[PATCH v2] mm: fix a race on nr_swap_pages

[PATCH v2] mm: fix a race on nr_swap_pages

[Zhaoyang Huang]

The scenario on which "Free swap = -4kB" happens in my system, which is caused
by several get_swap_pages racing with each other and show_swap_cache_info
happens simutaniously. No need to add a lock on get_swap_page_of_type as we
remove "Presub/PosAdd" here.

ProcessA			ProcessB			ProcessC
ngoals = 1			ngoals = 1
avail = nr_swap_pages(1)	avail = nr_swap_pages(1)
nr_swap_pages(1) -= ngoals
				nr_swap_pages(0) -= ngoals
								nr_swap_pages = -1

Signed-off-by: Zhaoyang Huang <zhaoyang.huang@unisoc.com>
---
change of v2: fix bug of unpaired of spin_lock
---
---
 mm/swapfile.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/mm/swapfile.c b/mm/swapfile.c
index cf63b5f..1212f17 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -974,9 +974,13 @@ int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
 	/* Only single cluster request supported */
 	WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
 
+	spin_lock(&swap_avail_lock);
+
 	avail_pgs = atomic_long_read(&nr_swap_pages) / size;
-	if (avail_pgs <= 0)
+	if (avail_pgs <= 0) {
+		spin_unlock(&swap_avail_lock);
 		goto noswap;
+	}
 
 	if (n_goal > SWAP_BATCH)
 		n_goal = SWAP_BATCH;
@@ -986,8 +990,6 @@ int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
 
 	atomic_long_sub(n_goal * size, &nr_swap_pages);
 
-	spin_lock(&swap_avail_lock);
-
 start_over:
 	node = numa_node_id();
 	plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
@@ -1061,14 +1063,13 @@ swp_entry_t get_swap_page_of_type(int type)
 
 	spin_lock(&si->lock);
 	if (si->flags & SWP_WRITEOK) {
-		atomic_long_dec(&nr_swap_pages);
 		/* This is called for allocating swap entry, not cache */
 		offset = scan_swap_map(si, 1);
 		if (offset) {
+			atomic_long_dec(&nr_swap_pages);
 			spin_unlock(&si->lock);
 			return swp_entry(type, offset);
 		}
-		atomic_long_inc(&nr_swap_pages);
 	}
 	spin_unlock(&si->lock);
 fail:
-- 
1.9.1

Re: [PATCH v2] mm: fix a race on nr_swap_pages

[Vlastimil Babka]

On 12/4/20 3:52 AM, Zhaoyang Huang wrote:
> The scenario on which "Free swap = -4kB" happens in my system, which is caused
> by several get_swap_pages racing with each other and show_swap_cache_info
> happens simutaniously. No need to add a lock on get_swap_page_of_type as we
> remove "Presub/PosAdd" here.
> 
> ProcessA			ProcessB			ProcessC
> ngoals = 1			ngoals = 1
> avail = nr_swap_pages(1)	avail = nr_swap_pages(1)
> nr_swap_pages(1) -= ngoals
> 				nr_swap_pages(0) -= ngoals
> 								nr_swap_pages = -1
> 
> Signed-off-by: Zhaoyang Huang <zhaoyang.huang@unisoc.com>

Better now.
Acked-by: Vlastimil Babka <vbabka@suse.cz>

> ---
> change of v2: fix bug of unpaired of spin_lock
> ---
> ---
>  mm/swapfile.c | 11 ++++++-----
>  1 file changed, 6 insertions(+), 5 deletions(-)
> 
> diff --git a/mm/swapfile.c b/mm/swapfile.c
> index cf63b5f..1212f17 100644
> --- a/mm/swapfile.c
> +++ b/mm/swapfile.c
> @@ -974,9 +974,13 @@ int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
>  	/* Only single cluster request supported */
>  	WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER);
>  
> +	spin_lock(&swap_avail_lock);
> +
>  	avail_pgs = atomic_long_read(&nr_swap_pages) / size;
> -	if (avail_pgs <= 0)
> +	if (avail_pgs <= 0) {
> +		spin_unlock(&swap_avail_lock);
>  		goto noswap;
> +	}
>  
>  	if (n_goal > SWAP_BATCH)
>  		n_goal = SWAP_BATCH;
> @@ -986,8 +990,6 @@ int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size)
>  
>  	atomic_long_sub(n_goal * size, &nr_swap_pages);
>  
> -	spin_lock(&swap_avail_lock);
> -
>  start_over:
>  	node = numa_node_id();
>  	plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) {
> @@ -1061,14 +1063,13 @@ swp_entry_t get_swap_page_of_type(int type)
>  
>  	spin_lock(&si->lock);
>  	if (si->flags & SWP_WRITEOK) {
> -		atomic_long_dec(&nr_swap_pages);
>  		/* This is called for allocating swap entry, not cache */
>  		offset = scan_swap_map(si, 1);
>  		if (offset) {
> +			atomic_long_dec(&nr_swap_pages);
>  			spin_unlock(&si->lock);
>  			return swp_entry(type, offset);
>  		}
> -		atomic_long_inc(&nr_swap_pages);
>  	}
>  	spin_unlock(&si->lock);
>  fail:
>