All of lore.kernel.org
 help / color / mirror / Atom feed
From: Hangbin Liu <hangbin.liu@linux.dev>
To: Andrea Mayer <andrea.mayer@uniroma2.it>
Cc: "David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@kernel.org>,
	Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
	Simon Horman <horms@kernel.org>, Shuah Khan <shuah@kernel.org>,
	netdev@vger.kernel.org, linux-kernel@vger.kernel.org,
	linux-kselftest@vger.kernel.org,
	Hangbin Liu <liuhangbin@kylinos.cn>,
	Stefano Salsano <stefano.salsano@uniroma2.it>,
	Justin Iurman <justin.iurman@gmail.com>,
	Andrea Mayer <andrea@common-net.org>
Subject: Re: [PATCH net-next v3 5/5] selftests: seg6: add PSP flavor testing for SRv6 End.X
Date: Tue, 17 Nov 2026 16:48:12 +0800	[thread overview]
Message-ID: <avwUzLyWxZl9cBIX@fedora> (raw)
In-Reply-To: <20261117035021.e724668d726ecce54e091288@uniroma2.it>

Hi Andrea,
On Tue, Nov 17, 2026 at 03:50:21AM +0100, Andrea Mayer wrote:
> On Thu, 12 Nov 2026 17:28:00 +0800
> Hangbin Liu <hangbin.liu@linux.dev> wrote:
> 
> > Add an End.X flavor scenario to srv6_end_flavors_test.sh that applies
> > the same PSP policies as the End scenario, alone and combined with
> > NEXT-C-SID, with every node bound through its adjacency (e.g. 4>2 for
> > rt-4 towards rt-2, 2>h towards the host), and exercise it with the
> > existing host connectivity tests.
> >
> > Teach the policy and endpoint helpers about <nodeid:op:act> elements:
> > __get_srv6_rtcfg_act() extracts the action, the __get_*_func()
> > helpers pick the Function for the action, __get_act_cfg() derives the
> > End.X next hop, and build_csid()/get_rt_sid() keep the adjacency so
> > it is encoded into the C-SIDs (e.g. fcbb:0:34a9:: is End.X on rt-3
> > towards rt-4).
> >
> > There is no need to re-check the ip command capability as it already
> > supports setting PSP flavor with End.X.
> 
> Hi Hangbin,
> thanks for adding the End.X scenario.
> 
> I see these main areas to work on:
> - How an element is read. The "router>adjacency" form is read in
>   different places, each with its own sed or cut. IMO, the selftest
>   would be easier to extend to other behaviors and flavor combinations
>   if the nodes were read in one place, with small helpers. What
>   depends on the behavior could also be chosen in one place.
> - A check of the elements, so that a mistake in a Policy does not go
>   unnoticed.
> - A check of the kernel support for End.X PSP.
> - The End.X tests with a consumed SRH.
> - One End.X SID per adjacency also without NEXT-C-SID, where today
>   the SIDs of a router do not depend on the adjacency.
> 
> The comments below are ideas on how to address each of them, plus
> some smaller points. There may also be other parts of the selftest
> to adapt.

Thanks for your kind review.

> > Signed-off-by: Hangbin Liu <liuhangbin@kylinos.cn>
> > ---
> > I didn't add `host_srv6_end_flv_consumed_srh_tests()` for End.X because
> > the extra policy insertion does not work with the End.X policy. The packet
> > resolves the destination via the next hop directly, rather than doing a
> > re-lookup in table 90.
> 
> The ingress router can insert the extra SRH instead, on the first SID
> of the Policy. As in the End tests, an End SID of the base setup can
> then consume it. This way, we can add these tests for End.X too. What
> do you think?

For the End.X Policy (i.a) testing, we have SID List=fcff:3::a,fcff:4::af1,fcff:2::af1,cafe::2.
And we configured the next hop policy for the SID on each rt. e.g

On rt3: fcff:3::000a table 90 encap seg6local action End.X nh6 fcf0:0:3:4::4
On rt4: fcff:4::0af1 table 90 encap seg6local action End.X nh6 fcf0:0:2:4::2 flavors psp
on rt2: fcff:2::0af1 table 90 encap seg6local action End.X nh6 cafe::2 flavors psp

When the packet arrive on rt4, it's dst is set to fcff:2::af1. And the new
insert policy (fcff:2::0af1 table 90 encap seg6 mode inline segs fcff:3::000e dev dum0)
doesn't works as the End.X policy will goto next hop fcf0:0:2:4::2 directly
instead of looking at fcff:2::af1 on table 90.

