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>
And switch bsockets to atomic_t since it might be changed in parallel.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Evgeniy Polyakov <zbr@ioremap.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 9db66bdcc8 (net: convert
TCP/DCCP ehash rwlocks to spinlocks), I forgot to change one
occurrence of rwlock_t to spinlock_t
I believe sizeof(raw_spinlock_t) might be > 0 on !CONFIG_SMP if
CONFIG_DEBUG_SPINLOCK while sizeof(raw_rwlock_t) should be 0 in this
case.
Fortunatly, CONFIG_DEBUG_SPINLOCK adds fields to both spinlock_t and
rwlock_t, but at this might change in the future (being able to debug
spinlocks but not rwlocks for example), better to be safe.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With simple extension to the binding mechanism, which allows to bind more
than 64k sockets (or smaller amount, depending on sysctl parameters),
we have to traverse the whole bind hash table to find out empty bucket.
And while it is not a problem for example for 32k connections, bind()
completion time grows exponentially (since after each successful binding
we have to traverse one bucket more to find empty one) even if we start
each time from random offset inside the hash table.
So, when hash table is full, and we want to add another socket, we have
to traverse the whole table no matter what, so effectivelly this will be
the worst case performance and it will be constant.
Attached picture shows bind() time depending on number of already bound
sockets.
Green area corresponds to the usual binding to zero port process, which
turns on kernel port selection as described above. Red area is the bind
process, when number of reuse-bound sockets is not limited by 64k (or
sysctl parameters). The same exponential growth (hidden by the green
area) before number of ports reaches sysctl limit.
At this time bind hash table has exactly one reuse-enbaled socket in a
bucket, but it is possible that they have different addresses. Actually
kernel selects the first port to try randomly, so at the beginning bind
will take roughly constant time, but with time number of port to check
after random start will increase. And that will have exponential growth,
but because of above random selection, not every next port selection
will necessary take longer time than previous. So we have to consider
the area below in the graph (if you could zoom it, you could find, that
there are many different times placed there), so area can hide another.
Blue area corresponds to the port selection optimization.
This is rather simple design approach: hashtable now maintains (unprecise
and racely updated) number of currently bound sockets, and when number
of such sockets becomes greater than predefined value (I use maximum
port range defined by sysctls), we stop traversing the whole bind hash
table and just stop at first matching bucket after random start. Above
limit roughly corresponds to the case, when bind hash table is full and
we turned on mechanism of allowing to bind more reuse-enabled sockets,
so it does not change behaviour of other sockets.
Signed-off-by: Evgeniy Polyakov <zbr@ioremap.net>
Tested-by: Denys Fedoryschenko <denys@visp.net.lb>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is the last step to be able to perform full RCU lookups
in __inet_lookup() : After established/timewait tables, we
add RCU lookups to listening hash table.
The only trick here is that a socket of a given type (TCP ipv4,
TCP ipv6, ...) can now flight between two different tables
(established and listening) during a RCU grace period, so we
must use different 'nulls' end-of-chain values for two tables.
We define a large value :
#define LISTENING_NULLS_BASE (1U << 29)
So that slots in listening table are guaranteed to have different
end-of-chain values than slots in established table. A reader can
still detect it finished its lookup in the right chain.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can avoid some useless instructions if !CONFIG_NET_NS
Because of RCU, we use INET_MATCH or INET_TW_MATCH twice for the found
socket, so thats six instructions less per incoming TCP packet.
Yet another tbench speedup :)
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now TCP & DCCP use RCU lookups, we can convert ehash rwlocks to spinlocks.
/proc/net/tcp and other seq_file 'readers' can safely be converted to 'writers'.
This should speedup writers, since spin_lock()/spin_unlock()
only use one atomic operation instead of two for write_lock()/write_unlock()
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch prepares RCU migration of listening_hash table for
TCP/DCCP protocols.
listening_hash table being small (32 slots per protocol), we add
a spinlock for each slot, instead of a single rwlock for whole table.
