Use "hlist_nulls" infrastructure we added in 2.6.29 for RCUification of UDP & TCP.
This permits an easy conversion from call_rcu() based hash lists to a
SLAB_DESTROY_BY_RCU one.
Avoiding call_rcu() delay at nf_conn freeing time has numerous gains.
First, it doesnt fill RCU queues (up to 10000 elements per cpu).
This reduces OOM possibility, if queued elements are not taken into account
This reduces latency problems when RCU queue size hits hilimit and triggers
emergency mode.
- It allows fast reuse of just freed elements, permitting better use of
CPU cache.
- We delete rcu_head from "struct nf_conn", shrinking size of this structure
by 8 or 16 bytes.
This patch only takes care of "struct nf_conn".
call_rcu() is still used for less critical conntrack parts, that may
be converted later if necessary.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Commit e1b4b9f ([NETFILTER]: {ip,ip6,arp}_tables: fix exponential worst-case
search for loops) introduced a regression in the loop detection algorithm,
causing sporadic incorrectly detected loops.
When a chain has already been visited during the check, it is treated as
having a standard target containing a RETURN verdict directly at the
beginning in order to not check it again. The real target of the first
rule is then incorrectly treated as STANDARD target and checked not to
contain invalid verdicts.
Fix by making sure the rule does actually contain a standard target.
Based on patch by Francis Dupont <Francis_Dupont@isc.org>
Signed-off-by: Patrick McHardy <kaber@trash.net>
We use same not trivial helper function in four places. We can factorize it.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
The ip_queue module is missing the net-pf-16-proto-3 alias that would
causae it to be auto-loaded when a socket of that type is opened. This
patch adds the alias.
Signed-off-by: Scott James Remnant <scott@canonical.com>
Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
This patch increases the statistics of packets drop if the sequence
adjustment fails in ipv4_confirm().
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Patrick McHardy <kaber@trash.net>
This patch modifies nf_log to use a linked list of loggers for each
protocol. This list of loggers is read and write protected with a
mutex.
This patch separates registration and binding. To be used as
logging module, a module has to register calling nf_log_register()
and to bind to a protocol it has to call nf_log_bind_pf().
This patch also converts the logging modules to the new API. For nfnetlink_log,
it simply switchs call to register functions to call to bind function and
adds a call to nf_log_register() during init. For other modules, it just
remove a const flag from the logger structure and replace it with a
__read_mostly.
Signed-off-by: Eric Leblond <eric@inl.fr>
Signed-off-by: Patrick McHardy <kaber@trash.net>
While doing oprofile tests I noticed two loops are not properly unrolled by gcc
Using a hand coded unrolled loop provides nice speedup : ipt_do_table
credited of 2.52 % of cpu instead of 3.29 % in tbench.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
The reader/writer lock in ip_tables is acquired in the critical path of
processing packets and is one of the reasons just loading iptables can cause
a 20% performance loss. The rwlock serves two functions:
1) it prevents changes to table state (xt_replace) while table is in use.
This is now handled by doing rcu on the xt_table. When table is
replaced, the new table(s) are put in and the old one table(s) are freed
after RCU period.
2) it provides synchronization when accesing the counter values.
This is now handled by swapping in new table_info entries for each cpu
then summing the old values, and putting the result back onto one
cpu. On a busy system it may cause sampling to occur at different
times on each cpu, but no packet/byte counts are lost in the process.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Sucessfully tested on my dual quad core machine too, but iptables only (no ipv6 here)
BTW, my new "tbench 8" result is 2450 MB/s, (it was 2150 MB/s not so long ago)
Acked-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Concern has been expressed about the changing Kconfig options.
Provide the old options that forward-select.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Suggested by: James King <t.james.king@gmail.com>
Similarly to commit c9fd496809, merge
TTL and HL. Since HL does not depend on any IPv6-specific function,
no new module dependencies would arise.
With slight adjustments to the Kconfig help text.
Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Signed-off-by: Patrick McHardy <kaber@trash.net>
x86 and powerpc can perform long word accesses in an efficient maner.
