[PATCH net] bonding: fix packets_per_slave showing

[PATCH net] bonding: fix packets_per_slave showing

[Nikolay Aleksandrov]

There's an issue when showing the value of packets_per_slave due to
using signed integer. The value may be < 0 and thus not put through
reciprocal_value() before showing. This patch makes it use unsigned
integer when showing it.

CC: Andy Gospodarek <andy@greyhouse.net>
CC: Jay Vosburgh <fubar@us.ibm.com>
CC: Veaceslav Falico <vfalico@redhat.com>
CC: David S. Miller <davem@davemloft.net>
Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
---
 drivers/net/bonding/bond_sysfs.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/net/bonding/bond_sysfs.c b/drivers/net/bonding/bond_sysfs.c
index abf5e10..0ae580b 100644
--- a/drivers/net/bonding/bond_sysfs.c
+++ b/drivers/net/bonding/bond_sysfs.c
@@ -1635,12 +1635,12 @@ static ssize_t bonding_show_packets_per_slave(struct device *d,
 					      char *buf)
 {
 	struct bonding *bond = to_bond(d);
-	int packets_per_slave = bond->params.packets_per_slave;
+	unsigned int packets_per_slave = bond->params.packets_per_slave;
 
 	if (packets_per_slave > 1)
 		packets_per_slave = reciprocal_value(packets_per_slave);
 
-	return sprintf(buf, "%d\n", packets_per_slave);
+	return sprintf(buf, "%u\n", packets_per_slave);
 }
 
 static ssize_t bonding_store_packets_per_slave(struct device *d,
-- 
1.8.1.4

Re: [PATCH net] bonding: fix packets_per_slave showing

[Veaceslav Falico]

On Thu, Dec 05, 2013 at 11:36:58AM +0100, Nikolay Aleksandrov wrote:
>There's an issue when showing the value of packets_per_slave due to
>using signed integer. The value may be < 0 and thus not put through
>reciprocal_value() before showing. This patch makes it use unsigned
>integer when showing it.

I was already checking my basic algebra knowledge here,
reciprocal_value(reciprocal_value(0..USHRT_MAX)) can become negative?!? :)

If anyone's also wondering...

packets_per_slave is reciprocal_value(0..USHRT_MAX), and thus can indeed be
negative, and then the code

	if (packets_per_slave > 1)
		packets_per_slave = reciprocal_value(packets_per_slave);

would fail to recognise that it's a reciprocal_divide() value, and not a
standard 0/1 option (in bond_rr_gen_slave_id() we verify it via a switch,
so we're safe there) - and thus output nonsense.

Acked-by: Veaceslav Falico <vfalico@redhat.com>

>
>CC: Andy Gospodarek <andy@greyhouse.net>
>CC: Jay Vosburgh <fubar@us.ibm.com>
>CC: Veaceslav Falico <vfalico@redhat.com>
>CC: David S. Miller <davem@davemloft.net>
>Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
>---
> drivers/net/bonding/bond_sysfs.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
>diff --git a/drivers/net/bonding/bond_sysfs.c b/drivers/net/bonding/bond_sysfs.c
>index abf5e10..0ae580b 100644
>--- a/drivers/net/bonding/bond_sysfs.c
>+++ b/drivers/net/bonding/bond_sysfs.c
>@@ -1635,12 +1635,12 @@ static ssize_t bonding_show_packets_per_slave(struct device *d,
> 					      char *buf)
> {
> 	struct bonding *bond = to_bond(d);
>-	int packets_per_slave = bond->params.packets_per_slave;
>+	unsigned int packets_per_slave = bond->params.packets_per_slave;
>
> 	if (packets_per_slave > 1)
> 		packets_per_slave = reciprocal_value(packets_per_slave);
>
>-	return sprintf(buf, "%d\n", packets_per_slave);
>+	return sprintf(buf, "%u\n", packets_per_slave);
> }
>
> static ssize_t bonding_store_packets_per_slave(struct device *d,
>-- 
>1.8.1.4
>

Re: [PATCH net] bonding: fix packets_per_slave showing

[Nikolay Aleksandrov]