This should reduce hold time of readers, and writers concurrency.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RCU was added to UDP lookups, using a fast infrastructure :
- sockets kmem_cache use SLAB_DESTROY_BY_RCU and dont pay the
price of call_rcu() at freeing time.
- hlist_nulls permits to use few memory barriers.
This patch uses same infrastructure for TCP/DCCP established
and timewait sockets.
Thanks to SLAB_DESTROY_BY_RCU, no slowdown for applications
using short lived TCP connections. A followup patch, converting
rwlocks to spinlocks will even speedup this case.
__inet_lookup_established() is pretty fast now we dont have to
dirty a contended cache line (read_lock/read_unlock)
Only established and timewait hashtable are converted to RCU
(bind table and listen table are still using traditional locking)
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We can shrink size of "struct inet_bind_bucket" by 50%, using
read_pnet() and write_pnet()
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the socket cached in the skb if it's present.
Signed-off-by: KOVACS Krisztian <hidden@sch.bme.hu>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
To be able to use the cached socket reference in the skb during input
processing we add a new set of lookup functions that receive the skb on
their argument list.
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: KOVACS Krisztian <hidden@sch.bme.hu>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are many possible ways to add this "salt", thus I made this
patch to be the last in the series to change it if required.
Currently I propose to use the struct net pointer itself as this
salt, but since this pointer is most often cache-line aligned, shift
this right to eliminate the bits, that are most often zeroed.
After this, simply add this mix to prepared hashfn-s.
For CONFIG_NET_NS=n case this salt is 0 and no changes in hashfn
appear.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Listening-on-one-port sockets in many namespaces produce long
chains in the listening_hash-es, so prepare the inet_lhashfn to
take struct net into account.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Binding to some port in many namespaces may create too long
chains in bhash-es, so prepare the hashfn to take struct net
into account.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This deblats ~200 bytes when ipv6 and dccp are 'y'.
Besides, this will ease compilation issues for patches
I'm working on to make inet hash tables more scalable
wrt net namespaces.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
As I can see from the code, two places (tcp_v6_syn_recv_sock and
dccp_v6_request_recv_sock) that call this one already run with
BHs disabled, so it's safe to call __inet_inherit_port there.
Besides (in case I missed smth with code review) the calltrace
tcp_v6_syn_recv_sock
`- tcp_v4_syn_recv_sock
`- __inet_inherit_port
and the similar for DCCP are valid, but assumes BHs to be disabled.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
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>
Inspired by the commit ab1e0a13 ([SOCK] proto: Add hashinfo member to
struct proto) from Arnaldo, I made similar thing for UDP/-Lite IPv4
and -v6 protocols.
The result is not that exciting, but it removes some levels of
indirection in udpxxx_get_port and saves some space in code and text.
The first step is to union existing hashinfo and new udp_hash on the
struct proto and give a name to this union, since future initialization
of tcpxxx_prot, dccp_vx_protinfo and udpxxx_protinfo will cause gcc
warning about inability to initialize anonymous member this way.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The port offset calculations depend on the protocol family, but, as
Adrian noticed, I broke this logic with the commit
5ee31fc1ec
[INET]: Consolidate inet(6)_hash_connect.
Return this logic back, by passing the port offset directly into the
consolidated function.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Noticed-by: Adrian Bunk <bunk@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This way we can remove TCP and DCCP specific versions of
sk->sk_prot->get_port: both v4 and v6 use inet_csk_get_port
sk->sk_prot->hash: inet_hash is directly used, only v6 need
a specific version to deal with mapped sockets
sk->sk_prot->unhash: both v4 and v6 use inet_hash directly
struct inet_connection_sock_af_ops also gets a new member, bind_conflict, so
that inet_csk_get_port can find the per family routine.
Now only the lookup routines receive as a parameter a struct inet_hashtable.
With this we further reuse code, reducing the difference among INET transport
protocols.
