Remove unexpected AltName after rename interface #6321

gran-vmv · 2024-05-11T06:14:46Z

gran-vmv · 2024-05-13T02:29:19Z

/test-e2e

pkg/agent/util/net_linux.go

wenyingd · 2024-05-13T03:22:06Z

pkg/agent/util/net_linux.go

@@ -378,6 +384,17 @@ func renameHostInterface(oriName string, newName string) error {
 	return nil
 }

+func removeInterfaceAltName(name string, altName string) error {
+	link, err := netlinkUtil.LinkByName(name)


I would prefer to add check first, if altname exists, and equals to the original name, we shall delete it, otherwise, ignore it.

Is it possible that time delays exist between adding the altname and our deletion on the target altname, e.g., the sequence is like this,

rename uplink

try to delete altname (introduced in this change)

system adds the altname.

If so, the change may not work as expected.

Is it possible that time delays exist between adding the altname and our deletion on the target altname, e.g., the sequence is like this,

rename uplink

try to delete altname (introduced in this change)

system adds the altname.

If so, the change may not work as expected.

AltName will be set very quickly.
If this process get delayed, antrea-agent will create internal port and occupy this name first, and this is not an issue.

pkg/agent/util/net_linux.go

gran-vmv · 2024-05-14T02:34:25Z

@tnqn @antoninbas

In this PR we bumped up netlink to latest dev version to support AltName operation. However, we found the default Route behavior is changed from the integration test failure, and we cannot guarantee that there is no other change which will break current logic.

Do you have any suggestion on this situation?

antoninbas · 2024-05-14T19:08:51Z

In this PR we bumped up netlink to latest dev version to support AltName operation. However, we found the default Route behavior is changed from the integration test failure

What is the behavior change?
Is it a matter of simply updating the tests?

we cannot guarantee that there is no other change which will break current logic.

That's true of a lot of dependency version updates.
It's unfortunate that the netlink library doesn't have a proper release process with release notes that we could refer to, and that the latest release tag is a "beta" tag from 2 years ago.
But we should still be able to update the library if we need to. Ideally, we would review the change and check if anything is likely to impact us. That's also why we have tests, to make sure that we don't break anything on our side when we bump up the dependency. In the case of the integration test failure, we need to investigate if it's a matter of updating the test or something deeper. Did an API change?

If we cannot update to a version of the netlink library that includes support for AltName while not introducing bugs that impact us, we will fork the repository and do our own release. (In that case, we should also open issues upstream as appropriate.)

gran-vmv · 2024-05-15T02:35:05Z

In this PR we bumped up netlink to latest dev version to support AltName operation. However, we found the default Route behavior is changed from the integration test failure

What is the behavior change? Is it a matter of simply updating the tests?

we cannot guarantee that there is no other change which will break current logic.

That's true of a lot of dependency version updates. It's unfortunate that the netlink library doesn't have a proper release process with release notes that we could refer to, and that the latest release tag is a "beta" tag from 2 years ago. But we should still be able to update the library if we need to. Ideally, we would review the change and check if anything is likely to impact us. That's also why we have tests, to make sure that we don't break anything on our side when we bump up the dependency. In the case of the integration test failure, we need to investigate if it's a matter of updating the test or something deeper. Did an API change?

If we cannot update to a version of the netlink library that includes support for AltName while not introducing bugs that impact us, we will fork the repository and do our own release. (In that case, we should also open issues upstream as appropriate.)

Previously the dafault route has nil Dst, but it is filled in latest dev version by below code in route_linux.go:

	// Same logic to generate "default" dst with iproute2 implementation
	if route.Dst == nil {
		var addLen int
		var ip net.IP
		switch msg.Family {
		case FAMILY_V4:
			addLen = net.IPv4len
			ip = net.IPv4zero
		case FAMILY_V6:
			addLen = net.IPv6len
			ip = net.IPv6zero
		}

		if addLen != 0 {
			route.Dst = &net.IPNet{
				IP:   ip,
				Mask: net.CIDRMask(int(msg.Dst_len), 8*addLen),
			}
		}
	}

Antrea uses Route.Dst == nil to check if this route is default route, and we need to fix the code and inteegration test.

tnqn · 2024-05-15T14:51:51Z

go.mod

@@ -236,3 +236,5 @@ require (
 	sigs.k8s.io/kustomize/kyaml v0.14.3-0.20230601165947-6ce0bf390ce3 // indirect
 	sigs.k8s.io/structured-merge-diff/v4 v4.4.1 // indirect
 )
+
+replace github.com/vishvananda/netlink => github.com/vishvananda/netlink v0.0.0-20240425164735-856e190dd707


why doesn't it set version in require directly?

