Skip to contentSkip to conversation

nftables: nft fails to add rules to chains

DISCUSSION Laurent Bercot 3 participants 11 messages Mar 19, 2015 – Mar 24, 2015

lore
Share & tools
lore
#0 lore raw

Hello,

(Platform: Intel Atom (x86_64), Linux 3.19.1, musl 1.1.7,
latest nftables/libnftnl/libmnl from git. All iptables modules out of the kernel, all necessary nftables modules in.)

I can flush tables, create tables and create chains with nft
without trouble; however, every time I try and add a rule to a chain, no matter what chain, no matter in what table, I get the following error:

netlink.c:182: Memory allocation failure

I dug a bit and found that the error always happens when
alloc_nft_expr() is called for the *first* time (which is also the last, since nft exits at that point...) and it is always called with the argument "payload".

What is happening ? Anything I could do to help fix it ? Thanks,

Signature · 1 line
 Laurent
#1 lore raw

Bumping, since it looks like there are people from the project
reading this list at this time.

Any idea on what is happening and how to solve it ? This is
preventing me from using nftables in my project. On the other hand, if nftables, which is supposed to be the shiny new thing, is still considered alpha and not to be used yet in real life, it would be nice to know it ASAP.

Thanks.

Quoted 22 lines · Laurent Bercot

Hello,

(Platform: Intel Atom (x86_64), Linux 3.19.1, musl 1.1.7,
latest nftables/libnftnl/libmnl from git. All iptables modules out of the kernel, all necessary nftables modules in.)

I can flush tables, create tables and create chains with nft
without trouble; however, every time I try and add a rule to a chain, no matter what chain, no matter in what table, I get the following error:

netlink.c:182: Memory allocation failure

I dug a bit and found that the error always happens when
alloc_nft_expr() is called for the *first* time (which is also the last, since nft exits at that point...) and it is always called with the argument "payload".

What is happening ? Anything I could do to help fix it ? Thanks,

Signature · 1 line
 Laurent
#2 lore raw
In reply to Laurent Bercot
Quoted 7 lines · Laurent Bercot

Hello,

(Platform: Intel Atom (x86_64), Linux 3.19.1, musl 1.1.7,
latest nftables/libnftnl/libmnl from git. All iptables modules out of the kernel, all necessary nftables modules in.)

Quoted 6 lines

I can flush tables, create tables and create chains with nft
without trouble; however, every time I try and add a rule to a chain, no matter what chain, no matter in what table, I get the following error:

netlink.c:182: Memory allocation failure

I think this error is bogus. Please, apply this patch to libnftnl and let us know. For some reason __init is being ignored, are you using gcc?

src/expr.c +4 -1
index 79782fa..db84d0b 100644
@@ -29,8 +29,11 @@ struct nft_rule_expr *nft_rule_expr_alloc(const char *name)
 	struct expr_ops *ops;
 
 	ops = nft_expr_ops_lookup(name);
-	if (ops == NULL)
+	if (ops == NULL) {
+		fprintf(stderr, "libnftnl: Expression '%s' not supported\n",
+			name);
 		return NULL;
+	}
 
 	expr = calloc(1, sizeof(struct nft_rule_expr) + ops->alloc_len);
 	if (expr == NULL)
#3 lore raw

Hi Pedro,

Thanks for answering.
Patch applied, and as expected nft now fails with the messages

libnftnl: Expression 'payload' not supported
netlink.c:182: Memory allocation failure

Quoted 3 lines

I think this error is bogus. Please, apply this patch to libnftnl and let us know. For some reason __init is being ignored, are you using gcc?

I am using gcc (version 4.8.3). However I am not using glibc, but musl,
and I link statically. This is probably the main difference between my system and the ones you guys are used to.
Is the "constructor" gcc attribute incompatible with static linking ?
Or is nft using glibc-specific constructs ?

