[PATCHSET 0/2] Fix DEFER_TASKRUN ring resize flag manipulation

[PATCHSET 0/2] Fix DEFER_TASKRUN ring resize flag manipulation

[Jens Axboe]

Hi,

Two patches here:

1) Fix adding local task_work during ring resize. There's a tiny gap
   where a NULL pointer would be used.

2) Same issue exists in the eventfd handling, so apply the same kind of
   fix there.

Thanks to Hao-Yu Yang for the report and initial fix attempt, and Pavel
for a good suggestion on how best to handle this.

-- 
Jens Axboe

[PATCH 1/2] io_uring: ensure ctx->rings is stable for task work flags manipulation

[Jens Axboe]

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: 79cfe9e59c2a ("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>
---
 include/linux/io_uring_types.h |  1 +
 io_uring/io_uring.c            |  2 ++
 io_uring/register.c            | 20 ++++++++++++++++++--
 io_uring/tw.c                  | 24 ++++++++++++++++++++++--
 4 files changed, 43 insertions(+), 4 deletions(-)

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 3e4a82a6f817..dd1420bfcb73 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -388,6 +388,7 @@ 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;
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index ccab8562d273..20fdc442e014 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -2066,6 +2066,7 @@ 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;
 }
 
@@ -2703,6 +2704,7 @@ 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);
 
diff --git a/io_uring/register.c b/io_uring/register.c
index a839b22fd392..5f2985ba0879 100644
--- a/io_uring/register.c
+++ b/io_uring/register.c
@@ -487,6 +487,18 @@ static void io_register_free_rings(struct io_ring_ctx *ctx,
 			 IORING_SETUP_CQE32 | IORING_SETUP_NO_MMAP | \
 			 IORING_SETUP_CQE_MIXED | IORING_SETUP_SQE_MIXED)
 
+static void io_resize_assign_rings(struct io_ring_ctx *ctx, struct io_rings *rings)
+{
+	/*
+	 * 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, &rings->sq_flags);
+	ctx->rings = rings;
+	rcu_assign_pointer(ctx->rings_rcu, rings);
+}
+
 static int io_register_resize_rings(struct io_ring_ctx *ctx, void __user *arg)
 {
 	struct io_ctx_config config;
@@ -579,6 +591,7 @@ static int io_register_resize_rings(struct io_ring_ctx *ctx, void __user *arg)
 	spin_lock(&ctx->completion_lock);
 	o.rings = ctx->rings;
 	ctx->rings = NULL;
+	RCU_INIT_POINTER(ctx->rings_rcu, NULL);
 	o.sq_sqes = ctx->sq_sqes;
 	ctx->sq_sqes = NULL;
 
@@ -604,7 +617,7 @@ static int io_register_resize_rings(struct io_ring_ctx *ctx, void __user *arg)
 	if (tail - old_head > p->cq_entries) {
 overflow:
 		/* restore old rings, and return -EOVERFLOW via cleanup path */
-		ctx->rings = o.rings;
+		io_resize_assign_rings(ctx, o.rings);
 		ctx->sq_sqes = o.sq_sqes;
 		to_free = &n;
 		ret = -EOVERFLOW;
@@ -633,7 +646,7 @@ 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;
 
-	ctx->rings = n.rings;
+	io_resize_assign_rings(ctx, n.rings);
 	ctx->sq_sqes = n.sq_sqes;
 	swap_old(ctx, o, n, ring_region);
 	swap_old(ctx, o, n, sq_region);
@@ -642,6 +655,9 @@ 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();
 	io_register_free_rings(ctx, to_free);
 
 	if (ctx->sq_data)
diff --git a/io_uring/tw.c b/io_uring/tw.c
index 1ee2b8ab07c8..c104e1e30d7c 100644
--- a/io_uring/tw.c
+++ b/io_uring/tw.c
@@ -152,6 +152,23 @@ 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.
+ * Must be called inside an RCU read-side critical section.
+ */
+static void io_ctx_mark_taskrun(struct io_ring_ctx *ctx)
+{
+	struct io_rings *rings;
+
+	if (!(ctx->flags & IORING_SETUP_TASKRUN_FLAG))
+		return;
+
+	rings = rcu_dereference(ctx->rings_rcu);
+	if (rings)
+		atomic_or(IORING_SQ_TASKRUN, &rings->sq_flags);
+}
+
 void io_req_local_work_add(struct io_kiocb *req, unsigned flags)
 {
 	struct io_ring_ctx *ctx = req->ctx;
@@ -206,8 +223,7 @@ 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);
 	}
@@ -231,6 +247,10 @@ 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);
 
-- 
2.53.0

[PATCH 2/2] io_uring/eventfd: use ctx->rings_rcu for flags checking

[Jens Axboe]

Similarly to what commit e78f7b70e837 did for local task work additions,
use ->rings_rcu under RCU rather than dereference ->rings directly. See
that commit for more details.

Cc: stable@vger.kernel.org
Fixes: 79cfe9e59c2a ("io_uring/register: add IORING_REGISTER_RESIZE_RINGS")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/eventfd.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/io_uring/eventfd.c b/io_uring/eventfd.c
index 78f8ab7db104..ab789e1ebe91 100644
--- a/io_uring/eventfd.c
+++ b/io_uring/eventfd.c
@@ -76,11 +76,15 @@ void io_eventfd_signal(struct io_ring_ctx *ctx, bool cqe_event)
 {
 	bool skip = false;
 	struct io_ev_fd *ev_fd;
-
-	if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
-		return;
+	struct io_rings *rings;
 
 	guard(rcu)();
+
+	rings = rcu_dereference(ctx->rings_rcu);
+	if (!rings)
+		return;
+	if (READ_ONCE(rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
+		return;
 	ev_fd = rcu_dereference(ctx->io_ev_fd);
 	/*
 	 * Check again if ev_fd exists in case an io_eventfd_unregister call
-- 
2.53.0