Questions regarding IORING_OP_SENDMSG

Questions regarding IORING_OP_SENDMSG

[Stefan Metzmacher]

[-- Attachment #1.1: Type: text/plain, Size: 792 bytes --]

Hi Jens,

In the "io_uring: add support for IORING_OP_IOCTL" thread I wrote this:

> I could use also use a generic way for an async fd-based syscall.
> I thought about using sendmsg() with special CMSG_ elements, but
> currently it's not possible with IORING_OP_SENDMSG to do an async
> io_kiocb based completion, using msg_iocb.

I just noticed that __sys_sendmsg_sock doesn't allow
msg->msg_control || msg->msg_controllen.

Why is that the case?
I could use that in several places for my smbdirect driver.

Optionally turning on msg_iocb usage would also be great,
I'd use that in combination with MSG_OOB and CMSG_
and this OOB message is not ordered like the typical
sendmsg stream socket flow.

Do you think we can add support for this usecase?

Thanks!
metze


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: Questions regarding IORING_OP_SENDMSG

[Jens Axboe]

On 1/16/20 4:13 PM, Stefan Metzmacher wrote:
> Hi Jens,
> 
> In the "io_uring: add support for IORING_OP_IOCTL" thread I wrote this:
> 
>> I could use also use a generic way for an async fd-based syscall.
>> I thought about using sendmsg() with special CMSG_ elements, but
>> currently it's not possible with IORING_OP_SENDMSG to do an async
>> io_kiocb based completion, using msg_iocb.
> 
> I just noticed that __sys_sendmsg_sock doesn't allow
> msg->msg_control || msg->msg_controllen.
> 
> Why is that the case?
> I could use that in several places for my smbdirect driver.
> 
> Optionally turning on msg_iocb usage would also be great,
> I'd use that in combination with MSG_OOB and CMSG_
> and this OOB message is not ordered like the typical
> sendmsg stream socket flow.
> 
> Do you think we can add support for this usecase?

It was done because of the CVE related to io_uring, but we really
don't need to since we have creds setup properly now. So it could
be allowed again.

-- 
Jens Axboe