Skip to contentSkip to conversation

inet: bring NLM_DONE out to a separate recv() again

PATCH Applied Jakub Kicinski 6 participants 15 messages Apr 11, 2024 – Jun 17, 2024

Subsystems (3)
lore
Share & tools
lore

Jakub Kicinski

author maintainer
#0 lore raw

PATCH 1/1 inet: bring NLM_DONE out to a separate recv() again

Commit under Fixes optimized the number of recv() calls needed during RTM_GETROUTE dumps, but we got multiple reports of applications hanging on recv() calls. Applications expect that a route dump will be terminated with a recv() reading an individual NLM_DONE message.

Coalescing NLM_DONE is perfectly legal in netlink, but even tho reporters fixed the code in respective projects, chances are it will take time for those applications to get updated. So revert to old behavior (for now)?

Old kernel (5.19):

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 692 bytes, 11 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
 Recv: read 20 bytes, 1 messages
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

Before (6.9-rc2):

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 712 bytes, 12 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

After:

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 692 bytes, 11 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
 Recv: read 20 bytes, 1 messages
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

---
CC: [email protected]
CC: [email protected]

---
 net/ipv4/fib_frontend.c | 5 +++++
 1 file changed, 5 insertions(+)
net/ipv4/fib_frontend.c +5 -0
index 48741352a88a..c484b1c0fc00 100644
@@ -1050,6 +1050,11 @@ static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 			e++;
 		}
 	}
+
+	/* Don't let NLM_DONE coalesce into a message, even if it could.
+	 * Some user space expects NLM_DONE in a separate recv().
+	 */
+	err = skb->len;
 out:
 
 	cb->args[1] = e;
Signature · 1 line
2.44.0
#1 lore raw
Quoted 34 lines · Jakub Kicinski

Commit under Fixes optimized the number of recv() calls needed during RTM_GETROUTE dumps, but we got multiple reports of applications hanging on recv() calls. Applications expect that a route dump will be terminated with a recv() reading an individual NLM_DONE message.

Coalescing NLM_DONE is perfectly legal in netlink, but even tho reporters fixed the code in respective projects, chances are it will take time for those applications to get updated. So revert to old behavior (for now)?

Old kernel (5.19):

---
CC: [email protected]
CC: [email protected]

---
 net/ipv4/fib_frontend.c | 5 +++++
 1 file changed, 5 insertions(+)
net/ipv4/fib_frontend.c +0 -0
index 48741352a88a..c484b1c0fc00 100644

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1057

Quoted 8 lines
net/ipv4/fib_frontend.c +5 -0
@@ -1050,6 +1050,11 @@ static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
                        e++;
                }
        }
+
+       /* Don't let NLM_DONE coalesce into a message, even if it could.
+        * Some user space expects NLM_DONE in a separate recv().
+        */
+       err = skb->len;

My plan was to perform this generically from netlink_dump_done().

This would still avoid calling a RTNL-enabled-dump() again if EOF has been met already.

A sysctl could opt-in for the coalescing, if there is interest.

net/netlink/af_netlink.c +0 -0
index dc8c3c01d51b709c132ff63a0c534c1cc286589a..cad1124393ac74f3d5bfa86556ed9028f5ec8f65

100644

net/netlink/af_netlink.c +5 -0
@@ -2282,7 +2282,12 @@ static int netlink_dump(struct sock *sk, bool lock_taken)
                cb->extack = NULL;
        }
 
+       /* Don't let NLM_DONE coalesce into a message, even if it could.
+        * Some user space expects NLM_DONE in a separate recv().
+        * Maybe opt-in this coalescing with a sysctl or socket option ?
+        */
        if (nlk->dump_done_errno > 0 ||
+           skb->len ||
            skb_tailroom(skb) <
nlmsg_total_size(sizeof(nlk->dump_done_errno))) {
                mutex_unlock(&nlk->nl_cb_mutex);
author maintainer
#2 lore raw
Quoted 6 lines · On Thu, 11 Apr 2024 20:42:45 +0200 Eric Dumazet wrote:

My plan was to perform this generically from netlink_dump_done().

