答复: 答复: 答复: 答复: [vpp-dev] about sctp
Hi Marco，
Thanks for your detail explanation.
1. I just saw the function ip4_sctp_compute_checksum(). The comment /*
Initialize checksum with ip header. */ seems that you include the IP header to
calculate the checksum. And also in this function, you call
ip_incremental_checksum(), so it seems that the algorithm not like crc32c.
2. Maybe I was wrong. But I didn't find any structure to save the cookie in the
sctp_state_cookie_param_t. In the function sctp_handle_init_ack(), you save it
like this:
case SCTP_STATE_COOKIE_TYPE:
{
sctp_state_cookie_param_t *state_cookie_param =
(sctp_state_cookie_param_t *) opt_params_hdr;
clib_memcpy (&(sctp_conn->cookie_param), state_cookie_param,
sizeof (sctp_state_cookie_param_t));
break;
}
And then in the function sctp_prepare_cookie_echo_chunk(), just copy it like
this:
clib_memcpy (&(cookie_echo_chunk->cookie), _conn->cookie_param,
sizeof (sctp_state_cookie_param_t));
Maybe I miss something, but indeed in the COOKIE_ECHO chunk from the standard,
it's only COOKIE, not optional state cookie parameter, i.e. there is no
Parameter type and Parameter length in the COOKIE_ECHO chunk.
Best regards
刘道林 (Daolin Liu)
T大连市共进科技有限公司
DALIAN GONGJIN TECHNOLOGY CO.,LTD
中国大连市高新园区软件园路1A-4-24层
Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China
直线(TEL)：(86-411)39996705 分机(EXT)：76824
手机(Mobile)：(86)13704090959
-邮件原件-
发件人: Marco Varlese [mailto:mvarl...@suse.de]
发送时间: 2018年11月23日 20:39
收件人: Liu Daolin (刘道林); vpp-dev@lists.fd.io
主题: Re: 答复: 答复: 答复: [vpp-dev] about sctp
On Fri, 2018-11-23 at 09:14 +, Liu Daolin (刘道林) wrote:
> Hi Marco，
>
> 1. If you indeed use crc32c, it's ok. But from my view, it should not include
> ip header.
And it does not!
>
> 2. the definition is below:
>
> #define SHA1_OUTPUT_LENGTH 20
> /*
> * 0 1 2 3
> * 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
> * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
> * | Parameter Type | Parameter Length|
> * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
> * \ \
> * / Parameter Value /
> * \ \
> * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
> */
> typedef struct
> {
> sctp_opt_params_hdr_t param_hdr;
>
> unsigned char mac[SHA1_OUTPUT_LENGTH];/* RFC 2104 */
> u64 creation_time;
> u32 cookie_lifespan;
>
> } sctp_state_cookie_param_t;
>
> The macro definition SHA1_OUTPUT_LENGTH 20 seems not enough.
In your earlier email you mentioned COOKIE LENGTH which is _NOT_ what you're
referring to now. The SHA1_OUTPUT_LENGTH has nothing to do with the COOKIE
LENGTH but the length of the hash calculated via a SHA1 algorithm!
Indeed the macro is called SHA1_OUTPUT_LENGTH which it's pretty self
explanatory.
I have no clue why you believe it is "not enough", maybe you were thinking about
the COOKIE LENGTH? I have no idea...
As you can see from the code there is a reference to an RFC (2104): if you spent
2 minutes looking that up you would see that it states (section 2, page 2, line
6): "16 bytes for MD5 hash output and 20 bytes for SHA-1 hash output".
Further, the RFC 2104 is referenced by the SCTP (RFC 4960) at section 5.1.3
where the topic State Cookie is discerned.
Having said all that, I won't go into any deeper details about the RFC/protocol
over the ML; you can read that up.
If you identify anything wrong with the implementation (which doesn't match the
RFC) please send a patch via gerrit with RFC references.
