Date:	Mon, 09 Dec 2013 09:55:55 -0500
From:	Vlad Yasevich <vyasevich@...il.com>
To:	Wang Weidong <wangweidong1@...wei.com>, nhorman@...driver.com,
	davem@...emloft.net
CC:	dborkman@...hat.com, netdev@...r.kernel.org,
	linux-sctp@...r.kernel.org
Subject: Re: [PATCH v5 1/2] sctp: check the rto_min and rto_max

On 12/08/2013 10:28 PM, Wang Weidong wrote:
> On 2013/12/9 10:40, Vlad Yasevich wrote:
>> On 12/08/2013 09:28 PM, Wang Weidong wrote:
>>> On 2013/12/9 10:19, Vlad Yasevich wrote:
>>>> On 12/08/2013 08:53 PM, Wang Weidong wrote:
>>>>> On 2013/12/8 2:54, Vlad Yasevich wrote:
>>>>>> On 12/07/2013 02:17 AM, Wang Weidong wrote:
>>>>>>> rto_min should be smaller than rto_max while rto_max should be larger
>>>>>>> than rto_min. Add two proc_handler for the checking. Add the check in
>>>>>>> sctp_setsockopt_rtoinfo.
>>>>>>>
>>>>>>> Suggested-by: Vlad Yasevich <vyasevich@...il.com>
>>>>>>> Signed-off-by: Wang Weidong <wangweidong1@...wei.com>
>>>>>>> ---
>>>>>>>  include/net/sctp/constants.h |  3 ++
>>>>>>>  net/sctp/socket.c            |  5 +++
>>>>>>>  net/sctp/sysctl.c            | 73 ++++++++++++++++++++++++++++++++++++++++----
>>>>>>>  3 files changed, 75 insertions(+), 6 deletions(-)
>>>>>>>
>>>>>>> diff --git a/include/net/sctp/constants.h b/include/net/sctp/constants.h
>>>>>>> index 2f0a565..d276978 100644
>>>>>>> --- a/include/net/sctp/constants.h
>>>>>>> +++ b/include/net/sctp/constants.h
>>>>>>> @@ -279,6 +279,9 @@ enum { SCTP_MAX_GABS = 16 };
>>>>>>>  #define SCTP_RTO_ALPHA          3   /* 1/8 when converted to right shifts. */
>>>>>>>  #define SCTP_RTO_BETA           2   /* 1/4 when converted to right shifts. */
>>>>>>>  
>>>>>>> +#define SCTP_ONE                1        /* 1 ms */
>>>>>>> +#define SCTP_TIMER_MAX          86400000 /* ms in one day */
>>>>>>> +
>>>>>>>  /* Maximum number of new data packets that can be sent in a burst.  */
>>>>>>>  #define SCTP_DEFAULT_MAX_BURST		4
>>>>>>>  
>>>>>>> diff --git a/net/sctp/socket.c b/net/sctp/socket.c
>>>>>>> index 72046b9..13411ad 100644
>>>>>>> --- a/net/sctp/socket.c
>>>>>>> +++ b/net/sctp/socket.c
>>>>>>> @@ -2818,6 +2818,11 @@ static int sctp_setsockopt_rtoinfo(struct sock *sk, char __user *optval, unsigne
>>>>>>>  	if (copy_from_user(&rtoinfo, optval, optlen))
>>>>>>>  		return -EFAULT;
>>>>>>>  
>>>>>>> +	if (rtoinfo.srto_min < SCTP_ONE ||
>>>>>>> +	    rtoinfo.srto_max > SCTP_TIMER_MAX ||
>>>>>>> +	    rtoinfo.srto_max < rtoinfo.srto_min)
>>>>>>> +		return -EINVAL;
>>>>>>> +
>>>>>>
>>>>>> You can not do the check for srto_min < 1.  The following is the text
>>>>>> from the spec:
>>>>>>    All times are given in milliseconds.  A value of 0, when modifying
>>>>>>    the parameters, indicates that the current value should not be
>>>>>>    changed.
>>>>>>
>>>>> Yes, Your are right, I found it in draft-ietf-tsvwg-sctpsocket-14.txt.
>>>>> Thanks.
>>>>>
>>>>>> So, it is valid for a user to pass in a value of 0.  Also, I am not sure
>>>>>> if it makes sense to bind the upper limit here, as well.
>>>>>>
>>>>>> -vlad
>>>>>>
>>>>> Here, I am not sure as well. I think it should like what we do to the
>>>>> init_net.sctp.rto_max when set larger than timer_max. Just not change the value.
>>>>>
>>>>
>>>> So, the basic reason that sysctl is limited is that it is a default
>>>> for all sctp association on the system.  It makes some sense to limit
>>>> what the max value here could be.  Limiting it to double suggested
>>>> RTO.MAX would only make it 2 minutes and may be insufficient for some
>>>> of the high latency low-throughput wireless links.  Making it about an
>>>> hour should be fine...  This would be a separate patch though...
>>>>
>>> Here, you mean that we should use 3600*1000 rather than 86400000? So
>>> we should use another patch to fix that after my patchs?
>>>
>>>> Limiting the user-supplied value is not as appropriate since the
>>>> assumption is that user application may know better what it's
>>>> requirements are and it is not up to the stack to limit those.  As
>>>> long as the user value is withing the usable range (and the kernel
>>>> will already knows how and does limit this range), we should not
>>>> limit this further.
>>>>
>>>> -vlad
>>>>
>>> Agree, So I should check like this:
>>> !srto_min || !srto_max || srto_min > srto_max ?
>>> And no need to add macros for checking.
>>
>> No, I think this would have to be a little more complicated :(
>> Remember it's ok to have srto_min == 0 and srto_max == 0.  It just
>> means that no change happens.
>>
>> You may need to do something like
>>
>>   unsigned long rto_max, rto_min;
>>
>>   if (rtoinfo.srto_max)
>>      rto_max = msecs_to_jiffies(rtoinfo.srto_max);
>>   else
>>      rto_max = asoc ? asoc->rto_max : sp->rto_max;
>>
>>   if (rtoinfo.srto_min)
>>      rto_min = msecs_to_jiffies(rtoinfo.srto_min);
>>   else
>>      rto_min = asoc ? asoc->rto_min : sp->rto_min;
>>
>>   if (rto_min > rto_max)
>>       return -EINVAL;
>>
>>   if (asoc) {
>>       asoc->rto_min = rto_min;
>>       asoc->rto_max = rto_max;
>>    ...
>>    etc....
>>
>> This way we make sure that the user that supplied just rto_min or just
>> rto_max didn't set them so that min > max.
>>
>> -vlad
> 
> Hi vald,
> 
> I found that we had checked the value of 0 in sctp_setsockopt_rtoinfo.
> So I only do this:
> 
> if (asoc) {
> +	if (msecs_to_jiffies(rtoinfo.srto_min) >
> +                   msecs_to_jiffies(rtoinfo.srto_max))
> +                       return -EINVAL;
> 	...

