Skip to contentSkip to conversation

list hooks refactoring

PATCH Applied Florian Westphal 3 participants 16 messages Jul 26, 2024 – Aug 19, 2024

Subsystems (2)
lore
Share & tools
lore

Florian Westphal

author maintainer
#0 lore raw

cover list hooks refactoring

First patch adds a bit more documentation for
"nft list hooks".

2nd patch removes decnet support, this was removed 2 years ago from the kernel.

3rd patch fixes 'nft list hooks device foo' from listing hooks for 'foo' as 'bridge' family instead of 'netdev'.

Last patch adds support for egress hooks, this was missing so far.

Florian Westphal (4):
doc: add documentation about list hooks feature
src: remove decnet support
src: mnl: clean up hook listing code
src: add egress support for 'list hooks'

 Makefile.am                      |   2 +-
 doc/additional-commands.txt      | 115 +++++++++++++++++++++++++++
 doc/nft.txt                      |  63 +--------------
 include/linux/netfilter_decnet.h |  72 -----------------
 src/mnl.c                        | 132 +++++++++----------------------
 5 files changed, 156 insertions(+), 228 deletions(-)
 create mode 100644 doc/additional-commands.txt
 delete mode 100644 include/linux/netfilter_decnet.h
Signature · 1 line
2.44.2
author maintainer
#1 lore raw

PATCH 1/4 doc: add documentation about list hooks feature

Add a brief segment about 'nft list hooks' and a summary of the output format.

As nft.txt is quite large, split the additonal commands into their own file.

The existing listing section is removed; list subcommand is already mentioned in the relevant statement sections.

---
 Makefile.am                 |   1 +
 doc/additional-commands.txt | 115 ++++++++++++++++++++++++++++++++++++
 doc/nft.txt                 |  63 +-------------------
 3 files changed, 117 insertions(+), 62 deletions(-)
 create mode 100644 doc/additional-commands.txt
Makefile.am +1 -0
index 9088170bfc68..ef198dafcbc8 100644
@@ -322,6 +322,7 @@ A2X_OPTS_MANPAGE = \
 ASCIIDOC_MAIN = doc/nft.txt
 
 ASCIIDOC_INCLUDES = \
+	doc/additional-commands.txt \
 	doc/data-types.txt \
 	doc/payload-expression.txt \
 	doc/primary-expression.txt \
doc/additional-commands.txt new file +115 -0

115 lines not on this page · raw message

doc/nft.txt +1 -62

75 lines not on this page · raw message

Signature · 1 line
2.44.2
#2 lore raw

Re: [PATCH nft 1/4] doc: add documentation about list hooks feature

Quoted 42 lines · Florian Westphal

Add a brief segment about 'nft list hooks' and a summary of the output format.

As nft.txt is quite large, split the additonal commands into their own file.

The existing listing section is removed; list subcommand is already mentioned in the relevant statement sections.

---
 Makefile.am                 |   1 +
 doc/additional-commands.txt | 115 ++++++++++++++++++++++++++++++++++++
 doc/nft.txt                 |  63 +-------------------
 3 files changed, 117 insertions(+), 62 deletions(-)
 create mode 100644 doc/additional-commands.txt
Makefile.am +1 -0
index 9088170bfc68..ef198dafcbc8 100644
@@ -322,6 +322,7 @@ A2X_OPTS_MANPAGE = \
 ASCIIDOC_MAIN = doc/nft.txt
 
 ASCIIDOC_INCLUDES = \
+	doc/additional-commands.txt \
 	doc/data-types.txt \
 	doc/payload-expression.txt \
 	doc/primary-expression.txt \
doc/additional-commands.txt new file +6 -0
new file mode 100644
index 000000000000..dd1b3d2d87d4
@@ -0,0 +1,115 @@
+LIST HOOKS
+~~~~~~~~~~
+
+This shows the low-level netfilter processing pipeline, including
+functions registered by kernel modules such as nf_conntrack. +
+

doc/additional-commands.txt line 16

Quoted 10 lines
doc/additional-commands.txt new file +10 -0
+[verse]
+____
+*list hooks* ['family']
+*list hooks netdev device* 'DEVICE_NAME'
+____
+
+*list hooks* is enough to display everything that is active
+on the system, however, it does currently omit hooks that are
+tied to a specific network device (netdev family). To obtain
+those, the network device needs to be queried by name.

IIRC, the idea is to display the ingress path pipeline according to the device (if specified)

list hooks netdev eth0

as for egress, as it is not possible to know where the packet is going, it is probably good to allow the user to specify the output device, so it gets the entire pipeline for ingress and egress paths, ie.

list hooks netdev eth0 eth1

Note that this is not implemented. This has limitations, discovering eth{0,1} belongs to bridge device would need more work (not asking to do this now, but it could be a nice usability feature to discover the pipeline?).

author maintainer
#3 lore raw

Re: [PATCH nft 1/4] doc: add documentation about list hooks feature

Quoted 8 lines · Pablo Neira Ayuso

doc/additional-commands.txt line 16

Quoted 4 lines
doc/additional-commands.txt new file +4 -0
+*list hooks* is enough to display everything that is active
+on the system, however, it does currently omit hooks that are
+tied to a specific network device (netdev family). To obtain
+those, the network device needs to be queried by name.

IIRC, the idea is to display the ingress path pipeline according to the device (if specified)

Quoted 8 lines

list hooks netdev eth0

as for egress, as it is not possible to know where the packet is going, it is probably good to allow the user to specify the output device, so it gets the entire pipeline for ingress and egress paths, ie.

list hooks netdev eth0 eth1

Not really, why would eth0 and eth1 be related here?