>
>
>
>
> Best regards
> 刘道林 (Daolin Liu)
> T大连市共进科技有限公司
> DALIAN GONGJIN TECHNOLOGY CO.,LTD
> 中国大连市高新园区软件园路1A-4-24层
> Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China
> 直线(TEL)：(86-411)39996705 分机(EXT)：76824
> 手机(Mobile)：(86)13704090959
>
>
> -邮件原件-
> 发件人: Marco Varlese [mailto:mvarl...@suse.de]
> 发送时间: 2018年11月23日 17:07
> 收件人: Liu Daolin (刘道林); vpp-dev@lists.fd.io
> 主题: Re: 答复: 答复: [vpp-dev] about sctp
>
>
> On Fri, 2018-11-23 at 00:57 +, Liu Daolin (刘道林) wrote:
> > Hi Macro,
> >
> > OK, I will try it when I have time.
> Well, as you wish.
>
> > I'm reading the sctp source code and found some issues. The most critical
> > one
> > is the checksum. The sctp checksum should not include the ip header, just
> > only
> > sctp data, and also it should use crc32c, not the current co
答复: 答复: 答复: [vpp-dev] about sctp
Hi Marco，
1. If you indeed use crc32c, it's ok. But from my view, it should not include
ip header.
2. the definition is below:
#define SHA1_OUTPUT_LENGTH 20
/*
* 0 1 2 3
* 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
* | Parameter Type | Parameter Length|
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
* \ \
* / Parameter Value /
* \ \
* +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
*/
typedef struct
{
sctp_opt_params_hdr_t param_hdr;
unsigned char mac[SHA1_OUTPUT_LENGTH];/* RFC 2104 */
u64 creation_time;
u32 cookie_lifespan;
} sctp_state_cookie_param_t;
The macro definition SHA1_OUTPUT_LENGTH 20 seems not enough.
Best regards
刘道林 (Daolin Liu)
T大连市共进科技有限公司
DALIAN GONGJIN TECHNOLOGY CO.,LTD
中国大连市高新园区软件园路1A-4-24层
Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China
直线(TEL)：(86-411)39996705 分机(EXT)：76824
手机(Mobile)：(86)13704090959
-邮件原件-
发件人: Marco Varlese [mailto:mvarl...@suse.de]
发送时间: 2018年11月23日 17:07
收件人: Liu Daolin (刘道林); vpp-dev@lists.fd.io
主题: Re: 答复: 答复: [vpp-dev] about sctp
On Fri, 2018-11-23 at 00:57 +, Liu Daolin (刘道林) wrote:
> Hi Macro,
>
> OK, I will try it when I have time.
Well, as you wish.
>
> I'm reading the sctp source code and found some issues. The most critical one
> is the checksum. The sctp checksum should not include the ip header, just only
> sctp data, and also it should use crc32c, not the current code shown.
I am not sure what you refer to with "code shown".
However the crc calculated is indeed the one you mention and it's using SSE4
instructions which are exactly using the polynomial depicted by RFC 4960.
> Another important issue is the max cookie length is 20 from the macro
> definition. Actually it should not be a fixed value (usually more than 200).
So I cannot find a definition for the actual length of the cookie.
It would help if you could point to the actual filename / definition when
reporting issues.
>
>
> Best regards
> 刘道林 (Daolin Liu)
> T大连市共进科技有限公司
> DALIAN GONGJIN TECHNOLOGY CO.,LTD
> 中国大连市高新园区软件园路1A-4-24层
> Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China
> 直线(TEL)：(86-411)39996705 分机(EXT)：76824
> 手机(Mobile)：(86)13704090959
>
> 发件人: Marco Varlese [mailto:mvarl...@suse.de]
> 发送时间: 2018年11月22日 20:52
> 收件人: Liu Daolin (刘道林); vpp-dev@lists.fd.io
> 主题: Re: 答复: [vpp-dev] about sctp
>
> Dear Liu,
>
> The reason why I asked you to take a look at the "test_sctp.py" is to see how
> things are setup and execute the exact steps in the CLI. You'll see things
> work.
>
> You cannot remove the MTU code because that's used to initialize the PMTU and
> calculate initially the MSS to be sent to the other peer.
>
>
> Cheers,
> Marco
>
> On Wed, 2018-11-21 at 01:50 +, Liu Daolin (刘道林) wrote:
> > Hi Marco,
> >
> > Thanks for your reply.