What if the value in rtoinfo is 0?  Right now, the doesn't do any
comparisons and just assigns values into the assoc or sp as long
as the user provided a non-0 value.

Now imagine the user did this:

    rtoinfo.srto_min = 0
    rtoinfo.srto_max = 5;
    setsockopt();

    ....  later on...

    rtoinfo.srto_min = 8;
    rtoinfo.srto_max = 0;
    setsockopt();

No you have a situation where min > max. However both calls were valid.

My suggestion to you, split the sysctl change into a separate patch and
and do socket option handling in its own patch.  Also, please be sure
to test it with different variants of the calls.

-vlad

> } else {
> 	...
> +	if (rtoinfo.srto_min > rtoinfo.srto_max)
> +                       return -EINVAL;
> 	...
> }
> 
> There because we set value to asoc and sp is not same. So I add the
> check into two path.
> 
> Regards.
> Wang
> 
>>>
>>> Thanks.
>>>
>>>>>
>>>>> Regards.
>>>>> Wang
>>>>>
>>>>>>>  	asoc = sctp_id2assoc(sk, rtoinfo.srto_assoc_id);
>>>>>>>  
>>>>>>>  	/* Set the values to the specific association */
>>>>>>> diff --git a/net/sctp/sysctl.c b/net/sctp/sysctl.c
>>>>>>> index 6b36561..33c56c6 100644
>>>>>>> --- a/net/sctp/sysctl.c
>>>>>>> +++ b/net/sctp/sysctl.c
>>>>>>> @@ -40,8 +40,8 @@
>>>>>>>  #include <linux/sysctl.h>
>>>>>>>  
>>>>>>>  static int zero = 0;
>>>>>>> -static int one = 1;
>>>>>>> -static int timer_max = 86400000; /* ms in one day */
>>>>>>> +static int one = SCTP_ONE;
>>>>>>> +static int timer_max = SCTP_TIMER_MAX;
>>>>>>>  static int int_max = INT_MAX;
>>>>>>>  static int sack_timer_min = 1;
>>>>>>>  static int sack_timer_max = 500;
>>>>>>> @@ -61,6 +61,13 @@ static int proc_sctp_do_hmac_alg(struct ctl_table *ctl,
>>>>>>>  				void __user *buffer, size_t *lenp,
>>>>>>>  
>>>>>>>  				loff_t *ppos);
>>>>>>> +static int proc_sctp_do_rto_min(struct ctl_table *ctl, int write,
>>>>>>> +				void __user *buffer, size_t *lenp,
>>>>>>> +				loff_t *ppos);
>>>>>>> +static int proc_sctp_do_rto_max(struct ctl_table *ctl, int write,
>>>>>>> +				void __user *buffer, size_t *lenp,
>>>>>>> +				loff_t *ppos);
>>>>>>> +
>>>>>>>  static struct ctl_table sctp_table[] = {
>>>>>>>  	{
>>>>>>>  		.procname	= "sctp_mem",
>>>>>>> @@ -102,17 +109,17 @@ static struct ctl_table sctp_net_table[] = {
>>>>>>>  		.data		= &init_net.sctp.rto_min,
>>>>>>>  		.maxlen		= sizeof(unsigned int),
>>>>>>>  		.mode		= 0644,
>>>>>>> -		.proc_handler	= proc_dointvec_minmax,
>>>>>>> +		.proc_handler	= proc_sctp_do_rto_min,
>>>>>>>  		.extra1         = &one,
>>>>>>> -		.extra2         = &timer_max
>>>>>>> +		.extra2         = &init_net.sctp.rto_max
>>>>>>>  	},
>>>>>>>  	{
>>>>>>>  		.procname	= "rto_max",
>>>>>>>  		.data		= &init_net.sctp.rto_max,
>>>>>>>  		.maxlen		= sizeof(unsigned int),
>>>>>>>  		.mode		= 0644,
>>>>>>> -		.proc_handler	= proc_dointvec_minmax,
>>>>>>> -		.extra1         = &one,
>>>>>>> +		.proc_handler	= proc_sctp_do_rto_max,