This would still avoid calling a RTNL-enabled-dump() again if EOF has been met already.

A sysctl could opt-in for the coalescing, if there is interest.

I'd prefer to keep hacks contained to the few places that need them. Netlink is full of strange quirks, I don't think this one deserves a sysctl.

Tested-by on v1 1/1
#3 lore raw
In reply to Jakub Kicinski
Quoted 63 lines · Jakub Kicinski

Commit under Fixes optimized the number of recv() calls needed during RTM_GETROUTE dumps, but we got multiple reports of applications hanging on recv() calls. Applications expect that a route dump will be terminated with a recv() reading an individual NLM_DONE message.

Coalescing NLM_DONE is perfectly legal in netlink, but even tho reporters fixed the code in respective projects, chances are it will take time for those applications to get updated. So revert to old behavior (for now)?

Old kernel (5.19):

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 692 bytes, 11 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
 Recv: read 20 bytes, 1 messages
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

Before (6.9-rc2):

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 712 bytes, 12 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

After:

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 692 bytes, 11 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
 Recv: read 20 bytes, 1 messages
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

---
CC: [email protected]
CC: [email protected]

---
 net/ipv4/fib_frontend.c | 5 +++++
 1 file changed, 5 insertions(+)
net/ipv4/fib_frontend.c +0 -0
index 48741352a88a..c484b1c0fc00 100644
@@ -1050,6 +1050,11 @@ static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 			e++;

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1060

Quoted 10 lines
net/ipv4/fib_frontend.c +5 -0
 		}
 	}
+
+	/* Don't let NLM_DONE coalesce into a message, even if it could.
+	 * Some user space expects NLM_DONE in a separate recv().
+	 */
+	err = skb->len;
 out:
 
 	cb->args[1] = e;

FWIW, on current net-next this fixes the issue with Libreswan and IPv4 (IPv6 issue remains, obviously). I also did a round of other OVS system tests and they worked fine.

maintainer Reviewed-by on v1 1/1
#7 lore raw
In reply to Jakub Kicinski
Quoted 62 lines · Jakub Kicinski

Commit under Fixes optimized the number of recv() calls needed during RTM_GETROUTE dumps, but we got multiple reports of applications hanging on recv() calls. Applications expect that a route dump will be terminated with a recv() reading an individual NLM_DONE message.

Coalescing NLM_DONE is perfectly legal in netlink, but even tho reporters fixed the code in respective projects, chances are it will take time for those applications to get updated. So revert to old behavior (for now)?

Old kernel (5.19):

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 692 bytes, 11 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
 Recv: read 20 bytes, 1 messages
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

Before (6.9-rc2):

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 712 bytes, 12 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

After:

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 692 bytes, 11 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
 Recv: read 20 bytes, 1 messages
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

---
CC: [email protected]
CC: [email protected]

---
 net/ipv4/fib_frontend.c | 5 +++++
 1 file changed, 5 insertions(+)
net/ipv4/fib_frontend.c +0 -0
index 48741352a88a..c484b1c0fc00 100644

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1055

Quoted 6 lines
net/ipv4/fib_frontend.c +3 -0
@@ -1050,6 +1050,11 @@ static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 			e++;
 		}
 	}
+
+	/* Don't let NLM_DONE coalesce into a message, even if it could.
+	 * Some user space expects NLM_DONE in a separate recv().

that's unfortunate

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1060

Quoted 5 lines
net/ipv4/fib_frontend.c +2 -0
+	 */
+	err = skb->len;
 out:
 
 	cb->args[1] = e;
author maintainer
#8 lore raw
Quoted 4 lines · On Thu, 11 Apr 2024 13:45:42 -0600 David Ahern wrote:

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1055

Quoted 2 lines
net/ipv4/fib_frontend.c +2 -0
+	/* Don't let NLM_DONE coalesce into a message, even if it could.
+	 * Some user space expects NLM_DONE in a separate recv().  