> >
> > Actually I'm trying to read the sctp code recently to want to know how it
> > works. My requirement is very simply. I need to run sctp server based on vpp
> > on one VM, and run sctp client without vpp (I have this now) on the other
> > VM. As my below email, there will be aborted (core dumped), and I found the
> > place yesterday, but I don’t know the root reason. You can see below
> > picture:
> >
> >
> >
> > The sw_if_index is -1, so crashed when get mtu. Maybe I miss something? When
> > I run actually is CLI command from echo_server.c. As you mentioned, if I run
> > test_sctp.py, it will work fine? My requirement is only C language as it
> > will merged some other code, so I don’t like to run it with python.
> >
> > Yesterday, I also try to remove this mtu code, and the crash was disappeared
> > although the sw_if_index is still -1, and then I start sctp client on the
> > other VM to send INIT, but the server with vpp seemed be dead (no crash but
> > cannot enter any command, seemed dead loop).
> >
> > Anyway, I think it's better to try CLI "test echo server" in your side and
> > find and fix all issues visually.
> >
> >
> >
> >
> >
> > Best regards
> > 刘道林 (Daolin Liu)
> > T大连市共进科技有限公司
> > DALIAN GONGJIN TECHNOLOGY CO.,LTD
> > 中国大连市高新园区软件园路1A-4-24层
> > Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China
> > 直线(TEL)：(86-411)39996705 分机(
答复: 答复: [vpp-dev] about sctp
Hi Macro, OK, I will try it when I have time. I'm reading the sctp source code and found some issues. The most critical one is the checksum. The sctp checksum should not include the ip header, just only sctp data, and also it should use crc32c, not the current code shown. Another important issue is the max cookie length is 20 from the macro definition. Actually it should not be a fixed value (usually more than 200). Best regards 刘道林 (Daolin Liu) T大连市共进科技有限公司 DALIAN GONGJIN TECHNOLOGY CO.,LTD 中国大连市高新园区软件园路1A-4-24层 Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China 直线(TEL)：(86-411)39996705 分机(EXT)：76824 手机(Mobile)：(86)13704090959 发件人: Marco Varlese [mailto:mvarl...@suse.de] 发送时间: 2018年11月22日 20:52 收件人: Liu Daolin (刘道林); vpp-dev@lists.fd.io 主题: Re: 答复: [vpp-dev] about sctp Dear Liu, The reason why I asked you to take a look at the "test_sctp.py" is to see how things are setup and execute the exact steps in the CLI. You'll see things work. You cannot remove the MTU code because that's used to initialize the PMTU and calculate initially the MSS to be sent to the other peer. Cheers, Marco On Wed, 2018-11-21 at 01:50 +, Liu Daolin (刘道林) wrote: Hi Marco, Thanks for your reply. Actually I'm trying to read the sctp code recently to want to know how it works. My requirement is very simply. I need to run sctp server based on vpp on one VM, and run sctp client without vpp (I have this now) on the other VM. As my below email, there will be aborted (core dumped), and I found the place yesterday, but I don’t know the root reason. You can see below picture: [cid:image001.jpg@01D4830A.123FCDF0] The sw_if_index is -1, so crashed when get mtu. Maybe I miss something? When I run actually is CLI command from echo_server.c. As you mentioned, if I run test_sctp.py, it will work fine? My requirement is only C language as it will merged some other code, so I don’t like to run it with python. Yesterday, I also try to remove this mtu code, and the crash was disappeared although the sw_if_index is still -1, and then I start sctp client on the other VM to send INIT, but the server with vpp seemed be dead (no crash but cannot enter any command, seemed dead loop). Anyway, I think it's better to try CLI "test echo server" in your side and find and fix all issues visually. Best regards 刘道林 (Daolin Liu) T大连市共进科技有限公司 DALIAN GONGJIN TECHNOLOGY CO.,LTD 中国大连市高新园区软件园路1A-4-24层 Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China 直线(TEL)：(86-411)39996705 分机(EXT)：76824 手机(Mobile)：(86)13704090959 -邮件原件- 发件人: Marco Varlese [mailto:mvarl...