When we scan a packet for GRO processing, we want to see the most
common packet types in the front of the offload_base list.
So add a priority field so we can handle this properly.
IPv4/IPv6 get the highest priority with the implicit zero priority
field.
Next comes ethernet with a priority of 10, and then we have the MPLS
types with a priority of 15.
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Suggested-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reduce ifdef pollution slightly, no functional change. We can simply
remove the extra alternative definition of handle_ing() and nf_ingress().
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
commit d2788d3488 ("net: sched: further simplify handle_ing")
removed the call to qdisc_enqueue_root().
However, after this removal we no longer set qdisc pkt length.
This breaks traffic policing on ingress.
This is the minimum fix: set qdisc pkt length before tc_classify.
Only setting the length does remove support for 'stab' on ingress, but
as Alexei pointed out:
"Though it was allowed to add qdisc_size_table to ingress, it's useless.
Nothing takes advantage of recomputed qdisc_pkt_len".
Jamal suggested to use qdisc_pkt_len_init(), but as Eric mentioned that
would result in qdisc_pkt_len_init to no longer get inlined due to the
additional 2nd call site.
ingress policing is rare and GRO doesn't really work that well with police
on ingress, as we see packets > mtu and drop skbs that -- without
aggregation -- would still have fitted the policier budget.
Thus to have reliable/smooth ingress policing GRO has to be turned off.
Cc: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Fixes: d2788d3488 ("net: sched: further simplify handle_ing")
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Eric Dumazet <edumazet@google.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds the Netfilter ingress hook just after the existing tc ingress
hook, that seems to be the consensus solution for this.
Note that the Netfilter hook resides under the global static key that enables
ingress filtering. Nonetheless, Netfilter still also has its own static key for
minimal impact on the existing handle_ing().
* Without this patch:
Result: OK: 6216490(c6216338+d152) usec, 100000000 (60byte,0frags)
16086246pps 7721Mb/sec (7721398080bps) errors: 100000000
42.46% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core
25.92% kpktgend_0 [kernel.kallsyms] [k] kfree_skb
7.81% kpktgend_0 [pktgen] [k] pktgen_thread_worker
5.62% kpktgend_0 [kernel.kallsyms] [k] ip_rcv
2.70% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal
2.34% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk
1.44% kpktgend_0 [kernel.kallsyms] [k] __build_skb
* With this patch:
Result: OK: 6214833(c6214731+d101) usec, 100000000 (60byte,0frags)
16090536pps 7723Mb/sec (7723457280bps) errors: 100000000
41.23% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core
26.57% kpktgend_0 [kernel.kallsyms] [k] kfree_skb
7.72% kpktgend_0 [pktgen] [k] pktgen_thread_worker
5.55% kpktgend_0 [kernel.kallsyms] [k] ip_rcv
2.78% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal
2.06% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk
1.43% kpktgend_0 [kernel.kallsyms] [k] __build_skb
* Without this patch + tc ingress:
tc filter add dev eth4 parent ffff: protocol ip prio 1 \
u32 match ip dst 4.3.2.1/32
Result: OK: 9269001(c9268821+d179) usec, 100000000 (60byte,0frags)
10788648pps 5178Mb/sec (5178551040bps) errors: 100000000
40.99% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core
17.50% kpktgend_0 [kernel.kallsyms] [k] kfree_skb
11.77% kpktgend_0 [cls_u32] [k] u32_classify
5.62% kpktgend_0 [kernel.kallsyms] [k] tc_classify_compat
5.18% kpktgend_0 [pktgen] [k] pktgen_thread_worker
3.23% kpktgend_0 [kernel.kallsyms] [k] tc_classify
2.97% kpktgend_0 [kernel.kallsyms] [k] ip_rcv
1.83% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal
1.50% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk
0.99% kpktgend_0 [kernel.kallsyms] [k] __build_skb
* With this patch + tc ingress:
tc filter add dev eth4 parent ffff: protocol ip prio 1 \
u32 match ip dst 4.3.2.1/32
Result: OK: 9308218(c9308091+d126) usec, 100000000 (60byte,0frags)
10743194pps 5156Mb/sec (5156733120bps) errors: 100000000
42.01% kpktgend_0 [kernel.kallsyms] [k] __netif_receive_skb_core
17.78% kpktgend_0 [kernel.kallsyms] [k] kfree_skb
11.70% kpktgend_0 [cls_u32] [k] u32_classify
5.46% kpktgend_0 [kernel.kallsyms] [k] tc_classify_compat
5.16% kpktgend_0 [pktgen] [k] pktgen_thread_worker
2.98% kpktgend_0 [kernel.kallsyms] [k] ip_rcv
2.84% kpktgend_0 [kernel.kallsyms] [k] tc_classify
1.96% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_internal
1.57% kpktgend_0 [kernel.kallsyms] [k] netif_receive_skb_sk
Note that the results are very similar before and after.
