netkit: Use NETKIT_SCRUB_NONE on primary device#35306
Merged
borkmann merged 1 commit intocilium:mainfrom Oct 22, 2024
Merged
Conversation
Contributor
Author
|
cc @borkmann . Tested this locally with a fresh kernel build and this change to |
eeb78f7 to
9afae90
Compare
Member
|
Looks great to me, thanks! Once the vishvananda/netlink#1022 PR gets merged, we need to pull in the go mod dependencies and then we can get this merged asap. |
48074fd to
b6fc704
Compare
When using endpoint routes with veth, cil_to_container runs before the veth driver scrubs the packet (i.e. clears skb->mark). In netkit mode, cil_to_container is executed after scrubbing the packet. This causes traffic from originating from the host to be miscategorized leading to cases like ciliumGH-34042 where enabling netkit causes liveness+readiness probes to fail. To address this, the IFLA_NETKIT_SCRUB and IFLA_NETKIT_PEER_SCRUB flags were recently introduced to configure whether or not netkit scrubs the packet when xmiting to the primary or peer devices, respectively. Set IFLA_NETKIT_SCRUB to NETKIT_SCRUB_NONE to ensure that any mark set in the host namespace is passed to cil_to_container when using netkit with endpoint routes. Set IFLA_NETKIT_PEER_SCRUB to NETKIT_SCRUB_DEFAULT to ensure that the mark is always cleared before running cil_from_container and passing the packet on to the host namespace. Signed-off-by: Jordan Rife <jrife@google.com> Link: https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net/T/#u Fixes: cilium#35060 Fixes: cilium#34042 Fixes: cilium#33875 Fixes: 6895341 ("cilium, connector: Add netkit connector")
b6fc704 to
e1f80c9
Compare
Contributor
Author
|
vishvananda/netlink#1022 was merged and I updated this PR. I ended up leaving out the new test case from |
borkmann
approved these changes
Oct 22, 2024
Member
|
/test |
YutaroHayakawa
approved these changes
Oct 22, 2024
Member
Sounds good, the repo for CI builds is here https://github.com/cilium/little-vm-helper-images/commits/main/ , and I think cilium/little-vm-helper-images@ef9d087 should already have it included (the bpf tree one). One more thing to follow-up apart from the e2e test addition is that the BPF program attached to the netkit primary device should also at the end clear prio+mark to 0, so that this doesn't go all the way into the Pod. Thanks @YutaroHayakawa for spotting! |
borkmann
added a commit
that referenced
this pull request
Oct 25, 2024
Jordan fixed Cilium with netkit and per-endpoint-routes in #35306. Given we have a more recent bpf image now, lets add it also to CI to regularly test for regressions. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
borkmann
added a commit
that referenced
this pull request
Oct 25, 2024
Jordan fixed Cilium with netkit and per-endpoint-routes in #35306. Given we have a more recent bpf image now, lets add it also to CI to regularly test for regressions. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
borkmann
added a commit
that referenced
this pull request
Oct 25, 2024
Jordan fixed Cilium with netkit and per-endpoint-routes in #35306. Given we have a more recent bpf image now, lets add it also to CI to regularly test for regressions. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
borkmann
added a commit
that referenced
this pull request
Oct 29, 2024
Jordan fixed Cilium with netkit and per-endpoint-routes in #35306. Given we have a more recent bpf image now, lets add it also to CI to regularly test for regressions. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
borkmann
added a commit
that referenced
this pull request
Nov 4, 2024
Jordan fixed Cilium with netkit and per-endpoint-routes in #35306. Given we have a more recent bpf image now, lets add it also to CI to regularly test for regressions. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
borkmann
added a commit
that referenced
this pull request
Nov 4, 2024
Jordan fixed Cilium with netkit and per-endpoint-routes in #35306. Given we have a more recent bpf image now, lets add it also to CI to regularly test for regressions. Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
3 tasks
3 tasks
ajmmm
added a commit
to ajmmm/cilium
that referenced
this pull request
Mar 23, 2026
Early versions of the netkit driver scrubbed skb metadata before BPF
programs are run, meaning identity metadata is not available when
applying network policy if per-endpoint routes are enabled. This was
fixed with a patch to the kernel driver [0] and an update to the Cilium
datapath connector [1].