What would make more sense to me is to allow

list hooks netdev

and then have nft fetch list of all network devices and then query them all.

If a packet coming in on devX will be forwarded to devY depends on the type of packet and the configuration, e.g. arp/ip vs. bridge/routing or even encapsulation...

Quoted 4 lines

Note that this is not implemented. This has limitations, discovering eth{0,1} belongs to bridge device would need more work (not asking to do this now, but it could be a nice usability feature to discover the pipeline?).

Bridge? I don't think we have bridge family support for netdev hooks? AFAIU its only netdev and inet.

This thing should only list the nf hooks registered for the device, and not start to guess. So for "list hooks br0", return ingress and egress hooks for the virtual device, not the bridge ports.

#4 lore raw

Re: [PATCH nft 1/4] doc: add documentation about list hooks feature

Hi Florian,

Quoted 11 lines · Florian Westphal
Quoted 10 lines · Pablo Neira Ayuso

doc/additional-commands.txt line 16

Quoted 4 lines
doc/additional-commands.txt new file +4 -0
+*list hooks* is enough to display everything that is active
+on the system, however, it does currently omit hooks that are
+tied to a specific network device (netdev family). To obtain
+those, the network device needs to be queried by name.

IIRC, the idea is to display the ingress path pipeline according to the device (if specified)

list hooks netdev eth0

Quoted 8 lines
Quoted 6 lines

as for egress, as it is not possible to know where the packet is going, it is probably good to allow the user to specify the output device, so it gets the entire pipeline for ingress and egress paths, ie.

list hooks netdev eth0 eth1

Not really, why would eth0 and eth1 be related here?

Note that you can specify:

list hooks ip device enp0s25

this shows the hooks that will be exercised for a given packet family, ie. IPv4 packets will exercise the following hooks.

family ip {
        hook ingress {
                 0000000000 chain netdev x y [nf_tables]
        }
        hook prerouting {
                -0000000400 ipv4_conntrack_defrag [nf_defrag_ipv4]
                -0000000200 ipv4_conntrack_in [nf_conntrack]
        }
        hook input {
                 0000000000 chain ip filter in [nf_tables]
                +2147483647 nf_confirm [nf_conntrack]
        }
        hook forward {
                -0000000225 selinux_ip_forward
        }
        hook output {
                -0000000400 ipv4_conntrack_defrag [nf_defrag_ipv4]
                -0000000225 selinux_ip_output
                -0000000200 ipv4_conntrack_local [nf_conntrack]
        }
        hook postrouting {
                +0000000225 selinux_ip_postroute
                +2147483647 nf_confirm [nf_conntrack]
        }
}

This is _not_ showing the list of hooks for a given family.

What I meant is that user could filter out by ingress and egress device to fetch the hooks that are traversed in such case, ie.

list hooks ip iifname eth0 oifname eth1

to get the traversal of hooks for IPv4 packets, assuming eth0 as ingress device and eth1 as egress device.

Quoted 6 lines

What would make more sense to me is to allow

list hooks netdev

and then have nft fetch list of all network devices and then query them all.

Makes sense, it currently fails with EINVAL because no device has been specified.

Quoted 9 lines

If a packet coming in on devX will be forwarded to devY depends on the type of packet and the configuration, e.g. arp/ip vs. bridge/routing or even encapsulation...

Quoted 4 lines

Note that this is not implemented. This has limitations, discovering eth{0,1} belongs to bridge device would need more work (not asking to do this now, but it could be a nice usability feature to discover the pipeline?).

Quoted 6 lines

Bridge? I don't think we have bridge family support for netdev hooks? AFAIU its only netdev and inet.

This thing should only list the nf hooks registered for the device, and not start to guess. So for "list hooks br0", return ingress and egress hooks for the virtual device, not the bridge ports.

OK

author maintainer
#5 lore raw

Re: [PATCH nft 1/4] doc: add documentation about list hooks feature

Quoted 35 lines · Pablo Neira Ayuso
Quoted 1 line

Not really, why would eth0 and eth1 be related here?

Note that you can specify:

list hooks ip device enp0s25

this shows the hooks that will be exercised for a given packet family, ie. IPv4 packets will exercise the following hooks.

family ip {
        hook ingress {
                 0000000000 chain netdev x y [nf_tables]
        }
        hook prerouting {
                -0000000400 ipv4_conntrack_defrag [nf_defrag_ipv4]
                -0000000200 ipv4_conntrack_in [nf_conntrack]
        }
        hook input {
                 0000000000 chain ip filter in [nf_tables]
                +2147483647 nf_confirm [nf_conntrack]
        }
        hook forward {
                -0000000225 selinux_ip_forward
        }
        hook output {
                -0000000400 ipv4_conntrack_defrag [nf_defrag_ipv4]
                -0000000225 selinux_ip_output
                -0000000200 ipv4_conntrack_local [nf_conntrack]
        }
        hook postrouting {
                +0000000225 selinux_ip_postroute
                +2147483647 nf_confirm [nf_conntrack]
        }
}
Quoted 1 line

This is _not_ showing the list of hooks for a given family.

I now realize that whats in the tree today is not what I wrote originally. So this is neither showing the hooks that will be execrised (packet can't be input and forward...). But ok. I don't know what to do now.

Quoted 7 lines

What I meant is that user could filter out by ingress and egress device to fetch the hooks that are traversed in such case, ie.

list hooks ip iifname eth0 oifname eth1

to get the traversal of hooks for IPv4 packets, assuming eth0 as ingress device and eth1 as egress device.

