Pablo Neira Ayuso says:
====================
Netfilter/IPVS updates for net-next
This small patchset contains three accumulated Netfilter/IPVS updates,
they are:
1) Refactorize common NAT code by encapsulating it into a helper
function, similarly to what we do in other conntrack extensions,
from Florian Westphal.
2) A minor format string mismatch fix for IPVS, from Masanari Iida.
3) Add quota support to the netfilter accounting infrastructure, now
you can add quotas to accounting objects via the nfnetlink interface
and use them from iptables. You can also listen to quota
notifications from userspace. This enhancement from Mathieu Poirier.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Pablo Neira Ayuso says:
====================
Netfilter/nftables updates for net-next
The following patchset contains Netfilter/nftables updates for net-next,
most relevantly they are:
1) Add set element update notification via netlink, from Arturo Borrero.
2) Put all object updates in one single message batch that is sent to
kernel-space. Before this patch only rules where included in the batch.
This series also introduces the generic transaction infrastructure so
updates to all objects (tables, chains, rules and sets) are applied in
an all-or-nothing fashion, these series from me.
3) Defer release of objects via call_rcu to reduce the time required to
commit changes. The assumption is that all objects are destroyed in
reverse order to ensure that dependencies betweem them are fulfilled
(ie. rules and sets are destroyed first, then chains, and finally
tables).
4) Allow to match by bridge port name, from Tomasz Bursztyka. This series
include two patches to prepare this new feature.
5) Implement the proper set selection based on the characteristics of the
data. The new infrastructure also allows you to specify your preferences
in terms of memory and computational complexity so the underlying set
type is also selected according to your needs, from Patrick McHardy.
6) Several cleanup patches for nft expressions, including one minor possible
compilation breakage due to missing mark support, also from Patrick.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that all objects are released in the reverse order via the
transaction infrastructure, we can enqueue the release via
call_rcu to save one synchronize_rcu. For small rule-sets loaded
via nft -f, it now takes around 50ms less here.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Instead of caching the original skbuff that contains the netlink
messages, this stores the netlink message sequence number, the
netlink portID and the report flag. This helps to prepare the
introduction of the object release via call_rcu.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Leave the set content in consistent state if we fail to load the
batch. Use the new generic transaction infrastructure to achieve
this.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This patch speeds up rule-set updates and it also provides a way
to revert updates and leave things in consistent state in case that
the batch needs to be aborted.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This patch speeds up rule-set updates and it also introduces a way to
revert chain updates if the batch is aborted. The idea is to store the
changes in the transaction to apply that in the commit step.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This patch reworks the nf_tables API so set updates are included in
the same batch that contains rule updates. This speeds up rule-set
updates since we skip a dialog of four messages between kernel and
user-space (two on each direction), from:
1) create the set and send netlink message to the kernel
2) process the response from the kernel that contains the allocated name.
3) add the set elements and send netlink message to the kernel.
4) process the response from the kernel (to check for errors).
To:
1) add the set to the batch.
2) add the set elements to the batch.
3) add the rule that points to the set.
4) send batch to the kernel.
This also introduces an internal set ID (NFTA_SET_ID) that is unique
in the batch so set elements and rules can refer to new sets.
Backward compatibility has been only retained in userspace, this
means that new nft versions can talk to the kernel both in the new
and the old fashion.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The patch adds message type to the transaction to simplify the
commit the and abort routines. Yet another step forward in the
generalisation of the transaction infrastructure.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This patch generalises the existing rule transaction infrastructure
so it can be used to handle set, table and chain object transactions
as well. The transaction provides a data area that stores private
information depending on the transaction type.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The new transaction infrastructure updates the family, table and chain
objects in the context structure, so let's deconstify them. While at it,
move the context structure initialization routine to the top of the
source file as it will be also used from the table and chain routines.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Reduce copy-past a bit by adding a common helper.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This will be useful to create network family dedicated META expression
as for NFPROTO_BRIDGE for instance.
Signed-off-by: Tomasz Bursztyka <tomasz.bursztyka@linux.intel.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
nft_cmp_fast is used for equality comparisions of size <= 4. For
comparisions of size < 4 byte a mask is calculated that is applied to
both the data from userspace (during initialization) and the register
value (during runtime). Both values are stored using (in effect) memcpy
to a memory area that is then interpreted as u32 by nft_cmp_fast.
This works fine on little endian since smaller types have the same base
address, however on big endian this is not true and the smaller types
are interpreted as a big number with trailing zero bytes.
The mask therefore must not include the lower bytes, but the higher bytes
on big endian. Add a helper function that does a cpu_to_le32 to switch
the bytes on big endian. Since we're dealing with a mask of just consequitive
bits, this works out fine.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
"len" contains sizeof(nf_ct_ext) and size of extensions. In a worst
case it can contain all extensions. Bellow you can find sizes for all
types of extensions. Their sum is definitely bigger than 256.
nf_ct_ext_types[0]->len = 24
nf_ct_ext_types[1]->len = 32
nf_ct_ext_types[2]->len = 24
nf_ct_ext_types[3]->len = 32
nf_ct_ext_types[4]->len = 152
nf_ct_ext_types[5]->len = 2
nf_ct_ext_types[6]->len = 16
nf_ct_ext_types[7]->len = 8
I have seen "len" up to 280 and my host has crashes w/o this patch.
The right way to fix this problem is reducing the size of the ecache
extension (4) and Florian is going to do this, but these changes will
be quite large to be appropriate for a stable tree.
