, vma->vm_start);
> -		vmend   = min(end, vma->vm_end);
> -		new = mpol_dup(vma_policy(vma));
> -		if (IS_ERR(new)) {
> -			err = PTR_ERR(new);
> -			break;
> -		}
> -		/*
> -		 * Only update home node if there is an existing vma policy
> -		 */
> -		if (!new)
> -			continue;
> -
>   		/*
>   		 * If any vma in the range got policy other than MPOL_BIND
>   		 * or MPOL_PREFERRED_MANY we return error. We don't reset
>   		 * the home node for vmas we already updated before.
>   		 */
> -		if (new->mode != MPOL_BIND && new->mode != MPOL_PREFERRED_MANY) {
> -			mpol_put(new);
> +		old = vma_policy(vma);
> +		if (!old)
> +			continue;
> +		if (old->mode != MPOL_BIND && old->mode != MPOL_PREFERRED_MANY) {
>   			err = -EOPNOTSUPP;
>   			break;
>   		}
> +		new = mpol_dup(old);
> +		if (IS_ERR(new)) {
> +			err = PTR_ERR(new);
> +			break;
> +		}
>   
>   		new->home_node = home_node;
> +		vmstart = max(start, vma->vm_start);
> +		vmend   = min(end, vma->vm_end);
>   		err = mbind_range(mm, vmstart, vmend, new);
>   		mpol_put(new);
>   		if (err)

-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com