@suse.de] 发送时间: 2018年11月21日 1:06 收件人: Liu Daolin (刘道林); vpp-dev@lists.fd.io 主题: Re: [vpp-dev] about sctp Hi, On Tue, 2018-11-20 at 02:50 +, Liu Daolin (刘道林) wrote: > Hi, > > I encountered below Aborted (core dumped) issue: > > > > When I run "test echo server uri", it's ok for tcp, but crashed immediately > for sctp. > > Please try this in your side and give me advice. I use 18.10. Thanks! Can you please try to run "make test TEST=test_sctp"? > > > Best regards > 刘道林 (Daolin Liu) > T大连市共进科技有限公司 > DALIAN GONGJIN TECHNOLOGY CO.,LTD > 中国大连市高新园区软件园路1A-4-24层 > Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China > 直线(TEL)：(86-411)39996705 分机(EXT)：76824 > 手机(Mobile)：(86)13704090959 > > 发件人: Liu Daolin (刘道林) > 发送时间: 2018年11月19日 19:23 > 收件人: 'vpp-dev@lists.fd.io' > 主题: about sctp > > Hi, > > I'd like to know some information about sctp. > > Is this fully functional? Or just partly? There are missing pieces to the SCTP implementations. > Actually, I want to try simply with CLI to verify sctp basic functions now. > But it seems that there is no CLI, and the binary APIs are also imperfect. What do you actually mean by imperfect? Any input (e.g. patch submission) would be greatly appreciated! > > Do you have any plan in the next release? Including the sample test. I am planning to create some JIRA ticket(s) so that people can see what's missing and contribute if they like. Would you be interested? > > Best regards > 刘道林 (Daolin Liu) > T大连市共进科技有限公司 > DALIAN GONGJIN TECHNOLOGY CO.,LTD > 中国大连市高新园区软件园路1A-4-24层 > Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China > 直线(TEL)：(86-411)39996705 分机(EXT)：76824 > 手机(Mobile)：(86)13704090959 > > 本电子邮件（包括任何的附件）为本公司保密文件。本文件仅仅可为以上指定的收件人或公司使用，如果阁下非电子邮件所指定之收件人，那么阁下对该邮件部分或全部的泄漏、 > 阅览、复印、变更、散布或对邮件内容的使用都是被严格禁止的。如果阁下接收了该错误传送的电子邮件，敬请阁下通过回复该邮件的方式立即通知寄件人，同时删除你所接收到 > 的文本。 This e-mail may contain confidential and/or privileged information. If > you are not the intended recipient (or have received this e-mail in error) > please notify the sender immediately and destroy this e-mail. Any unauthorized > copying, disclosure or distribution of the material in this e-mail is strictly > forbidden. > -=-=-=-=-=-=-=-=-=-=-=- > Links: You receive all
Re: 答复: 答复: [vpp-dev] about sctp
On Fri, 2018-11-23 at 00:57 +, Liu Daolin (刘道林) wrote: > Hi Macro, > > OK, I will try it when I have time. Well, as you wish. > > I'm reading the sctp source code and found some issues. The most critical one > is the checksum. The sctp checksum should not include the ip header, just only > sctp data, and also it should use crc32c, not the current code shown. I am not sure what you refer to with "code shown". However the crc calculated is indeed the one you mention and it's using SSE4 instructions which are exactly using the polynomial depicted by RFC 4960. > Another important issue is the max cookie length is 20 from the macro > definition. Actually it should not be a fixed value (usually more than 200). So I cannot find a definition for the actual length of the cookie. It would help if you could point to the actual filename / definition when reporting issues. > > > Best regards > 刘道林 (Daolin Liu) > T大连市共进科技有限公司 > DALIAN GONGJIN TECHNOLOGY CO.,LTD > 中国大连市高新园区软件园路1A-4-24层 > Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China > 直线(TEL)：(86-411)39996705 分机(EXT)：76824 > 手机(Mobile)：(86)13704090959 > > 发件人: Marco Varlese [mailto:mvarl...