No idea how to make this, or I fail to understand.

Quoted 9 lines
Quoted 6 lines

What would make more sense to me is to allow

list hooks netdev

and then have nft fetch list of all network devices and then query them all.

Makes sense, it currently fails with EINVAL because no device has been specified.

I'll try to add it, but I don't know if I should toss these patches first or not :/

#6 lore raw

Re: [PATCH nft 1/4] doc: add documentation about list hooks feature

Quoted 36 lines · Florian Westphal
Quoted 35 lines · Pablo Neira Ayuso
Quoted 1 line

Not really, why would eth0 and eth1 be related here?

Note that you can specify:

list hooks ip device enp0s25

this shows the hooks that will be exercised for a given packet family, ie. IPv4 packets will exercise the following hooks.

family ip {
        hook ingress {
                 0000000000 chain netdev x y [nf_tables]
        }
        hook prerouting {
                -0000000400 ipv4_conntrack_defrag [nf_defrag_ipv4]
                -0000000200 ipv4_conntrack_in [nf_conntrack]
        }
        hook input {
                 0000000000 chain ip filter in [nf_tables]
                +2147483647 nf_confirm [nf_conntrack]
        }
        hook forward {
                -0000000225 selinux_ip_forward
        }
        hook output {
                -0000000400 ipv4_conntrack_defrag [nf_defrag_ipv4]
                -0000000225 selinux_ip_output
                -0000000200 ipv4_conntrack_local [nf_conntrack]
        }
        hook postrouting {
                +0000000225 selinux_ip_postroute
                +2147483647 nf_confirm [nf_conntrack]
        }
}
Quoted 3 lines
Quoted 1 line

This is _not_ showing the list of hooks for a given family.

I now realize that whats in the tree today is not what I wrote originally.

We agreed to change it.

Quoted 2 lines

So this is neither showing the hooks that will be execrised (packet can't be input and forward...). But ok. I don't know what to do now.

As it is not possible to know where the packet is going (input / forward), this listing represents what hooks can be potentially exercised for such packet family, hence, netdev hooks are included in the ip family view.

If you don't like this behaviour and prefer a strict view per hook family it should be possible to revisit. User will have to get all the pieces together to understand what the hook pipeline is instead. This command has not been documented so far.

Quoted 9 lines
Quoted 7 lines

What I meant is that user could filter out by ingress and egress device to fetch the hooks that are traversed in such case, ie.

list hooks ip iifname eth0 oifname eth1

to get the traversal of hooks for IPv4 packets, assuming eth0 as ingress device and eth1 as egress device.

No idea how to make this, or I fail to understand.

Not important for this series, this can be extended later.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

What would make more sense to me is to allow

list hooks netdev

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

and then have nft fetch list of all network devices and then query them all.

Makes sense, it currently fails with EINVAL because no device has been specified.

I'll try to add it, but I don't know if I should toss these patches first or not :/

I think patchset looks fine, but documentation update needs a revamp if you agree in the existing behaviour.

author maintainer
#7 lore raw

Re: [PATCH nft 1/4] doc: add documentation about list hooks feature

Quoted 37 lines · Pablo Neira Ayuso
Quoted 36 lines · Florian Westphal
Quoted 35 lines · Pablo Neira Ayuso
Quoted 1 line

Not really, why would eth0 and eth1 be related here?

Note that you can specify:

list hooks ip device enp0s25

this shows the hooks that will be exercised for a given packet family, ie. IPv4 packets will exercise the following hooks.

family ip {
        hook ingress {
                 0000000000 chain netdev x y [nf_tables]
        }
        hook prerouting {
                -0000000400 ipv4_conntrack_defrag [nf_defrag_ipv4]
                -0000000200 ipv4_conntrack_in [nf_conntrack]
        }
        hook input {
                 0000000000 chain ip filter in [nf_tables]
                +2147483647 nf_confirm [nf_conntrack]
        }
        hook forward {
                -0000000225 selinux_ip_forward
        }
        hook output {
                -0000000400 ipv4_conntrack_defrag [nf_defrag_ipv4]
                -0000000225 selinux_ip_output
                -0000000200 ipv4_conntrack_local [nf_conntrack]
        }
        hook postrouting {
                +0000000225 selinux_ip_postroute
                +2147483647 nf_confirm [nf_conntrack]
        }
}
Quoted 5 lines
Quoted 3 lines
Quoted 1 line

This is _not_ showing the list of hooks for a given family.

I now realize that whats in the tree today is not what I wrote originally.

We agreed to change it.

Sorry. I'm old and do not remember.
TL;DR: I think we should revert to strict behaviour, i.e. nft list hooks foo

queries hooks registered as NFPROTO_FOO.

NFPROTO_INET expands to shorthand for 'list hooks ip and ip6'.

AFAICS the problem is that ip, ip6 and arp are both NFPROTO_ values (protocol families), but also l3 protocols, whereas netdev and bridge are only families.

Hence, what to do on bridge becomes murky, there is just a large number of possible l3 protocols that can pass through these hooks.

Netdev is simple because its scoped only to one single interface.

Sorry for the wall of text below, but maybe it helps to understand why i think the above (no-guesses, treat strictly as nfproto) makes sense.

Quoted 7 lines
Quoted 2 lines

So this is neither showing the hooks that will be execrised (packet can't be input and forward...). But ok. I don't know what to do now.

As it is not possible to know where the packet is going (input / forward), this listing represents what hooks can be potentially exercised for such packet family, hence, netdev hooks are included in the ip family view.