We can use this to unroll two loops in arp_packet_match(), to
perform arithmetic on long words instead of bytes. This is a win
on x86_64 for example.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Later patches change the locking on xt_table and the initialization of
the lock element is not needed since the lock is always initialized in
xt_table_register anyway.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Impact: syntax fix
Interestingly enough this compiles w/o any complaints:
orphans = percpu_counter_sum_positive(&tcp_orphan_count),
sockets = percpu_counter_sum_positive(&tcp_sockets_allocated),
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
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>
gro: Optimise TCP packet reception
As this function can be called more than half a million times for
10GbE, it's important to optimise it as much as we can.
This patch uses bit ops to logical ops, as well as open coding
memcmp to exploit alignment properties.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
As this function can be called more than half a million times for
10GbE, it's important to optimise it as much as we can.
This patch does some obvious changes to use 2-byte and 4-byte
operations instead of byte-oriented ones where possible. Bit
ops are also used to replace logical ops to reduce branching.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Like the UDP header fix, pskb_may_pull() can potentially
alter the SKB buffer. Thus the saddr and daddr, pointers
may point to the old skb->data buffer.
I haven't seen corruptions, as its only seen if the old
skb->data buffer were reallocated by another user and
written into very quickly (or poison'd by SLAB debugging).
Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
The UDP header pointer assignment must happen after calling
pskb_may_pull(). As pskb_may_pull() can potentially alter the SKB
buffer.
This was exposted by running multicast traffic through the NIU driver,
as it won't prepull the protocol headers into the linear area on
receive.
Signed-off-by: Jesper Dangaard Brouer <hawk@comx.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 93821778de (udp: Fix rcv socket
locking) accidentally removed sk_drops increments for UDP IPV4
sockets.
This field can be used to detect incorrect sizing of socket receive
buffers.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sk_alloc now sets sk_family so this is redundant. In fact it caught
my eye because sock_init_data already uses sk_family so this is too
late anyway.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
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>
From: Stephen Hemminger <shemminger@vyatta.com>
Fix regression introduced by a9d8f9110d
("inet: Allowing more than 64k connections and heavily optimize
bind(0) time.")
Based upon initial patches and feedback from Evegniy Polyakov and
Eric Dumazet.
From Eric Dumazet:
--------------------
Also there might be a problem at line 175
if (sk->sk_reuse && sk->sk_state != TCP_LISTEN && --attempts >= 0) {
spin_unlock(&head->lock);
goto again;
If we entered inet_csk_get_port() with a non null snum, we can "goto again"
while it was not expected.
--------------------
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds another inet device option to enable gratuitous ARP
when device is brought up or address change. This is handy for
clusters or virtualization.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Base versions handle constant folding now.
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Unfortunately simplicity isn't always the best. The fraginfo
interface turned out to be suboptimal. The problem was quite
obvious. For every packet, we have to copy the headers from
the frags structure into skb->head, even though for 99% of the
packets this part is immediately thrown away after the merge.
LRO didn't have this problem because it directly read the headers
from the frags structure.
This patch attempts to address this by creating an interface
that allows GRO to access the headers in the first frag without
having to copy it. Because all drivers that use frags place the
headers in the first frag this optimisation should be enough.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
tcp_splice_data_recv has two lengths to consider: the len parameter it
gets from tcp_read_sock, which specifies the amount of data in the skb,
and rd_desc->count, which is the amount of data the splice caller still
wants. Currently it passes just the latter to skb_splice_bits, which then
splices min(rd_desc->count, skb->len - offset) bytes.
Most of the time this is fine, except when the skb contains urgent data.
In that case len goes only up to the urgent byte and is less than
skb->len - offset. By ignoring len tcp_splice_data_recv may a) splice
data tcp_read_sock told it not to, b) return to tcp_read_sock a value > len.
Now, tcp_read_sock doesn't handle used > len and leaves the socket in a
bad state (both sk_receive_queue and copied_seq are bad at that point)
resulting in duplicated data and corruption.
Fix by passing min(rd_desc->count, len) to skb_splice_bits.
Signed-off-by: Dimitris Michailidis <dm@chelsio.com>
Acked-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
commit 9088c56095
(udp: Improve port randomization) introduced a regression for UDP bind() syscall
to null port (getting a random port) in case lot of ports are already in use.
This is because we do about 28000 scans of very long chains (220 sockets per chain),
with many spin_lock_bh()/spin_unlock_bh() calls.
Fix this using a bitmap (64 bytes for current value of UDP_HTABLE_SIZE)
so that we scan chains at most once.