On 12/05/2013 12:08 PM, Veaceslav Falico wrote:
> On Thu, Dec 05, 2013 at 11:36:58AM +0100, Nikolay Aleksandrov wrote:
>> There's an issue when showing the value of packets_per_slave due to
>> using signed integer. The value may be < 0 and thus not put through
>> reciprocal_value() before showing. This patch makes it use unsigned
>> integer when showing it.
> 
> I was already checking my basic algebra knowledge here,
> reciprocal_value(reciprocal_value(0..USHRT_MAX)) can become negative?!? :)
> 
> If anyone's also wondering...
> 
> packets_per_slave is reciprocal_value(0..USHRT_MAX), and thus can indeed be
> negative, and then the code
> 
>     if (packets_per_slave > 1)
>         packets_per_slave = reciprocal_value(packets_per_slave);
> 
> would fail to recognise that it's a reciprocal_divide() value, and not a
> standard 0/1 option (in bond_rr_gen_slave_id() we verify it via a switch,
> so we're safe there) - and thus output nonsense.
> 
> Acked-by: Veaceslav Falico <vfalico@redhat.com>
> 
Thanks for the review. Indeed, I should've made it unsigned from the beginning
but thought to be consistent with the other options (just don't ask why :) ).
Anyhow, once net-next opens up I can convert it to unsigned completely as it's
supposed to be.

Cheers,
 Nik

Re: [PATCH net] bonding: fix packets_per_slave showing

[Eric Dumazet]

On Thu, 2013-12-05 at 12:08 +0100, Veaceslav Falico wrote:
> On Thu, Dec 05, 2013 at 11:36:58AM +0100, Nikolay Aleksandrov wrote:
> >There's an issue when showing the value of packets_per_slave due to
> >using signed integer. The value may be < 0 and thus not put through
> >reciprocal_value() before showing. This patch makes it use unsigned
> >integer when showing it.
> 
> I was already checking my basic algebra knowledge here,
> reciprocal_value(reciprocal_value(0..USHRT_MAX)) can become negative?!? :)
> 
> If anyone's also wondering...
> 
> packets_per_slave is reciprocal_value(0..USHRT_MAX), and thus can indeed be
> negative, and then the code
> 
> 	if (packets_per_slave > 1)
> 		packets_per_slave = reciprocal_value(packets_per_slave);
> 
> would fail to recognise that it's a reciprocal_divide() value, and not a
> standard 0/1 option (in bond_rr_gen_slave_id() we verify it via a switch,
> so we're safe there) - and thus output nonsense.
> 
> Acked-by: Veaceslav Falico <vfalico@redhat.com>

This code is very confusing.

Please rename bond->params.packets_per_slave
to bond->params.reciprocal_packets_per_slave 

To make clear that its the reciprocal value.

Also the module parameter is named packets_per_slave, it would be nice
if same name was not reused as local variable in bond_rr_gen_slave_id()

bond_check_params() reads the sys value several times.

This is racy with /sys access.

You should use ACCESS_ONCE() to make sure nothing bad happens.

Re: [PATCH net] bonding: fix packets_per_slave showing

[Nikolay Aleksandrov]

On 12/05/2013 01:33 PM, Eric Dumazet wrote:
> On Thu, 2013-12-05 at 12:08 +0100, Veaceslav Falico wrote:
>> On Thu, Dec 05, 2013 at 11:36:58AM +0100, Nikolay Aleksandrov wrote:
>>> There's an issue when showing the value of packets_per_slave due to
>>> using signed integer. The value may be < 0 and thus not put through
>>> reciprocal_value() before showing. This patch makes it use unsigned
>>> integer when showing it.
>>
>> I was already checking my basic algebra knowledge here,
>> reciprocal_value(reciprocal_value(0..USHRT_MAX)) can become negative?!? :)
>>
>> If anyone's also wondering...
>>
>> packets_per_slave is reciprocal_value(0..USHRT_MAX), and thus can indeed be
>> negative, and then the code
>>
>> 	if (packets_per_slave > 1)
>> 		packets_per_slave = reciprocal_value(packets_per_slave);
>>
>> would fail to recognise that it's a reciprocal_divide() value, and not a
>> standard 0/1 option (in bond_rr_gen_slave_id() we verify it via a switch,
>> so we're safe there) - and thus output nonsense.
>>
>> Acked-by: Veaceslav Falico <vfalico@redhat.com>
> 
> This code is very confusing.
> 
> Please rename bond->params.packets_per_slave
> to bond->params.reciprocal_packets_per_slave 
> 
> To make clear that its the reciprocal value.
> 
> Also the module parameter is named packets_per_slave, it would be nice
> if same name was not reused as local variable in bond_rr_gen_slave_id()
> 
> bond_check_params() reads the sys value several times.
> 
> This is racy with /sys access.
> 
IIRC bond_check_params() runs before sysfs is initialized for the bond device.