Would you add more info about the ingress route?

> 
> > [snip]
> > diff --git a/tools/testing/selftests/net/srv6_end_flavors_test.sh b/tools/testing/selftests/net/srv6_end_flavors_test.sh
> > index 374b43661c28..5b1b13586df0 100755
> > --- a/tools/testing/selftests/net/srv6_end_flavors_test.sh
> > +++ b/tools/testing/selftests/net/srv6_end_flavors_test.sh
> > [snip]
> >  __get_srv6_rtcfg_id()
> >  {
> >  	local element="$1"
> >  
> > +	echo "${element}" | sed 's/:.*//; s/>[1234h]//g'
> > +}
> > [snip]
> > +__get_srv6_rtcfg_all_id()
> > +{
> > +	local element="$1"
> > +
> >  	echo "${element}" | cut -d':' -f1
> >  }
> >  
> 
> The same node is read again in build_csid(), get_rt_sid() and
> setup_rt_endpoints(), each time in its own way. Two small helpers,
> one returning the router of a node and one its adjacency, could
> replace these sed and cut. __get_srv6_rtcfg_id() above would then not
> need the sed, and __get_srv6_rtcfg_all_id() would probably not be
> needed at all. Without NEXT-C-SID, __setup_rt_policy() could then get
> each SID of the Policy from get_rt_sid(), as setup_rt_endpoints()
> does.

For a single End.X node, e.g. "4>2", we can return the router (4) and adjacent
router (2). But how to get multi nodes. e.g. "4>2,2>3", with theses 2 small
helpers?

> 
> > [snip]
> > +__get_noflv_func()
> > +{
> > +	local act="$1"
> > +
> > +	case "${act}" in
> > +	"End")
> > +		echo "${END_FUNC}" ;;
> > +	"End.X")
> > +		echo "${ENDX_FUNC}" ;;
> > +	esac
> > +}
> > +
> > +__get_psp_func()
> > +{
> > +	local act="$1"
> > +
> > +	case "${act}" in
> > +	"End")
> > +		echo "${END_PSP_FUNC}" ;;
> > +	"End.X")
> > +		echo "${ENDX_PSP_FUNC}" ;;
> > +	esac
> > +}
> > +
> > +__get_csid_psp_func()
> > +{
> > +	local act="$1"
> > +
> > +	case "${act}" in
> > +	"End")
> > +		echo "${END_CSID_PSP_FUNC}" ;;
> > +	"End.X")
> > +		echo "${ENDX_CSID_PSP_FUNC}" ;;
> > +	esac
> > +}
> 
> The three __get_*_func() helpers differ only in the constants. As a
> possible simplification, they could be one function with, for example,
> a case on the behavior and the operation ("End noflv", "End psp", ...,
> "End.X next-csid,psp"). Then the Functions are chosen in one place,
> also for a new behavior.

Thanks, I will merge them together.

> 
> > [snip]
> > @@ -713,10 +814,10 @@ op_has_next_csid()
> >  #  $1 - router
> >  #  $2 - operation <op> of the element (e.g. "next-csid,psp")
> >  #  $3 - Function (e.g. "0ef1", or "e9" with NEXT-C-SID)
> > -# e.g. input: "4" "psp" "0ef1"
> > -#      output: "fcff:4::0ef1"
> > -#      input: "4" "next-csid,psp" "e9"
> > -#      output: "fcbb:0:40e9::/48"
> > +# e.g. input: "4" "psp" "0ef1", output: "fcff:4::0ef1"
> > +#      input: "4>2" "psp" "0af1", output: "fcff:4::0af1"
> > +#      input: "4" "next-csid,psp" "e9", output: "fcbb:0:40e9::/48"
> > +#      input: "4>2" "next-csid,psp" "a9", output: "fcbb:0:42a9::/48"
> >  get_rt_sid()
> >  {
> >  	local rt="$1"
> > @@ -730,6 +831,7 @@ get_rt_sid()
> >  		return
> >  	fi
> >  
> > +	rt=$(echo "${rt}" | sed 's/>.*//')
> >  	echo "${LOCATOR_SERVICE}:${rt}::${func}"
> >  }
> 
> Without NEXT-C-SID, get_rt_sid() drops the adjacency, as the Sashiko
> review also notes: "4>2:psp:End.X" and "4>3:psp:End.X" would both get
> fcff:4::0af1. I think each adjacency should have its own SID. The
> SID could carry the adjacency, as the C-SID already does. Then
> __setup_rt_policy() needs the new SID. Calling get_rt_sid() there,
> as suggested above, would do that.

