[RFC 3/4] io_uring: support futex wake requests

[Pavel Begunkov]

Add support for futex wake requests, which also modifies the addr and
checks against it with encoded operation as FUTEX_WAKE_OP does, but only
operates with a single address as may be problematic to squeeze into SQE
and io_kiocb otherwise.

Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
---
 fs/io_uring.c                 | 48 +++++++++++++++++++++++++++++++++--
 include/uapi/linux/io_uring.h | 10 +++++++-
 2 files changed, 55 insertions(+), 3 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 2c6b14a3a4f6..99f4f8d9f685 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -668,6 +668,12 @@ struct io_unlink {
 
 struct io_futex {
 	struct file			*file;
+	unsigned int			futex_op;
+
+	unsigned int			nr_wake;
+	unsigned int			wake_op_arg;
+	unsigned int			flags;
+	void __user			*uaddr;
 };
 
 struct io_completion {
@@ -5874,12 +5880,50 @@ static int io_files_update(struct io_kiocb *req, unsigned int issue_flags)
 
 static int io_futex_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
-	return -EINVAL;
+	struct io_futex *f = &req->futex;
+	u64 v;
+
+	if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
+		return -EINVAL;
+	if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
+		return -EINVAL;
+	if (sqe->len)
+		return -EINVAL;
+	f->flags = READ_ONCE(sqe->futex_flags);
+	if (f->flags & ~IORING_FUTEX_SHARED)
+		return -EINVAL;
+
+	v = READ_ONCE(sqe->off);
+	f->nr_wake = (u32)v;
+	f->wake_op_arg = (u32)(v >> 32);
+	f->futex_op = READ_ONCE(sqe->futex_op);
+	f->uaddr = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	return 0;
 }
 
 static int io_futex(struct io_kiocb *req, unsigned int issue_flags)
 {
-	return -EINVAL;
+	bool nonblock = issue_flags & IO_URING_F_NONBLOCK;
+	struct io_futex *f = &req->futex;
+	int ret;
+
+	switch (f->futex_op) {
+	case IORING_FUTEX_WAKE_OP:
+		ret = futex_wake_op_single(f->uaddr, f->nr_wake, f->wake_op_arg,
+					   !(f->flags & IORING_FUTEX_SHARED),
+					   nonblock);
+		/* retry from blocking context */
+		if (nonblock && ret == -EAGAIN)
+			return -EAGAIN;
+		break;
+	default:
+		ret = -EINVAL;
+	}
+
+	if (ret < 0)
+		req_set_fail(req);
+	__io_req_complete(req, issue_flags, ret, 0);
+	return 0;
 }
 
 static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 6a1af5bb2ddf..6fa5a6e59934 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -17,7 +17,10 @@
 struct io_uring_sqe {
 	__u8	opcode;		/* type of operation for this sqe */
 	__u8	flags;		/* IOSQE_ flags */
-	__u16	ioprio;		/* ioprio for the request */
+	union {
+		__u16	ioprio;		/* ioprio for the request */
+		__u16	futex_op;	/* futex operation */
+	} __attribute__((packed));
 	__s32	fd;		/* file descriptor to do IO on */
 	union {
 		__u64	off;	/* offset into file */
@@ -161,6 +164,11 @@ enum {
  */
 #define SPLICE_F_FD_IN_FIXED	(1U << 31) /* the last bit of __u32 */
 
+/*
+ * sqe->futex_flags
+ */
+#define IORING_FUTEX_SHARED	(1U << 0)
+
 /*
  * POLL_ADD flags. Note that since sqe->poll_events is the flag space, the
  * command flags for POLL_ADD are stored in sqe->len.
-- 
2.31.1