Signature · 1 line
 Laurent
#4 lore raw

And my apologies for mangling your name. -_- (I just wrote a mail to a Pedro before replying. Sigh.)

#5 lore raw
In reply to Laurent Bercot
Quoted 9 lines · Laurent Bercot

Hi Pedro,

Thanks for answering.
Patch applied, and as expected nft now fails with the messages

libnftnl: Expression 'payload' not supported
netlink.c:182: Memory allocation failure

Quoted 8 lines
Quoted 3 lines

I think this error is bogus. Please, apply this patch to libnftnl and let us know. For some reason __init is being ignored, are you using gcc?

I am using gcc (version 4.8.3). However I am not using glibc, but musl,
and I link statically. This is probably the main difference between my system and the ones you guys are used to.
Is the "constructor" gcc attribute incompatible with static linking ?

The constructor seems to be ignored with static links, libnftnl needs a fix.

#6 lore raw
Quoted 2 lines · Pablo Neira Ayuso

The constructor seems to be ignored with static links, libnftnl needs a fix.

OK, I'll try linking dynamically until it's fixed then. Thank you!

Signature · 1 line
 Laurent
#7 lore raw
Quoted 5 lines · Laurent Bercot
Quoted 2 lines · Pablo Neira Ayuso

The constructor seems to be ignored with static links, libnftnl needs a fix.

OK, I'll try linking dynamically until it's fixed then.

Please, manually apply this:

http://patchwork.ozlabs.org/patch/453392/

And provide feedback. Thank you.

#8 lore raw

nftables feature request: don't fail "flush" on nonexistent tables (was: nftables: nft fails to add rules to chains)

Quoted 5 lines · Pablo Neira Ayuso

Please, manually apply this:

http://patchwork.ozlabs.org/patch/453392/

And provide feedback. Thank you.

Done. It's working beautifully. Thank you.

Now that I can play with nft, I have a feature request:

I'm saving my rule set in a file, called whenever the rule
set must be applied/reapplied via nft -f. (It's to be applied whenever my DHCP client obtains a new lease.)
I would like the rule set file to be the same for the first
time and the subsequent times the rules are applied. It's only logical.
I have to "flush table nat" and "flush table filter" at the
beginning of the file, so nft does not duplicate rules on the second and later invocations.
Problem is, the first invocation fails on those "flush" lines,
because the tables are not defined yet!

Is there a way for me to tell nft -f to ignore failures on "flush" ?
I'm ok with an option to nft if you so choose. I'm also ok with a warning in my logs, provided nft keeps reading the ruleset, does the job, and exits 0.

Thanks,

Signature · 1 line
 Laurent
#9 lore raw

Re: nftables feature request: don't fail "flush" on nonexistent tables (was: nftables: nft fails to add rules to chains)

Quoted 13 lines · Laurent Bercot
Quoted 6 lines · Pablo Neira Ayuso

Please, manually apply this:

http://patchwork.ozlabs.org/patch/453392/

And provide feedback. Thank you.

Done. It's working beautifully. Thank you.

Now that I can play with nft, I have a feature request:

Quoted 11 lines

I'm saving my rule set in a file, called whenever the rule
set must be applied/reapplied via nft -f. (It's to be applied whenever my DHCP client obtains a new lease.)
I would like the rule set file to be the same for the first
time and the subsequent times the rules are applied. It's only logical.
I have to "flush table nat" and "flush table filter" at the
beginning of the file, so nft does not duplicate rules on the second and later invocations.
Problem is, the first invocation fails on those "flush" lines,
because the tables are not defined yet!

perhaps `flush ruleset'.

That doesn't fail if there is no ruleset.

Signature · 1 line
Arturo Borrero González
#10 lore raw

Re: nftables feature request: don't fail "flush" on nonexistent tables

Quoted 3 lines · Arturo Borrero Gonzalez

perhaps `flush ruleset'.

That doesn't fail if there is no ruleset.

It's exactly what I was looking for, thanks.

Signature · 1 line
 Laurent