Hmm, do you want to set the adjacency like xaf1 or xxa1?
e.g. ffcf:4::2af1 for 4>2:psp:End.X, ffcf:4::3af1 for 4>3:psp:End.X
or
ffcf:4::42a1 for 4>2:psp:End.X, ffcf:4::43a1 for 4>3:psp:End.X?

The second one may also need to adjust the End address. e.g.
fcff:x::ef1 -> fcff:x::e1

> 
> > [snip]
> > @@ -765,26 +867,34 @@ setup_rt_endpoint()
> >  
> >  # Add an SRv6 Endpoint behavior to each router of a Policy element <nodeid:op>.
> >  # args:
> > -#  $1 - <nodeid> of the element: one router, or a comma-separated list
> > +#  $1 - <nodeid:op:act> element
> >  #  $2 - operation <op> of the element (e.g. "next-csid,psp")
> >  #  $3 - Function (e.g. "0ef1", or "e9" with NEXT-C-SID)
> > -#  $4 - SRv6 Endpoint behavior and flavors (e.g. "End flavors psp")
> > -# e.g. input: "4,2" "next-csid,psp" "e9"
> > -#             "End flavors next-csid,psp lblen 32 nflen 16"
> > +#  $4 - flavor configuration of the behavior, without the action
> > +#       (e.g. "flavors psp")
> > +# e.g. input: "4,2:next-csid,psp" "next-csid,psp" "e9"
> > +#             "flavors next-csid,psp lblen 32 nflen 16"
> >  #      result: the behavior is added on rt-4, bound to fcbb:0:40e9::/48, and
> >  #              on rt-2, bound to fcbb:0:20e9::/48
> >  setup_rt_endpoints()
> >  {
> > -	local node="$1"
> > +	local nodeid="$1"
> >  	local op_type="$2"
> >  	local func="$3"
> >  	local behavior_cfg="$4"
> > +	local nodes node next_node
> > +	local act act_cfg
> >  	local sid
> >  	local rt
> >  
> > -	for rt in $(echo "${node}" | sed 's/,/ /g'); do
> > -		sid="$(get_rt_sid "${rt}" "${op_type}" "${func}")"
> > -		setup_rt_endpoint "${rt}" "${sid}" "${behavior_cfg}"
> > +	nodes="$(__get_srv6_rtcfg_all_id "${nodeid}")"
> > +	act="$(__get_srv6_rtcfg_act "${nodeid}")"
> > +	for node in $(echo "${nodes}" | sed 's/,/ /g'); do
> > +		rt=$(echo "${node}" | cut -d'>' -f1)
> > +		next_node=$(echo "${node}" | cut -d'>' -f2)
> > +		act_cfg="$(__get_act_cfg "${act}" "${rt}" "${next_node}")"
> > +		sid="$(get_rt_sid "${node}" "${op_type}" "${func}")"
> > +		setup_rt_endpoint "${rt}" "${sid}" "${act_cfg} ${behavior_cfg}"
> >  	done
> >  }
> 
> setup_rt_endpoints() now gets the whole element, and reads it again
> with the same helpers as __setup_rt_policy(). I think
> __setup_rt_policy() can read the element once, and pass only what
> setup_rt_endpoints() needs, for example the nodes and the behavior.
> 
> Nit: I would declare one local per line in the new code, as in the
> rest of the selftest.
> 
> > [snip]
> > @@ -824,7 +934,8 @@ setup_rt_host_proxy()
> >  #       cafe::x:yy on the loopback of the host (e.g. 2:e9)
> >  #  $2 - SRv6 router configured for enforcing the SRv6 Policy
> >  #  $3 - compact way to represent a list of SRv6 routers with their operations
> > -#       (i.e. behaviors) that each of them needs to perform. Every <nodeid:op>
> > +#       (i.e. behaviors) that each of them needs to perform. Every
> > +#       <nodeid:op:act>
> >  #       element constructs a SID that is associated with the behavior <op> on
> >  #       the <nodeid> node. The list of such elements forms an SRv6 Policy.
> >  #       For "next-csid,psp", <nodeid> is a comma-separated list of node IDs.
> > @@ -841,34 +952,38 @@ __setup_rt_policy()
> >  	local container
> >  	local function
> >  	local op_type
> > +	local nodes
> >  	local node
> > +	local act
> >  	local n
> >  
> >  	in_nsname="$(get_rtname "${encap_rt}")"
> >  
> >  	for n in ${policy_rts}; do
> >  		node="$(__get_srv6_rtcfg_id "${n}")"
> > +		nodes="$(__get_srv6_rtcfg_all_id "${n}")"
> >  		op_type="$(__get_srv6_rtcfg_op "${n}")"
> > +		act="$(__get_srv6_rtcfg_act "${n}")"
> 
> I would also describe the new syntax in the comment of
> __setup_rt_policy() above: the ">" with the adjacency, "h" for the
> host, and the ":End.X" field.
> 
> A mistake in an element can also go unnoticed. With "1>h:psp",
> rt-1 gets End instead of End.X, and the End.X tests still pass. Could
> a check at the top of this loop make the test fail on such an element?
> The check could also refuse an End.X element without an adjacency,
> e.g. "4:psp:End.X".
> On a setup error, cleanup() reports SKIP, unless the check sets ret
> to ksft_fail first.

