* [RFC PATCH] fix IP_ECN_set_ce
@ 2012-12-19 6:21 roy.qing.li
2012-12-19 8:11 ` Julian Anastasov
2012-12-19 16:14 ` Eric Dumazet
0 siblings, 2 replies; 7+ messages in thread
From: roy.qing.li @ 2012-12-19 6:21 UTC (permalink / raw)
To: netdev
From: Li RongQing <roy.qing.li@gmail.com>
1. ECN uses the two least significant (right-most) bits of the DiffServ
field in the IPv4, so it should be in iph->tos, not in (iph->tos+1)
2. When setting CE, we should check if ECN Capable Transport supports,
both 10 and 01 mean ECN Capable Transport, so only check 10 is not enough
00: Non ECN-Capable Transport — Non-ECT
10: ECN Capable Transport — ECT(0)
01: ECN Capable Transport — ECT(1)
11: Congestion Encountered — CE
3. Remove the misunderstand comment
4. fix the checksum computation
Signed-off-by: Li RongQing <roy.qing.li@gmail.com>
---
include/net/inet_ecn.h | 22 ++++------------------
1 file changed, 4 insertions(+), 18 deletions(-)
diff --git a/include/net/inet_ecn.h b/include/net/inet_ecn.h
index aab7375..545a683 100644
--- a/include/net/inet_ecn.h
+++ b/include/net/inet_ecn.h
@@ -73,27 +73,13 @@ static inline void INET_ECN_dontxmit(struct sock *sk)
static inline int IP_ECN_set_ce(struct iphdr *iph)
{
- u32 check = (__force u32)iph->check;
- u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
-
- /*
- * After the last operation we have (in binary):
- * INET_ECN_NOT_ECT => 01
- * INET_ECN_ECT_1 => 10
- * INET_ECN_ECT_0 => 11
- * INET_ECN_CE => 00
- */
- if (!(ecn & 2))
+ u32 ecn = iph->tos & INET_ECN_MASK;
+
+ if (INET_ECN_is_ce(ecn) || INET_ECN_is_not_ect(ecn))
return !ecn;
- /*
- * The following gives us:
- * INET_ECN_ECT_1 => check += htons(0xFFFD)
- * INET_ECN_ECT_0 => check += htons(0xFFFE)
- */
- check += (__force u16)htons(0xFFFB) + (__force u16)htons(ecn);
+ csum_replace2(&iph->check, iph->tos, iph->tos|INET_ECN_CE);
- iph->check = (__force __sum16)(check + (check>=0xFFFF));
iph->tos |= INET_ECN_CE;
return 1;
}
--
1.7.10.4
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: [RFC PATCH] fix IP_ECN_set_ce
2012-12-19 6:21 [RFC PATCH] fix IP_ECN_set_ce roy.qing.li
@ 2012-12-19 8:11 ` Julian Anastasov
2012-12-19 8:41 ` RongQing Li
2012-12-19 16:14 ` Eric Dumazet
1 sibling, 1 reply; 7+ messages in thread
From: Julian Anastasov @ 2012-12-19 8:11 UTC (permalink / raw)
To: roy.qing.li; +Cc: netdev
[-- Attachment #1: Type: TEXT/PLAIN, Size: 2288 bytes --]
Hello,
On Wed, 19 Dec 2012, roy.qing.li@gmail.com wrote:
> From: Li RongQing <roy.qing.li@gmail.com>
>
> 1. ECN uses the two least significant (right-most) bits of the DiffServ
> field in the IPv4, so it should be in iph->tos, not in (iph->tos+1)
>
> 2. When setting CE, we should check if ECN Capable Transport supports,
> both 10 and 01 mean ECN Capable Transport, so only check 10 is not enough
> 00: Non ECN-Capable Transport — Non-ECT
> 10: ECN Capable Transport — ECT(0)
> 01: ECN Capable Transport — ECT(1)
> 11: Congestion Encountered — CE
>
> 3. Remove the misunderstand comment
>
> 4. fix the checksum computation
>
> Signed-off-by: Li RongQing <roy.qing.li@gmail.com>
> ---
> include/net/inet_ecn.h | 22 ++++------------------
> 1 file changed, 4 insertions(+), 18 deletions(-)
>
> diff --git a/include/net/inet_ecn.h b/include/net/inet_ecn.h
> index aab7375..545a683 100644
> --- a/include/net/inet_ecn.h
> +++ b/include/net/inet_ecn.h
> @@ -73,27 +73,13 @@ static inline void INET_ECN_dontxmit(struct sock *sk)
>
> static inline int IP_ECN_set_ce(struct iphdr *iph)
> {
> - u32 check = (__force u32)iph->check;
> - u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
> -
> - /*
> - * After the last operation we have (in binary):
> - * INET_ECN_NOT_ECT => 01
> - * INET_ECN_ECT_1 => 10
> - * INET_ECN_ECT_0 => 11
> - * INET_ECN_CE => 00
> - */
I think, the above comment explains how an
increment (iph->tos + 1) serves the purpose to check
for ECT_1 and ECT_0, there is no such thing as
addressing the next byte from header. It is just an
optimized logic that avoids complex INET_ECN_is_XXX
checks.