> You should use ACCESS_ONCE() to make sure nothing bad happens.
> 
Actually I think ACCESS_ONCE() should be added to bond_show_packets_per_slave
and to bond_rr_gen_slave_id() just as a precaution. It'll also serve the purpose
to show what's intended.
What do you think ?
> 
> 
Thanks for the feedback, I'll address the ACCESS_ONCE() in a separate patch for
net and leave the renaming to my net-next patch which will take care of the
types as well.

Cheers,
 Nik

Re: [PATCH net] bonding: fix packets_per_slave showing

[Veaceslav Falico]

On Thu, Dec 05, 2013 at 04:33:55AM -0800, Eric Dumazet wrote:
>On Thu, 2013-12-05 at 12:08 +0100, Veaceslav Falico wrote:
>> On Thu, Dec 05, 2013 at 11:36:58AM +0100, Nikolay Aleksandrov wrote:
>> >There's an issue when showing the value of packets_per_slave due to
>> >using signed integer. The value may be < 0 and thus not put through
>> >reciprocal_value() before showing. This patch makes it use unsigned
>> >integer when showing it.
>>
>> I was already checking my basic algebra knowledge here,
>> reciprocal_value(reciprocal_value(0..USHRT_MAX)) can become negative?!? :)
>>
>> If anyone's also wondering...
>>
>> packets_per_slave is reciprocal_value(0..USHRT_MAX), and thus can indeed be
>> negative, and then the code
>>
>> 	if (packets_per_slave > 1)
>> 		packets_per_slave = reciprocal_value(packets_per_slave);
>>
>> would fail to recognise that it's a reciprocal_divide() value, and not a
>> standard 0/1 option (in bond_rr_gen_slave_id() we verify it via a switch,
>> so we're safe there) - and thus output nonsense.
>>
>> Acked-by: Veaceslav Falico <vfalico@redhat.com>
>
>This code is very confusing.
>
>Please rename bond->params.packets_per_slave
>to bond->params.reciprocal_packets_per_slave
>
>To make clear that its the reciprocal value.
>
>Also the module parameter is named packets_per_slave, it would be nice
>if same name was not reused as local variable in bond_rr_gen_slave_id()

Agreed, but I think it'd be rather net-next material. Here we have a clear
bug...

>
>bond_check_params() reads the sys value several times.
>
>This is racy with /sys access.
>
>You should use ACCESS_ONCE() to make sure nothing bad happens.

Hrm? bond_check_params() isn't involved in sysfs at all :-/, it's called
only via bonding_init(). And bond->params.packets_per_slave isn't read
there at all, only assigned.

Or, given the naming confusions, I'm again missing something?

Re: [PATCH net] bonding: fix packets_per_slave showing

[Eric Dumazet]

On Thu, 2013-12-05 at 13:42 +0100, Veaceslav Falico wrote:
> .
> 
> Hrm? bond_check_params() isn't involved in sysfs at all :-/, it's called
> only via bonding_init(). And bond->params.packets_per_slave isn't read
> there at all, only assigned.
> 
> Or, given the naming confusions, I'm again missing something?

This is so confusing.

If the check was done properly, we could expose the value
in /sys/module/bonding/parameters/packets_per_slave

So that you can change the setting without reloading the module.

Re: [PATCH net] bonding: fix packets_per_slave showing

[David Miller]

From: Nikolay Aleksandrov <nikolay@redhat.com>
Date: Thu,  5 Dec 2013 11:36:58 +0100

> There's an issue when showing the value of packets_per_slave due to
> using signed integer. The value may be < 0 and thus not put through
> reciprocal_value() before showing. This patch makes it use unsigned
> integer when showing it.
> 
> CC: Andy Gospodarek <andy@greyhouse.net>
> CC: Jay Vosburgh <fubar@us.ibm.com>
> CC: Veaceslav Falico <vfalico@redhat.com>
> CC: David S. Miller <davem@davemloft.net>
> Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>

Applied, thanks.