@hongliangl have you figured out whether the new version really affects the e2e test?

Not yet, still testing it on my forked Antrea repo.

I tried to change the version in require directly, but it is changed back to v1.2.1-beta.2 after Goland resynced the dependency.

Hongliang's PR is merged. No need to bump up netlink now.

tnqn · 2024-05-15T14:52:05Z

pkg/agent/util/net_linux.go

+	// which is also occupied by an interface name, after the interface name changed from generated value to another one,
+	// the interface altname will be set to previous name immediately.
+	// This altname must be removed to avoid unexpected conflict.
+	removeInterfaceAltName(to, from)


miss error handling

tnqn · 2024-05-15T15:01:34Z

Previously the dafault route has nil Dst, but it is filled in latest dev version by below code in route_linux.go:

If this is the only known case, we should adapt the code to it. But I remember @hongliangl mentioned other test failures that may be related to it.

antoninbas

I had started a review yesterday, but didn't submit it...

antoninbas · 2024-05-14T18:59:05Z

go.mod

@@ -236,3 +236,5 @@ require (
 	sigs.k8s.io/kustomize/kyaml v0.14.3-0.20230601165947-6ce0bf390ce3 // indirect
 	sigs.k8s.io/structured-merge-diff/v4 v4.4.1 // indirect
 )
+
+replace github.com/vishvananda/netlink => github.com/vishvananda/netlink v0.0.0-20240425164735-856e190dd707


I don't think you need a replace directive. Can't you just use the desired "version" in the require block directly?

I tried to change the version in require directly, but it is changed back to v1.2.1-beta.2 after Goland resynced the dependency.

Hongliang's PR is merged. No need to bump up netlink now.

antoninbas · 2024-05-14T19:01:20Z

pkg/agent/util/net_linux.go

@@ -290,6 +290,12 @@ func RenameInterface(from, to string) error {
 	if pollErr != nil {
 		return fmt.Errorf("failed to rename host interface name %s to %s", from, to)
 	}
+	// Fix for the issue https://github.com/antrea-io/antrea/issues/6301.
+	// In some new Linux versions which support AltName, if there is only one valid value in AlternativeNamesPolicy


what do you mean by "one valid value" here? In the original issue, I see that AlternativeNamesPolicy=database onboard slot path. It looks like 4 valid values to me based on https://access.redhat.com/solutions/6964829

Although we can set many values here, it may not be valid for an specific interface.

For example, here is the default value in ubuntu 23.10

[Link] NamePolicy=keep kernel database onboard slot path AlternativeNamesPolicy=database onboard slot path

If slot and path is valid for this interface (it depends on the hardware information), we'll get below name/altname for a new interface. Name is slot and altname is path:

2: ens192: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 00:50:56:a6:68:79 brd ff:ff:ff:ff:ff:ff altname enp11s0

If only path is valid, we'll get below name for a new interface. Name is path and no altname:

3: enp11s1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 00:50:56:a6:68:7A brd ff:ff:ff:ff:ff:ff

If we change the name of above interface, altname will be set to path immediately. And antrea will failed to create internal OVS port for this case.

3: enp11s1~: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 00:50:56:a6:68:7A brd ff:ff:ff:ff:ff:ff altname enp11s1

Maybe adjust it a little:

In some new Linux versions which support AltName, if the only valid altname of the interface is the same as the interface name, it would be left empty when the name is occupied by the interface name; after we rename the interface name to another value, the altname of the interface would be set to the original interface name by the system. This altname must be removed as we need to reserve the name for an OVS internal port.

hongliangl · 2024-05-16T03:35:57Z

Previously the dafault route has nil Dst, but it is filled in latest dev version by below code in route_linux.go:

If this is the only known case, we should adapt the code to it. But I remember @hongliangl mentioned other test failures that may be related to it.

