[PATCH v2] io_uring/sqpoll: protect task-work publication with RCU

[PATCH v2] io_uring/sqpoll: protect task-work publication with RCU

[Jérémy Jean]

SQPOLL can consume a request immediately after mpscq_push(). If this
drops the final ring reference, ctx and tctx may be freed before
io_req_normal_work_add() finishes using them. KASAN reports:

  BUG: KASAN: slab-use-after-free in io_req_normal_work_add+0x439/0x510
  Read of size 4 at addr ff11000000ca0000 by task repro/55
  (...)
  BUG: KASAN: slab-use-after-free in queue_work_on+0x25/0x70
  Write of size 8 at addr ff11000000c3bd00 by task repro/55

Add guard(rcu)() to io_req_normal_work_add() and wait for RCU readers
when freeing an SQPOLL ring, as done for DEFER_TASKRUN. Save
tctx->task before queueing the request so notification can use the
saved pointer under RCU even if SQPOLL cancellation has freed tctx.

Fixes: af5d68f8892f ("io_uring/sqpoll: manage task_work privately")
Assisted-by: LLM
Signed-off-by: Jérémy Jean <Jeremy.Jean@oss.cyber.gouv.fr>
---
Changes in v2:
- Add guard(rcu)() to io_req_normal_work_add() (Jens Axboe).
- Extend io_ring_exit_work()'s synchronize_rcu() to SQPOLL (Jens Axboe).
- Cache tctx->task before publication for SQPOLL cancellation.

v1: https://lore.kernel.org/all/20260924205056.3759980-2-Jeremy.Jean@oss.cyber.gouv.fr/

 io_uring/io_uring.c | 4 ++--
 io_uring/tw.c       | 7 ++++++-
 2 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 61053421d809..a1e961549468 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -2403,8 +2403,8 @@ static __cold void io_ring_exit_work(struct work_struct *work)
 	spin_lock(&ctx->completion_lock);
 	spin_unlock(&ctx->completion_lock);
 
-	/* pairs with RCU read section in io_req_local_work_add() */
-	if (ctx->flags & IORING_SETUP_DEFER_TASKRUN)
+	/* pairs with RCU read sections in io_req_{local,normal}_work_add() */
+	if (ctx->flags & (IORING_SETUP_DEFER_TASKRUN | IORING_SETUP_SQPOLL))
 		synchronize_rcu();
 
 	io_ring_ctx_free(ctx);
diff --git a/io_uring/tw.c b/io_uring/tw.c
index f573bcc3af6a..0985b6ab6dfe 100644
--- a/io_uring/tw.c
+++ b/io_uring/tw.c
@@ -209,6 +209,10 @@ void io_req_normal_work_add(struct io_kiocb *req)
 {
 	struct io_uring_task *tctx = req->tctx;
 	struct io_ring_ctx *ctx = req->ctx;
+	struct task_struct *task = tctx->task;
+
+	/* pairs with synchronize_rcu() in io_ring_exit_work() */
+	guard(rcu)();
 
 	/* tw run already pending, nothing else to do */
 	if (!mpscq_push(&tctx->task_list, &req->io_task_work.node))
@@ -223,7 +227,8 @@ void io_req_normal_work_add(struct io_kiocb *req)
 
 	/* SQPOLL doesn't need the task_work added, it'll run it itself */
 	if (ctx->flags & IORING_SETUP_SQPOLL) {
-		__set_notify_signal(tctx->task);
+		/* SQPOLL cancellation may have freed tctx after the push. */
+		__set_notify_signal(task);
 		return;
 	}
 
-- 
2.47.3