Check if the destination register already contains the data that this
meta store expression performs. This allows to skip this redundant
operation. If the destination contains a different selector, update
the register tracking information.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Check if the destination register already contains the data that this
payload store expression performs. This allows to skip this redundant
operation. If the destination contains a different selector, update
the register tracking information.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

This patch adds new infrastructure to skip redundant selector store
operations on the same register to achieve a performance boost from
the packet path.

This is particularly noticeable in pure linear rulesets but it also
helps in rulesets which are already heaving relying in maps to avoid
ruleset linear inspection.

The idea is to keep data of the most recurrent store operations on
register to reuse them with cmp and lookup expressions.

This infrastructure allows for dynamic ruleset updates since the ruleset
blob reduction happens from the kernel.

Userspace still needs to be updated to maximize register utilization to
cooperate to improve register data reuse / reduce number of store on
register operations.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Add a definition including the maximum number of 32-bits registers that
are used a scratchpad memory area to store data.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

This patch adds a blob layout per chain to represent the ruleset in the
packet datapath.

	size (unsigned long)
	struct nft_rule_dp
	  struct nft_expr
	  ...
        struct nft_rule_dp
          struct nft_expr
          ...
        struct nft_rule_dp (is_last=1)

The new structure nft_rule_dp represents the rule in a more compact way
(smaller memory footprint) compared to the control-plane nft_rule
structure.

The ruleset blob is a read-only data structure. The first field contains
the blob size, then the rules containing expressions. There is a trailing
rule which is used by the tracing infrastructure which is equivalent to
the NULL rule marker in the previous representation. The blob size field
does not include the size of this trailing rule marker.

The ruleset blob is generated from the commit path.

This patch reuses the infrastructure available since 0cbc06b3
("netfilter: nf_tables: remove synchronize_rcu in commit phase") to
build the array of rules per chain.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

In preparation for the rule blob representation.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

From struct nft_limit to nft_limit_priv.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

In preparation for the rule blob representation.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

In preparation for the rule blob representation.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

In preparation for the rule blob representation.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

In preparation for the rule blob representation.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

include/linux/netfilter_netdev.h:97 suspicious rcu_dereference_check() usage!
2 locks held by sd-resolve/1100:
 0: ..(rcu_read_lock_bh){1:3}, at: ip_finish_output2
 1: ..(rcu_read_lock_bh){1:3}, at: __dev_queue_xmit
 __dev_queue_xmit+0 ..

The helper has two callers, one uses rcu_read_lock, the other
rcu_read_lock_bh().  Annotate the dereference to reflect this.

Fixes: 42df6e1d ("netfilter: Introduce egress hook")
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Its the same as nf_conntrack_put(), but without the
need for an indirect call.  The downside is a module dependency on
nf_conntrack, but all of these already depend on conntrack anyway.

Cc: Paul Blakey <paulb@mellanox.com>
Cc: dev@openvswitch.org
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

nf_ct_put() results in a usesless indirection:

nf_ct_put -> nf_conntrack_put -> nf_conntrack_destroy -> rcu readlock +
indirect call of ct_hooks->destroy().

There are two _put helpers:
nf_ct_put and nf_conntrack_put.  The latter is what should be used in
code that MUST NOT cause a linker dependency on the conntrack module
(e.g. calls from core network stack).

Everyone else should call nf_ct_put() instead.

A followup patch will convert a few nf_conntrack_put() calls to
nf_ct_put(), in particular from modules that already have a conntrack
dependency such as act_ct or even nf_conntrack itself.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

No functional changes, these structures should be const.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

ip_ct_attach predates struct nf_ct_hook, we can place it there and
remove the exported symbol.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Convert nf_conn reference counting from atomic_t to refcount_t based api.
refcount_t api provides more runtime sanity checks and will warn on
certain constructs, e.g. refcount_inc() on a zero reference count, which
usually indicates use-after-free.

For this reason template allocation is changed to init the refcount to
1, the subsequenct add operations are removed.

Likewise, init_conntrack() is changed to set the initial refcount to 1
instead refcount_inc().