that's unfortunate

Do you have an opinion on the sysfs/opt-in question?
Feels to me like there shouldn't be that much user space doing raw netlink, without a library. Old crufty code usually does ioctls, right? So maybe we can periodically reintroduce this bug to shake out all the bad apps? :D

maintainer
#9 lore raw
Quoted 7 lines · Jakub Kicinski
Quoted 4 lines · On Thu, 11 Apr 2024 13:45:42 -0600 David Ahern wrote:

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1055

Quoted 2 lines
net/ipv4/fib_frontend.c +2 -0
+	/* Don't let NLM_DONE coalesce into a message, even if it could.
+	 * Some user space expects NLM_DONE in a separate recv().  

that's unfortunate

Do you have an opinion on the sysfs/opt-in question?

Not a performance critical path, so I would not add it right now.

Quoted 4 lines

Feels to me like there shouldn't be that much user space doing raw netlink, without a library. Old crufty code usually does ioctls, right? So maybe we can periodically reintroduce this bug to shake out all the bad apps? :D

:-) On the one hand, yea, push apps to improve. On the other, Donald (FRR) is one who complains about nightmares chasing nuances across kernels.

#10 lore raw
In reply to Jakub Kicinski
Quoted 9 lines · On Thu, 11 Apr 2024 14:01:54 -0700 Jakub Kicinski <[email protected]> wrote:
Quoted 4 lines · On Thu, 11 Apr 2024 13:45:42 -0600 David Ahern wrote:

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1055

Quoted 2 lines
net/ipv4/fib_frontend.c +2 -0
+	/* Don't let NLM_DONE coalesce into a message, even if it could.
+	 * Some user space expects NLM_DONE in a separate recv().    

that's unfortunate

Do you have an opinion on the sysfs/opt-in question?
Feels to me like there shouldn't be that much user space doing raw netlink, without a library. Old crufty code usually does ioctls, right?

I think so too -- if there were more (maintained) applications with this issue, we would have noticed by now.

Quoted 2 lines

So maybe we can periodically reintroduce this bug to shake out all the bad apps? :D

Actually, I had half a mind of proposing something on these lines: add a TODO comment here and revisit in, say, two years.

I guess it's definitely more painful for libreswan, but for passt, I think it's quite unlikely that distribution users could get the "breaking" kernel change without a fixed version of the application: we made a new release relatively close to the NLM_DONE change.

There might be substantial value in keeping this type of short netlink exchanges fast, for example for container engines that need to be able to spawn a bazillion containers per second.

Signature · 1 line
Stefano
#11 lore raw
Quoted 9 lines · Stefano Brivio
Quoted 6 lines · On Thu, 11 Apr 2024 14:01:54 -0700 Jakub Kicinski <[email protected]> wrote:
Quoted 4 lines · On Thu, 11 Apr 2024 13:45:42 -0600 David Ahern wrote:

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1055

Quoted 2 lines
net/ipv4/fib_frontend.c +2 -0
+	/* Don't let NLM_DONE coalesce into a message, even if it could.
+	 * Some user space expects NLM_DONE in a separate recv().    

that's unfortunate

Quoted 6 lines
Quoted 3 lines

Do you have an opinion on the sysfs/opt-in question?
Feels to me like there shouldn't be that much user space doing raw netlink, without a library. Old crufty code usually does ioctls, right?

I think so too -- if there were more (maintained) applications with this issue, we would have noticed by now.

It depends on how you define "maintained". Most application devs do not test with unreleased kernels.

Quoted 16 lines
Quoted 3 lines

So maybe we can periodically reintroduce this bug to shake out all the bad apps? :D

Actually, I had half a mind of proposing something on these lines: add a TODO comment here and revisit in, say, two years.

I guess it's definitely more painful for libreswan, but for passt, I think it's quite unlikely that distribution users could get the "breaking" kernel change without a fixed version of the application: we made a new release relatively close to the NLM_DONE change.

There might be substantial value in keeping this type of short netlink exchanges fast, for example for container engines that need to be able to spawn a bazillion containers per second.

