>> +
>> +	if (!vec)
>> +		ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
>> +			GFP_KERNEL);
>> +	else {
>> +		struct iovec fast_iov[UIO_FASTIOV];
>> +		struct iovec *iov = fast_iov;
>> +		struct iov_iter iter;
>> +
>> +		ret = import_iovec(rq_data_dir(req), ubuffer, bufflen,
>> +				UIO_FASTIOV, &iov, &iter);
>> +		if (ret < 0)
>>  			goto out;
>> +		ret = blk_rq_map_user_iov(q, req, NULL, &iter, GFP_KERNEL);
>> +		kfree(iov);
>
>To me some of this almost screams like lifting the vectored vs
>not to the block layer into a separate helper.
>
So I skipped doing this, as cleanup is effective when we have the
elephant; only a part is visible here. The last patch (nvme fixedbufs
support) also changes this region. 
I can post a cleanup when all these moving pieces get settled.

>> +	}
>> +	bio = req->bio;
>> +	if (ret)
>> +		goto out_unmap;
>
>This seems incorrect, we don't need to unmap if blk_rq_map_user*
>failed.
>
>> +	if (bdev)
>> +		bio_set_dev(bio, bdev);
>
>I think we can actually drop this now - bi_bdev should only be used
>by the non-passthrough path these days.
Not sure if I am missing something, but this seemed necessary. bi_bdev was
null otherwise.

Did all other changes.