> - if (!(ecn & 2))
> + u32 ecn = iph->tos & INET_ECN_MASK;
> +
> + if (INET_ECN_is_ce(ecn) || INET_ECN_is_not_ect(ecn))
> return !ecn;
May be return INET_ECN_is_ce(ecn) ?
>
> - /*
> - * The following gives us:
> - * INET_ECN_ECT_1 => check += htons(0xFFFD)
> - * INET_ECN_ECT_0 => check += htons(0xFFFE)
> - */
> - check += (__force u16)htons(0xFFFB) + (__force u16)htons(ecn);
> + csum_replace2(&iph->check, iph->tos, iph->tos|INET_ECN_CE);
>
> - iph->check = (__force __sum16)(check + (check>=0xFFFF));
> iph->tos |= INET_ECN_CE;
> return 1;
> }
> --
> 1.7.10.4
Regards
--
Julian Anastasov <ja@ssi.bg>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [RFC PATCH] fix IP_ECN_set_ce
2012-12-19 8:11 ` Julian Anastasov
@ 2012-12-19 8:41 ` RongQing Li
2012-12-19 8:58 ` Julian Anastasov
0 siblings, 1 reply; 7+ messages in thread
From: RongQing Li @ 2012-12-19 8:41 UTC (permalink / raw)
To: Julian Anastasov; +Cc: netdev
2012/12/19 Julian Anastasov <ja@ssi.bg>:
>
> Hello,
>
> On Wed, 19 Dec 2012, roy.qing.li@gmail.com wrote:
>
>> From: Li RongQing <roy.qing.li@gmail.com>
>>
>> 1. ECN uses the two least significant (right-most) bits of the DiffServ
>> field in the IPv4, so it should be in iph->tos, not in (iph->tos+1)
>>
>> 2. When setting CE, we should check if ECN Capable Transport supports,
>> both 10 and 01 mean ECN Capable Transport, so only check 10 is not enough
>> 00: Non ECN-Capable Transport — Non-ECT
>> 10: ECN Capable Transport — ECT(0)
>> 01: ECN Capable Transport — ECT(1)
>> 11: Congestion Encountered — CE
>>
>> 3. Remove the misunderstand comment
>>
>> 4. fix the checksum computation
>>
>> Signed-off-by: Li RongQing <roy.qing.li@gmail.com>
>> ---
>> include/net/inet_ecn.h | 22 ++++------------------
>> 1 file changed, 4 insertions(+), 18 deletions(-)
>>
>> diff --git a/include/net/inet_ecn.h b/include/net/inet_ecn.h
>> index aab7375..545a683 100644
>> --- a/include/net/inet_ecn.h
>> +++ b/include/net/inet_ecn.h
>> @@ -73,27 +73,13 @@ static inline void INET_ECN_dontxmit(struct sock *sk)
>>
>> static inline int IP_ECN_set_ce(struct iphdr *iph)
>> {
>> - u32 check = (__force u32)iph->check;
>> - u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
>> -
>> - /*
>> - * After the last operation we have (in binary):
>> - * INET_ECN_NOT_ECT => 01
>> - * INET_ECN_ECT_1 => 10
>> - * INET_ECN_ECT_0 => 11
>> - * INET_ECN_CE => 00
>> - */
>
> I think, the above comment explains how an
> increment (iph->tos + 1) serves the purpose to check
> for ECT_1 and ECT_0, there is no such thing as
> addressing the next byte from header. It is just an
> optimized logic that avoids complex INET_ECN_is_XXX
> checks.
Thanks for your reply.
Do you mean this comment are valuable?
>
>> - if (!(ecn & 2))
>> + u32 ecn = iph->tos & INET_ECN_MASK;
>> +
>> + if (INET_ECN_is_ce(ecn) || INET_ECN_is_not_ect(ecn))
>> return !ecn;
>
> May be return INET_ECN_is_ce(ecn) ?
>
I like to set the return value to void, since noone cares about the
return value.