I was confused because it lists netdev hooks when asking for bridge, and when I made this facility my intent was for this to return hooks that were registered with NFPROTO_$QUERY.

So i did not expect to see netdev hooks.

list hooks arp device eth0
my expectation:
1. no output OR
2. NFPROTO_ARP in and output hooks OR
3. error, e.g. "device not supported for arp family".

But I get this instead:

  family arp {
          hook ingress {
                   0000000000 chain netdev ingress in_public [nf_tables]
          }

}

Which I considered a bug but if we say that the query should return hooks that this protocol family is exposed to, then that might be ok.

Should it list bridge hooks too? Same rationale would apply as to why it shows the netdev family ingress hook:
ARP packets will be picked up by bridge hooks if they arrive on interfaces that serve as a bridge port.

I think that original code was simpler, less guesswork on userspace side, what-you-ask-for-is-what-you-get, i.e. for "list hooks arp" shows only hooks registered with NFPROTO_ARP on kernel side, and nothing else.

For "list hooks arp device eth0", we could either return an error, or pass this to kernel in case we gain arp+device at some point, even though i don't see why anyone would add that.

inet ingress is already awkward in my opinion, I'm not sure why it got added.

Quoted 4 lines

If you don't like this behaviour and prefer a strict view per hook family it should be possible to revisit. User will have to get all the pieces together to understand what the hook pipeline is instead. This command has not been documented so far.

Yes. In theory it should be possible to do this, so to go with arp example:

nft list hooks arp device foo

would list:

1. netdev ingress and egress hook for the queried device
2. arp in and output hooks (there is no forward for arp)
3. bridge pre,in,forward,output and postrouting

but does that make sense? I don't think so, because it gets more complicated for bridge query:

nft list hooks bridge

1. can't show netdev, we lack device to query -> query all interfaces? But why would one clutter output with netdev in/egress when bridge was asked for?
2. show pre/in/fwd/out/postrouting NFPROTO_BRIDGE hooks
3. should it show ip/ip6 hooks? They could be relevant in case of broute expression in nftables.
ip/ip6 Output+postrouting are travesed in case of local packets.

and it would have to show arp hooks, no? for arp packet, we might pass them up to local stack. Local arp query pass through arp:output.

So I'm just worried that this gets complicated/murky as to what is shown (and what is omitted). For bridge, we would have to end up dumping everything and treat it like 'list hooks'....

I do admit that it would be nice to have something like:

nft list pipeline meta input eth0 ip saddr 1.2.3.4 ip daddr 5.6.7.8

and then have nft:
1. list NFPROTO_NETDEV for eth0 ingress only (no egress)
2. list NFPROTO_INET hooks for ingress eth0
3. list NFPROTO_IPV4 hooks for prerouting
4. query FIB to get output device for 1.2.3.4->5.6.7.8
5. list NFPROTO_IPV4 for EITHER input or forward+postrouting
6. for forward case, list NFPROTO_NETDEV egress hooks for the fib-derviced output device

But thats hard, because there might be rules in place that alter ip daddr or ip saddr, or packet mark etc etc so the pipeline shown can be a wrong.

And bridge needs extra work, we need to figure out if eth0 is a bridge port and then dump NFPROTO_BRIDGE.
And query l2 address of the ip address to query bridge fib to figure out of this enters local delivery or forwarding case, or both...

Or, require use of 'ether daddr' for bridge. But still, its a lot of work to make this.

Quoted 2 lines

I think patchset looks fine, but documentation update needs a revamp if you agree in the existing behaviour.

I agree that we first need to figure out what 'nft list hooks xxxx' should do. I would prefer 'no guesses' approach, i.e.

1. nft list hooks
dump everything EXCEPT netdev families/devices
2. nft list hooks netdev device foo
dump ingress/egress netdev hooks,
INCLUDING inet ingress (its scoped to the device).
3. nft list hooks arp
dump arp input and output, if any
4. nft list hooks bridge
dump bridge pre/input/forward/out/postrouting
5. nft list hooks ip
dump ip pre/input/forward/out/postrouting
6. nft list hooks ip6 -> see above
7. nft list hooks inet -> alias for dumping both ip+ip6 hooks.

This also means that i'd make 'inet device foo' return a warning that the device will be ignored because "inet ingress" is syntactic frontend sugar similar to inet pseudo-family.

We could try the 'detect pipeline' later. But, as per example above, I don't think its easy, unless one omits all packet rewrites (stateless nat, dnat, redirect) and everything else that causes re-routing, e.g. policy routing, tproxy, tc(x), etc.

And then there is l3 and l2 multicasting...

But, admittingly, it might be nice to have something and it might be good enough if pipeline alterations by ruleset and other entities are omitted.

#8 lore raw

Re: [PATCH nft 1/4] doc: add documentation about list hooks feature

On Mon, Jul 29, 2024 at 05:32:11PM +0200, Florian Westphal wrote:
[...]

Quoted 2 lines

TL;DR: I think we should revert to strict behaviour, i.e. nft list hooks foo

Agreed.

Quoted 8 lines

queries hooks registered as NFPROTO_FOO.

NFPROTO_INET expands to shorthand for 'list hooks ip and ip6'.

AFAICS the problem is that ip, ip6 and arp are both NFPROTO_ values (protocol families), but also l3 protocols, whereas netdev and bridge are only families.

Quoted 7 lines

Hence, what to do on bridge becomes murky, there is just a large number of possible l3 protocols that can pass through these hooks.

Netdev is simple because its scoped only to one single interface.

Sorry for the wall of text below, but maybe it helps to understand why i think the above (no-guesses, treat strictly as nfproto) makes sense.

