summaryrefslogtreecommitdiff
path: root/fs
diff options
context:
space:
mode:
authorJens Axboe <axboe@kernel.dk>2020-03-08 20:07:28 -0600
committerJens Axboe <axboe@kernel.dk>2020-03-08 20:07:28 -0600
commit805b13adde3964c78cba125a15527e88c19f87b3 (patch)
treec79c1839ac977bc84f7b8f6bed0038bf3f95b8e2 /fs
parentf0e20b8943509d81200cef5e30af2adfddba0f5c (diff)
io_uring: ensure RCU callback ordering with rcu_barrier()
After more careful studying, Paul informs me that we cannot rely on ordering of RCU callbacks in the way that the the tagged commit did. The current construct looks like this: void C(struct rcu_head *rhp) { do_something(rhp); call_rcu(&p->rh, B); } call_rcu(&p->rh, A); call_rcu(&p->rh, C); and we're relying on ordering between A and B, which isn't guaranteed. Make this explicit instead, and have a work item issue the rcu_barrier() to ensure that A has run before we manually execute B. While thorough testing never showed this issue, it's dependent on the per-cpu load in terms of RCU callbacks. The updated method simplifies the code as well, and eliminates the need to maintain an rcu_head in the fileset data. Fixes: c1e2148f8ecb ("io_uring: free fixed_file_data after RCU grace period") Reported-by: Paul E. McKenney <paulmck@kernel.org> Signed-off-by: Jens Axboe <axboe@kernel.dk>
Diffstat (limited to 'fs')
-rw-r--r--fs/io_uring.c29
1 files changed, 13 insertions, 16 deletions
diff --git a/fs/io_uring.c b/fs/io_uring.c
index c06082bb039a..1b2517291b78 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -191,7 +191,6 @@ struct fixed_file_data {
struct llist_head put_llist;
struct work_struct ref_work;
struct completion done;
- struct rcu_head rcu;
};
struct io_ring_ctx {
@@ -5331,24 +5330,21 @@ static void io_file_ref_kill(struct percpu_ref *ref)
complete(&data->done);
}
-static void __io_file_ref_exit_and_free(struct rcu_head *rcu)
+static void io_file_ref_exit_and_free(struct work_struct *work)
{
- struct fixed_file_data *data = container_of(rcu, struct fixed_file_data,
- rcu);
- percpu_ref_exit(&data->refs);
- kfree(data);
-}
+ struct fixed_file_data *data;
+
+ data = container_of(work, struct fixed_file_data, ref_work);
-static void io_file_ref_exit_and_free(struct rcu_head *rcu)
-{
/*
- * We need to order our exit+free call against the potentially
- * existing call_rcu() for switching to atomic. One way to do that
- * is to have this rcu callback queue the final put and free, as we
- * could otherwise have a pre-existing atomic switch complete _after_
- * the free callback we queued.
+ * Ensure any percpu-ref atomic switch callback has run, it could have
+ * been in progress when the files were being unregistered. Once
+ * that's done, we can safely exit and free the ref and containing
+ * data structure.
*/
- call_rcu(rcu, __io_file_ref_exit_and_free);
+ rcu_barrier();
+ percpu_ref_exit(&data->refs);
+ kfree(data);
}
static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
@@ -5369,7 +5365,8 @@ static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
for (i = 0; i < nr_tables; i++)
kfree(data->table[i].files);
kfree(data->table);
- call_rcu(&data->rcu, io_file_ref_exit_and_free);
+ INIT_WORK(&data->ref_work, io_file_ref_exit_and_free);
+ queue_work(system_wq, &data->ref_work);
ctx->file_data = NULL;
ctx->nr_user_files = 0;
return 0;