This is safe because the new entry is not (yet) visible to other cpus.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Fix following coccicheck warning:

./include/net/netfilter/nf_conntrack.h:282:16-17: WARNING opportunity
for max().
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

The pointer lt is being assigned a value and then later
updated but that value is never read. The pointer is
redundant and can be removed.
Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Just place the structs and registration in the inet module.
nf_flow_table_ipv6, nf_flow_table_ipv4 and nf_flow_table_inet share
same module dependencies: nf_flow_table, nf_tables.

before:
   text	   data	    bss	    dec	    hex	filename
   2278	   1480	      0	   3758	    eae	nf_flow_table_inet.ko
   1159	   1352	      0	   2511	    9cf	nf_flow_table_ipv6.ko
   1154	   1352	      0	   2506	    9ca	nf_flow_table_ipv4.ko

after:
   2369	   1672	      0	   4041	    fc9	nf_flow_table_inet.ko
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

If destination port is above 32k and source port below 16k
assume this might cause 'port shadowing' where a 'new' inbound
connection matches an existing one, e.g.

inbound X:41234 -> Y:53 matches existing conntrack entry
        Z:53 -> X:4123, where Z got natted to X.

In this case, new packet is natted to Z:53 which is likely
unwanted.

We avoid the rewrite for connections that originate from local host:
port-shadowing is only possible with forwarded connections.

Also adjust test case.

v3: no need to call tuple_force_port_remap if already in random mode (Phil)
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Phil Sutter <phil@nwl.cc>
Acked-by: Eric Garver <eric@garver.life>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

This allows to identify flows that originate from local machine
in a followup patch.

It would be possible to make this a ->status bit instead.
For now I did not do that yet because I don't have a use-case for
exposing this info to userspace.

If one comes up the toggle can be replaced with a status bit.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Make counter support built-in to allow for direct call in case of
CONFIG_RETPOLINE.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Bug might trigger warning for each packet, call WARN_ON_ONCE instead.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Add function to consolidate verdict tracing.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

BUG() is too harsh for unknown payload base, use WARN_ON_ONCE() instead.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Chain stats are updated from the Netfilter hook path which already run
under rcu read-size lock section.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

If compiled with CONFIG_NET_NS_REFCNT_TRACKER=y,
using put_net_track() in iterate_cleanup_work()
and netns_tracker_alloc() in nf_nat_masq_schedule()
might help us finding netns refcount imbalances.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

If compiled with CONFIG_NET_NS_REFCNT_TRACKER=y,
using put_net_track() in nfulnl_instance_free_rcu()
and get_net_track() in instance_create()
might help us finding netns refcount imbalances.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>

Add user template explicit support. At this moment, max
TCAM rule size is utilized for all rules, doesn't matter
which and how much flower matches are provided by user. It
means that some of TCAM space is wasted, which impacts
the number of filters that can be offloaded.

Introducing the template, allows to have more HW offloaded
filters by specifying the template explicitly.

Example:
  tc qd add dev PORT clsact
  tc chain add dev PORT ingress protocol ip \
    flower dst_ip 0.0.0.0/16
  tc filter add dev PORT ingress protocol ip \
    flower skip_sw dst_ip 1.2.3.4/16 action drop

NOTE: chain 0 is the default chain id for "tc chain" & "tc filter"
      command, so it is omitted in the example above.

This patch adds only template support for default chain 0 suppoerted
by prestera driver at this moment. Chains are not supported yet,
and will be added later.
Signed-off-by: Volodymyr Mytnyk <vmytnyk@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Recent net-next fails to initialize ports with:

 realtek-smi switch: phy mode gmii is unsupported on port 0
 realtek-smi switch lan5 (uninitialized): validation of gmii with
 support 0000000,00000000,000062ef and advertisement
 0000000,00000000,000062ef failed: -22
 realtek-smi switch lan5 (uninitialized): failed to connect to PHY:
 -EINVAL
 realtek-smi switch lan5 (uninitialized): error -22 setting up PHY
 for tree 1, switch 0, port 0

Current net branch(3dd7d40b) is not
affected.

