Merge tag 'io_uring-6.2-2022-12-19' of git://git.kernel.dk/linux

Pull io_uring fixes from Jens Axboe:

 - Improve the locking for timeouts. This was originally queued up for
   the initial pull, but I messed up and it got missed. (Pavel)

 - Fix an issue with running task_work from the wait path, causing some
   inefficiencies (me)

 - Add a clear of ->free_iov upfront in the 32-bit compat data
   importing, so we ensure that it's always sane at completion time (me)

 - Use call_rcu_hurry() for the eventfd signaling (Dylan)

 - Ordering fix for multishot recv completions (Pavel)

 - Add the io_uring trace header to the MAINTAINERS entry (Ammar)

* tag 'io_uring-6.2-2022-12-19' of git://git.kernel.dk/linux:
  MAINTAINERS: io_uring: Add include/trace/events/io_uring.h
  io_uring/net: fix cleanup after recycle
  io_uring/net: ensure compat import handlers clear free_iov
  io_uring: include task_work run after scheduling in wait for events
  io_uring: don't use TIF_NOTIFY_SIGNAL to test for availability of task_work
  io_uring: use call_rcu_hurry if signaling an eventfd
  io_uring: fix overflow handling regression
  io_uring: ease timeout flush locking requirements
  io_uring: revise completion_lock locking
  io_uring: protect cq_timeouts with timeout_lock

MAINTAINERS

@@ -10878,6 +10878,7 @@ T:	git git://git.kernel.dk/liburing
 F:	io_uring/
 F:	include/linux/io_uring.h
 F:	include/linux/io_uring_types.h
+F:	include/trace/events/io_uring.h
 F:	include/uapi/linux/io_uring.h
 F:	tools/io_uring/
 

io_uring/io_uring.c

@@ -538,7 +538,7 @@ static void io_eventfd_signal(struct io_ring_ctx *ctx)
 	} else {
 		atomic_inc(&ev_fd->refs);
 		if (!atomic_fetch_or(BIT(IO_EVENTFD_OP_SIGNAL_BIT), &ev_fd->ops))
-			call_rcu(&ev_fd->rcu, io_eventfd_ops);
+			call_rcu_hurry(&ev_fd->rcu, io_eventfd_ops);
 		else
 			atomic_dec(&ev_fd->refs);
 	}
@@ -572,12 +572,11 @@ static void io_eventfd_flush_signal(struct io_ring_ctx *ctx)
 
 void __io_commit_cqring_flush(struct io_ring_ctx *ctx)
 {
-	if (ctx->off_timeout_used || ctx->drain_active) {
+	if (ctx->off_timeout_used)
+		io_flush_timeouts(ctx);
+	if (ctx->drain_active) {
 		spin_lock(&ctx->completion_lock);
-		if (ctx->off_timeout_used)
-			io_flush_timeouts(ctx);
-		if (ctx->drain_active)
-			io_queue_deferred(ctx);
+		io_queue_deferred(ctx);
 		spin_unlock(&ctx->completion_lock);
 	}
 	if (ctx->has_evfd)
@@ -597,6 +596,18 @@ static inline void __io_cq_unlock(struct io_ring_ctx *ctx)
 		spin_unlock(&ctx->completion_lock);
 }
 
+static inline void io_cq_lock(struct io_ring_ctx *ctx)
+	__acquires(ctx->completion_lock)
+{
+	spin_lock(&ctx->completion_lock);
+}
+
+static inline void io_cq_unlock(struct io_ring_ctx *ctx)
+	__releases(ctx->completion_lock)
+{
+	spin_unlock(&ctx->completion_lock);
+}
+
 /* keep it inlined for io_submit_flush_completions() */
 static inline void __io_cq_unlock_post(struct io_ring_ctx *ctx)
 	__releases(ctx->completion_lock)
@@ -916,7 +927,7 @@ static void __io_req_complete_post(struct io_kiocb *req)
 
 	io_cq_lock(ctx);
 	if (!(req->flags & REQ_F_CQE_SKIP))
-		__io_fill_cqe_req(ctx, req);
+		io_fill_cqe_req(ctx, req);
 
 	/*
 	 * If we're the last reference to this request, add to our locked
@@ -1074,9 +1085,9 @@ static void __io_req_find_next_prep(struct io_kiocb *req)
 {
 	struct io_ring_ctx *ctx = req->ctx;
 
-	io_cq_lock(ctx);
+	spin_lock(&ctx->completion_lock);
 	io_disarm_next(req);
-	io_cq_unlock_post(ctx);
+	spin_unlock(&ctx->completion_lock);
 }
 
 static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req)
@@ -2470,7 +2481,14 @@ static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
 	}
 	if (!schedule_hrtimeout(&timeout, HRTIMER_MODE_ABS))
 		return -ETIME;
-	return 1;
+
+	/*
+	 * Run task_work after scheduling. If we got woken because of
+	 * task_work being processed, run it now rather than let the caller
+	 * do another wait loop.
+	 */
+	ret = io_run_task_work_sig(ctx);
+	return ret < 0 ? ret : 1;
 }
 