It's still possible to run Cilium with netkit on a kernel that doesn't
support the new scrub attributes, which is confusing [2] and may still
result in strange behavior.
This commit introduces a runtime probe that will actively check for the
presence of scrub attribute support if running with per-endpoint routes.
If the probe fails, we error out.
- With bpf.datapathMode={netkit,netkit-l2} this will error and the agent
will fail to start.
- with bpf.datapathMode=auto, this will log a warning and fall-back to
veth connector.
This commit also introduces additional test permutations to cover this.
[0] https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net
[1] cilium#35306
[2] cilium#40021
Signed-off-by: Alasdair McWilliam <alasdair.mcwilliam@isovalent.com>
ajmmm
added a commit
to ajmmm/cilium
that referenced
this pull request
Mar 23, 2026
Early versions of the netkit driver scrubbed skb metadata before BPF
programs are run, meaning identity metadata is not available when
applying network policy if per-endpoint routes are enabled. This was
fixed with a patch to the kernel driver [0] and an update to the Cilium
datapath connector [1].
It's still possible to run Cilium with netkit on a kernel that doesn't
support the new scrub attributes, which is confusing [2] and may still
result in strange behavior.
This commit introduces a runtime probe that will actively check for the
presence of scrub attribute support if running with per-endpoint routes.
If the probe fails, we error out.
- With bpf.datapathMode={netkit,netkit-l2} this will error and the agent
will fail to start.
- with bpf.datapathMode=auto, this will log a warning and fall-back to
veth connector.
This commit also introduces additional test permutations to cover this.
[0] https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net
[1] cilium#35306
[2] cilium#40021
Signed-off-by: Alasdair McWilliam <alasdair.mcwilliam@isovalent.com>
ajmmm
added a commit
to ajmmm/cilium
that referenced
this pull request
Mar 23, 2026
Early versions of the netkit driver scrubbed skb metadata before BPF
programs are run, meaning identity metadata is not available in BPF.
This can cause network policy mis-classification if per-endpoint routes
are enabled.
A kernel patch landed in the driver [0] that adds a new scrub attributes
to manage this, and a subsequent change was made in the Cilium datapath
connector [1] to make use of the new attributes. However, it's still
possible to run Cilium with netkit on a kernel that doesn't support the
new scrub attributes. This is confusing [2] and may still result in
strange behavior.
This commit introduces a runtime probe that will actively probe the
underlying host for netkit scrub attributes, if running with per-endpoint
routes. If the probe fails, we error out.
- With bpf.datapathMode={netkit,netkit-l2} this will error and the agent
will fail to start.
- with bpf.datapathMode=auto, this will log a warning and fall-back to
veth connector.
This commit also introduces additional test permutations to cover this.
[0] https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net
[1] cilium#35306
[2] cilium#40021
Signed-off-by: Alasdair McWilliam <alasdair.mcwilliam@isovalent.com>
ajmmm
added a commit
to ajmmm/cilium
that referenced
this pull request
Mar 23, 2026
Early versions of the netkit driver scrubbed skb metadata before BPF
programs are run, meaning identity metadata is not available in BPF.
This can cause network policy mis-classification if per-endpoint routes
are enabled.
A kernel patch landed in the driver [0] that adds a new scrub attributes
to manage this, and a subsequent change was made in the Cilium datapath
connector [1] to make use of the new attributes. However, it's still
possible to run Cilium with netkit on a kernel that doesn't support the
new scrub attributes. This is confusing [2] and may still result in
strange behavior.
This commit introduces a runtime probe that will actively probe the
underlying host for netkit scrub attributes, if running with per-endpoint
routes. If the probe fails, we error out.
- With bpf.datapathMode={netkit,netkit-l2} this will error and the agent
will fail to start.
- with bpf.datapathMode=auto, this will log a warning and fall-back to
veth connector.
This commit also introduces additional test permutations to cover this.