#12 lore raw
Quoted 14 lines · On Fri, 12 Apr 2024 19:38:53 +0200 Ilya Maximets <[email protected]> wrote:
Quoted 13 lines · Stefano Brivio
Quoted 9 lines · On Thu, 11 Apr 2024 14:01:54 -0700 Jakub Kicinski <[email protected]> wrote:
Quoted 4 lines · On Thu, 11 Apr 2024 13:45:42 -0600 David Ahern wrote:

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1055

Quoted 2 lines
net/ipv4/fib_frontend.c +2 -0
+	/* Don't let NLM_DONE coalesce into a message, even if it could.
+	 * Some user space expects NLM_DONE in a separate recv().      

that's unfortunate

Do you have an opinion on the sysfs/opt-in question?
Feels to me like there shouldn't be that much user space doing raw netlink, without a library. Old crufty code usually does ioctls, right?

Quoted 5 lines
Quoted 2 lines

I think so too -- if there were more (maintained) applications with this issue, we would have noticed by now.

It depends on how you define "maintained". Most application devs do not test with unreleased kernels.

I haven't, either, but users started shouting: we have nowadays plenty of distributions shipping unreleased kernels.

Signature · 1 line
Stefano
#13 lore raw
Quoted 17 lines · Stefano Brivio
Quoted 14 lines · On Fri, 12 Apr 2024 19:38:53 +0200 Ilya Maximets <[email protected]> wrote:
Quoted 13 lines · Stefano Brivio
Quoted 9 lines · On Thu, 11 Apr 2024 14:01:54 -0700 Jakub Kicinski <[email protected]> wrote:
Quoted 4 lines · On Thu, 11 Apr 2024 13:45:42 -0600 David Ahern wrote:

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1055

Quoted 2 lines
net/ipv4/fib_frontend.c +2 -0
+	/* Don't let NLM_DONE coalesce into a message, even if it could.
+	 * Some user space expects NLM_DONE in a separate recv().      

that's unfortunate

Do you have an opinion on the sysfs/opt-in question?
Feels to me like there shouldn't be that much user space doing raw netlink, without a library. Old crufty code usually does ioctls, right?

Quoted 9 lines
Quoted 5 lines
Quoted 2 lines

I think so too -- if there were more (maintained) applications with this issue, we would have noticed by now.

It depends on how you define "maintained". Most application devs do not test with unreleased kernels.

I haven't, either, but users started shouting: we have nowadays plenty of distributions shipping unreleased kernels.

In OVS we typically don't get any bug reports from users until the issue hits a major distribution like RHEL or Ubuntu.

Slightly different crowd, I guess. :)

Best regards, Ilya Maximets.

#14 lore raw
In reply to Jakub Kicinski
This patch was applied to netdev/net.git (main) by David S. Miller <[email protected]>:

Hello:

This patch was applied to netdev/net.git (main)
by David S. Miller <[email protected]>:

Quoted 6 lines · On Thu, 11 Apr 2024 11:02:02 -0700 you wrote:

Commit under Fixes optimized the number of recv() calls needed during RTM_GETROUTE dumps, but we got multiple reports of applications hanging on recv() calls. Applications expect that a route dump will be terminated with a recv() reading an individual NLM_DONE message.

Quoted 7 lines

Coalescing NLM_DONE is perfectly legal in netlink, but even tho reporters fixed the code in respective projects, chances are it will take time for those applications to get updated. So revert to old behavior (for now)?

[...]

Here is the summary with links:
- [net] inet: bring NLM_DONE out to a separate recv() again

    https://git.kernel.org/netdev/net/c/460b0d33cf10

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
#4 lore raw
In reply to Ilya Maximets
Quoted 73 lines · Ilya Maximets
Quoted 72 lines · Jakub Kicinski

Commit under Fixes optimized the number of recv() calls needed during RTM_GETROUTE dumps, but we got multiple reports of applications hanging on recv() calls. Applications expect that a route dump will be terminated with a recv() reading an individual NLM_DONE message.