Eventually work has to be done on UDP and SCTP to make them share this
infrastructure and get as a bonus inet_diag interfaces so that iproute can be
used with these protocols.
net-2.6/net/ipv4/inet_hashtables.c:
struct proto | +8
struct inet_connection_sock_af_ops | +8
2 structs changed
__inet_hash_nolisten | +18
__inet_hash | -210
inet_put_port | +8
inet_bind_bucket_create | +1
__inet_hash_connect | -8
5 functions changed, 27 bytes added, 218 bytes removed, diff: -191
net-2.6/net/core/sock.c:
proto_seq_show | +3
1 function changed, 3 bytes added, diff: +3
net-2.6/net/ipv4/inet_connection_sock.c:
inet_csk_get_port | +15
1 function changed, 15 bytes added, diff: +15
net-2.6/net/ipv4/tcp.c:
tcp_set_state | -7
1 function changed, 7 bytes removed, diff: -7
net-2.6/net/ipv4/tcp_ipv4.c:
tcp_v4_get_port | -31
tcp_v4_hash | -48
tcp_v4_destroy_sock | -7
tcp_v4_syn_recv_sock | -2
tcp_unhash | -179
5 functions changed, 267 bytes removed, diff: -267
net-2.6/net/ipv6/inet6_hashtables.c:
__inet6_hash | +8
1 function changed, 8 bytes added, diff: +8
net-2.6/net/ipv4/inet_hashtables.c:
inet_unhash | +190
inet_hash | +242
2 functions changed, 432 bytes added, diff: +432
vmlinux:
16 functions changed, 485 bytes added, 492 bytes removed, diff: -7
/home/acme/git/net-2.6/net/ipv6/tcp_ipv6.c:
tcp_v6_get_port | -31
tcp_v6_hash | -7
tcp_v6_syn_recv_sock | -9
3 functions changed, 47 bytes removed, diff: -47
/home/acme/git/net-2.6/net/dccp/proto.c:
dccp_destroy_sock | -7
dccp_unhash | -179
dccp_hash | -49
dccp_set_state | -7
dccp_done | +1
5 functions changed, 1 bytes added, 242 bytes removed, diff: -241
/home/acme/git/net-2.6/net/dccp/ipv4.c:
dccp_v4_get_port | -31
dccp_v4_request_recv_sock | -2
2 functions changed, 33 bytes removed, diff: -33
/home/acme/git/net-2.6/net/dccp/ipv6.c:
dccp_v6_get_port | -31
dccp_v6_hash | -7
dccp_v6_request_recv_sock | +5
3 functions changed, 5 bytes added, 38 bytes removed, diff: -33
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add a net argument to inet_lookup and propagate it further
into lookup calls. Plus tune the __inet_check_established.
The dccp and inet_diag, which use that lookup functions
pass the init_net into them.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This tags the inet_bind_bucket struct with net pointer,
initializes it during creation and makes a filtering
during lookup.
A better hashfn, that takes the net into account is to
be done in the future, but currently all bind buckets
with similar port will be in one hash chain.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
These two functions are the same except for what they call
to "check_established" and "hash" for a socket.
This saves half-a-kilo for ipv4 and ipv6.
add/remove: 1/0 grow/shrink: 1/4 up/down: 582/-1128 (-546)
function old new delta
__inet_hash_connect - 577 +577
arp_ignore 108 113 +5
static.hint 8 4 -4
rt_worker_func 376 372 -4
inet6_hash_connect 584 25 -559
inet_hash_connect 586 25 -561
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
1) Cleanups (all functions are prefixed by sock_prot_inuse)
sock_prot_inc_use(prot) -> sock_prot_inuse_add(prot,-1)
sock_prot_dec_use(prot) -> sock_prot_inuse_add(prot,-1)
sock_prot_inuse() -> sock_prot_inuse_get()
New functions :
sock_prot_inuse_init() and sock_prot_inuse_free() to abstract pcounter use.
2) if CONFIG_PROC_FS=n, we can zap 'inuse' member from "struct proto",
since nobody wants to read the inuse value.
This saves 1372 bytes on i386/SMP and some cpu cycles.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is -700 bytes from the net/ipv4/built-in.o
add/remove: 1/0 grow/shrink: 1/3 up/down: 340/-1040 (-700)
function old new delta
__inet_lookup_established - 339 +339
tcp_sacktag_write_queue 2254 2255 +1
tcp_v4_err 1304 973 -331
tcp_v4_rcv 2089 1744 -345
tcp_v4_do_rcv 826 462 -364
Exporting is for dccp module (used via e.g. inet_lookup).