-Roy
>>
>> - /*
>> - * The following gives us:
>> - * INET_ECN_ECT_1 => check += htons(0xFFFD)
>> - * INET_ECN_ECT_0 => check += htons(0xFFFE)
>> - */
>> - check += (__force u16)htons(0xFFFB) + (__force u16)htons(ecn);
>> + csum_replace2(&iph->check, iph->tos, iph->tos|INET_ECN_CE);
>>
>> - iph->check = (__force __sum16)(check + (check>=0xFFFF));
>> iph->tos |= INET_ECN_CE;
>> return 1;
>> }
>> --
>> 1.7.10.4
>
> Regards
>
> --
> Julian Anastasov <ja@ssi.bg>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [RFC PATCH] fix IP_ECN_set_ce
2012-12-19 8:41 ` RongQing Li
@ 2012-12-19 8:58 ` Julian Anastasov
2012-12-19 9:11 ` RongQing Li
0 siblings, 1 reply; 7+ messages in thread
From: Julian Anastasov @ 2012-12-19 8:58 UTC (permalink / raw)
To: RongQing Li; +Cc: netdev
Hello,
On Wed, 19 Dec 2012, RongQing Li wrote:
> >> static inline int IP_ECN_set_ce(struct iphdr *iph)
> >> {
> >> - u32 check = (__force u32)iph->check;
> >> - u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
> >> -
> >> - /*
> >> - * After the last operation we have (in binary):
> >> - * INET_ECN_NOT_ECT => 01
> >> - * INET_ECN_ECT_1 => 10
> >> - * INET_ECN_ECT_0 => 11
> >> - * INET_ECN_CE => 00
> >> - */
> >
> > I think, the above comment explains how an
> > increment (iph->tos + 1) serves the purpose to check
> > for ECT_1 and ECT_0, there is no such thing as
> > addressing the next byte from header. It is just an
> > optimized logic that avoids complex INET_ECN_is_XXX
> > checks.
> Thanks for your reply.
> Do you mean this comment are valuable?
It looks better to me with the comment and the
original checks. But I can't comment the correctness of
the other changes in your patch.
> >> - if (!(ecn & 2))
> >> + u32 ecn = iph->tos & INET_ECN_MASK;
> >> +
> >> + if (INET_ECN_is_ce(ecn) || INET_ECN_is_not_ect(ecn))
> >> return !ecn;
> >
> > May be return INET_ECN_is_ce(ecn) ?
> >
>
> I like to set the return value to void, since noone cares about the
> return value.
It is used by INET_ECN_set_ce and its users in
net/sched/
> -Roy
Regards
--
Julian Anastasov <ja@ssi.bg>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [RFC PATCH] fix IP_ECN_set_ce
2012-12-19 8:58 ` Julian Anastasov
@ 2012-12-19 9:11 ` RongQing Li
2012-12-19 9:31 ` David Miller
0 siblings, 1 reply; 7+ messages in thread
From: RongQing Li @ 2012-12-19 9:11 UTC (permalink / raw)
To: Julian Anastasov; +Cc: netdev
2012/12/19 Julian Anastasov <ja@ssi.bg>:
>
> Hello,
>
> On Wed, 19 Dec 2012, RongQing Li wrote:
>
>> >> static inline int IP_ECN_set_ce(struct iphdr *iph)
>> >> {
>> >> - u32 check = (__force u32)iph->check;
>> >> - u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
>> >> -
>> >> - /*
>> >> - * After the last operation we have (in binary):
>> >> - * INET_ECN_NOT_ECT => 01
>> >> - * INET_ECN_ECT_1 => 10
>> >> - * INET_ECN_ECT_0 => 11
>> >> - * INET_ECN_CE => 00
>> >> - */
>> >
>> > I think, the above comment explains how an
>> > increment (iph->tos + 1) serves the purpose to check
>> > for ECT_1 and ECT_0, there is no such thing as
>> > addressing the next byte from header. It is just an
>> > optimized logic that avoids complex INET_ECN_is_XXX
>> > checks.
>> Thanks for your reply.
>> Do you mean this comment are valuable?
>
> It looks better to me with the comment and the
> original checks. But I can't comment the correctness of
> the other changes in your patch.
I do not know how they are useful, and how the original check
works, since the value in comments are wrong, the correct is:
enum {
INET_ECN_NOT_ECT = 0,
INET_ECN_ECT_1 = 1,
INET_ECN_ECT_0 = 2,
INET_ECN_CE = 3,
INET_ECN_MASK = 3,
};
00: Non ECN-Capable Transport — Non-ECT
10: ECN Capable Transport — ECT(0)
01: ECN Capable Transport — ECT(1)
11: Congestion Encountered — CE
-Roy
>
>> >> - if (!(ecn & 2))
>> >> + u32 ecn = iph->tos & INET_ECN_MASK;
>> >> +
>> >> + if (INET_ECN_is_ce(ecn) || INET_ECN_is_not_ect(ecn))
>> >> return !ecn;
>> >
>> > May be return INET_ECN_is_ce(ecn) ?
>> >
>>
>> I like to set the return value to void, since noone cares about the
>> return value.
