[Differential] D12457: Complete INADDR_HASH lock protection

Fri Sep 22 12:18:49 UTC 2017

mav added inline comments.

INLINE COMMENTS

> if_stf.c:383
>  				break;
> +		IN_IFADDR_RUNLOCK(&in_ifa_tracker);
>  		if (ia4 == NULL)

Wouldn't it be better to use in_localip() here instead of code duplication?  Or I miss something?

> in_mcast.c:1383
> +			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
> +		}
>  		if (sopt->sopt_name == IP_BLOCK_SOURCE)

Do I miss some other protection means or this and few below places just scream about possible races due to missing interface pointer pulled out of the lock without taking reference?

I suspect INADDR_TO_IFP KPI is not safe now in general.

> ip_input.c:697
>  	}
> -	/* IN_IFADDR_RUNLOCK(); */
> +	IN_IFADDR_RUNLOCK(&in_ifa_tracker);
>  

Looking on r194951 commit message and later r286001 commit this should be uncommented now, but comments from active networking people are welcome.

REPOSITORY
  rS FreeBSD src repository

REVISION DETAIL
  https://reviews.freebsd.org/D12457

EMAIL PREFERENCES
  https://reviews.freebsd.org/settings/panel/emailpreferences/

To: eugen_grosbein.net, ae, avg, mav, rwatson
Cc: imp, freebsd-net-list