Re: wine fails to start with seccomp updates for v5.9-rc1

[Alex Xu (Hello71)]

Excerpts from Thadeu Lima de Souza Cascardo's message of August 7, 2020 1:36 pm:
> On Fri, Aug 07, 2020 at 08:48:46AM -0700, Linus Torvalds wrote:
>> On Fri, Aug 7, 2020 at 8:19 AM Alex Xu (Hello71) <alex_y_xu@xxxxxxxx> wrote:
>> >
>> > On Linus' master, wine fails to start with the following error:
>> >
>> > wine client error:0: write: Bad file descriptor
>> >
>> > This issue is not present on 5.8. It appears to be caused by failure to
>> > write to a pipe FD received via SCM_RIGHTS. Therefore, I tried reverting
>> > 9ecc6ea491f0, which resolved the issue.
>> 
>> Would you mind trying to bisect exactly where it happens?
>> 
> 
> This report [1] seemed related and pointed out at c0029de50982 ("net/scm:
> Regularize compat handling of scm_detach_fds()"). The use of CMSG_USER_DATA
> instead of CMSG_COMPAT_DATA seems fishy.
> 
> Alex, can you try applying the patch below?
> 
> Cascardo.
> 
> [1] https://lists.ozlabs.org/pipermail/linuxppc-dev/2020-August/216156.html
> 
>> I don't think any of the commits in that pull are supposed to change
>> semantics, and while reverting the whole merge shows that yes, that's
>> what brought in the problems, it would be good to pinpoint just which
>> change breaks so that we can fix just that thing.
>> 
>> Kees, ideas?
>> 
>>                  Linus
> 
> ---
> diff --git a/net/compat.c b/net/compat.c
> index 703acb51c698..95ce707a30a3 100644
> --- a/net/compat.c
> +++ b/net/compat.c
> @@ -294,7 +294,7 @@ void scm_detach_fds_compat(struct msghdr *msg, struct scm_cookie *scm)
>  		(struct compat_cmsghdr __user *)msg->msg_control;
>  	unsigned int o_flags = (msg->msg_flags & MSG_CMSG_CLOEXEC) ? O_CLOEXEC : 0;
>  	int fdmax = min_t(int, scm_max_fds_compat(msg), scm->fp->count);
> -	int __user *cmsg_data = CMSG_USER_DATA(cm);
> +	int __user *cmsg_data = CMSG_COMPAT_DATA(cm);
>  	int err = 0, i;
>  
>  	for (i = 0; i < fdmax; i++) {
> 

Yes, this seems to work.

Tested-by: Alex Xu (Hello71) <alex_y_xu@xxxxxxxx>

Thanks!