Fixes: 5b423f6a40 (netfilter: nf_conntrack: fix racy timer handling with reliable)
Cc: Pablo Neira Ayuso <pablo@netfilter.org>
Cc: Patrick McHardy <kaber@trash.net>
Cc: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The current set selection simply choses the first set type that provides
the requested features, which always results in the rbtree being chosen
by virtue of being the first set in the list.
What we actually want to do is choose the implementation that can provide
the requested features and is optimal from either a performance or memory
perspective depending on the characteristics of the elements and the
preferences specified by the user.
The elements are not known when creating a set. Even if we would provide
them for anonymous (literal) sets, we'd still have standalone sets where
the elements are not known in advance. We therefore need an abstract
description of the data charcteristics.
The kernel already knows the size of the key, this patch starts by
introducing a nested set description which so far contains only the maximum
amount of elements. Based on this the set implementations are changed to
provide an estimate of the required amount of memory and the lookup
complexity class.
The set ops have a new callback ->estimate() that is invoked during set
selection. It receives a structure containing the attributes known to the
kernel and is supposed to populate a struct nft_set_estimate with the
complexity class and, in case the size is known, the complete amount of
memory required, or the amount of memory required per element otherwise.
Based on the policy specified by the user (performance/memory, defaulting
to performance) the kernel will then select the best suited implementation.
Even if the set implementation would allow to add more than the specified
maximum amount of elements, they are enforced since new implementations
might not be able to add more than maximum based on which they were
selected.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
In order to fix set destruction notifications and get rid of unnecessary
members in private data structures, pass the context to expressions'
destructor functions again.
In order to do so, replace various members in the nft_rule_trans structure
by the full context.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
nf_conntrack_lock is a monolithic lock and suffers from huge contention
on current generation servers (8 or more core/threads).
Perf locking congestion is clear on base kernel:
- 72.56% ksoftirqd/6 [kernel.kallsyms] [k] _raw_spin_lock_bh
- _raw_spin_lock_bh
+ 25.33% init_conntrack
+ 24.86% nf_ct_delete_from_lists
+ 24.62% __nf_conntrack_confirm
+ 24.38% destroy_conntrack
+ 0.70% tcp_packet
+ 2.21% ksoftirqd/6 [kernel.kallsyms] [k] fib_table_lookup
+ 1.15% ksoftirqd/6 [kernel.kallsyms] [k] __slab_free
+ 0.77% ksoftirqd/6 [kernel.kallsyms] [k] inet_getpeer
+ 0.70% ksoftirqd/6 [nf_conntrack] [k] nf_ct_delete
+ 0.55% ksoftirqd/6 [ip_tables] [k] ipt_do_table
This patch change conntrack locking and provides a huge performance
improvement. SYN-flood attack tested on a 24-core E5-2695v2(ES) with
10Gbit/s ixgbe (with tool trafgen):
Base kernel: 810.405 new conntrack/sec
After patch: 2.233.876 new conntrack/sec
Notice other floods attack (SYN+ACK or ACK) can easily be deflected using:
# iptables -A INPUT -m state --state INVALID -j DROP
# sysctl -w net/netfilter/nf_conntrack_tcp_loose=0
Use an array of hashed spinlocks to protect insertions/deletions of
conntracks into the hash table. 1024 spinlocks seem to give good
results, at minimal cost (4KB memory). Due to lockdep max depth,
1024 becomes 8 if CONFIG_LOCKDEP=y
The hash resize is a bit tricky, because we need to take all locks in
the array. A seqcount_t is used to synchronize the hash table users
with the resizing process.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reviewed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Netfilter expectations are protected with the same lock as conntrack
entries (nf_conntrack_lock). This patch split out expectations locking
to use it's own lock (nf_conntrack_expect_lock).
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reviewed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
One spinlock per cpu to protect dying/unconfirmed/template special lists.
(These lists are now per cpu, a bit like the untracked ct)
Add a @cpu field to nf_conn, to make sure we hold the appropriate
spinlock at removal time.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reviewed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Changes while reading through the netfilter code.
Added hint about how conntrack nf_conn refcnt is accessed.
And renamed repl_hash to reply_hash for readability
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reviewed-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This allows us to store user comment strings, but it could be also
used to store any kind of information that the user application needs
to link to the rule.
Scratch 8 bits for the new ulen field that indicates the length the
user data area. 4 bits from the handle (so it's 42 bits long, according
to Patrick, it would last 139 years with 1000 new rules per second)
and 4 bits from dlen (so the expression data area is 4K, which seems
sufficient by now even considering the compatibility layer).
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Patrick McHardy <kaber@trash.net>
This also adds NF_CT_LABELS_MAX_SIZE so it can be re-used
as BUILD_BUG_ON in nft_ct.
At this time, nft doesn't yet support writing to the label area;
when this changes the label->words handling needs to be moved
out of xt_connlabel.c into nf_conntrack_labels.c.
Also removes a useless run-time check: words cannot grow beyond
4 (32 bit) or 2 (64bit) since xt_connlabel enforces a maximum of
128 labels.
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
We may lost race if we flush the rule-set (which happens asynchronously
via call_rcu) and we try to remove the table (that userspace assumes
to be empty).
Fix this by recovering synchronous rule and chain deletion. This was
introduced time ago before we had no batch support, and synchronous
rule deletion performance was not good. Now that we have the batch
support, we can just postpone the purge of old rule in a second step
in the commit phase. All object deletions are synchronous after this
patch.
As a side effect, we save memory as we don't need rcu_head per rule
anymore.
Cc: Patrick McHardy <kaber@trash.net>
Reported-by: Arturo Borrero Gonzalez <arturo.borrero.glez@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