 /*
@@ -2535,6 +2553,8 @@ static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
 		prepare_to_wait_exclusive(&ctx->cq_wait, &iowq.wq,
 						TASK_INTERRUPTIBLE);
 		ret = io_cqring_wait_schedule(ctx, &iowq, timeout);
+		if (__io_cqring_events_user(ctx) >= min_events)
+			break;
 		cond_resched();
 	} while (ret > 0);
 

io_uring/io_uring.h

@@ -87,17 +87,6 @@ static inline void io_req_task_work_add(struct io_kiocb *req)
 #define io_for_each_link(pos, head) \
 	for (pos = (head); pos; pos = pos->link)
 
-static inline void io_cq_lock(struct io_ring_ctx *ctx)
-	__acquires(ctx->completion_lock)
-{
-	spin_lock(&ctx->completion_lock);
-}
-
-static inline void io_cq_unlock(struct io_ring_ctx *ctx)
-{
-	spin_unlock(&ctx->completion_lock);
-}
-
 void io_cq_unlock_post(struct io_ring_ctx *ctx);
 
 static inline struct io_uring_cqe *io_get_cqe_overflow(struct io_ring_ctx *ctx,
@@ -277,8 +266,7 @@ static inline int io_run_task_work(void)
 
 static inline bool io_task_work_pending(struct io_ring_ctx *ctx)
 {
-	return test_thread_flag(TIF_NOTIFY_SIGNAL) ||
-		!wq_list_empty(&ctx->work_llist);
+	return task_work_pending(current) || !wq_list_empty(&ctx->work_llist);
 }
 
 static inline int io_run_task_work_ctx(struct io_ring_ctx *ctx)

io_uring/net.c

@@ -494,6 +494,7 @@ static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
 	if (req->flags & REQ_F_BUFFER_SELECT) {
 		compat_ssize_t clen;
 
+		iomsg->free_iov = NULL;
 		if (msg.msg_iovlen == 0) {
 			sr->len = 0;
 		} else if (msg.msg_iovlen > 1) {
@@ -819,10 +820,10 @@ int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
 		goto retry_multishot;
 
 	if (mshot_finished) {
-		io_netmsg_recycle(req, issue_flags);
 		/* fast path, check for non-NULL to avoid function call */
 		if (kmsg->free_iov)
 			kfree(kmsg->free_iov);
+		io_netmsg_recycle(req, issue_flags);
 		req->flags &= ~REQ_F_NEED_CLEANUP;
 	}
 

io_uring/rw.c

@@ -1062,7 +1062,7 @@ int io_do_iopoll(struct io_ring_ctx *ctx, bool force_nonspin)
 			continue;
 
 		req->cqe.flags = io_put_kbuf(req, 0);
-		__io_fill_cqe_req(req->ctx, req);
+		io_fill_cqe_req(req->ctx, req);
 	}
 
 	if (unlikely(!nr_events))

io_uring/timeout.c

@@ -50,7 +50,6 @@ static inline void io_put_req(struct io_kiocb *req)
 }
 
 static bool io_kill_timeout(struct io_kiocb *req, int status)
-	__must_hold(&req->ctx->completion_lock)
 	__must_hold(&req->ctx->timeout_lock)
 {
 	struct io_timeout_data *io = req->async_data;
@@ -70,12 +69,13 @@ static bool io_kill_timeout(struct io_kiocb *req, int status)
 }
 
 __cold void io_flush_timeouts(struct io_ring_ctx *ctx)
-	__must_hold(&ctx->completion_lock)
 {
-	u32 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
+	u32 seq;
 	struct io_timeout *timeout, *tmp;
 
 	spin_lock_irq(&ctx->timeout_lock);
+	seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
+
 	list_for_each_entry_safe(timeout, tmp, &ctx->timeout_list, list) {
 		struct io_kiocb *req = cmd_to_io_kiocb(timeout);
 		u32 events_needed, events_got;
@@ -622,7 +622,11 @@ __cold bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk,
 	struct io_timeout *timeout, *tmp;
 	int canceled = 0;
 
-	io_cq_lock(ctx);
+	/*
+	 * completion_lock is needed for io_match_task(). Take it before
+	 * timeout_lockfirst to keep locking ordering.
+	 */
+	spin_lock(&ctx->completion_lock);
 	spin_lock_irq(&ctx->timeout_lock);
 	list_for_each_entry_safe(timeout, tmp, &ctx->timeout_list, list) {
 		struct io_kiocb *req = cmd_to_io_kiocb(timeout);
@@ -632,6 +636,6 @@ __cold bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk,
 			canceled++;
 	}
 	spin_unlock_irq(&ctx->timeout_lock);
-	io_cq_unlock_post(ctx);
+	spin_unlock(&ctx->completion_lock);
 	return canceled != 0;
 }