Thanks, I will add a check for this.

> 
> > [snip]
> > @@ -1479,4 +1623,12 @@ host_srv6_end_flv_next_csid_psp_tests
> >  host_srv6_end_flv_consumed_srh_tests
> >  defer_scope_pop
> >  
> > +defer_scope_push
> > +set -e
> > +setup_endx_flv_psp
> > +set +e
> > +host_srv6_end_flv_psp_tests "End.X"
> > +host_srv6_end_flv_next_csid_psp_tests "End.X"
> > +defer_scope_pop
> > +
> >  print_log_test_results
> 
> On a kernel with End PSP but without End.X PSP, the whole script ends
> as SKIP, after the End tests passed. This can happen without any
> backport, since the selftests from mainline can also be run on older
> stable kernels. The selftest already probes End PSP with
> test_kernel_supp_or_ksft_skip(), for kernels without it. We could
> probe End.X PSP first in the same way, with an nh6, but returning
> instead of exiting. If each scenario is moved into its own function,
> the End.X one can return early, and only that scenario is skipped.
> 
> Sashiko asks whether the End.X tests would also pass if End.X worked
> as End. The "1>h:psp" case above suggests that they would. I think it
> is worth a look.

Also a check for this.

Thanks
Hangbin

  reply	other threads:[~2026-11-17  8:48 UTC|newest]

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-12  9:27 [PATCH net-next v3 0/5] seg6: add End.X PSP flavor support and selftests Hangbin Liu
2026-11-12  9:27 ` [PATCH net-next v3 1/5] seg6: split final End process out of flavor processing Hangbin Liu
2026-11-12  9:27 ` [PATCH net-next v3 2/5] seg6: add RFC8986 flavor support for End.X Hangbin Liu
2026-11-12  9:27 ` [PATCH net-next v3 3/5] selftests: seg6: move SRv6 End PSP setup into its own helper Hangbin Liu
2026-11-16  4:26   ` Andrea Mayer
2026-11-16  6:20     ` Hangbin Liu
2026-11-12  9:27 ` [PATCH net-next v3 4/5] selftest: seg6: explicit cleanup for SRv6 End PSP test Hangbin Liu
2026-11-16  4:34   ` Andrea Mayer
2026-11-16  6:27     ` Hangbin Liu
2026-11-16 12:29   ` netdev-bot+sashiko
2026-11-12  9:28 ` [PATCH net-next v3 5/5] selftests: seg6: add PSP flavor testing for SRv6 End.X Hangbin Liu
2026-11-16 12:29   ` netdev-bot+sashiko
2026-11-17  2:50   ` Andrea Mayer
2026-11-17  8:48     ` Hangbin Liu [this message]
2026-11-18  0:51       ` Hangbin Liu
2026-11-18  2:57         ` Andrea Mayer
2026-11-18  2:50       ` Andrea Mayer
2026-11-18  7:06         ` Hangbin Liu

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=avwUzLyWxZl9cBIX@fedora \
    --to=hangbin.liu@linux.dev \
    --cc=andrea.mayer@uniroma2.it \
    --cc=andrea@common-net.org \
    --cc=davem@davemloft.net \
    --cc=edumazet@kernel.org \
    --cc=horms@kernel.org \
    --cc=justin.iurman@gmail.com \
    --cc=kuba@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-kselftest@vger.kernel.org \
    --cc=liuhangbin@kylinos.cn \
    --cc=netdev@vger.kernel.org \
    --cc=pabeni@redhat.com \
    --cc=shuah@kernel.org \
    --cc=stefano.salsano@uniroma2.it \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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.