Makes sense, thanks for explaining.

[...]

Quoted 48 lines
Quoted 4 lines

If you don't like this behaviour and prefer a strict view per hook family it should be possible to revisit. User will have to get all the pieces together to understand what the hook pipeline is instead. This command has not been documented so far.

Yes. In theory it should be possible to do this, so to go with arp example:

nft list hooks arp device foo

would list:

1. netdev ingress and egress hook for the queried device
2. arp in and output hooks (there is no forward for arp)
3. bridge pre,in,forward,output and postrouting

but does that make sense? I don't think so, because it gets more complicated for bridge query:

nft list hooks bridge

1. can't show netdev, we lack device to query -> query all interfaces? But why would one clutter output with netdev in/egress when bridge was asked for?
2. show pre/in/fwd/out/postrouting NFPROTO_BRIDGE hooks
3. should it show ip/ip6 hooks? They could be relevant in case of broute expression in nftables.
ip/ip6 Output+postrouting are travesed in case of local packets.

and it would have to show arp hooks, no? for arp packet, we might pass them up to local stack. Local arp query pass through arp:output.

So I'm just worried that this gets complicated/murky as to what is shown (and what is omitted). For bridge, we would have to end up dumping everything and treat it like 'list hooks'....

I do admit that it would be nice to have something like:

nft list pipeline meta input eth0 ip saddr 1.2.3.4 ip daddr 5.6.7.8

and then have nft:
1. list NFPROTO_NETDEV for eth0 ingress only (no egress)
2. list NFPROTO_INET hooks for ingress eth0
3. list NFPROTO_IPV4 hooks for prerouting
4. query FIB to get output device for 1.2.3.4->5.6.7.8
5. list NFPROTO_IPV4 for EITHER input or forward+postrouting
6. for forward case, list NFPROTO_NETDEV egress hooks for the fib-derviced output device

Quoted 3 lines

But thats hard, because there might be rules in place that alter ip daddr or ip saddr, or packet mark etc etc so the pipeline shown can be a wrong.

Yes, providing a pipeline description is a much wider task than just listing hooks. Please, move on with restoring list hooks.

Thanks Florian.

reviewer
#9 lore raw

Re: [PATCH nft 1/4] doc: add documentation about list hooks feature

In reply to Florian Westphal

Hi,

On Mon, Jul 29, 2024 at 05:32:11PM +0200, Florian Westphal wrote:
[...]

Quoted 2 lines

inet ingress is already awkward in my opinion, I'm not sure why it got added.

I suppose the ability to reuse other inet family ruleset parts for ingress was a sought-after feature.

In hindsight, making tables family-agnostic and thus pure "containers" for grouping ruleset parts might be a superior design choice. (Or maybe I just miss the rub in doing that. :)

[...]

Quoted 2 lines

I agree that we first need to figure out what 'nft list hooks xxxx' should do. I would prefer 'no guesses' approach, i.e.

Let me suggest a deviation which seems more user-friendly:

Quoted 2 lines

1. nft list hooks
dump everything EXCEPT netdev families/devices

Include netdev here, make it really list *all* hooks. Iterating over the list of currently existing NICs in this netns is no big deal, is it?

Quoted 3 lines

2. nft list hooks netdev device foo
dump ingress/egress netdev hooks, INCLUDING inet ingress (its scoped to the device).

Drop 'netdev' from the syntax here. The output really is "all hooks specific to that NIC", not necessarily only netdev ones. (And "device" is a distinct identifier for network interfaces in nftables syntax.)

Quoted 8 lines

3. nft list hooks arp
dump arp input and output, if any
4. nft list hooks bridge
dump bridge pre/input/forward/out/postrouting
5. nft list hooks ip
dump ip pre/input/forward/out/postrouting
6. nft list hooks ip6 -> see above
7. nft list hooks inet -> alias for dumping both ip+ip6 hooks.

I wonder if this is more confusing than not - on the netfilter hooks layer, it doesn't quite exist. The only thing which persists a tad longer is inet ingress, indicated by nf_register_net_hook() passing it down to __nf_register_net_hook for nf_hook_entry_head() to return the same value as for netdev ingress. I guess they could be spliced even further up.

Quoted 3 lines

This also means that i'd make 'inet device foo' return a warning that the device will be ignored because "inet ingress" is syntactic frontend sugar similar to inet pseudo-family.

Iff my claim holds true, there is no such thing as an inet hook and thus also no inet device one. :)

Quoted 10 lines

We could try the 'detect pipeline' later. But, as per example above, I don't think its easy, unless one omits all packet rewrites (stateless nat, dnat, redirect) and everything else that causes re-routing, e.g. policy routing, tproxy, tc(x), etc.

And then there is l3 and l2 multicasting...

But, admittingly, it might be nice to have something and it might be good enough if pipeline alterations by ruleset and other entities are omitted.

I seem to recall us discussing something like that on NFWS, but to simulate traversal of a packet with given properties through the ruleset. Which would also identify the hooks involved, I guess?

Cheers, Phil

#10 lore raw

Re: [PATCH nft 1/4] doc: add documentation about list hooks feature

Hi Phil, Florian,

@Florian, could you push out what you have to flush your queue in this front?

Quoted 10 lines · Phil Sutter

On Mon, Jul 29, 2024 at 05:32:11PM +0200, Florian Westphal wrote:
[...]
Let me suggest a deviation which seems more user-friendly:

Quoted 2 lines

1. nft list hooks
dump everything EXCEPT netdev families/devices

Include netdev here, make it really list *all* hooks. Iterating over the list of currently existing NICs in this netns is no big deal, is it?