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This one is used in quite many places in the networking code and
seems to big to be inline.
After the patch net/ipv4/build-in.o loses ~650 bytes:
add/remove: 2/0 grow/shrink: 0/5 up/down: 461/-1114 (-653)
function old new delta
__inet_hash_nolisten - 282 +282
__inet_hash - 179 +179
tcp_sacktag_write_queue 2255 2254 -1
__inet_lookup_listener 284 274 -10
tcp_v4_syn_recv_sock 755 493 -262
tcp_v4_hash 389 35 -354
inet_hash_connect 1086 599 -487
This version addresses the issue pointed by Eric, that
while being inline this function was optimized by gcc
in respect to the 'listen_possible' argument.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The inet_ehash_locks_alloc() looks like this:
#ifdef CONFIG_NUMA
if (size > PAGE_SIZE)
x = vmalloc(...);
else
#endif
x = kmalloc(...);
Unlike it, the inet_ehash_locks_alloc() looks like this:
#ifdef CONFIG_NUMA
if (size > PAGE_SIZE)
vfree(x);
else
#else
kfree(x);
#endif
The error is obvious - if the NUMA is on and the size
is less than the PAGE_SIZE we leak the pointer (kfree is
inside the #else branch).
Compiler doesn't warn us because after the kfree(x) there's
a "x = NULL" assignment, so here's another (minor?) bug: we
don't set x to NULL under certain circumstances.
Boring explanation, I know... Patch explains it better.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
As done two years ago on IP route cache table (commit
22c047ccbc) , we can avoid using one
lock per hash bucket for the huge TCP/DCCP hash tables.
On a typical x86_64 platform, this saves about 2MB or 4MB of ram, for
litle performance differences. (we hit a different cache line for the
rwlock, but then the bucket cache line have a better sharing factor
among cpus, since we dirty it less often). For netstat or ss commands
that want a full scan of hash table, we perform fewer memory accesses.
Using a 'small' table of hashed rwlocks should be more than enough to
provide correct SMP concurrency between different buckets, without
using too much memory. Sizing of this table depends on
num_possible_cpus() and various CONFIG settings.
This patch provides some locking abstraction that may ease a future
work using a different model for TCP/DCCP table.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
UDP currently uses skb->dev->ifindex which may provide the wrong
information when the socket bound to a specific interface.
This patch makes inet_iif() accessible to UDP and makes UDP use it.
The scenario we are trying to fix is when a client is running on
the same system and the server and both client and server bind to
a non-loopback device.
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Acked-by: David L Stevens <dlstevens@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix a bunch of sparse warnings. Mostly about 0 used as
NULL pointer, and shadowed variable declarations.
One notable case was that hash size should have been unsigned.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
ehash table layout is currently this one :
First half of this table is used by sockets not in TIME_WAIT state
Second half of it is used by sockets in TIME_WAIT state.
This is non optimal because of for a given hash or socket, the two chain heads
are located in separate cache lines.
Moreover the locks of the second half are never used.
If instead of this halving, we use two list heads in inet_ehash_bucket instead
of only one, we probably can avoid one cache miss, and reduce ram usage,
particularly if sizeof(rwlock_t) is big (various CONFIG_DEBUG_SPINLOCK,
CONFIG_DEBUG_LOCK_ALLOC settings). So we still halves the table but we keep
together related chains to speedup lookups and socket state change.
In this patch I did not try to align struct inet_ehash_bucket, but a future
patch could try to make this structure have a convenient size (a power of two
or a multiple of L1_CACHE_SIZE).
I guess rwlock will just vanish as soon as RCU is plugged into ehash :) , so
maybe we dont need to scratch our heads to align the bucket...
Note : In case struct inet_ehash_bucket is not a power of two, we could
probably change alloc_large_system_hash() (in case it use __get_free_pages())
to free the unused space. It currently allocates a big zone, but the last
quarter of it could be freed. Again, this should be a temporary 'problem'.