Yes, I always got the same failure from the following test after bumping the lib netlink.

 --- FAIL: TestAntreaPolicy/TestGroupValidationWebhook/Case=CreateInvalidTier (18.07s)

gran-vmv · 2024-05-16T05:48:02Z

Previously the dafault route has nil Dst, but it is filled in latest dev version by below code in route_linux.go:

If this is the only known case, we should adapt the code to it. But I remember @hongliangl mentioned other test failures that may be related to it.

Yes, I always got the same failure from the following test after bumping the lib netlink.
 --- FAIL: TestAntreaPolicy/TestGroupValidationWebhook/Case=CreateInvalidTier (18.07s)

Which version you bumped up and which pipeline you get the failure?

In this PR, all finished e2e pipelines are passed.

hongliangl · 2024-05-16T06:13:51Z

Previously the dafault route has nil Dst, but it is filled in latest dev version by below code in route_linux.go:

If this is the only known case, we should adapt the code to it. But I remember @hongliangl mentioned other test failures that may be related to it.

Yes, I always got the same failure from the following test after bumping the lib netlink.
 --- FAIL: TestAntreaPolicy/TestGroupValidationWebhook/Case=CreateInvalidTier (18.07s)
Which version you bumped up and which pipeline you get the failure?

In this PR, all finished e2e pipelines are passed.

All e2e tests might pass sometimes, but it seems that the failure is always the same. I saw the same failure from the tests in this PR:

This is from another PR I made:

https://github.com/antrea-io/antrea/actions/runs/8812876040/job/24189848588?pr=6193

gran-vmv · 2024-05-17T02:52:20Z

#6193 also bumped up netlink lib. Need rebase after #6193 merged.

gran-vmv · 2024-05-29T08:35:59Z

/test-all
/test-flexible-ipam-e2e

tnqn · 2024-05-29T15:08:09Z

pkg/agent/util/net_linux.go

@@ -290,6 +290,12 @@ func RenameInterface(from, to string) error {
 	if pollErr != nil {
 		return fmt.Errorf("failed to rename host interface name %s to %s", from, to)
 	}
+	// Fix for the issue https://github.com/antrea-io/antrea/issues/6301.
+	// In some new Linux versions which support AltName, if there is only one valid value in AlternativeNamesPolicy


Maybe adjust it a little:

In some new Linux versions which support AltName, if the only valid altname of the interface is the same as the interface name, it would be left empty when the name is occupied by the interface name; after we rename the interface name to another value, the altname of the interface would be set to the original interface name by the system. This altname must be removed as we need to reserve the name for an OVS internal port.

tnqn · 2024-05-29T15:10:57Z

pkg/agent/util/net_linux.go

+		klog.ErrorS(err, "Failed to remove AltName after interface renamed")
+		return fmt.Errorf("failed to remove AltName %s on interface %s", from, to)


Suggested change

klog.ErrorS(err, "Failed to remove AltName after interface renamed")

return fmt.Errorf("failed to remove AltName %s on interface %s", from, to)

return fmt.Errorf("failed to remove AltName %s on interface %s: %w", from, to, err)

don't handle the same error twice

Fixed. Thanks.

Fix antrea-io#6301 Signed-off-by: gran <gran@vmware.com> Co-authored-by: Lan <luola@vmware.com>

gran-vmv · 2024-05-30T09:18:49Z

/test-all
/test-flexible-ipam-e2e

gran-vmv · 2024-05-31T02:36:11Z

/test-all

tnqn

LGTM

tnqn · 2024-06-05T02:52:50Z

@gran-vmv could you backport this to 1.14-2.0?