[0] https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net
[1] cilium#35306
[2] cilium#40021
Signed-off-by: Alasdair McWilliam <alasdair.mcwilliam@isovalent.com>
ajmmm
added a commit
to ajmmm/cilium
that referenced
this pull request
Mar 23, 2026
Early versions of the netkit driver scrubbed skb metadata before BPF
programs are run, meaning identity metadata is not available in BPF.
This can cause network policy mis-classification if per-endpoint routes
are enabled.
A kernel patch landed in the driver [0] which adds new scrub attributes
to manage this, and a subsequent change was made in the Cilium datapath
connector [1] to make use of these new attributes. However, it's still
possible to run Cilium with netkit on a kernel that doesn't support
them, which is confusing [2] and may still result in strange behavior.
This commit introduces a runtime probe that will actively probe the
underlying host for netkit scrub attributes, if running with per-endpoint
routes. If the probe fails, we error out.
- With bpf.datapathMode={netkit,netkit-l2} this will error and the agent
will fail to start.
- with bpf.datapathMode=auto, this will log a warning and fall-back to
veth connector.
This commit also introduces additional test permutations to cover this.
[0] https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net
[1] cilium#35306
[2] cilium#40021
Signed-off-by: Alasdair McWilliam <alasdair.mcwilliam@isovalent.com>
ajmmm
added a commit
to ajmmm/cilium
that referenced
this pull request
Mar 23, 2026
Early versions of the netkit driver scrubbed skb metadata before BPF
programs are run, meaning identity metadata is not available in BPF.
This can cause network policy mis-classification if per-endpoint routes
are enabled.
A kernel patch landed in the driver [0] which adds new scrub attributes
to manage this, and a subsequent change was made in the Cilium datapath
connector [1] to make use of these new attributes. However, it's still
possible to run Cilium with netkit on a kernel that doesn't support
them, which is confusing [2] and may still result in strange behavior.
This commit introduces a runtime probe that will actively probe the
underlying host for netkit scrub attributes, if running with per-endpoint
routes. If the probe fails, we error out.
- With bpf.datapathMode={netkit,netkit-l2} this will error and the agent
will fail to start.
- with bpf.datapathMode=auto, this will log a warning and fall-back to
veth connector.
This commit also introduces additional test permutations to cover this.
[0] https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net
[1] cilium#35306
[2] cilium#40021
Signed-off-by: Alasdair McWilliam <alasdair.mcwilliam@isovalent.com>
ajmmm
added a commit
to ajmmm/cilium
that referenced
this pull request
Mar 23, 2026
Early versions of the netkit driver scrubbed skb metadata before BPF
programs are run, meaning identity metadata is not available in BPF.
This can cause network policy mis-classification if per-endpoint routes
are enabled.
A kernel patch landed in the driver [0] which adds new scrub attributes
to manage this, and a subsequent change was made in the Cilium datapath
connector [1] to make use of these new attributes. However, it's still
possible to run Cilium with netkit on a kernel that doesn't support
them, which is confusing [2] and may still result in strange behavior.
This commit introduces a runtime probe that will actively probe the
underlying host for netkit scrub attributes, if running with per-endpoint
routes. If the probe fails, we error out.
- With bpf.datapathMode={netkit,netkit-l2} this will error and the agent
will fail to start.
- with bpf.datapathMode=auto, this will log a warning and fall-back to
veth connector.
This commit also introduces additional test permutations to cover this.
[0] https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net
[1] cilium#35306
[2] cilium#40021
Signed-off-by: Alasdair McWilliam <alasdair.mcwilliam@isovalent.com>
ajmmm
added a commit
to ajmmm/cilium
that referenced
this pull request
Mar 23, 2026
Early versions of the netkit driver scrubbed skb metadata before BPF
programs are run, meaning identity metadata is not available in BPF.
This can cause network policy mis-classification if per-endpoint routes
are enabled.
A kernel patch landed in the driver [0] which adds new scrub attributes
to manage this, and a subsequent change was made in the Cilium datapath
connector [1] to make use of these new attributes. However, it's still
possible to run Cilium with netkit on a kernel that doesn't support
them, which is confusing [2] and may still result in strange behavior.
This commit introduces a runtime probe that will actively probe the
underlying host for netkit scrub attributes, if running with per-endpoint
routes. If the probe fails, we error out.
- With bpf.datapathMode={netkit,netkit-l2} this will error and the agent
will fail to start.
- with bpf.datapathMode=auto, this will log a warning and fall-back to
veth connector.
This commit also introduces additional test permutations to cover this.
[0] https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net
[1] cilium#35306
[2] cilium#40021
Signed-off-by: Alasdair McWilliam <alasdair.mcwilliam@isovalent.com>
ajmmm
added a commit
to ajmmm/cilium
that referenced
this pull request
Mar 23, 2026
Early versions of the netkit driver scrubbed skb metadata before BPF
programs are run, meaning identity metadata is not available in BPF.
This can cause network policy mis-classification if per-endpoint routes
are enabled.
A kernel patch landed in the driver [0] which adds new scrub attributes
to manage this, and a subsequent change was made in the Cilium datapath
connector [1] to make use of these new attributes. However, it's still
possible to run Cilium with netkit on a kernel that doesn't support
them, which is confusing [2] and may still result in strange behavior.
This commit introduces a runtime probe that will actively probe the
underlying host for netkit scrub attributes, if running with per-endpoint
routes. If the probe fails, we error out.
- With bpf.datapathMode={netkit,netkit-l2} this will error and the agent
will fail to start.
- with bpf.datapathMode=auto, this will log a warning and fall-back to
veth connector.
This commit also introduces additional test permutations to cover this.
[0] https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net
[1] cilium#35306
[2] cilium#40021
Signed-off-by: Alasdair McWilliam <alasdair.mcwilliam@isovalent.com>
github-merge-queue bot
pushed a commit
that referenced
this pull request
Mar 25, 2026
Early versions of the netkit driver scrubbed skb metadata before BPF
programs are run, meaning identity metadata is not available in BPF.
This can cause network policy mis-classification if per-endpoint routes
are enabled.
A kernel patch landed in the driver [0] which adds new scrub attributes
to manage this, and a subsequent change was made in the Cilium datapath
connector [1] to make use of these new attributes. However, it's still
possible to run Cilium with netkit on a kernel that doesn't support
them, which is confusing [2] and may still result in strange behavior.
This commit introduces a runtime probe that will actively probe the
underlying host for netkit scrub attributes, if running with per-endpoint
routes. If the probe fails, we error out.
- With bpf.datapathMode={netkit,netkit-l2} this will error and the agent
will fail to start.
- with bpf.datapathMode=auto, this will log a warning and fall-back to
veth connector.
This commit also introduces additional test permutations to cover this.
[0] https://lore.kernel.org/bpf/20241004101335.117711-1-daniel@iogearbox.net
[1] #35306
[2] #40021
Signed-off-by: Alasdair McWilliam <alasdair.mcwilliam@isovalent.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
When using endpoint routes with veth, cil_to_container runs before the veth driver scrubs the packet (i.e. clears skb->mark). In netkit mode, cil_to_container is executed after scrubbing the packet. This causes traffic from originating from the host to be miscategorized leading to cases like GH-34042 where enabling netkit causes liveness+readiness probes to fail.
To address this, the
IFLA_NETKIT_SCRUBandIFLA_NETKIT_PEER_SCRUBflags were recently introduced to configure whether or not netkit scrubs the packet when xmiting to the primary or peer devices, respectively. SetIFLA_NETKIT_SCRUBtoNETKIT_SCRUB_NONEto ensure that any mark set in the host namespace is passed to cil_to_container when using netkit with endpoint routes. SetIFLA_NETKIT_PEER_SCRUBtoNETKIT_SCRUB_DEFAULTto ensure that the mark is always cleared before running cil_from_container and passing the packet on to the host namespace.This PR relies on this change which adds support for these new flags to
github.com/vishvananda/netlink.Fixes: #35060
Fixes: #34042
Fixes: #33875
Fixes: 6895341 ("cilium, connector: Add netkit connector")