I like this suggestion.

I think it should be possible to incrementally extend with these suggestions, it should be possible to submit patches for your review.

Quoted 7 lines
Quoted 3 lines

2. nft list hooks netdev device foo
dump ingress/egress netdev hooks, INCLUDING inet ingress (its scoped to the device).

Drop 'netdev' from the syntax here. The output really is "all hooks specific to that NIC", not necessarily only netdev ones. (And "device" is a distinct identifier for network interfaces in nftables syntax.)

I think allowing 'device foo' without family would be good.

Quoted 9 lines
Quoted 8 lines

3. nft list hooks arp
dump arp input and output, if any
4. nft list hooks bridge
dump bridge pre/input/forward/out/postrouting
5. nft list hooks ip
dump ip pre/input/forward/out/postrouting
6. nft list hooks ip6 -> see above
7. nft list hooks inet -> alias for dumping both ip+ip6 hooks.

Quoted 6 lines

I wonder if this is more confusing than not - on the netfilter hooks layer, it doesn't quite exist. The only thing which persists a tad longer is inet ingress, indicated by nf_register_net_hook() passing it down to __nf_register_net_hook for nf_hook_entry_head() to return the same value as for netdev ingress. I guess they could be spliced even further up.

inet is an "alias". I think dumpg both ip+ip6 hooks should be fine.

Quoted 14 lines
Quoted 3 lines

This also means that i'd make 'inet device foo' return a warning that the device will be ignored because "inet ingress" is syntactic frontend sugar similar to inet pseudo-family.

Iff my claim holds true, there is no such thing as an inet hook and thus also no inet device one. :)

Quoted 7 lines

We could try the 'detect pipeline' later. But, as per example above, I don't think its easy, unless one omits all packet rewrites (stateless nat, dnat, redirect) and everything else that causes re-routing, e.g. policy routing, tproxy, tc(x), etc.

And then there is l3 and l2 multicasting...

Quoted 7 lines
Quoted 3 lines

But, admittingly, it might be nice to have something and it might be good enough if pipeline alterations by ruleset and other entities are omitted.

I seem to recall us discussing something like that on NFWS, but to simulate traversal of a packet with given properties through the ruleset. Which would also identify the hooks involved, I guess?

Yes, this all is more complicated and it needs a whole lot of work to add this feature in a complete way, which makes sense to schedule this for the _future_.

author maintainer
#11 lore raw

Re: [PATCH nft 1/4] doc: add documentation about list hooks feature

Quoted 3 lines · Pablo Neira Ayuso

Hi Phil, Florian,

@Florian, could you push out what you have to flush your queue in this front?

OK, I pushed the patches to nftables.git.

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

1. nft list hooks
dump everything EXCEPT netdev families/devices

Include netdev here, make it really list *all* hooks. Iterating over the list of currently existing NICs in this netns is no big deal, is it?

I like this suggestion.

Fail enough, I will send a patch for this later this week.

Quoted 9 lines
Quoted 7 lines
Quoted 3 lines

2. nft list hooks netdev device foo
dump ingress/egress netdev hooks, INCLUDING inet ingress (its scoped to the device).

Drop 'netdev' from the syntax here. The output really is "all hooks specific to that NIC", not necessarily only netdev ones. (And "device" is a distinct identifier for network interfaces in nftables syntax.)

I think allowing 'device foo' without family would be good.

OK, I'm still unclear however because internally only netdev families exist at the device level, so I'm not sure how to represent this.

But dumping the existing network devices and querying them all is not and issue so I will make a patch for this.

author maintainer
#12 lore raw

PATCH 2/4 src: remove decnet support

In reply to Florian Westphal

Removed 2 years ago with v6.1, so we can ditch this from hook list code too.

---
 Makefile.am                      |  1 -
 include/linux/netfilter_decnet.h | 72 --------------------------------
 src/mnl.c                        | 27 ------------
 3 files changed, 100 deletions(-)
 delete mode 100644 include/linux/netfilter_decnet.h
Makefile.am +0 -1
index ef198dafcbc8..fb64105dda88 100644
@@ -57,7 +57,6 @@ noinst_HEADERS = \
 	include/linux/netfilter_arp/arp_tables.h \
 	include/linux/netfilter_bridge.h \
 	include/linux/netfilter_bridge/ebtables.h \
-	include/linux/netfilter_decnet.h \
 	include/linux/netfilter_ipv4.h \
 	include/linux/netfilter_ipv4/ip_tables.h \
 	include/linux/netfilter_ipv6.h \
include/linux/netfilter_decnet.h deleted +0 -72

72 lines not on this page · raw message

src/mnl.c +0 -27
index 9e4bfcd9a030..ec7d2bd5defc 100644
@@ -2581,29 +2581,6 @@ static int mnl_nft_dump_nf_netdev(struct netlink_ctx *ctx, int family, int hook,
 	return err;
 }
 