I can see gcc gets the code under the ingress static key out of the hot path.
Then, on that cold branch, it generates the code to accomodate the netfilter
ingress static key. My explanation for this is that this reduces the pressure
on the instruction cache for non-users as the new code is out of the hot path,
and it comes with minimal impact for tc ingress users.
Using gcc version 4.8.4 on:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 8
[...]
L1d cache: 16K
L1i cache: 64K
L2 cache: 2048K
L3 cache: 8192K
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This new config switch enables the ingress filtering infrastructure that is
controlled through the ingress_needed static key. This prepares the
introduction of the Netfilter ingress hook that resides under this unique
static key.
Note that CONFIG_SCH_INGRESS automatically selects this, that should be no
problem since this also depends on CONFIG_NET_CLS_ACT.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
next to its user. No relation to flow_dissector so it makes no sense to
have it in flow_dissector.c
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
__skb_tx_hash function has no relation to flow_dissect so just move it
to dev.c
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
Four minor merge conflicts:
1) qca_spi.c renamed the local variable used for the SPI device
from spi_device to spi, meanwhile the spi_set_drvdata() call
got moved further up in the probe function.
2) Two changes were both adding new members to codel params
structure, and thus we had overlapping changes to the
initializer function.
3) 'net' was making a fix to sk_release_kernel() which is
completely removed in 'net-next'.
4) In net_namespace.c, the rtnl_net_fill() call for GET operations
had the command value fixed, meanwhile 'net-next' adjusted the
argument signature a bit.
This also matches example merge resolutions posted by Stephen
Rothwell over the past two days.
Signed-off-by: David S. Miller <davem@davemloft.net>
These functions compile to 60 bytes of machine code each.
With this .config: http://busybox.net/~vda/kernel_config
there are 617 calls of netif_tx_stop_queue()
and 49 calls of netif_tx_stop_all_queues() in vmlinux.
To fix this, remove WARN_ON in netif_tx_stop_queue()
as suggested by davem, and deinline netif_tx_stop_all_queues().
Change in code size is about 20k:
text data bss dec hex filename
82426986 22255416 20627456 125309858 77813a2 vmlinux.before
82406248 22255416 20627456 125289120 777c2a0 vmlinux
gcc-4.7.2 still creates deinlined version of netif_tx_stop_queue
sometimes:
$ nm --size-sort vmlinux | grep netif_tx_stop_queue | wc -l
190
ffffffff81b558a8 <netif_tx_stop_queue>:
ffffffff81b558a8: 55 push %rbp
ffffffff81b558a9: 48 89 e5 mov %rsp,%rbp
ffffffff81b558ac: f0 80 8f e0 01 00 00 lock orb $0x1,0x1e0(%rdi)
ffffffff81b558b3: 01
ffffffff81b558b4: 5d pop %rbp
ffffffff81b558b5: c3 retq
This needs additional fixing.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Alexei Starovoitov <alexei.starovoitov@gmail.com>
CC: Alexander Duyck <alexander.duyck@gmail.com>
CC: Joe Perches <joe@perches.com>
CC: David S. Miller <davem@davemloft.net>
CC: Jiri Pirko <jpirko@redhat.com>
CC: linux-kernel@vger.kernel.org
CC: netdev@vger.kernel.org
CC: netfilter-devel@vger.kernel.org
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Ingress qdisc has no other purpose than calling into tc_classify()
that executes attached classifier(s) and action(s).
