Skip to contentSkip to conversation

doc: Document that kernel may accept unimplemented expressions

PATCH In review Topi Miettinen 3 participants 8 messages Apr 9, 2022 – Apr 10, 2022 netfilter

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 doc: Document that kernel may accept unimplemented expressions

Kernel silently accepts input chain filters using meta skuid, meta skgid, meta cgroup or socket cgroupv2 expressions but they don't work yet. Warn the users of this possibility.

---
 doc/nft.txt | 5 +++++
 1 file changed, 5 insertions(+)
doc/nft.txt +5 -0
index f7a53ac9..4820b4ae 100644
@@ -932,6 +932,11 @@ filter output oif wlan0
 ^^^^^^^^^^^^^^^^^^^^^^^
 ---------------------------------
 
+Note that the kernel may accept expressions without errors even if it
+doesn't implement the feature. For example, input chain filters using
+*meta skuid*, *meta skgid*, *meta cgroup* or *socket cgroupv2*
+expressions are silently accepted but they don't work yet.
+
 EXIT STATUS
 -----------
 On success, nft exits with a status of 0. Unspecified errors cause it to exit

base-commit: 6fa4ff56385831f01bd9d993178969a4eddbcdbf

Signature · 1 line
2.35.1
maintainer
#1 lore raw
Quoted 14 lines · Topi Miettinen

Kernel silently accepts input chain filters using meta skuid, meta skgid, meta cgroup or socket cgroupv2 expressions but they don't work yet. Warn the users of this possibility.

---
 doc/nft.txt | 5 +++++
 1 file changed, 5 insertions(+)
doc/nft.txt +0 -0
index f7a53ac9..4820b4ae 100644

doc/nft.txt filter output oif wlan0 · line 938

Quoted 7 lines
doc/nft.txt +4 -0
@@ -932,6 +932,11 @@ filter output oif wlan0
 ^^^^^^^^^^^^^^^^^^^^^^^
 ---------------------------------
 
+Note that the kernel may accept expressions without errors even if it
+doesn't implement the feature. For example, input chain filters using
+*meta skuid*, *meta skgid*, *meta cgroup* or *socket cgroupv2*
+expressions are silently accepted but they don't work yet.

Thats not correct.

Those expressions load values from skb->sk, i.e. the socket associated with the packet.

In input, such socket may exist, either because of tproxy rules, early demux, or bpf programs.

#2 lore raw
Quoted 25 lines · Florian Westphal
Quoted 14 lines · Topi Miettinen

Kernel silently accepts input chain filters using meta skuid, meta skgid, meta cgroup or socket cgroupv2 expressions but they don't work yet. Warn the users of this possibility.

---
  doc/nft.txt | 5 +++++
  1 file changed, 5 insertions(+)
doc/nft.txt +0 -0
index f7a53ac9..4820b4ae 100644

doc/nft.txt filter output oif wlan0 · line 938

Quoted 7 lines
doc/nft.txt +4 -0
@@ -932,6 +932,11 @@ filter output oif wlan0
  ^^^^^^^^^^^^^^^^^^^^^^^
  ---------------------------------
 
+Note that the kernel may accept expressions without errors even if it
+doesn't implement the feature. For example, input chain filters using
+*meta skuid*, *meta skgid*, *meta cgroup* or *socket cgroupv2*
+expressions are silently accepted but they don't work yet.

Thats not correct.

Quoted 5 lines

Those expressions load values from skb->sk, i.e. the socket associated with the packet.

In input, such socket may exist, either because of tproxy rules, early demux, or bpf programs.

Thanks. How about:
Note that the kernel may accept expressions without errors even if it
doesn't implement the feature. For example, input chain filters using
*meta skuid*, *meta skgid*, *meta cgroup* or *socket cgroupv2*
expressions are silently accepted but they don't work yet, except when used with *tproxy* rules, early demultiplexing or BPF programs.

Could you please explain this 'early demux' concept? Is this something that can be triggered with NFT rules, kernel configuration etc? I can't find any documentation.

-Topi

maintainer
#3 lore raw
Quoted 5 lines · Topi Miettinen

Note that the kernel may accept expressions without errors even if it doesn't implement the feature. For example, input chain filters using *meta skuid*, *meta skgid*, *meta cgroup* or *socket cgroupv2* expressions are silently accepted but they don't work yet, except when used with *tproxy* rules, early demultiplexing or BPF programs.

This is what iptables-extensions(8) says:

IMPORTANT: when being used in the INPUT chain, the cgroup matcher is currently only of limited functionality, meaning it will only match on packets that are processed for local sockets through early socket demuxing. Therefore, general usage on the INPUT chain is not advised unless the implications are well understood.

For nftables, this is true for all meta types that use skb->sk internally, such as skuid, skgid, cgroup, ...

Quoted 3 lines

Could you please explain this 'early demux' concept? Is this something that can be triggered with NFT rules, kernel configuration etc? I can't find any documentation.

sysctl.
net.ipv4.ip_early_demux = 1
net.ipv4.tcp_early_demux = 1
net.ipv4.udp_early_demux = 1

This is a performance optimization, tcp edemux only works for sockets in established state, udp demux has restrictions as well.

