One of the problem with sock memory accounting is it uses
a pair of sock_hold()/sock_put() for each transmitted packet.
This slows down bidirectional flows because the receive path
also needs to take a refcount on socket and might use a different
cpu than transmit path or transmit completion path. So these
two atomic operations also trigger cache line bounces.
We can see this in tx or tx/rx workloads (media gateways for example),
where sock_wfree() can be in top five functions in profiles.
We use this sock_hold()/sock_put() so that sock freeing
is delayed until all tx packets are completed.
As we also update sk_wmem_alloc, we could offset sk_wmem_alloc
by one unit at init time, until sk_free() is called.
Once sk_free() is called, we atomic_dec_and_test(sk_wmem_alloc)
to decrement initial offset and atomicaly check if any packets
are in flight.
skb_set_owner_w() doesnt call sock_hold() anymore
sock_wfree() doesnt call sock_put() anymore, but check if sk_wmem_alloc
reached 0 to perform the final freeing.
Drawback is that a skb->truesize error could lead to unfreeable sockets, or
even worse, prematurely calling __sk_free() on a live socket.
Nice speedups on SMP. tbench for example, going from 2691 MB/s to 2711 MB/s
on my 8 cpu dev machine, even if tbench was not really hitting sk_refcnt
contention point. 5 % speedup on a UDP transmit workload (depends
on number of flows), lowering TX completion cpu usage.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Define three accessors to get/set dst attached to a skb
struct dst_entry *skb_dst(const struct sk_buff *skb)
void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
void skb_dst_drop(struct sk_buff *skb)
This one should replace occurrences of :
dst_release(skb->dst)
skb->dst = NULL;
Delete skb->dst field
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Define skb_rtable(const struct sk_buff *skb) accessor to get rtable from skb
Delete skb->rtable field
Setting rtable is not allowed, just set dst instead as rtable is an alias.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The IP MIB (RFC 4293) defines stats for InOctets, OutOctets, InMcastOctets and
OutMcastOctets:
http://tools.ietf.org/html/rfc4293
But it seems we don't track those in any way that easy to separate from other
protocols. This patch adds those missing counters to the stats file. Tested
successfully by me
With help from Eric Dumazet.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instructions for time stamping outgoing packets are take from the
socket layer and later copied into the new skb.
Signed-off-by: Patrick Ohly <patrick.ohly@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can reduce pressure on dst entry refcount that slowdown UDP transmit
path on SMP machines. This pressure is visible on RTP servers when
delivering content to mediagateways, especially big ones, handling
thousand of streams. Several cpus send UDP frames to the same
destination, hence use the same dst entry.
This patch makes ip_push_pending_frames() steal the refcount its
callers had to take when filling inet->cork.dst.
This doesnt avoid all refcounting, but still gives speedups on SMP,
on UDP/RAW transmit path.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can reduce pressure on dst entry refcount that slowdown UDP transmit
path on SMP machines. This pressure is visible on RTP servers when
delivering content to mediagateways, especially big ones, handling
thousand of streams. Several cpus send UDP frames to the same
destination, hence use the same dst entry.
This patch makes ip_append_data() eventually steal the refcount its
callers had to take on the dst entry.
This doesnt avoid all refcounting, but still gives speedups on SMP,
on UDP/RAW transmit path
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Netfilter's ip_route_me_harder() tries to re-route packets either
generated or re-routed by Netfilter. This patch changes
ip_route_me_harder() to handle packets from non-locally-bound sockets
with IP_TRANSPARENT set as local and to set the appropriate flowi
flags when re-doing the routing lookup.
Signed-off-by: KOVACS Krisztian <hidden@sch.bme.hu>
Signed-off-by: David S. Miller <davem@davemloft.net>
Removes legacy reinvent-the-wheel type thing. The generic
machinery integrates much better to automated debugging aids
such as kerneloops.org (and others), and is unambiguous due to
better naming. Non-intuively BUG_TRAP() is actually equal to
WARN_ON() rather than BUG_ON() though some might actually be
promoted to BUG_ON() but I left that to future.
I could make at least one BUILD_BUG_ON conversion.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
All the callers already have either the net itself, or the place
where to get it from.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This routine deals with ICMP statistics, but doesn't have a
struct net at hands, so add one.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch removes CVS keywords that weren't updated for a long time
from comments.