Patch tested on ipv4 tcp only, but should be OK for IPV6 and DCCP.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Replace all uses of kmem_cache_t with struct kmem_cache.
The patch was generated using the following script:
#!/bin/sh
#
# Replace one string by another in all the kernel sources.
#
set -e
for file in `find * -name "*.c" -o -name "*.h"|xargs grep -l $1`; do
quilt add $file
sed -e "1,\$s/$1/$2/g" $file >/tmp/$$
mv /tmp/$$ $file
quilt refresh
done
The script was run like this
sh replace kmem_cache_t "struct kmem_cache"
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
inet_lookup() annotated along with helper functions (__inet_lookup(),
__inet_lookup_established(), inet_lookup_established(),
inet_lookup_listener(), __inet_lookup_listener() and inet_ehashfn())
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
INET_MATCH() and friends depend on an interesting set of kludges:
* there's a pair of adjacent fields in struct inet_sock - __be16 dport
followed by __u16 num. We want to search by pair, so we combine the keys into
a single 32bit value and compare with 32bit value read from &...->dport.
* on 64bit targets we combine comparisons with pair of adjacent __be32
fields in the same way.
Make sure that we don't mix those values with anything else and that pairs
we form them from have correct types.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Right now most inet_lookup_* functions take a host-order hnum instead
of a network-order dport because that's how it is represented
internally.
This means that users of these functions have to be careful about
using the right byte-order. To add more confusion, inet_lookup takes
a network-order dport unlike all other functions.
So this patch changes all visible inet_lookup functions to take a
dport and move all dport->hnum conversion inside them.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
By modern standards this function is way too big to be inlined. It's
even bigger than __inet_lookup_listener :)
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
To help in reducing the number of include dependencies, several files were
touched as they were getting needed headers indirectly for stuff they use.
Thanks also to Alan Menegotto for pointing out that net/dccp/proto.c had
linux/dccp.h include twice.
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Renaming it to inet_hash_connect, making it possible to ditch
dccp_v4_hash_connect and share the same code with TCP instead.
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch randomizes the port selected on bind() for connections
to help with possible security attacks. It should also be faster
in most cases because there is no need for a global lock.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
No need to align struct inet_ehash_bucket on a 8 bytes boundary.
On 32 bits Uniprocessor, that's a waste of 4 bytes per struct (50 %)
On other platforms, the attribute is useless, natual alignement is already 8.
platform | Size before | Size after patch
-------------+-------------+------------------
32 bits, UP | 8 | 4
32 bits, SMP | 8 | 8
64 bits, UP | 8 | 8
64 bits, SMP | 16 | 16
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Arnaldo and I agreed it could be applied now, because I have other
pending patches depending on this one (Thank you Arnaldo)
(The other important patch moves skc_refcnt in a separate cache line,
so that the SMP/NUMA performance doesnt suffer from cache line ping pongs)
1) First some performance data :
--------------------------------
tcp_v4_rcv() wastes a *lot* of time in __inet_lookup_established()
The most time critical code is :
sk_for_each(sk, node, &head->chain) {
if (INET_MATCH(sk, acookie, saddr, daddr, ports, dif))
goto hit; /* You sunk my battleship! */
}
The sk_for_each() does use prefetch() hints but only the begining of
"struct sock" is prefetched.
As INET_MATCH first comparison uses inet_sk(__sk)->daddr, wich is far
away from the begining of "struct sock", it has to bring into CPU
cache cold cache line. Each iteration has to use at least 2 cache
lines.
This can be problematic if some chains are very long.
2) The goal
-----------
The idea I had is to change things so that INET_MATCH() may return
FALSE in 99% of cases only using the data already in the CPU cache,
using one cache line per iteration.