>
> It is used by INET_ECN_set_ce and its users in
> net/sched/
>
>> -Roy
>
> Regards
>
> --
> Julian Anastasov <ja@ssi.bg>
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [RFC PATCH] fix IP_ECN_set_ce
2012-12-19 9:11 ` RongQing Li
@ 2012-12-19 9:31 ` David Miller
0 siblings, 0 replies; 7+ messages in thread
From: David Miller @ 2012-12-19 9:31 UTC (permalink / raw)
To: roy.qing.li; +Cc: ja, netdev
From: RongQing Li <roy.qing.li@gmail.com>
Date: Wed, 19 Dec 2012 17:11:59 +0800
> 2012/12/19 Julian Anastasov <ja@ssi.bg>:
>>
>> Hello,
>>
>> On Wed, 19 Dec 2012, RongQing Li wrote:
>>
>>> >> static inline int IP_ECN_set_ce(struct iphdr *iph)
>>> >> {
>>> >> - u32 check = (__force u32)iph->check;
>>> >> - u32 ecn = (iph->tos + 1) & INET_ECN_MASK;
>>> >> -
>>> >> - /*
>>> >> - * After the last operation we have (in binary):
>>> >> - * INET_ECN_NOT_ECT => 01
>>> >> - * INET_ECN_ECT_1 => 10
>>> >> - * INET_ECN_ECT_0 => 11
>>> >> - * INET_ECN_CE => 00
>>> >> - */
>>> >
>>> > I think, the above comment explains how an
>>> > increment (iph->tos + 1) serves the purpose to check
>>> > for ECT_1 and ECT_0, there is no such thing as
>>> > addressing the next byte from header. It is just an
>>> > optimized logic that avoids complex INET_ECN_is_XXX
>>> > checks.
>>> Thanks for your reply.
>>> Do you mean this comment are valuable?
>>
>> It looks better to me with the comment and the
>> original checks. But I can't comment the correctness of
>> the other changes in your patch.
>
> I do not know how they are useful, and how the original check
> works, since the value in comments are wrong, the correct is:
>
> enum {
> INET_ECN_NOT_ECT = 0,
> INET_ECN_ECT_1 = 1,
> INET_ECN_ECT_0 = 2,
> INET_ECN_CE = 3,
> INET_ECN_MASK = 3,
> };
>
>
> 00: Non ECN-Capable Transport ― Non-ECT
> 10: ECN Capable Transport ― ECT(0)
> 01: ECN Capable Transport ― ECT(1)
> 11: Congestion Encountered ― CE
You really don't understand the comment, it is saying what
the incremented value corresponds to, ECN wise.
If iph->tos + 1 is 01, we had INET_ECN_NOT_ECT in iph->tos to
begine with, and so on an so forth.
Because you are having so much trouble with this most fundamental
aspect of this code, I have zero confidence in your being able to
make reasonable changes here.
I am not applying this patch.
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [RFC PATCH] fix IP_ECN_set_ce
2012-12-19 6:21 [RFC PATCH] fix IP_ECN_set_ce roy.qing.li
2012-12-19 8:11 ` Julian Anastasov
@ 2012-12-19 16:14 ` Eric Dumazet
1 sibling, 0 replies; 7+ messages in thread
From: Eric Dumazet @ 2012-12-19 16:14 UTC (permalink / raw)
To: roy.qing.li; +Cc: netdev
On Wed, 2012-12-19 at 14:21 +0800, roy.qing.li@gmail.com wrote:
> From: Li RongQing <roy.qing.li@gmail.com>
>
> 1. ECN uses the two least significant (right-most) bits of the DiffServ
> field in the IPv4, so it should be in iph->tos, not in (iph->tos+1)
>
> 2. When setting CE, we should check if ECN Capable Transport supports,
> both 10 and 01 mean ECN Capable Transport, so only check 10 is not enough
> 00: Non ECN-Capable Transport — Non-ECT
> 10: ECN Capable Transport — ECT(0)
> 01: ECN Capable Transport — ECT(1)
> 11: Congestion Encountered — CE
>
> 3. Remove the misunderstand comment
>
> 4. fix the checksum computation
>
> Signed-off-by: Li RongQing <roy.qing.li@gmail.com>
This is total crap.
Its perfectly clear to me and compiler generates fast code.
If you don't understand this code, please don't touch it.
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2012-12-19 16:14 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2012-12-19 6:21 [RFC PATCH] fix IP_ECN_set_ce roy.qing.li
2012-12-19 8:11 ` Julian Anastasov
2012-12-19 8:41 ` RongQing Li
2012-12-19 8:58 ` Julian Anastasov
2012-12-19 9:11 ` RongQing Li
2012-12-19 9:31 ` David Miller
2012-12-19 16:14 ` Eric Dumazet
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.