Signed-off-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Problem: ip_append_data() could wrongly generate a chained skb for
devices which support UFO. When sk_write_queue is not empty
(e.g. MSG_MORE), __instead__ of appending data into the next nr_frag
of the queued skb, a new chained skb is created.
I would normally assume UFO device should get data in nr_frags and not
in frag_list. Later the udp4_hwcsum_outgoing() resets csum to NONE
and skb_gso_segment() has oops.
Proposal:
1. Even length is less than mtu, employ ip_ufo_append_data()
and append data to the __existed__ skb in the sk_write_queue.
2. ip_ufo_append_data() is fixed due to a wrong manipulation of
peek-ing and later enqueue-ing of the same skb. Now, enqueuing is
always performed, because on error the further
ip_flush_pending_frames() would release the queued skb.
Signed-off-by: Kostya B <bkostya@hotmail.com>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Introduce per-sock inlines: sock_net(), sock_net_set()
and per-inet_timewait_sock inlines: twsk_net(), twsk_net_set().
Without CONFIG_NET_NS, no namespace other than &init_net exists.
Let's explicitly define them to help compiler optimizations.
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Replace all the rest of the init_net with a proper net on the IP layer.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
(Anonymous) unions can help us to avoid ugly casts.
A common cast it the (struct rtable *)skb->dst one.
Defining an union like :
union {
struct dst_entry *dst;
struct rtable *rtable;
};
permits to use skb->rtable in place.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
A userspace program may wish to set the mark for each packets its send
without using the netfilter MARK target. Changing the mark can be used
for mark based routing without netfilter or for packet filtering.
It requires CAP_NET_ADMIN capability.
Signed-off-by: Laszlo Attila Toth <panther@balabit.hu>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
When ip_fragment has to hit the slow path the value of skb->truesize
may go out of sync because we would have updated it without changing
the packet length. This violates the constraints on truesize.
This patch postpones the update of skb->truesize to prevent this.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
All needed API is done, the namespace is available when required from
the device on the DST entry from the incoming packet. So, just replace
init_net with proper namespace.
Other protocols will follow.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Needed to propagate it down to the ip_route_output_flow.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Needed to propagate it down to the __ip_route_output_key.
Signed_off_by: Denis V. Lunev <den@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
It seems that ip_build_xmit is no longer used in here and
ip_append_data is used.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
The IPv4 and IPv6 hook values are identical, yet some code tries to figure
out the "correct" value by looking at the address family. Introduce NF_INET_*
values for both IPv4 and IPv6. The old values are kept in a #ifndef __KERNEL__
section for userspace compatibility.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Acked-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Most callers of the LOCAL_OUT chain will set the IP packet length and
header checksum before doing so. They also share the same output
function dst_output.
This patch creates a new function called ip_local_out which does all
of that and converts the appropriate users over to it.
Apart from removing duplicate code, it will also help in merging the
IPsec output path once the same thing is done for IPv6.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
As it is ip_append_data only counts page fragments to the skb that
allocated it. As such it means that the first skb gets hit with a
4K charge even though it might have only used a fraction of it while
all subsequent skb's that use the same page gets away with no charge
at all.
This bug was exposed by the UDP accounting patch.
[ The wmem_alloc bumping needs to be moved with the truesize,
noticed by Takahiro Yasui. -DaveM ]
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
The ip_push_pending_frames and ip_flush_pending_frames do the
same things to flush the sock's cork. Move this into a separate
function and save ~80 bytes from the .text
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In some places, the result of skb_headroom() is compared to an unsigned
integer, and in others, the result is compared to a signed integer. Make
the comparisons consistent and correct.
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that we don't pass double skb pointers to nf_hook_slow anymore, gcc
can generate tail calls for some of the netfilter hook okfn invocations,
so there is no need to inline the functions anymore. This caused huge
code bloat since we ended up with one inlined version and one out-of-line
version since we pass the address to nf_hook_slow.
