Skip to contentSkip to conversation

doc: fib: explain example in more detail

PATCH In review Florian Westphal 4 participants 15 messages Jul 21, 2019 – Jul 24, 2019 netfilter

lore
Share & tools
lore

Florian Westphal

author maintainer
#0 lore raw

PATCH 1/1 doc: fib: explain example in more detail

As noted by Felix Dreissig, fib documentation is quite terse, so explain the 'saddr . iif' example with a few more words.

---
 doc/primary-expression.txt | 6 ++++++
 1 file changed, 6 insertions(+)
doc/primary-expression.txt +6 -0
index 6eb9583ac9e9..124193626aa7 100644
@@ -274,6 +274,12 @@ fib_addrtype
 # drop packets without a reverse path
 filter prerouting fib saddr . iif oif missing drop
 
+In this example, 'saddr . iif' lookups up routing information based on the source address and the input interface.
+oif picks the output interface index from the routing information.
+If no route was found for the source address/input interface combination, the output interface index is zero.
+In case the input interface is specified as part of the input key, the output interface index is always the same as the input interface index or zero.
+If only 'saddr oif' is given, then oif can be any interface index or zero.
+
 # drop packets to address not configured on ininterface
 filter prerouting fib daddr . iif type != { local, broadcast, multicast } drop
 
Signature · 1 line
2.21.0
#1 lore raw

On Sun, Jul 21, 2019 at 12:48 PM Florian Westphal <[email protected]> wrote:
...

doc/primary-expression.txt fib_addrtype · line 277

Quoted 10 lines
doc/primary-expression.txt +1 -0
index 6eb9583ac9e9..124193626aa7 100644
@@ -274,6 +274,12 @@ fib_addrtype
 # drop packets without a reverse path
 filter prerouting fib saddr . iif oif missing drop
 
+In this example, 'saddr . iif' lookups up routing information based on the source address and the input interface.

s/lookups up/looks up/

#2 lore raw
In reply to Florian Westphal

Cc'ing Phil.

Quoted 2 lines · Florian Westphal

As noted by Felix Dreissig, fib documentation is quite terse, so explain the 'saddr . iif' example with a few more words.

There are patches to disallow ifindex 0 from Phil, probably we should leave them behind given that fib allows matching on zero in case of no matchings. This is also done from userspace, anyone using the API directly may also skip this.

Telling all this because we should be consistent on this.

author maintainer
#3 lore raw
Quoted 7 lines · Pablo Neira Ayuso

Cc'ing Phil.

Quoted 2 lines · Florian Westphal

As noted by Felix Dreissig, fib documentation is quite terse, so explain the 'saddr . iif' example with a few more words.

There are patches to disallow ifindex 0 from Phil

WHich ones?
I only see those that make meta write 0 in case iface doesn't exist, so it does exactly what fib does.

reviewer
#4 lore raw
Quoted 4 lines · Florian Westphal
Quoted 3 lines · Pablo Neira Ayuso

Cc'ing Phil.

On Sun, Jul 21, 2019 at 12:43:05PM +0200, Florian Westphal wrote:

Quoted 8 lines
Quoted 4 lines
Quoted 2 lines

As noted by Felix Dreissig, fib documentation is quite terse, so explain the 'saddr . iif' example with a few more words.

There are patches to disallow ifindex 0 from Phil

WHich ones?
I only see those that make meta write 0 in case iface doesn't exist, so it does exactly what fib does.

It is message-ID [email protected] and follow-ups, trying to prevent users from matching against the values we decided to use for "no data available" situations. This whole attempt feels a bit futile. Maybe we should introduce something to signal "no value" so that cmp expression will never match for '==' and always for '!='? Not sure how to realize this via registers. Also undecided about '<' and '>' ops.

Cheers, Phil

author maintainer
#5 lore raw
Quoted 4 lines · Phil Sutter

use for "no data available" situations. This whole attempt feels a bit futile. Maybe we should introduce something to signal "no value" so that cmp expression will never match for '==' and always for '!='? Not sure how to realize this via registers. Also undecided about '<' and '>' ops.

Whats the point?

reviewer
#6 lore raw
Quoted 7 lines · Florian Westphal
Quoted 4 lines · Phil Sutter

use for "no data available" situations. This whole attempt feels a bit futile. Maybe we should introduce something to signal "no value" so that cmp expression will never match for '==' and always for '!='? Not sure how to realize this via registers. Also undecided about '<' and '>' ops.

Whats the point?

IIRC, Pablo's demand for not aborting in nft_meta in case of insufficient data was to insert a value into dreg which will never match. I think the idea was to avoid accidental matching in situations where a match doesn't make sense.

For ifindex or ifname I can't come up with a good example, but let's assume we set dreg to 0 for 'meta l4proto' if pkt->tprot is not initialized (i.e., pkt->tprot_set is false for whatever reason). A rule 'meta l4proto == 0' would start to match even if l4proto of the packet is not 0.

Remember, the original problem was that with iptables-legacy, I can do '! -i foobar' in POSTROUTING chain and it will always match. With iptables-nft, the same rule will never match.

Maybe we should just go with fixing for iifname/oifname only and leave the rest as-is?

Cheers, Phil

author maintainer
#7 lore raw
Quoted 7 lines · Phil Sutter
Quoted 6 lines · Florian Westphal
Quoted 4 lines · Phil Sutter

use for "no data available" situations. This whole attempt feels a bit futile. Maybe we should introduce something to signal "no value" so that cmp expression will never match for '==' and always for '!='? Not sure how to realize this via registers. Also undecided about '<' and '>' ops.

Quoted 6 lines
Quoted 1 line

Whats the point?

IIRC, Pablo's demand for not aborting in nft_meta in case of insufficient data was to insert a value into dreg which will never match. I think the idea was to avoid accidental matching in situations where a match doesn't make sense.

I think the only contraint is that it must not overlap with a legitimate ifindex.

But 0 cannot occur, so 'meta iif 0' will only match in case no input interface existed -- I think thats fine and might even be desired.

Quoted 5 lines

For ifindex or ifname I can't come up with a good example, but let's assume we set dreg to 0 for 'meta l4proto' if pkt->tprot is not initialized (i.e., pkt->tprot_set is false for whatever reason). A rule 'meta l4proto == 0' would start to match even if l4proto of the packet is not 0.

Thats my point, we can't use a value that can occur normally.

Quoted 3 lines

Remember, the original problem was that with iptables-legacy, I can do '! -i foobar' in POSTROUTING chain and it will always match. With iptables-nft, the same rule will never match.

O know.

Quoted 2 lines

Maybe we should just go with fixing for iifname/oifname only and leave the rest as-is?

I would propose to go with '0' dreg for ifindex, "" for name and leave rest as-is.

reviewer
#8 lore raw
Quoted 15 lines · Florian Westphal
Quoted 7 lines · Phil Sutter
Quoted 6 lines · Florian Westphal
Quoted 4 lines · Phil Sutter

use for "no data available" situations. This whole attempt feels a bit futile. Maybe we should introduce something to signal "no value" so that cmp expression will never match for '==' and always for '!='? Not sure how to realize this via registers. Also undecided about '<' and '>' ops.

Quoted 6 lines
Quoted 1 line

Whats the point?

IIRC, Pablo's demand for not aborting in nft_meta in case of insufficient data was to insert a value into dreg which will never match. I think the idea was to avoid accidental matching in situations where a match doesn't make sense.

Quoted 5 lines

I think the only contraint is that it must not overlap with a legitimate ifindex.

But 0 cannot occur, so 'meta iif 0' will only match in case no input interface existed -- I think thats fine and might even be desired.

OK, so we just drop my patch to reject ifindex 0 from userspace to keep fib working?

[...]

Quoted 2 lines

I would propose to go with '0' dreg for ifindex, "" for name and leave rest as-is.

My kernel patch also changes iftype to set ARPHRD_VOID and ifkind to set an empty string as well.

For iftype, I also sent a userspace patch to disallow ARPHRD_VOID. Do you think it should be dropped as well?

Thanks, Phil

author maintainer
#9 lore raw
Quoted 25 lines · Phil Sutter
Quoted 15 lines · Florian Westphal
Quoted 7 lines · Phil Sutter
Quoted 6 lines · Florian Westphal
Quoted 4 lines · Phil Sutter

use for "no data available" situations. This whole attempt feels a bit futile. Maybe we should introduce something to signal "no value" so that cmp expression will never match for '==' and always for '!='? Not sure how to realize this via registers. Also undecided about '<' and '>' ops.

Quoted 6 lines
Quoted 1 line

Whats the point?

IIRC, Pablo's demand for not aborting in nft_meta in case of insufficient data was to insert a value into dreg which will never match. I think the idea was to avoid accidental matching in situations where a match doesn't make sense.

Quoted 5 lines

I think the only contraint is that it must not overlap with a legitimate ifindex.

But 0 cannot occur, so 'meta iif 0' will only match in case no input interface existed -- I think thats fine and might even be desired.

OK, so we just drop my patch to reject ifindex 0 from userspace to keep fib working?

Quoted 6 lines

[...]

Quoted 2 lines

I would propose to go with '0' dreg for ifindex, "" for name and leave rest as-is.

My kernel patch also changes iftype to set ARPHRD_VOID and ifkind to set an empty string as well.

I would keep both with current semantics, i.e. 'break'/no match until we get more evidence that we need this ARPHDR_VOID store.

For iifkind, I am not sure. Perhaps leave it as-is?

Kernel doesn't allow "" iifname (so we can reuse it for 'no interface'), but what about ifkind?

Quoted 2 lines

For iftype, I also sent a userspace patch to disallow ARPHRD_VOID. Do you think it should be dropped as well?

I think we should leave userspace alone.

reviewer
#10 lore raw
Quoted 33 lines · Florian Westphal
Quoted 25 lines · Phil Sutter
Quoted 15 lines · Florian Westphal
Quoted 7 lines · Phil Sutter
Quoted 6 lines · Florian Westphal
Quoted 4 lines · Phil Sutter

use for "no data available" situations. This whole attempt feels a bit futile. Maybe we should introduce something to signal "no value" so that cmp expression will never match for '==' and always for '!='? Not sure how to realize this via registers. Also undecided about '<' and '>' ops.

Quoted 6 lines
Quoted 1 line

Whats the point?

IIRC, Pablo's demand for not aborting in nft_meta in case of insufficient data was to insert a value into dreg which will never match. I think the idea was to avoid accidental matching in situations where a match doesn't make sense.

Quoted 5 lines

I think the only contraint is that it must not overlap with a legitimate ifindex.

But 0 cannot occur, so 'meta iif 0' will only match in case no input interface existed -- I think thats fine and might even be desired.

OK, so we just drop my patch to reject ifindex 0 from userspace to keep fib working?

Quoted 6 lines

[...]

Quoted 2 lines

I would propose to go with '0' dreg for ifindex, "" for name and leave rest as-is.

My kernel patch also changes iftype to set ARPHRD_VOID and ifkind to set an empty string as well.

Quoted 7 lines

I would keep both with current semantics, i.e. 'break'/no match until we get more evidence that we need this ARPHDR_VOID store.

For iifkind, I am not sure. Perhaps leave it as-is?

Kernel doesn't allow "" iifname (so we can reuse it for 'no interface'), but what about ifkind?

Well, at least every implementation of rtnl_link_ops I found in current kernel sources initializes 'kind' field but there's indeed no guarantee.

Quoted 4 lines
Quoted 2 lines

For iftype, I also sent a userspace patch to disallow ARPHRD_VOID. Do you think it should be dropped as well?

I think we should leave userspace alone.

OK.

Thanks, Phil

#11 lore raw
In reply to Phil Sutter
Quoted 16 lines · Phil Sutter
Quoted 15 lines · Florian Westphal
Quoted 13 lines · Phil Sutter
Quoted 7 lines · Florian Westphal
Quoted 4 lines · Phil Sutter

use for "no data available" situations. This whole attempt feels a bit futile. Maybe we should introduce something to signal "no value" so that cmp expression will never match for '==' and always for '!='? Not sure how to realize this via registers. Also undecided about '<' and '>' ops.

Whats the point?

IIRC, Pablo's demand for not aborting in nft_meta in case of insufficient data was to insert a value into dreg which will never match. I think the idea was to avoid accidental matching in situations where a match doesn't make sense.

Quoted 8 lines
Quoted 5 lines

I think the only contraint is that it must not overlap with a legitimate ifindex.

But 0 cannot occur, so 'meta iif 0' will only match in case no input interface existed -- I think thats fine and might even be desired.

OK, so we just drop my patch to reject ifindex 0 from userspace to keep fib working?

I think so, yes.

#12 lore raw
In reply to Florian Westphal
Quoted 10 lines · Florian Westphal
Quoted 7 lines · Pablo Neira Ayuso

Cc'ing Phil.

Quoted 2 lines · Florian Westphal

As noted by Felix Dreissig, fib documentation is quite terse, so explain the 'saddr . iif' example with a few more words.

There are patches to disallow ifindex 0 from Phil

WHich ones?

https://patchwork.ozlabs.org/patch/1133521/

Quoted 2 lines

I only see those that make meta write 0 in case iface doesn't exist, so it does exactly what fib does.

This one disallows matching on zero, while you can match on ifindex 0 from fib to check for no routes, correct?

author maintainer
#13 lore raw
Quoted 5 lines · Pablo Neira Ayuso
Quoted 4 lines · Florian Westphal
Quoted 3 lines · Pablo Neira Ayuso

Cc'ing Phil.

On Sun, Jul 21, 2019 at 12:43:05PM +0200, Florian Westphal wrote:

Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines

As noted by Felix Dreissig, fib documentation is quite terse, so explain the 'saddr . iif' example with a few more words.

There are patches to disallow ifindex 0 from Phil

WHich ones?

https://patchwork.ozlabs.org/patch/1133521/

This is insane. "nft meta iif 0 counter" seems totally fine to me.

I think we should never try to be clever here.