It has a 1:1 relationship to dev->ingress_queue. After having commit
087c1a601a ("net: sched: run ingress qdisc without locks") removed
the central ingress lock, one major contention point is gone.
The extra indirection layers however, are not necessary for calling
into ingress qdisc. pktgen calling locally into netif_receive_skb()
with a dummy u32, single CPU result on a Supermicro X10SLM-F, Xeon
E3-1240: before ~21,1 Mpps, after patch ~22,9 Mpps.
We can redirect the private classifier list to the netdev directly,
without changing any classifier API bits (!) and execute on that from
handle_ing() side. The __QDISC_STATE_DEACTIVATE test can be removed,
ingress qdisc doesn't have a queue and thus dev_deactivate_queue()
is also not applicable, ingress_cl_list provides similar behaviour.
In other words, ingress qdisc acts like TCQ_F_BUILTIN qdisc.
One next possible step is the removal of the dev's ingress (dummy)
netdev_queue, and to only have the list member in the netdevice
itself.
Note, the filter chain is RCU protected and individual filter elements
are being kfree'd by sched subsystem after RCU grace period. RCU read
lock is being held by __netif_receive_skb_core().
Joint work with Alexei Starovoitov.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Given quite some code has been removed from ing_filter(), we can just
consolidate that function into handle_ing() and get rid of a few
instructions at the same time.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The code in __netdev_upper_dev_link() has an over-stringent
loop detection logic that actually prevents valid configurations
from working correctly.
In particular, the logic returns an error if an upper device
is already in the list of all upper devices for a given dev.
This particular check seems to be a overzealous as it disallows
perfectly valid configurations. For example:
# ip l a link eth0 name eth0.10 type vlan id 10
# ip l a dev br0 typ bridge
# ip l s eth0.10 master br0
# ip l s eth0 master br0 <--- Will fail
If you switch the last two commands (add eth0 first), then both
will succeed. If after that, you remove eth0 and try to re-add
it, it will fail!
It appears to be enough to simply check adj_list to keeps things
safe.
I've tried stacking multiple devices multiple times in all different
combinations, and either rx_handler registration prevented the stacking
of the device linking cought the error.
Signed-off-by: Vladislav Yasevich <vyasevic@redhat.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Veaceslav Falico <vfalico@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
TC classifiers/actions were converted to RCU by John in the series:
http://thread.gmane.org/gmane.linux.network/329739/focus=329739
and many follow on patches.
This is the last patch from that series that finally drops
ingress spin_lock.
Single cpu ingress+u32 performance goes from 22.9 Mpps to 24.5 Mpps.
In two cpu case when both cores are receiving traffic on the same
device and go into the same ingress+u32 the performance jumps
from 4.5 + 4.5 Mpps to 23.5 + 23.5 Mpps
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 567e4b7973 ("net: rfs: add hash collision detection") had one
mistake :
RPS_NO_CPU is no longer the marker for invalid cpu in set_rps_cpu()
and get_rps_cpu(), as @next_cpu was the result of an AND with
rps_cpu_mask
This bug showed up on a host with 72 cpus :
next_cpu was 0x7f, and the code was trying to access percpu data of an
non existent cpu.
In a follow up patch, we might get rid of compares against nr_cpu_ids,
if we init the tables with 0. This is silly to test for a very unlikely
condition that exists only shortly after table initialization, as
we got rid of rps_reset_sock_flow() and similar functions that were
writing this RPS_NO_CPU magic value at flow dismantle : When table is
old enough, it never contains this value anymore.
Fixes: 567e4b7973 ("net: rfs: add hash collision detection")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Tom Herbert <tom@herbertland.com>
Cc: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 04ffcb255f ("net: Add ndo_gso_check") Tom originally
added the 'dev' argument to be able to call ndo_gso_check().
Then later, when generalizing this in commit 5f35227ea3
("net: Generalize ndo_gso_check to ndo_features_check")
Jesse removed the call to ndo_gso_check() in netif_needs_gso()
by calling the new ndo_features_check() in a different place.
This made the 'dev' argument unused.
Remove the unused argument and go back to the code as before.
Cc: Tom Herbert <therbert@google.com>
Cc: Jesse Gross <jesse@nicira.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Even if we make use of classifier and actions from the egress
path, we're going into handle_ing() executing additional code
on a per-packet cost for ingress qdisc, just to realize that
nothing is attached on ingress.
Instead, this can just be blinded out as a no-op entirely with
the use of a static key. On input fast-path, we already make
use of static keys in various places, e.g. skb time stamping,
in RPS, etc. It makes sense to not waste time when we're assured
that no ingress qdisc is attached anywhere.
Enabling/disabling of that code path is being done via two
helpers, namely net_{inc,dec}_ingress_queue(), that are being
invoked under RTNL mutex when a ingress qdisc is being either
initialized or destructed.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On the output paths in particular, we have to sometimes deal with two
socket contexts. First, and usually skb->sk, is the local socket that
generated the frame.
And second, is potentially the socket used to control a tunneling
socket, such as one the encapsulates using UDP.
We do not want to disassociate skb->sk when encapsulating in order
to fix this, because that would break socket memory accounting.
The most extreme case where this can cause huge problems is an
AF_PACKET socket transmitting over a vxlan device. We hit code
paths doing checks that assume they are dealing with an ipv4
socket, but are actually operating upon the AF_PACKET one.
Signed-off-by: David S. Miller <davem@davemloft.net>
Conflicts:
drivers/net/ethernet/mellanox/mlx4/cmd.c
net/core/fib_rules.c
net/ipv4/fib_frontend.c
The fib_rules.c and fib_frontend.c conflicts were locking adjustments
in 'net' overlapping addition and removal of code in 'net-next'.
The mlx4 conflict was a bug fix in 'net' happening in the same
place a constant was being replaced with a more suitable macro.
Signed-off-by: David S. Miller <davem@davemloft.net>
We should not consult skb->sk for output decisions in xmit recursion
levels > 0 in the stack. Otherwise local socket settings could influence
the result of e.g. tunnel encapsulation process.
ipv6 does not conform with this in three places:
1) ip6_fragment: we do consult ipv6_npinfo for frag_size
2) sk_mc_loop in ipv6 uses skb->sk and checks if we should
loop the packet back to the local socket
3) ip6_skb_dst_mtu could query the settings from the user socket and
force a wrong MTU
Furthermore:
In sk_mc_loop we could potentially land in WARN_ON(1) if we use a
PF_PACKET socket ontop of an IPv6-backed vxlan device.
Reuse xmit_recursion as we are currently only interested in protecting
tunnel devices.
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Hannes Frederic Sowa <hannes@stressinduktion.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Virtual interfaces are supposed to set an iflink value != of their ifindex.
It was not the case for some of them, like vxlan, bond or bridge.
Let's set iflink to 0 when dev->rtnl_link_ops is set.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that all users of iflink have the ndo_get_iflink handler available, it's
possible to remove this field.
By default, dev_get_iflink() returns the ifindex of the interface.
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The goal of this patch is to prepare the removal of the iflink field. It
introduces a new ndo function, which will be implemented by virtual interfaces.
There is no functional change into this patch. All readers of iflink field
now call dev_get_iflink().
Signed-off-by: Nicolas Dichtel <nicolas.dichtel@6wind.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Unlike other places, this function uses name "dev" for what should be
"orig_dev", which might be a bit confusing. So fix this.
Signed-off-by: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
As there are a number of (especially virtual) devices that don't
need the multiple vlan check, introduce passthru_features_check() for
convenience.
Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
