Re: [PATCH] io_uring: don't mark S_ISBLK async work as unbounded

[Pavel Begunkov <asml.silence@gmail.com>]

On 02/04/2021 11:32, Pavel Begunkov wrote:
> On 02/04/2021 09:52, Hao Xu wrote:
>> 在 2021/4/1 下午10:57, Jens Axboe 写道:
>>> S_ISBLK is marked as unbounded work for async preparation, because it
>>> doesn't match S_ISREG. That is incorrect, as any read/write to a block
>>> device is also a bounded operation. Fix it up and ensure that S_ISBLK
>>> isn't marked unbounded.
>>>
>>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>>
>> Hi Jens, I saw a (un)bounded work is for a (un)bounded worker to
>> execute. What is the difference between bounded and unbounded?
> 
> Unbounded works are not bounded in execution time, i.e. they may take
> forever to complete. E.g. recv depends on the other end to send something,
> that not necessarily will ever happen.

To elaborate a bit, one example of how it's used: because unbounded may
stay for long, it always spawns a new worker thread for each of them.

If app submits SQEs as below, and send's are not actually sent for execution
but stashed somewhere internally in a list, e.g. waiting for a worker thread
to get free, it would just hang from the userspace perspective.

recv(fd1), recv(fd1), send(fd1), send(fd1)


>>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>>> index 6d7a1b69712b..a16b7df934d1 100644
>>> --- a/fs/io_uring.c
>>> +++ b/fs/io_uring.c
>>> @@ -1213,7 +1213,7 @@ static void io_prep_async_work(struct io_kiocb *req)
>>>       if (req->flags & REQ_F_ISREG) {
>>>           if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
>>>               io_wq_hash_work(&req->work, file_inode(req->file));
>>> -    } else {
>>> +    } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) {
>>>           if (def->unbound_nonreg_file)
>>>               req->work.flags |= IO_WQ_WORK_UNBOUND;
>>>       }

-- 
Pavel Begunkov