io_uring: fix IOPOLL -EAGAIN retries

This normally isn't hit, as polling is mostly done on NVMe with deep
queue depths. But if we do run into request starvation, we need to
ensure that retries are properly serialized.
Reported-by: Andres Freund <andres@anarazel.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs/io_uring.c

@@ -1150,7 +1150,7 @@ static void io_prep_async_work(struct io_kiocb *req)
 	io_req_init_async(req);
 
 	if (req->flags & REQ_F_ISREG) {
-		if (def->hash_reg_file)
+		if (def->hash_reg_file || (req->ctx->flags & IORING_SETUP_IOPOLL))
 			io_wq_hash_work(&req->work, file_inode(req->file));
 	} else {
 		if (def->unbound_nonreg_file)
@@ -3132,6 +3132,7 @@ static int io_read(struct io_kiocb *req, bool force_nonblock,
 	ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock);
 	if (ret < 0)
 		return ret;
+	iov_count = iov_iter_count(iter);
 	io_size = ret;
 	req->result = io_size;
 	ret = 0;
@@ -3144,7 +3145,6 @@ static int io_read(struct io_kiocb *req, bool force_nonblock,
 	if (force_nonblock && !io_file_supports_async(req->file, READ))
 		goto copy_iov;
 
-	iov_count = iov_iter_count(iter);
 	ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), iov_count);
 	if (unlikely(ret))
 		goto out_free;
@@ -3157,7 +3157,8 @@ static int io_read(struct io_kiocb *req, bool force_nonblock,
 		ret = 0;
 		goto out_free;
 	} else if (ret == -EAGAIN) {
-		if (!force_nonblock)
+		/* IOPOLL retry should happen for io-wq threads */
+		if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
 			goto done;
 		/* some cases will consume bytes even on error returns */
 		iov_iter_revert(iter, iov_count - iov_iter_count(iter));
@@ -3251,6 +3252,7 @@ static int io_write(struct io_kiocb *req, bool force_nonblock,
 	ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock);
 	if (ret < 0)
 		return ret;
+	iov_count = iov_iter_count(iter);
 	io_size = ret;
 	req->result = io_size;
 
@@ -3267,7 +3269,6 @@ static int io_write(struct io_kiocb *req, bool force_nonblock,
 	    (req->flags & REQ_F_ISREG))
 		goto copy_iov;
 
-	iov_count = iov_iter_count(iter);
 	ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), iov_count);
 	if (unlikely(ret))
 		goto out_free;
@@ -3301,11 +3302,14 @@ static int io_write(struct io_kiocb *req, bool force_nonblock,
 	if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
 		ret2 = -EAGAIN;
 	if (!force_nonblock || ret2 != -EAGAIN) {
+		/* IOPOLL retry should happen for io-wq threads */
+		if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
+			goto copy_iov;
 		kiocb_done(kiocb, ret2, cs);
 	} else {
+copy_iov:
 		/* some cases will consume bytes even on error returns */
 		iov_iter_revert(iter, iov_count - iov_iter_count(iter));
-copy_iov:
 		ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false);
 		if (!ret)
 			return -EAGAIN;