-static int mnl_nft_dump_nf_decnet(struct netlink_ctx *ctx, int family, int hook,
-				  const char *devname, struct list_head *hook_list,
-				  int *ret)
-{
-	int i, err;
-
-	/* show ingress in first place in hook listing. */
-	err = __mnl_nft_dump_nf_hooks(ctx, family, NFPROTO_NETDEV, NF_NETDEV_INGRESS, devname, hook_list);
-	if (err < 0)
-		*ret = err;
-
-#define NF_DN_NUMHOOKS		7
-	for (i = 0; i < NF_DN_NUMHOOKS; i++) {
-		err = __mnl_nft_dump_nf_hooks(ctx, family, family, i, devname, hook_list);
-		if (err < 0) {
-			*ret = err;
-			return err;
-		}
-	}
-
-	return err;
-}
-
 static void release_hook_list(struct list_head *hook_list)
 {
 	struct basehook *hook, *next;
@@ -2626,7 +2603,6 @@ int mnl_nft_dump_nf_hooks(struct netlink_ctx *ctx, int family, int hook, const c
 		mnl_nft_dump_nf(ctx, NFPROTO_IPV4, hook, devname, &hook_list, &ret);
 		mnl_nft_dump_nf(ctx, NFPROTO_IPV6, hook, devname, &hook_list, &ret);
 		mnl_nft_dump_nf(ctx, NFPROTO_BRIDGE, hook, devname, &hook_list, &ret);
-		mnl_nft_dump_nf_decnet(ctx, NFPROTO_DECNET, hook, devname, &hook_list, &ret);
 		break;
 	case NFPROTO_INET:
 		mnl_nft_dump_nf(ctx, NFPROTO_IPV4, hook, devname, &hook_list, &ret);
@@ -2643,9 +2619,6 @@ int mnl_nft_dump_nf_hooks(struct netlink_ctx *ctx, int family, int hook, const c
 	case NFPROTO_NETDEV:
 		mnl_nft_dump_nf_netdev(ctx, family, hook, devname, &hook_list, &ret);
 		break;
-	case NFPROTO_DECNET:
-		mnl_nft_dump_nf_decnet(ctx, family, hook, devname, &hook_list, &ret);
-		break;
 	}
 
 	switch (family) {
Signature · 1 line
2.44.2
author maintainer Applied
#13 lore raw

Re: [PATCH nft 2/4] src: remove decnet support

Quoted 2 lines · Florian Westphal

Removed 2 years ago with v6.1, so we can ditch this from hook list code too.

I've applied this one as its unrelated to the 'hook list' discussion.

author maintainer
#14 lore raw

PATCH 3/4 src: mnl: clean up hook listing code

In reply to Florian Westphal

mnl_nft_dump_nf_hooks() can call itself for the UNSPEC case, this avoids the second switch/case to handle printing for inet/unspec.

As for the error handling, 'nft list hooks' should not print an error, even if nothing is printed, UNLESS there was also a lowlevel (syscall) error from the kernel.

We don't want to indicate failure just because e.g. kernel doesn't support NFPROTO_ARP.

This also fixes a display bug, 'nft list hooks device foo' would show hooks registered for that device as 'bridge' family instead of the expected 'netdev' family.

This was because UNSPEC handling did not query 'netdev' family and did pass the device name to the lowlevel function. Add it, and pass NULL device name for those families that don't support device attachment.

The lowelevel function always queries NFPROTO_NETDEV to handle the 'inet' ingress case.

---
 src/mnl.c | 98 ++++++++++++++++++++-----------------------------------
 1 file changed, 35 insertions(+), 63 deletions(-)
src/mnl.c +35 -63
index ec7d2bd5defc..9de3d34cfdbd 100644
@@ -2518,65 +2518,42 @@ static void print_hooks(struct netlink_ctx *ctx, int family, struct list_head *h
 	fprintf(fp, "}\n");
 }
 
-#define HOOK_FAMILY_MAX	5
-
-static uint8_t hook_family[HOOK_FAMILY_MAX] = {
-	NFPROTO_IPV4,
-	NFPROTO_IPV6,
-	NFPROTO_BRIDGE,
-	NFPROTO_ARP,
-};
-
 static int mnl_nft_dump_nf(struct netlink_ctx *ctx, int family, int hook,
-			   const char *devname, struct list_head *hook_list,
-			   int *ret)
+			   const char *devname, struct list_head *hook_list)
 {
 	int i, err;
 
 	/* show ingress in first place in hook listing. */
 	err = __mnl_nft_dump_nf_hooks(ctx, family, NFPROTO_NETDEV, NF_NETDEV_INGRESS, devname, hook_list);
-	if (err < 0)
-		*ret = err;
 
 	for (i = 0; i <= NF_INET_POST_ROUTING; i++) {
-		err = __mnl_nft_dump_nf_hooks(ctx, family, family, i, devname, hook_list);
-		if (err < 0)
-			*ret = err;
+		int tmp;
+
+		tmp = __mnl_nft_dump_nf_hooks(ctx, family, family, i, devname, hook_list);
+		if (tmp == 0)
+			err = 0;
 	}
 
 	return err;
 }
 
 static int mnl_nft_dump_nf_arp(struct netlink_ctx *ctx, int family, int hook,
-			       const char *devname, struct list_head *hook_list,
-			       int *ret)
+			       const char *devname, struct list_head *hook_list)
 {
-	int err;
-
-	/* show ingress in first place in hook listing. */
-	err = __mnl_nft_dump_nf_hooks(ctx, family, NFPROTO_NETDEV, NF_NETDEV_INGRESS, devname, hook_list);
-	if (err < 0)
-		*ret = err;
+	int err1, err2;
 
-	err = __mnl_nft_dump_nf_hooks(ctx, family, family, NF_ARP_IN, devname, hook_list);
-	if (err < 0)
-		*ret = err;
-	err = __mnl_nft_dump_nf_hooks(ctx, family, family, NF_ARP_OUT, devname, hook_list);
-	if (err < 0)
-		*ret = err;
+	err1 = __mnl_nft_dump_nf_hooks(ctx, family, family, NF_ARP_IN, devname, hook_list);
+	err2 = __mnl_nft_dump_nf_hooks(ctx, family, family, NF_ARP_OUT, devname, hook_list);
 