Fix antrea-io#6301 Signed-off-by: gran <gran@vmware.com> Co-authored-by: Lan <luola@vmware.com>

Fix #6301 Signed-off-by: gran <gran@vmware.com>

…trea-io#6402) Fix antrea-io#6301 Signed-off-by: gran <gran@vmware.com>

Fix #6301 Signed-off-by: gran <gran@vmware.com>

gran-vmv self-assigned this May 11, 2024

gran-vmv requested review from wenyingd and luolanzone May 13, 2024 03:36

gran-vmv added kind/bug Categorizes issue or PR as related to a bug. area/interface Issues or PRs related to network interfaces area/OS/linux Issues or PRs related to the Linux operating system. labels May 13, 2024

wenyingd reviewed May 13, 2024

View reviewed changes

gran-vmv force-pushed the ipam-renamefix branch 4 times, most recently from 303b114 to 58dbcc3 Compare May 13, 2024 06:18

luolanzone reviewed May 13, 2024

View reviewed changes

pkg/agent/util/net_linux.go Outdated Show resolved Hide resolved

pkg/agent/util/net_linux.go Outdated Show resolved Hide resolved

luolanzone added the action/release-note Indicates a PR that should be included in release notes. label May 13, 2024

gran-vmv force-pushed the ipam-renamefix branch from b489727 to 4dbddbb Compare May 13, 2024 09:05

tnqn reviewed May 15, 2024

View reviewed changes

antoninbas reviewed May 15, 2024

View reviewed changes

antoninbas added the action/backport Indicates a PR that requires backports. label May 15, 2024

gran-vmv force-pushed the ipam-renamefix branch 2 times, most recently from 43a83cd to 41a3bbe Compare May 16, 2024 09:20

antoninbas mentioned this pull request May 21, 2024

Delay removal of flow-restore-wait #6342

Merged

gran-vmv force-pushed the ipam-renamefix branch from 41a3bbe to 4d49aa3 Compare May 29, 2024 07:40

tnqn reviewed May 29, 2024

View reviewed changes

gran-vmv force-pushed the ipam-renamefix branch from 4d49aa3 to 7245b73 Compare May 30, 2024 03:16

Remove unexpected AltName after rename interface

dd33d2f

Fix antrea-io#6301 Signed-off-by: gran <gran@vmware.com> Co-authored-by: Lan <luola@vmware.com>

gran-vmv force-pushed the ipam-renamefix branch from 7245b73 to dd33d2f Compare May 30, 2024 08:16

tnqn approved these changes Jun 3, 2024

View reviewed changes

tnqn merged commit d895124 into antrea-io:main Jun 5, 2024
52 of 56 checks passed

gran-vmv added a commit to gran-vmv/antrea that referenced this pull request Jun 5, 2024

Remove unexpected AltName after rename interface (antrea-io#6321)

21fa5c3

Fix antrea-io#6301 Signed-off-by: gran <gran@vmware.com> Co-authored-by: Lan <luola@vmware.com>

gran-vmv mentioned this pull request Jun 5, 2024

Automated cherry pick of #6321: Remove unexpected AltName after rename interface (#6321) #6401

Merged

gran-vmv added a commit to gran-vmv/antrea that referenced this pull request Jun 5, 2024

Remove unexpected AltName after rename interface (antrea-io#6321)

3f880c1

Fix antrea-io#6301 Signed-off-by: gran <gran@vmware.com> Co-authored-by: Lan <luola@vmware.com>

This was referenced Jun 5, 2024

Automated cherry pick of #6321: Remove unexpected AltName after rename interface (#6321) #6402

Merged

Fix that AntreaProxy could unintentionally delete conntrack entries in zone 0 #6193

Merged

gran-vmv added a commit to gran-vmv/antrea that referenced this pull request Jun 7, 2024

Remove unexpected AltName after rename interface (antrea-io#6321)

c5e5627

Fix antrea-io#6301 Signed-off-by: gran <gran@vmware.com> Co-authored-by: Lan <luola@vmware.com>

gran-vmv added a commit to gran-vmv/antrea that referenced this pull request Jun 7, 2024

Remove unexpected AltName after rename interface (antrea-io#6321)

5b0bc27

Fix antrea-io#6301 Signed-off-by: gran <gran@vmware.com> Co-authored-by: Lan <luola@vmware.com>

gran-vmv added a commit to gran-vmv/antrea that referenced this pull request Jun 7, 2024

