On 02/18, Eric Dumazet wrote:
From: Eric Dumazet <edumazet@xxxxxxxxxx>
As stated in the comment found in maybe_wait_bpf_programs(),
the synchronize_rcu() barrier is only needed before returning
to userspace, not after each deletion in the batch.
Signed-off-by: Eric Dumazet <edumazet@xxxxxxxxxx>
Cc: Stanislav Fomichev <sdf@xxxxxxxxxx>
Makes sense. Probably a copy-paste from the non-batch case...
Reviewed-by: Stanislav Fomichev <sdf@xxxxxxxxxx>
Cc: Brian Vazquez <brianvv@xxxxxxxxxx>
---
kernel/bpf/syscall.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
index
a72f63d5a7daee057bcec3fa6119aca32e2945f7..9c7a72b65eee0ec8d54d36e2c0ab9ff4962091af
100644
--- a/kernel/bpf/syscall.c
+++ b/kernel/bpf/syscall.c
@@ -1352,7 +1352,6 @@ int generic_map_delete_batch(struct bpf_map *map,
err = map->ops->map_delete_elem(map, key);
rcu_read_unlock();
bpf_enable_instrumentation();
- maybe_wait_bpf_programs(map);
if (err)
break;
cond_resched();
@@ -1361,6 +1360,8 @@ int generic_map_delete_batch(struct bpf_map *map,
err = -EFAULT;
kvfree(key);
+
+ maybe_wait_bpf_programs(map);
return err;
}
--
2.35.1.473.g83b2b277ed-goog