-	return err;
+	return err1 ? err1 : err2;
 }
 
 static int mnl_nft_dump_nf_netdev(struct netlink_ctx *ctx, int family, int hook,
-				  const char *devname, struct list_head *hook_list,
-				  int *ret)
+				  const char *devname, struct list_head *hook_list)
 {
 	int err;
 
 	err = __mnl_nft_dump_nf_hooks(ctx, family, NFPROTO_NETDEV, NF_NETDEV_INGRESS, devname, hook_list);
-	if (err < 0)
-		*ret = err;
 
 	return err;
 }
@@ -2592,51 +2569,46 @@ static void release_hook_list(struct list_head *hook_list)
 int mnl_nft_dump_nf_hooks(struct netlink_ctx *ctx, int family, int hook, const char *devname)
 {
 	LIST_HEAD(hook_list);
-	unsigned int i;
-	int ret;
+	int ret = -1, tmp;
 
 	errno = 0;
-	ret = 0;
 
 	switch (family) {
 	case NFPROTO_UNSPEC:
-		mnl_nft_dump_nf(ctx, NFPROTO_IPV4, hook, devname, &hook_list, &ret);
-		mnl_nft_dump_nf(ctx, NFPROTO_IPV6, hook, devname, &hook_list, &ret);
-		mnl_nft_dump_nf(ctx, NFPROTO_BRIDGE, hook, devname, &hook_list, &ret);
-		break;
+		ret = mnl_nft_dump_nf_hooks(ctx, NFPROTO_ARP, hook, NULL);
+		tmp = mnl_nft_dump_nf_hooks(ctx, NFPROTO_INET, hook, devname);
+		if (tmp == 0)
+			ret = 0;
+		tmp = mnl_nft_dump_nf_hooks(ctx, NFPROTO_BRIDGE, hook, NULL);
+		if (tmp == 0)
+			ret = 0;
+		tmp = mnl_nft_dump_nf_hooks(ctx, NFPROTO_NETDEV, hook, devname);
+		if (tmp == 0)
+			ret = 0;
+
+		return ret;
 	case NFPROTO_INET:
-		mnl_nft_dump_nf(ctx, NFPROTO_IPV4, hook, devname, &hook_list, &ret);
-		mnl_nft_dump_nf(ctx, NFPROTO_IPV6, hook, devname, &hook_list, &ret);
-		break;
+		ret = mnl_nft_dump_nf_hooks(ctx, NFPROTO_IPV4, hook, devname);
+		tmp = mnl_nft_dump_nf_hooks(ctx, NFPROTO_IPV6, hook, devname);
+		if (tmp == 0)
+			ret = 0;
+
+		return ret;
 	case NFPROTO_IPV4:
 	case NFPROTO_IPV6:
 	case NFPROTO_BRIDGE:
-		mnl_nft_dump_nf(ctx, family, hook, devname, &hook_list, &ret);
+		ret = mnl_nft_dump_nf(ctx, family, hook, devname, &hook_list);
 		break;
 	case NFPROTO_ARP:
-		mnl_nft_dump_nf_arp(ctx, family, hook, devname, &hook_list, &ret);
+		ret = mnl_nft_dump_nf_arp(ctx, family, hook, devname, &hook_list);
 		break;
 	case NFPROTO_NETDEV:
-		mnl_nft_dump_nf_netdev(ctx, family, hook, devname, &hook_list, &ret);
-		break;
-	}
-
-	switch (family) {
-	case NFPROTO_UNSPEC:
-		for (i = 0; i < HOOK_FAMILY_MAX; i++)
-			print_hooks(ctx, hook_family[i], &hook_list);
-		break;
-	case NFPROTO_INET:
-		print_hooks(ctx, NFPROTO_IPV4, &hook_list);
-		print_hooks(ctx, NFPROTO_IPV6, &hook_list);
-		break;
-	default:
-		print_hooks(ctx, family, &hook_list);
+		ret = mnl_nft_dump_nf_netdev(ctx, family, hook, devname, &hook_list);
 		break;
 	}
 
+	print_hooks(ctx, family, &hook_list);
 	release_hook_list(&hook_list);
-	ret = 0;
 
 	return ret;
 }
Signature · 1 line
2.44.2
author maintainer
#15 lore raw

PATCH 4/4 src: add egress support for 'list hooks'

In reply to Florian Westphal

This was missing: Also include the egress hooks when listing the netdev family (or unspec).

---
 src/mnl.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
src/mnl.c +4 -3
index 9de3d34cfdbd..8703bcbf88c8 100644
@@ -2551,11 +2551,12 @@ static int mnl_nft_dump_nf_arp(struct netlink_ctx *ctx, int family, int hook,
 static int mnl_nft_dump_nf_netdev(struct netlink_ctx *ctx, int family, int hook,
 				  const char *devname, struct list_head *hook_list)
 {
-	int err;
+	int err1, err2;
 
-	err = __mnl_nft_dump_nf_hooks(ctx, family, NFPROTO_NETDEV, NF_NETDEV_INGRESS, devname, hook_list);
+	err1 = __mnl_nft_dump_nf_hooks(ctx, family, NFPROTO_NETDEV, NF_NETDEV_INGRESS, devname, hook_list);
+	err2 = __mnl_nft_dump_nf_hooks(ctx, family, NFPROTO_NETDEV, NF_NETDEV_EGRESS, devname, hook_list);
 
-	return err;
+	return err1 ? err1 : err2;
 }
 
 static void release_hook_list(struct list_head *hook_list)
Signature · 1 line
2.44.2