io_uring: ensure ctx->rings is stable for task work flags manipulation

Commit 9618908 upstream.

If DEFER_TASKRUN | SETUP_TASKRUN is used and task work is added while
the ring is being resized, it's possible for the OR'ing of
IORING_SQ_TASKRUN to happen in the small window of swapping into the
new rings and the old rings being freed.

Prevent this by adding a 2nd ->rings pointer, ->rings_rcu, which is
protected by RCU. The task work flags manipulation is inside RCU
already, and if the resize ring freeing is done post an RCU synchronize,
then there's no need to add locking to the fast path of task work
additions.

Note: this is only done for DEFER_TASKRUN, as that's the only setup mode
that supports ring resizing. If this ever changes, then they too need to
use the io_ctx_mark_taskrun() helper.

Link: https://lore.kernel.org/io-uring/20260309062759.482210-1-naup96721@gmail.com/
Cc: stable@vger.kernel.org
Fixes: 79cfe9e ("io_uring/register: add IORING_REGISTER_RESIZE_RINGS")
Reported-by: Hao-Yu Yang <naup96721@gmail.com>
Suggested-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Author: axboe

include/linux/io_uring_types.h

@@ -371,6 +371,7 @@ struct io_ring_ctx {
 	 * regularly bounce b/w CPUs.
 	 */
 	struct {
+		struct io_rings	__rcu	*rings_rcu;
 		struct llist_head	work_llist;
 		struct llist_head	retry_llist;
 		unsigned long		check_cq;

io_uring/io_uring.c

@@ -1202,6 +1202,21 @@ void tctx_task_work(struct callback_head *cb)
 	WARN_ON_ONCE(ret);
 }
 
+/*
+ * Sets IORING_SQ_TASKRUN in the sq_flags shared with userspace, using the
+ * RCU protected rings pointer to be safe against concurrent ring resizing.
+ */
+static void io_ctx_mark_taskrun(struct io_ring_ctx *ctx)
+{
+	lockdep_assert_in_rcu_read_lock();
+
+	if (ctx->flags & IORING_SETUP_TASKRUN_FLAG) {
+		struct io_rings *rings = rcu_dereference(ctx->rings_rcu);
+
+		atomic_or(IORING_SQ_TASKRUN, &rings->sq_flags);
+	}
+}
+
 static void io_req_local_work_add(struct io_kiocb *req, unsigned flags)
 {
 	struct io_ring_ctx *ctx = req->ctx;
@@ -1256,8 +1271,7 @@ static void io_req_local_work_add(struct io_kiocb *req, unsigned flags)
 	 */
 
 	if (!head) {
-		if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
-			atomic_or(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
+		io_ctx_mark_taskrun(ctx);
 		if (ctx->has_evfd)
 			io_eventfd_signal(ctx, false);
 	}
@@ -1281,6 +1295,10 @@ static void io_req_normal_work_add(struct io_kiocb *req)
 	if (!llist_add(&req->io_task_work.node, &tctx->task_list))
 		return;
 
+	/*
+	 * Doesn't need to use ->rings_rcu, as resizing isn't supported for
+	 * !DEFER_TASKRUN.
+	 */
 	if (ctx->flags & IORING_SETUP_TASKRUN_FLAG)
 		atomic_or(IORING_SQ_TASKRUN, &ctx->rings->sq_flags);
 
@@ -2760,6 +2778,7 @@ static void io_rings_free(struct io_ring_ctx *ctx)
 	io_free_region(ctx->user, &ctx->sq_region);
 	io_free_region(ctx->user, &ctx->ring_region);
 	ctx->rings = NULL;
+	RCU_INIT_POINTER(ctx->rings_rcu, NULL);
 	ctx->sq_sqes = NULL;
 }
 
@@ -3389,6 +3408,7 @@ static __cold int io_allocate_scq_urings(struct io_ring_ctx *ctx,
 	if (ret)
 		return ret;
 	ctx->rings = rings = io_region_get_ptr(&ctx->ring_region);
+	rcu_assign_pointer(ctx->rings_rcu, rings);
 	if (!(ctx->flags & IORING_SETUP_NO_SQARRAY))
 		ctx->sq_array = (u32 *)((char *)rings + rl->sq_array_offset);
 

io_uring/register.c

@@ -545,7 +545,15 @@ static int io_register_resize_rings(struct io_ring_ctx *ctx, void __user *arg)
 	ctx->sq_entries = p->sq_entries;
 	ctx->cq_entries = p->cq_entries;
 
+	/*
+	 * Just mark any flag we may have missed and that the application
+	 * should act on unconditionally. Worst case it'll be an extra
+	 * syscall.
+	 */
+	atomic_or(IORING_SQ_TASKRUN | IORING_SQ_NEED_WAKEUP, &n.rings->sq_flags);
 	ctx->rings = n.rings;
+	rcu_assign_pointer(ctx->rings_rcu, n.rings);
+
 	ctx->sq_sqes = n.sq_sqes;
 	swap_old(ctx, o, n, ring_region);
 	swap_old(ctx, o, n, sq_region);
@@ -554,6 +562,10 @@ static int io_register_resize_rings(struct io_ring_ctx *ctx, void __user *arg)
 out:
 	spin_unlock(&ctx->completion_lock);
 	mutex_unlock(&ctx->mmap_lock);
+
+	/* Wait for concurrent io_ctx_mark_taskrun() */
+	if (to_free == &o)
+		synchronize_rcu_expedited();
 	io_register_free_rings(ctx, to_free);
 
 	if (ctx->sq_data)