I also noticed the same issue before with older versions but using
a MDIO interface driver, not realtek-smi.
Tested-by: Arınç ÜNAL <arinc.unal@arinc9.com>
Signed-off-by: Luiz Angelo Daros de Luca <luizluca@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jeroen de Borst says:

====================
gve improvements

This patchset consists of unrelated changes:

A bug fix for an issue that disabled jumbo-frame support, a few code
improvements and minor funcitonal changes and 3 new features:
  Supporting tx|rx-coalesce-usec for DQO
  Suspend/resume/shutdown
  Optional metadata descriptors
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Adding ethtool support for changing rx-coalesce-usec and tx-coalesce-usec
when using the DQO queue format.
Signed-off-by: Tao Liu <xliutaox@google.com>
Signed-off-by: Jeroen de Borst <jeroendb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Being able to see how many descriptors are in-use is helpful
when diagnosing certain issues.
Signed-off-by: Jeroen de Borst <jeroendb@google.com>
Signed-off-by: Jordan Kim <jrkim@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add support for suspend, resume and shutdown.
Signed-off-by: Catherine Sullivan <csully@google.com>
Signed-off-by: David Awogbemila <awogbemila@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Allow drivers to pass metadata along with packet data to the device.
Introduce a new metadata descriptor type

* GVE_TXD_MTD

This descriptor is optional. If present it immediate follows the
packet descriptor and precedes the segment descriptor.

This descriptor may be repeated. Multiple metadata descriptors may
follow. There are no immediate uses for this, this is for future
proofing. At present devices allow only 1 MTD descriptor.

The lower four bits of the type_flags field encode GVE_TXD_MTD.
The upper four bits of the type_flags field encodes a *sub*type.

Introduce one such metadata descriptor subtype

* GVE_MTD_SUBTYPE_PATH

This shares path information with the device for network failure
discovery and robust response:

Linux derives ipv6 flowlabel and ECMP multipath from sk->sk_txhash,
and updates this field on error with sk_rethink_txhash. Allow the host
stack to do the same. Pass the tx_hash value if set. Also communicate
whether the path hash is set, or more exactly, what its type is. Define
two common types

  GVE_MTD_PATH_HASH_NONE
  GVE_MTD_PATH_HASH_L4

Concrete examples of error conditions that are resolved are
mentioned in the commits that add sk_rethink_txhash calls. Such as
commit 7788174e ("tcp: change IPv6 flow-label upon receiving
spurious retransmission").

Experimental results mirror what the theory suggests: where IPv6
FlowLabel is included in path selection (e.g., LAG/ECMP), flowlabel
rotation on TCP timeout avoids the vast majority of TCP disconnects
that would otherwise have occurred during link failures in long-haul
backbones, when an alternative path is available.

Rotation can be applied to various bad connection signals, such as
timeouts and spurious retransmissions. In aggregate, such flow level
signals can help locate network issues. Define initial common states:

  GVE_MTD_PATH_STATE_DEFAULT
  GVE_MTD_PATH_STATE_TIMEOUT
  GVE_MTD_PATH_STATE_CONGESTION
  GVE_MTD_PATH_STATE_RETRANSMIT
Signed-off-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: David Awogbemila <awogbemila@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

No longer needed after we introduced the barrier in gve_napi_poll.
Signed-off-by: Catherine Sullivan <csully@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The id field should be a u32 not a signed int.
Signed-off-by: Catherine Sullivan <csully@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Giving the device access to other kernel structs is not ideal.
Move the indexes into their own array and just keep pointers to
them in the ntfy block struct.
Signed-off-by: Catherine Sullivan <csully@google.com>
Signed-off-by: David Awogbemila <awogbemila@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The legacy raw addressing device option was processed before the
new RDA queue format option.  This caused the supported features mask,
which is provided only on the RDA queue format option, not to be set.

This disabled jumbo-frame support when using raw adressing.

Fixes: 255489f5 ("gve: Add a jumbo-frame device option")
Signed-off-by: Jeroen de Borst <jeroendb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>