Remove unexpected AltName after rename interface (antrea-io#6321)

8679b04

Fix antrea-io#6301 Signed-off-by: gran <gran@vmware.com> Co-authored-by: Lan <luola@vmware.com>

tnqn pushed a commit that referenced this pull request Jun 11, 2024

Remove unexpected AltName after rename interface (#6321) (#6402)

d8555ea

Fix #6301 Signed-off-by: gran <gran@vmware.com>

tnqn pushed a commit that referenced this pull request Jun 11, 2024

Remove unexpected AltName after rename interface (#6321) (#6401)

99b732b

Fix #6301 Signed-off-by: gran <gran@vmware.com>

gran-vmv added a commit to gran-vmv/antrea that referenced this pull request Jun 12, 2024

Remove unexpected AltName after rename interface (antrea-io#6321) (an…

4e3d54c

…trea-io#6402) Fix antrea-io#6301 Signed-off-by: gran <gran@vmware.com>

gran-vmv mentioned this pull request Jun 12, 2024

Automated cherry pick of #6402: Remove unexpected AltName after rename interface (#6321) #6434

Merged

tnqn pushed a commit that referenced this pull request Jun 12, 2024

Remove unexpected AltName after rename interface (#6321) (#6402) (#6434)

0b20962

Fix #6301 Signed-off-by: gran <gran@vmware.com>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Remove unexpected AltName after rename interface #6321

Remove unexpected AltName after rename interface #6321

gran-vmv commented May 11, 2024

gran-vmv commented May 13, 2024

wenyingd May 13, 2024

wenyingd May 13, 2024 •

edited

Loading

gran-vmv May 13, 2024

gran-vmv May 13, 2024

gran-vmv commented May 14, 2024

antoninbas commented May 14, 2024 •

edited

Loading

gran-vmv commented May 15, 2024

tnqn May 15, 2024

hongliangl May 16, 2024

gran-vmv May 16, 2024

gran-vmv May 30, 2024

tnqn May 15, 2024

tnqn commented May 15, 2024

antoninbas left a comment

antoninbas May 14, 2024

gran-vmv May 16, 2024

gran-vmv May 30, 2024

antoninbas May 14, 2024

gran-vmv May 16, 2024 •

edited

Loading

tnqn May 29, 2024

gran-vmv May 30, 2024

hongliangl commented May 16, 2024

gran-vmv commented May 16, 2024

hongliangl commented May 16, 2024 •

edited

Loading

gran-vmv commented May 17, 2024

gran-vmv commented May 29, 2024

tnqn May 29, 2024

tnqn May 29, 2024

gran-vmv May 30, 2024

gran-vmv commented May 30, 2024

gran-vmv commented May 31, 2024

tnqn left a comment

tnqn commented Jun 5, 2024

		klog.ErrorS(err, "Failed to remove AltName after interface renamed")
		return fmt.Errorf("failed to remove AltName %s on interface %s", from, to)

Remove unexpected AltName after rename interface #6321

Remove unexpected AltName after rename interface #6321

Conversation

gran-vmv commented May 11, 2024

gran-vmv commented May 13, 2024

Choose a reason for hiding this comment

wenyingd May 13, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gran-vmv commented May 14, 2024

antoninbas commented May 14, 2024 • edited Loading

gran-vmv commented May 15, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

tnqn commented May 15, 2024

antoninbas left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gran-vmv May 16, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

hongliangl commented May 16, 2024

gran-vmv commented May 16, 2024

hongliangl commented May 16, 2024 • edited Loading

gran-vmv commented May 17, 2024

gran-vmv commented May 29, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gran-vmv commented May 30, 2024

gran-vmv commented May 31, 2024

tnqn left a comment

Choose a reason for hiding this comment

tnqn commented Jun 5, 2024

wenyingd May 13, 2024 •

edited

Loading

antoninbas commented May 14, 2024 •

edited

Loading

gran-vmv May 16, 2024 •

edited

Loading

hongliangl commented May 16, 2024 •

edited

Loading