Before:
text data bss dec hex filename
8997385 1016524 524652 10538561 a0ce41 vmlinux
After:
text data bss dec hex filename
8994009 1016524 524652 10535185 a0c111 vmlinux
-------------------------------------------------------
-3376
All cases have been verified to generate tail-calls with and without
netfilter. The okfns in ipmr and xfrm4_input still remain inline because
gcc can't generate tail-calls for them.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since hardware header operations are part of the protocol class
not the device instance, make them into a separate object and
save memory.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Background: RFC 4293 deprecates existing individual, named ICMP
type counters to be replaced with the ICMPMsgStatsTable. This table
includes entries for both IPv4 and IPv6, and requires counting of all
ICMP types, whether or not the machine implements the type.
These patches "remove" (but not really) the existing counters, and
replace them with the ICMPMsgStats tables for v4 and v6.
It includes the named counters in the /proc places they were, but gets the
values for them from the new tables. It also counts packets generated
from raw socket output (e.g., OutEchoes, MLD queries, RA's from
radvd, etc).
Changes:
1) create icmpmsg_statistics mib
2) create icmpv6msg_statistics mib
3) modify existing counters to use these
4) modify /proc/net/snmp to add "IcmpMsg" with all ICMP types
listed by number for easy SNMP parsing
5) modify /proc/net/snmp printing for "Icmp" to get the named data
from new counters.
Signed-off-by: David L Stevens <dlstevens@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch cleans up duplicate includes in
net/ipv4/
Signed-off-by: Jesper Juhl <jesper.juhl@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The TRACE target can be used to follow IP and IPv6 packets through
the ruleset.
Signed-off-by: Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>
Signed-off-by: Patrick NcHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
The existing model for checksum offload does not correctly handle
devices that can offload IPV4 and IPV6 only. The NETIF_F_HW_CSUM flag
implies device can do any arbitrary protocol.
This patch:
* adds NETIF_F_IPV6_CSUM for those devices
* fixes bnx2 and tg3 devices that need it
* add NETIF_F_IPV6_CSUM to ipv6 output (incl GSO)
* fixes assumptions about NETIF_F_ALL_CSUM in nat
* adjusts bridge union of checksumming computation
Signed-off-by: David S. Miller <davem@davemloft.net>
A time_wait socket inherits sk_bound_dev_if from the original socket,
but it is not used when sending ACK packets using ip_send_reply.
Fix by passing the oif to ip_send_reply in struct ip_reply_arg and
use it for output routing.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
A transmitted IP multicast datagram should be counted as OutMcastPkts.
By the same token, a transmitted IP broadcast datagram should be
counted as OutBcastPkts.
Signed-off-by: Mitsuru Chinen <mitch@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add IP(V6)_PMTUDISC_PROBE value for IP(V6)_MTU_DISCOVER. This option forces
us not to fragment, but does not make use of the kernel path MTU discovery.
That is, it allows for user-mode MTU probing (or, packetization-layer path
MTU discovery). This is particularly useful for diagnostic utilities, like
traceroute/tracepath.
Signed-off-by: John Heffner <jheffner@psc.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>
To clearly state the intent of copying from linear sk_buffs, _offset being a
overly long variant but interesting for the sake of saving some bytes.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Renaming skb->h to skb->transport_header, skb->nh to skb->network_header and
skb->mac to skb->mac_header, to match the names of the associated helpers
(skb[_[re]set]_{transport,network,mac}_header).
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the common sequence "skb->h.raw - skb->nh.raw", similar to skb->mac_len,
that is precalculated tho, don't think we need to bloat skb with one more
member, so just use this new helper, reducing the number of non-skbuff.h
references to the layer headers even more.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This unifies the codes to copy netfilter related datas. Before copying,
nf_copy() puts original members in destination skb.
Signed-off-by: Yasuyuki Kozakai <yasuyuki.kozakai@toshiba.co.jp>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the places where we need a pointer to the transport header, it is
still legal to touch skb->h.raw directly if just adding to,
subtracting from or setting it to another layer header.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the cases where the transport header is being set to a offset from
skb->data.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the common, open coded 'skb->h.raw = skb->data' operation, so that we can
later turn skb->h.raw into a offset, reducing the size of struct sk_buff in
64bit land while possibly keeping it as a pointer on 32bit.
This one touches just the most simple cases:
skb->h.raw = skb->data;
skb->h.raw = {skb_push|[__]skb_pull}()
The next ones will handle the slightly more "complex" cases.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet