Skip to contentSkip to conversation

netfilter: nft_compat: run checkentry() from .validate

PATCH Superseded Pablo Neira Ayuso 2 participants 7 messages Apr 20

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Pablo Neira Ayuso

author maintainer
#0 lore raw

PATCH 1/1 netfilter: nft_compat: run checkentry() from .validate

Several matches and one target check that the hook is correct from checkentry(), however, the basechain is only available from nft_table_validate().

This patch calls checkentry() for matches and targets from the nft_compat expression .validate path for the following matches/target:

- addrtype
- devgroup
- physdev
- policy
- set
- TCPMSS

The xt_match .destroy indirection is also called to restore the xt_set refcounts that is performed by .checkentry. The remaining extensions provide no .destroy interface.

This patch sets the table in the nft_ctx struct in nft_table_validate() which is required by this patch.

The nft_compat_check_match() and nft_compat_check_target() helper functions are added to wrap common code used from .init and .validate path.

The protocol and inverse flags are set to always match from the expression .validate path, this is already checked from the init path.

Fixes 0ca743a55991 ("netfilter: nf_tables: add compatibility layer for x_tables")

---
A more self-contained alternative to Florian's:
https://patchwork.ozlabs.org/project/netfilter-devel/patch/[email protected]/

 net/netfilter/nf_tables_api.c |   1 +
 net/netfilter/nft_compat.c    | 118 ++++++++++++++++++++++++++++------
 2 files changed, 100 insertions(+), 19 deletions(-)
net/netfilter/nf_tables_api.c +1 -0
index 00ccc0734ac9..070e97efe7d1 100644
@@ -4196,6 +4196,7 @@ static int nft_table_validate(struct net *net, const struct nft_table *table)
 	struct nft_chain *chain;
 	struct nft_ctx ctx = {
 		.net	= net,
+		.table	= (struct nft_table *)table,
 		.family	= table->family,
 	};
 	int err = 0;
net/netfilter/nft_compat.c +99 -19

188 lines not on this page · raw message

Signature · 1 line
2.47.3
maintainer
#1 lore raw
Quoted 6 lines · Pablo Neira Ayuso

Several matches and one target check that the hook is correct from checkentry(), however, the basechain is only available from nft_table_validate().

This patch calls checkentry() for matches and targets from the nft_compat expression .validate path for the following matches/target:

I worry that this is fragile. Not all ->checkentry callbacks are pure. Some create /proc entries or bump reference counts.

net/netfilter/nft_compat.c static int nft_target_validate(const struct nft_ctx *ctx, · line 401

Quoted 1 line
net/netfilter/nft_compat.c +1 -0
+		if (!strcmp(target->name, "TCPMSS")) {

This is missing "SET".

author maintainer
#2 lore raw
Quoted 10 lines · Florian Westphal
Quoted 6 lines · Pablo Neira Ayuso

Several matches and one target check that the hook is correct from checkentry(), however, the basechain is only available from nft_table_validate().

This patch calls checkentry() for matches and targets from the nft_compat expression .validate path for the following matches/target:

I worry that this is fragile. Not all ->checkentry callbacks are pure. Some create /proc entries or bump reference counts.

xt_set does bump the reference count. This calls xt.destroy to restore it. I am only calling them for the list of expression you mentioned.

Quoted 3 lines

net/netfilter/nft_compat.c static int nft_target_validate(const struct nft_ctx *ctx, · line 401

Quoted 1 line
net/netfilter/nft_compat.c +1 -0
+		if (!strcmp(target->name, "TCPMSS")) {

This is missing "SET".

I can add this.

maintainer
#3 lore raw
Quoted 6 lines · Pablo Neira Ayuso
Quoted 5 lines · Florian Westphal
Quoted 4 lines · Pablo Neira Ayuso

Several matches and one target check that the hook is correct from checkentry(), however, the basechain is only available from nft_table_validate().

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

This patch calls checkentry() for matches and targets from the nft_compat expression .validate path for the following matches/target:

I worry that this is fragile. Not all ->checkentry callbacks are pure. Some create /proc entries or bump reference counts.

xt_set does bump the reference count. This calls xt.destroy to restore it. I am only calling them for the list of expression you mentioned.

I worry this will lead to trouble later, e.g. info->priv = kmalloc( ...) -> memory leak.

But OK, at least there is a test case in iptables.git for this.

author maintainer
#4 lore raw
Quoted 10 lines · Florian Westphal
Quoted 9 lines · Pablo Neira Ayuso
Quoted 8 lines · Florian Westphal
Quoted 6 lines · Pablo Neira Ayuso

Several matches and one target check that the hook is correct from checkentry(), however, the basechain is only available from nft_table_validate().

This patch calls checkentry() for matches and targets from the nft_compat expression .validate path for the following matches/target:

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

I worry that this is fragile. Not all ->checkentry callbacks are pure. Some create /proc entries or bump reference counts.

xt_set does bump the reference count. This calls xt.destroy to restore it. I am only calling them for the list of expression you mentioned.

I worry this will lead to trouble later, e.g. info->priv = kmalloc( ...) -> memory leak.

If someone needs to cover for more extensions, they will have to update the list of extensions covered by the strcmp() check on the extension name.

Quoted 1 line

But OK, at least there is a test case in iptables.git for this.

Yes.

Your approach duplicates .checkentry in some way, you have to make sure what your .validate and .checkentry perform the same check, ie. they are in sync.

The approach proposed in this patch is not universal, because checkentry() is a place where many things happen as you suggested (/proc entries being registered, reference count being bumped,i kmalloc...).

If this needs to be generalized further, maybe checkentry() needs to extended to improve integration with nftables.

maintainer
#5 lore raw
Quoted 3 lines · Pablo Neira Ayuso

Your approach duplicates .checkentry in some way, you have to make sure what your .validate and .checkentry perform the same check, ie. they are in sync.

Thats why I updated the affected .checkentry functions to use the validate functions internally -- to make sure the code is called even for classic iptables.

Quoted 2 lines

If this needs to be generalized further, maybe checkentry() needs to extended to improve integration with nftables.

I hope not. But I don't care, if you prefer your patch then so be it.

I just find it sad we duplicate efforts all the time.

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

Your approach duplicates .checkentry in some way, you have to make sure what your .validate and .checkentry perform the same check, ie. they are in sync.

Quoted 8 lines

Thats why I updated the affected .checkentry functions to use the validate functions internally -- to make sure the code is called even for classic iptables.

Quoted 2 lines

If this needs to be generalized further, maybe checkentry() needs to extended to improve integration with nftables.

I hope not. But I don't care, if you prefer your patch then so be it.

I can toss this patch if you prefer, I will post it complete then decide what to do.

Quoted 1 line

I just find it sad we duplicate efforts all the time.

I thought I could provide a more self-contained approach for this nft_compat specific bug.