So, no guarantee that this will set the socket reliably, hence the paragraph in the iptables-extensions manpage.

#4 lore raw
Quoted 24 lines · Florian Westphal
Quoted 5 lines · Topi Miettinen

Note that the kernel may accept expressions without errors even if it doesn't implement the feature. For example, input chain filters using *meta skuid*, *meta skgid*, *meta cgroup* or *socket cgroupv2* expressions are silently accepted but they don't work yet, except when used with *tproxy* rules, early demultiplexing or BPF programs.

This is what iptables-extensions(8) says:

IMPORTANT: when being used in the INPUT chain, the cgroup matcher is currently only of limited functionality, meaning it will only match on packets that are processed for local sockets through early socket demuxing. Therefore, general usage on the INPUT chain is not advised unless the implications are well understood.

For nftables, this is true for all meta types that use skb->sk internally, such as skuid, skgid, cgroup, ...

Quoted 3 lines

Could you please explain this 'early demux' concept? Is this something that can be triggered with NFT rules, kernel configuration etc? I can't find any documentation.

sysctl.
net.ipv4.ip_early_demux = 1
net.ipv4.tcp_early_demux = 1
net.ipv4.udp_early_demux = 1

Quoted 5 lines

This is a performance optimization, tcp edemux only works for sockets in established state, udp demux has restrictions as well.

So, no guarantee that this will set the socket reliably, hence the paragraph in the iptables-extensions manpage.

Thanks. From this blog post I suppose the problem is that NFT rules aren't checked after final demux:
https://www.privateinternetaccess.com/blog/linux-networking-stack-from-the-ground-up-part-4-2/

Would it be possible to add such checks in the future?

What about:

Note that the kernel may accept expressions without errors even if it doesn't implement the feature. For example, input chain filters using expressions such as *meta skuid*, *meta skgid*, *meta cgroup* or *socket cgroupv2* are silently accepted but they don't work reliably yet, except when used with *tproxy* rules, early demultiplexing (available only for TCP for sockets in established state and UDP demux has restrictions as well) or BPF programs.

-Topi

maintainer
#5 lore raw
Quoted 1 line · Topi Miettinen

Would it be possible to add such checks in the future?

We could add socket skuid, socket skgid, its not hard.

Quoted 3 lines

Note that the kernel may accept expressions without errors even if it doesn't implement the feature. For example, input chain filters using expressions such as *meta skuid*, *meta skgid*, *meta cgroup* or

Those can not be made to work.

Quoted 1 line

*socket cgroupv2* are silently accepted but they don't work reliably

socket should work, at least for tcp and udp.
The cgroupv2 is buggy. I sent a patch, feel free to test it.

#6 lore raw
Quoted 4 lines · Florian Westphal
Quoted 1 line · Topi Miettinen

Would it be possible to add such checks in the future?

We could add socket skuid, socket skgid, its not hard.

That would be nice. Could the syntax still remain 'meta skuid' even though the credentials come from a socket for compatibility?

Quoted 10 lines
Quoted 3 lines

Note that the kernel may accept expressions without errors even if it doesn't implement the feature. For example, input chain filters using expressions such as *meta skuid*, *meta skgid*, *meta cgroup* or

Those can not be made to work.

Quoted 1 line

*socket cgroupv2* are silently accepted but they don't work reliably

socket should work, at least for tcp and udp.
The cgroupv2 is buggy. I sent a patch, feel free to test it.

Once the patch is applied, the warnings in manual page wrt. cgroupv2 would only apply to old kernels. How about the following:

Note that different kernel versions may accept expressions without errors even if they don't implement the feature. For example, input chain filters using expressions such as *meta skuid*, *meta skgid*, *meta cgroup* or *socket cgroupv2* are silently accepted but they may not work reliably or at all.

-Topi

#7 lore raw
Quoted 20 lines · Topi Miettinen
Quoted 4 lines · Florian Westphal
Quoted 1 line · Topi Miettinen

Would it be possible to add such checks in the future?

We could add socket skuid, socket skgid, its not hard.

That would be nice. Could the syntax still remain 'meta skuid' even though the credentials come from a socket for compatibility?

Quoted 10 lines
Quoted 3 lines

Note that the kernel may accept expressions without errors even if it doesn't implement the feature. For example, input chain filters using expressions such as *meta skuid*, *meta skgid*, *meta cgroup* or

Those can not be made to work.

Quoted 1 line

*socket cgroupv2* are silently accepted but they don't work reliably

socket should work, at least for tcp and udp.
The cgroupv2 is buggy. I sent a patch, feel free to test it.

Quoted 8 lines

Once the patch is applied, the warnings in manual page wrt. cgroupv2 would only apply to old kernels. How about the following:

Note that different kernel versions may accept expressions without errors even if they don't implement the feature. For example, input chain filters using expressions such as *meta skuid*, *meta skgid*, *meta cgroup* or *socket cgroupv2* are silently accepted but they may not work reliably or at all.

Wrt this fix, it will be passed to -stable.

Regarding general use of socket match from input: Probably more documentation on what kind of sockets early demux is actually being attached to might help understand how this is working.