@suse.de] > 发送时间: 2018年11月22日 20:52 > 收件人: Liu Daolin (刘道林); vpp-dev@lists.fd.io > 主题: Re: 答复: [vpp-dev] about sctp > > Dear Liu, > > The reason why I asked you to take a look at the "test_sctp.py" is to see how > things are setup and execute the exact steps in the CLI. You'll see things > work. > > You cannot remove the MTU code because that's used to initialize the PMTU and > calculate initially the MSS to be sent to the other peer. > > > Cheers, > Marco > > On Wed, 2018-11-21 at 01:50 +, Liu Daolin (刘道林) wrote: > > Hi Marco, > > > > Thanks for your reply. > > > > Actually I'm trying to read the sctp code recently to want to know how it > > works. My requirement is very simply. I need to run sctp server based on vpp > > on one VM, and run sctp client without vpp (I have this now) on the other > > VM. As my below email, there will be aborted (core dumped), and I found the > > place yesterday, but I don’t know the root reason. You can see below > > picture: > > > > > > > > The sw_if_index is -1, so crashed when get mtu. Maybe I miss something? When > > I run actually is CLI command from echo_server.c. As you mentioned, if I run > > test_sctp.py, it will work fine? My requirement is only C language as it > > will merged some other code, so I don’t like to run it with python. > > > > Yesterday, I also try to remove this mtu code, and the crash was disappeared > > although the sw_if_index is still -1, and then I start sctp client on the > > other VM to send INIT, but the server with vpp seemed be dead (no crash but > > cannot enter any command, seemed dead loop). > > > > Anyway, I think it's better to try CLI "test echo server" in your side and > > find and fix all issues visually. > > > > > > > > > > > > Best regards > > 刘道林 (Daolin Liu) > > T大连市共进科技有限公司 > > DALIAN GONGJIN TECHNOLOGY CO.,LTD > > 中国大连市高新园区软件园路1A-4-24层 > > Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China > > 直线(TEL)：(86-411)39996705 分机(EXT)：76824 > > 手机(Mobile)：(86)13704090959 > > > > -邮件原件- > > 发件人: Marco Varlese [mailto:mvarl...@suse.de] > > 发送时间: 2018年11月21日 1:06 > > 收件人: Liu Daolin (刘道林); vpp-dev@lists.fd.io > > 主题: Re: [vpp-dev] about sctp > > > > Hi, > > > > On Tue, 2018-11-20 at 02:50 +, Liu Daolin (刘道林) wrote: > > > Hi, > > > > > > I encountered below Aborted (core dumped) issue: > > > > > > > > > > > > When I run "test echo server uri", it's ok for tcp, but crashed > > immediately > > > for sctp. > > > > > > Please try this in your side and give me advice. I use 18.10. Thanks! > > Can you please try to run "make test TEST=test_sctp"? > > > > > > > > > Best regards > > > 刘道林 (Daolin Liu) > > > T大连市共进科技有限公司 > > > DALIAN GONGJIN TECHNOLOGY CO.,LTD > > > 中国大连市高新园区软件园路1A-4-24层 > > > Floor 24th, 1A-4 Software Park Road, Hi-tech Zone, Dalian, Liaoning, China > > > 直线(TEL)：(86-411)39996705 分机(EXT)：76824 > > > 手机(Mobile)：(86)13704090959 > > > > > > 发件人: Liu Daolin (刘道林) > > > 发送时间: 2018年11月19日 19:23 > > > 收件人: 'vpp-dev@lists.fd.io' > > > 主题: about sctp > > > > > > Hi, > > > > > > I'd like to know some information about sctp. > > > > > > Is this fully functional? Or just partly? > > There are missing pieces to the SCTP implementations. > > > > > Actually, I want to try simply with CLI to verify sctp basic functions > > now. > > > But it seems that there is no CLI, and the binary APIs are also imperfect. > > What do you actually mean by imperfect? Any input (e.g. patch submission) > > would > > be greatly appreciated! > > > > > > Do you have any plan in the next release? Including the sample test. > > I am planning to create some JIRA ticket(s) so that people can see what's > > missing and contribute if they like. Would you be interested? > > > > > > > > Best