>>>>>>> +		.extra1         = &init_net.sctp.rto_min,
>>>>>>>  		.extra2         = &timer_max
>>>>>>>  	},
>>>>>>>  	{
>>>>>>> @@ -342,6 +349,60 @@ static int proc_sctp_do_hmac_alg(struct ctl_table *ctl,
>>>>>>>  	return ret;
>>>>>>>  }
>>>>>>>  
>>>>>>> +static int proc_sctp_do_rto_min(struct ctl_table *ctl, int write,
>>>>>>> +				void __user*buffer, size_t *lenp,
>>>>>>> +				loff_t *ppos)
>>>>>>> +{
>>>>>>> +	struct net *net = current->nsproxy->net_ns;
>>>>>>> +	int new_value;
>>>>>>> +	struct ctl_table tbl;
>>>>>>> +	unsigned int min = *(unsigned int *) ctl->extra1;
>>>>>>> +	unsigned int max = *(unsigned int *) ctl->extra2;
>>>>>>> +	int ret;
>>>>>>> +
>>>>>>> +	memset(&tbl, 0, sizeof(struct ctl_table));
>>>>>>> +	tbl.maxlen = sizeof(unsigned int);
>>>>>>> +
>>>>>>> +	if (write)
>>>>>>> +		tbl.data = &new_value;
>>>>>>> +	else
>>>>>>> +		tbl.data = &net->sctp.rto_min;
>>>>>>> +	ret = proc_dointvec(&tbl, write, buffer, lenp, ppos);
>>>>>>> +	if (write) {
>>>>>>> +		if (ret || new_value > max || new_value < min)
>>>>>>> +			return -EINVAL;
>>>>>>> +		net->sctp.rto_min = new_value;
>>>>>>> +	}
>>>>>>> +	return ret;
>>>>>>> +}
>>>>>>> +
>>>>>>> +static int proc_sctp_do_rto_max(struct ctl_table *ctl, int write,
>>>>>>> +				void __user*buffer, size_t *lenp,
>>>>>>> +				loff_t *ppos)
>>>>>>> +{
>>>>>>> +	struct net *net = current->nsproxy->net_ns;
>>>>>>> +	int new_value;
>>>>>>> +	struct ctl_table tbl;
>>>>>>> +	unsigned int min = *(unsigned int *) ctl->extra1;
>>>>>>> +	unsigned int max = *(unsigned int *) ctl->extra2;
>>>>>>> +	int ret;
>>>>>>> +
>>>>>>> +	memset(&tbl, 0, sizeof(struct ctl_table));
>>>>>>> +	tbl.maxlen = sizeof(unsigned int);
>>>>>>> +
>>>>>>> +	if (write)
>>>>>>> +		tbl.data = &new_value;
>>>>>>> +	else
>>>>>>> +		tbl.data = &net->sctp.rto_max;
>>>>>>> +	ret = proc_dointvec(&tbl, write, buffer, lenp, ppos);
>>>>>>> +	if (write) {
>>>>>>> +                if (ret || new_value > max || new_value < min)
>>>>>>> +			return -EINVAL;
>>>>>>> +		net->sctp.rto_max = new_value;
>>>>>>> +	}
>>>>>>> +	return ret;
>>>>>>> +}
>>>>>>> +
>>>>>>>  int sctp_sysctl_net_register(struct net *net)
>>>>>>>  {
>>>>>>>  	struct ctl_table *table;
>>>>>>>
>>>>>>
>>>>>> --
>>>>>> To unsubscribe from this list: send the line "unsubscribe netdev" in
>>>>>> the body of a message to majordomo@...r.kernel.org
>>>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>>>>
>>>>>> .
>>>>>>
>>>>>
>>>>>
>>>>
>>>> --
>>>> To unsubscribe from this list: send the line "unsubscribe linux-sctp" in
>>>> the body of a message to majordomo@...r.kernel.org
>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>>
>>>> .
>>>>
>>>
>>>
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-sctp" in
>> the body of a message to majordomo@...r.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>> .
>>
> 
> 

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists