Merge tag 'io_uring-6.1-2022-11-11' of git://git.kernel.dk/linux

Pull io_uring fixes from Jens Axboe:
 "Nothing major, just a few minor tweaks:

   - Tweak for the TCP zero-copy io_uring self test (Pavel)

   - Rather than use our internal cached value of number of CQ events
     available, use what the user can see (Dylan)

   - Fix a typo in a comment, added in this release (me)

   - Don't allow wrapping while adding provided buffers (me)

   - Fix a double poll race, and add a lockdep assertion for it too
     (Pavel)"

* tag 'io_uring-6.1-2022-11-11' of git://git.kernel.dk/linux:
  io_uring/poll: lockdep annote io_poll_req_insert_locked
  io_uring/poll: fix double poll req->flags races
  io_uring: check for rollover of buffer ID when providing buffers
  io_uring: calculate CQEs from the user visible value
  io_uring: fix typo in io_uring.h comment
  selftests/net: don't tests batched TCP io_uring zc

include/uapi/linux/io_uring.h

@@ -222,7 +222,7 @@ enum io_uring_op {
 
 /*
  * sqe->uring_cmd_flags
- * IORING_URING_CMD_FIXED	use registered buffer; pass thig flag
+ * IORING_URING_CMD_FIXED	use registered buffer; pass this flag
  *				along with setting sqe->buf_index.
  */
 #define IORING_URING_CMD_FIXED	(1U << 0)

io_uring/io_uring.c

@@ -176,6 +176,11 @@ static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx)
 	return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head);
 }
 
+static inline unsigned int __io_cqring_events_user(struct io_ring_ctx *ctx)
+{
+	return READ_ONCE(ctx->rings->cq.tail) - READ_ONCE(ctx->rings->cq.head);
+}
+
 static bool io_match_linked(struct io_kiocb *head)
 {
 	struct io_kiocb *req;
@@ -2315,7 +2320,7 @@ static inline bool io_has_work(struct io_ring_ctx *ctx)
 static inline bool io_should_wake(struct io_wait_queue *iowq)
 {
 	struct io_ring_ctx *ctx = iowq->ctx;
-	int dist = ctx->cached_cq_tail - (int) iowq->cq_tail;
+	int dist = READ_ONCE(ctx->rings->cq.tail) - (int) iowq->cq_tail;
 
 	/*
 	 * Wake up if we have enough events, or if a timeout occurred since we
@@ -2399,7 +2404,8 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 			return ret;
 		io_cqring_overflow_flush(ctx);
 
-		if (io_cqring_events(ctx) >= min_events)
+		/* if user messes with these they will just get an early return */
+		if (__io_cqring_events_user(ctx) >= min_events)
 			return 0;
 	} while (ret > 0);
 

io_uring/kbuf.c

@@ -346,6 +346,8 @@ int io_provide_buffers_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe
 	tmp = READ_ONCE(sqe->off);
 	if (tmp > USHRT_MAX)
 		return -E2BIG;
+	if (tmp + p->nbufs >= USHRT_MAX)
+		return -EINVAL;
 	p->bid = tmp;
 	return 0;
 }

io_uring/poll.c

@@ -116,6 +116,8 @@ static void io_poll_req_insert_locked(struct io_kiocb *req)
 	struct io_hash_table *table = &req->ctx->cancel_table_locked;
 	u32 index = hash_long(req->cqe.user_data, table->hash_bits);
 
+	lockdep_assert_held(&req->ctx->uring_lock);
+
 	hlist_add_head(&req->hash_node, &table->hbs[index].list);
 }
 
@@ -394,7 +396,8 @@ static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 	return 1;
 }
 
-static void io_poll_double_prepare(struct io_kiocb *req)
+/* fails only when polling is already completing by the first entry */
+static bool io_poll_double_prepare(struct io_kiocb *req)
 {
 	struct wait_queue_head *head;
 	struct io_poll *poll = io_poll_get_single(req);
@@ -403,20 +406,20 @@ static void io_poll_double_prepare(struct io_kiocb *req)
 	rcu_read_lock();
 	head = smp_load_acquire(&poll->head);
 	/*
-	 * poll arm may not hold ownership and so race with
-	 * io_poll_wake() by modifying req->flags. There is only one
-	 * poll entry queued, serialise with it by taking its head lock.
+	 * poll arm might not hold ownership and so race for req->flags with
+	 * io_poll_wake(). There is only one poll entry queued, serialise with
+	 * it by taking its head lock. As we're still arming the tw hanlder
+	 * is not going to be run, so there are no races with it.
 	 */
-	if (head)
+	if (head) {
 		spin_lock_irq(&head->lock);
-
 		req->flags |= REQ_F_DOUBLE_POLL;
 		if (req->opcode == IORING_OP_POLL_ADD)
 			req->flags |= REQ_F_ASYNC_DATA;
-
-	if (head)
 		spin_unlock_irq(&head->lock);
+	}
 	rcu_read_unlock();
+	return !!head;
 }
 
 static void __io_queue_proc(struct io_poll *poll, struct io_poll_table *pt,
@@ -454,7 +457,11 @@ static void __io_queue_proc(struct io_poll *poll, struct io_poll_table *pt,
 		/* mark as double wq entry */
 		wqe_private |= IO_WQE_F_DOUBLE;
 		io_init_poll_iocb(poll, first->events, first->wait.func);
-		io_poll_double_prepare(req);
+		if (!io_poll_double_prepare(req)) {
+			/* the request is completing, just back off */
+			kfree(poll);
+			return;
+		}
 		*poll_ptr = poll;
 	} else {
 		/* fine to modify, there is no poll queued to race with us */

tools/testing/selftests/net/io_uring_zerocopy_tx.sh

@@ -29,7 +29,7 @@ if [[ "$#" -eq "0" ]]; then
 	for IP in "${IPs[@]}"; do
 		for mode in $(seq 1 3); do
 			$0 "$IP" udp -m "$mode" -t 1 -n 32
-			$0 "$IP" tcp -m "$mode" -t 1 -n 32
+			$0 "$IP" tcp -m "$mode" -t 1 -n 1
 		done
 	done