3) Description of the patch
---------------------------
Adds a new 'unsigned int skc_hash' field in 'struct sock_common',
filling a 32 bits hole on 64 bits platform.
struct sock_common {
unsigned short skc_family;
volatile unsigned char skc_state;
unsigned char skc_reuse;
int skc_bound_dev_if;
struct hlist_node skc_node;
struct hlist_node skc_bind_node;
atomic_t skc_refcnt;
+ unsigned int skc_hash;
struct proto *skc_prot;
};
Store in this 32 bits field the full hash, not masked by (ehash_size -
1) Using this full hash as the first comparison done in INET_MATCH
permits us immediatly skip the element without touching a second cache
line in case of a miss.
Suppress the sk_hashent/tw_hashent fields since skc_hash (aliased to
sk_hash and tw_hash) already contains the slot number if we mask with
(ehash_size - 1)
File include/net/inet_hashtables.h
64 bits platforms :
#define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
(((__sk)->sk_hash == (__hash))
((*((__u64 *)&(inet_sk(__sk)->daddr)))== (__cookie)) && \
((*((__u32 *)&(inet_sk(__sk)->dport))) == (__ports)) && \
(!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
32bits platforms:
#define TCP_IPV4_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\
(((__sk)->sk_hash == (__hash)) && \
(inet_sk(__sk)->daddr == (__saddr)) && \
(inet_sk(__sk)->rcv_saddr == (__daddr)) && \
(!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif))))
- Adds a prefetch(head->chain.first) in
__inet_lookup_established()/__tcp_v4_check_established() and
__inet6_lookup_established()/__tcp_v6_check_established() and
__dccp_v4_check_established() to bring into cache the first element of the
list, before the {read|write}_lock(&head->lock);
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This also moved inet_iif from tcp to inet_hashtables.h, as it is
needed by the inet_lookup callers, perhaps this needs a bit of
polishing, but for now seems fine.
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This creates struct inet_connection_sock, moving members out of struct
tcp_sock that are shareable with other INET connection oriented
protocols, such as DCCP, that in my private tree already uses most of
these members.
The functions that operate on these members were renamed, using a
inet_csk_ prefix while not being moved yet to a new file, so as to
ease the review of these changes.
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
And also some TIME_WAIT functions.
[acme@toy net-2.6.14]$ grep built-in /tmp/before.size /tmp/after.size
/tmp/before.size: 282955 13122 9312 305389 4a8ed net/ipv4/built-in.o
/tmp/after.size: 281566 13122 9312 304000 4a380 net/ipv4/built-in.o
[acme@toy net-2.6.14]$
I kept them still inlined, will uninline at some point to see what
would be the performance difference.
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This paves the way to generalise the rest of the sock ID lookup
routines and saves some bytes in TCPv4 TIME_WAIT sockets on distro
kernels (where IPv6 is always built as a module):
[root@qemu ~]# grep tw_sock /proc/slabinfo
tw_sock_TCPv6 0 0 128 31 1
tw_sock_TCP 0 0 96 41 1
[root@qemu ~]#
Now if a protocol wants to use the TIME_WAIT generic infrastructure it
only has to set the sk_prot->twsk_obj_size field with the size of its
inet_timewait_sock derived sock and proto_register will create
sk_prot->twsk_slab, for now its only for INET sockets, but we can
introduce timewait_sock later if some non INET transport protocolo
wants to use this stuff.
Next changesets will take advantage of this new infrastructure to
generalise even more TCP code.
[acme@toy net-2.6.14]$ grep built-in /tmp/before.size /tmp/after.size
/tmp/before.size: 188646 11764 5068 205478 322a6 net/ipv4/built-in.o
/tmp/after.size: 188144 11764 5068 204976 320b0 net/ipv4/built-in.o
[acme@toy net-2.6.14]$
Tested with both IPv4 & IPv6 (::1 (localhost) & ::ffff:172.20.0.1
(qemu host)).
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
[acme@toy net-2.6.14]$ grep built-in /tmp/before /tmp/after
/tmp/before: 282560 13122 9312 304994 4a762 net/ipv4/built-in.o
/tmp/after: 282560 13122 9312 304994 4a762 net/ipv4/built-in.o
Will be used in DCCP, not exporting it right now not to get in Adrian
Bunk's exported-but-not-used-on-modules radar 8)
Signed-off-by: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet