mm: fix a race on nr_swap_pages
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
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Zhaoyang Huang <[email protected]>
Acked-by: Vlastimil Babka <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
diff --git a/mm/swapfile.c b/mm/swapfile.c
index 1c0a829..9fffc5a 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -1042,16 +1042,18 @@ 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;
+ }
n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs);
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]) {
@@ -1125,14 +1127,13 @@ 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:
