svn commit: r320761 - head/sbin/init

[Xin Li]

Hi, Kostantin,

On 7/6/17 22:43, Konstantin Belousov wrote:
>> Modified: head/sbin/init/init.c
>> ==============================================================================
>> --- head/sbin/init/init.c	Fri Jul  7 00:34:51 2017	(r320760)
>> +++ head/sbin/init/init.c	Fri Jul  7 02:48:55 2017	(r320761)
>> @@ -1271,8 +1271,8 @@ new_session(session_t *sprev, struct ttyent *typ)
>>  
>>  	sp->se_flags |= SE_PRESENT;
>>  
>> -	sp->se_device = malloc(sizeof(_PATH_DEV) + strlen(typ->ty_name));
>> -	sprintf(sp->se_device, "%s%s", _PATH_DEV, typ->ty_name);
>> +	if (asprintf(&sp->se_device, "%s%s", _PATH_DEV, typ->ty_name) < 0)
>> +		err(1, "asprintf");
>>  
> IMO this is wrong.  init(8) too important for the system operations,
> and panicing the machine due to error from attempt creating getty
> session is not worth it.
> 
> Either session should be disabled, or retried after some time, or
> some other measures taken, but please do not kill init just due to a
> local error.
> 
> I would even argue that using snprintf() there and ignoring truncation
> is much better than err(), not least because the problem probably can
> only practically appear due to a misconfiguration.

Note that the previous code would just crash (due to NULL pointer
deference) so I think this change is an improvement over the status quo.

I do agree that the reliability of init(8) is critical and will see what
we can do with the extreme situation and submit a new CR.

Cheers,

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: OpenPGP digital signature
URL: <http://lists.freebsd.org/pipermail/svn-src-head/attachments/20170706/10e226bc/attachment.sig>