[PATCH RFC -next] io_uring: add support for level triggered poll

[Jens Axboe <axboe@xxxxxxxxx>]

By default, the POLL_ADD command does edge triggered poll - if we get
a non-zero mask on the initial poll attempt, we complete the request
successfully.

Support level triggered by always waiting for a notification, regardless
of whether or not the initial mask matches the file state.

Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>

---

Wrote a test case and it seems to work as it should, and the usual
regressions pass as well.

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 53e7dae92e42..1d176f935f5d 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -226,10 +226,13 @@ enum io_uring_op {
  *
  * IORING_POLL_UPDATE		Update existing poll request, matching
  *				sqe->addr as the old user_data field.
+ *
+ * IORING_POLL_LEVEL		Level triggered poll.
  */
 #define IORING_POLL_ADD_MULTI	(1U << 0)
 #define IORING_POLL_UPDATE_EVENTS	(1U << 1)
 #define IORING_POLL_UPDATE_USER_DATA	(1U << 2)
+#define IORING_POLL_ADD_LEVEL		(1U << 3)
 
 /*
  * ASYNC_CANCEL flags.
diff --git a/io_uring/poll.c b/io_uring/poll.c
index ed9f74403d89..7a98d934428e 100644
--- a/io_uring/poll.c
+++ b/io_uring/poll.c
@@ -425,11 +425,13 @@ static int __io_arm_poll_handler(struct io_kiocb *req,
 	atomic_set(&req->poll_refs, 1);
 	mask = vfs_poll(req->file, &ipt->pt) & poll->events;
 
-	if (mask && (poll->events & EPOLLONESHOT)) {
+	if (mask &&
+	   ((poll->events & (EPOLLET|EPOLLONESHOT)) == (EPOLLET|EPOLLONESHOT))) {
 		io_poll_remove_entries(req);
 		/* no one else has access to the req, forget about the ref */
 		return mask;
 	}
+
 	if (!mask && unlikely(ipt->error || !ipt->nr_entries)) {
 		io_poll_remove_entries(req);
 		if (!ipt->error)
@@ -441,7 +443,7 @@ static int __io_arm_poll_handler(struct io_kiocb *req,
 	io_poll_req_insert(req);
 	spin_unlock(&ctx->completion_lock);
 
-	if (mask) {
+	if (mask && (poll->events & EPOLLET)) {
 		/* can't multishot if failed, just queue the event we've got */
 		if (unlikely(ipt->error || !ipt->nr_entries))
 			poll->events |= EPOLLONESHOT;
@@ -474,7 +476,7 @@ int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 	struct io_ring_ctx *ctx = req->ctx;
 	struct async_poll *apoll;
 	struct io_poll_table ipt;
-	__poll_t mask = POLLPRI | POLLERR;
+	__poll_t mask = POLLPRI | POLLERR | EPOLLET;
 	int ret;
 
 	if (!def->pollin && !def->pollout)
@@ -636,7 +638,10 @@ static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe,
 #endif
 	if (!(flags & IORING_POLL_ADD_MULTI))
 		events |= EPOLLONESHOT;
-	return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT));
+	if (!(flags & IORING_POLL_ADD_LEVEL))
+		events |= EPOLLET;
+	return demangle_poll(events) |
+		(events & (EPOLLEXCLUSIVE|EPOLLONESHOT|EPOLLET));
 }
 
 int io_poll_remove_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
@@ -677,7 +682,7 @@ int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	if (sqe->buf_index || sqe->off || sqe->addr)
 		return -EINVAL;
 	flags = READ_ONCE(sqe->len);
-	if (flags & ~IORING_POLL_ADD_MULTI)
+	if (flags & ~(IORING_POLL_ADD_MULTI|IORING_POLL_ADD_LEVEL))
 		return -EINVAL;
 	if ((flags & IORING_POLL_ADD_MULTI) && (req->flags & REQ_F_CQE_SKIP))
 		return -EINVAL;

-- 
Jens Axboe