Coalescing NLM_DONE is perfectly legal in netlink, but even tho reporters fixed the code in respective projects, chances are it will take time for those applications to get updated. So revert to old behavior (for now)?

Old kernel (5.19):

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 692 bytes, 11 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
 Recv: read 20 bytes, 1 messages
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

Before (6.9-rc2):

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 712 bytes, 12 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

After:

 $ ./cli.py --dbg-small-recv 4096 --spec netlink/specs/rt_route.yaml \
            --dump getroute --json '{"rtm-family": 2}'
 Recv: read 692 bytes, 11 messages
   nl_len = 68 (52) nl_flags = 0x22 nl_type = 24
 ...
   nl_len = 60 (44) nl_flags = 0x22 nl_type = 24
 Recv: read 20 bytes, 1 messages
   nl_len = 20 (4) nl_flags = 0x2 nl_type = 3

---
CC: [email protected]
CC: [email protected]

---
 net/ipv4/fib_frontend.c | 5 +++++
 1 file changed, 5 insertions(+)
net/ipv4/fib_frontend.c +5 -0
index 48741352a88a..c484b1c0fc00 100644
@@ -1050,6 +1050,11 @@ static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb)
 			e++;
 		}
 	}
+
+	/* Don't let NLM_DONE coalesce into a message, even if it could.
+	 * Some user space expects NLM_DONE in a separate recv().
+	 */
+	err = skb->len;
 out:
 
Quoted 7 lines

net/ipv4/fib_frontend.c static int inet_dump_fib(struct sk_buff *skb, struct netlink_callback *cb) · line 1060

Quoted 1 line
net/ipv4/fib_frontend.c +0 -0
 	cb->args[1] = e;

FWIW, on current net-next this fixes the issue with Libreswan and IPv4 (IPv6 issue remains, obviously). I also did a round of other OVS system tests and they worked fine.

Hi, Jakub. Now that IPv6 change is in 6.10-rc, do you plan to submit a similar fix for it as well? (Sorry if I missed it.) Libreswan is getting stuck on IPv6 route lookups with 6.10-rc4.

Note: Libreswan fixed the issue on their main branch, but it is not available in any release yet, and I'm not sure if the fix is going to make it into stable releases.

Best regards, Ilya Maximets.

author maintainer
#5 lore raw
Quoted 6 lines · On Mon, 17 Jun 2024 17:09:44 +0200 Ilya Maximets wrote:
Quoted 5 lines

FWIW, on current net-next this fixes the issue with Libreswan and IPv4 (IPv6 issue remains, obviously). I also did a round of other OVS system tests and they worked fine.

Quoted 7 lines

Hi, Jakub. Now that IPv6 change is in 6.10-rc, do you plan to submit a similar fix for it as well? (Sorry if I missed it.) Libreswan is getting stuck on IPv6 route lookups with 6.10-rc4.

Note: Libreswan fixed the issue on their main branch, but it is not available in any release yet, and I'm not sure if the fix is going to make it into stable releases.

Sorry for the delay, we'll get it resolved before this week's PR.

#6 lore raw
Quoted 11 lines · Jakub Kicinski
Quoted 10 lines · On Mon, 17 Jun 2024 17:09:44 +0200 Ilya Maximets wrote:
Quoted 5 lines

FWIW, on current net-next this fixes the issue with Libreswan and IPv4 (IPv6 issue remains, obviously). I also did a round of other OVS system tests and they worked fine.

Hi, Jakub. Now that IPv6 change is in 6.10-rc, do you plan to submit a similar fix for it as well? (Sorry if I missed it.) Libreswan is getting stuck on IPv6 route lookups with 6.10-rc4.

Quoted 5 lines
Quoted 3 lines

Note: Libreswan fixed the issue on their main branch, but it is not available in any release yet, and I'm not sure if the fix is going to make it into stable releases.

Sorry for the delay, we'll get it resolved before this week's PR.

Ack. Thanks!

Best regards, Ilya Maximets.