Instead of 250 ms per bind() call, we get after patch a time of 2.9 ms
Based on a report from Vitaly Mayatskikh
Reported-by: Vitaly Mayatskikh <v.mayatskih@gmail.com>
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Tested-by: Vitaly Mayatskikh <v.mayatskih@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Instead of keeping candidate tunnel device from all categories,
keep only one candidate with best score. This optimizes stack
usage and speeds up exit code.
Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
This last patch makes the appropriate changes to use and propagate the
network namespace where needed in IPv4 multicast routing code.
This consists mainly in replacing all the remaining init_net occurences
with current netns pointer retrieved from sockets, net devices or
mfc_caches depending on the routines' contexts.
Some routines receive a new 'struct net' parameter to propagate the current
netns:
* vif_add/vif_delete
* ipmr_new_tunnel
* mroute_clean_tables
* ipmr_cache_find
* ipmr_cache_report
* ipmr_cache_unresolved
* ipmr_mfc_add/ipmr_mfc_delete
* ipmr_get_route
* rt_fill_info (in route.c)
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Declare IPv4 multicast forwarding /proc/net entries per-namespace:
/proc/net/ip_mr_vif
/proc/net/ip_mr_cache
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Preliminary work to make IPv4 multicast routing netns-aware.
Declare variable 'reg_vif_num' per-namespace, move into struct netns_ipv4.
At the moment, this variable is only referenced in init_net.
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Preliminary work to make IPv4 multicast routing netns-aware.
Declare IPv multicast routing variables 'mroute_do_assert' and
'mroute_do_pim' per-namespace in struct netns_ipv4.
At the moment, these variables are only referenced in init_net.
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Preliminary work to make IPv4 multicast routing netns-aware.
Declare variable cache_resolve_queue_len per-namespace: move it into
struct netns_ipv4.
This variable counts the number of unresolved cache entries queued in the
list mfc_unres_queue. This list is kept global to all netns as the number
of entries per namespace is limited to 10 (hardcoded in routine
ipmr_cache_unresolved).
Entries belonging to different namespaces in mfc_unres_queue will be
identified by matching the mfc_net member introduced previously in
struct mfc_cache.
Keeping this list global to all netns, also allows us to keep a single
timer (ipmr_expire_timer) to handle their expiration.
In some places cache_resolve_queue_len value was tested for arming
or deleting the timer. These tests were equivalent to testing
mfc_unres_queue value instead and are replaced in this patch.
At the moment, cache_resolve_queue_len is only referenced in init_net.
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Preliminary work to make IPv4 multicast routing netns-aware.
Dynamically allocate IPv4 multicast forwarding cache, mfc_cache_array,
and move it to struct netns_ipv4.
At the moment, mfc_cache_array is only referenced in init_net.
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch stores into struct mfc_cache the network namespace each
mfc_cache belongs to. The new member is mfc_net.
mfc_net is assigned at cache allocation and doesn't change during
the rest of the cache entry life.
A new net parameter is added to ipmr_cache_alloc/ipmr_cache_alloc_unres.
This will help to retrieve the current netns around the IPv4 multicast
routing code.
At the moment, all mfc_cache are allocated in init_net.
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Preliminary work to make IPv6 multicast routing netns-aware.
Dynamically allocate interface table vif_table and move it to
struct netns_ipv4, and update MIF_EXISTS() macro.
At the moment, vif_table is only referenced in init_net.
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Preliminary work to make IPv4 multicast routing netns-aware.
Make IPv4 multicast routing mroute_socket per-namespace,
moves it into struct netns_ipv4.
At the moment, mroute_socket is only referenced in init_net.
Signed-off-by: Benjamin Thery <benjamin.thery@bull.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Check the device on receive path and allow otherwise identical devices
as long as the physical device differs.
This is useful for NBMA tunnels, where you want to use different gre IP
for each public IP available via different physical devices.
Signed-off-by: Timo Teras <timo.teras@iki.fi>
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>
When we get a GSO packet from an untrusted source, we need to
ensure that it is sufficiently long so that we don't end up
crashing.
Based on discovery and patch by Ian Campbell.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Tested-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As spotted by Willy Tarreau, current splice() from tcp socket to pipe is not
optimal. It processes at most one segment per call.
This results in low performance and very high overhead due to syscall rate
when splicing from interfaces which do not support LRO.
Willy provided a patch inside tcp_splice_read(), but a better fix
is to let tcp_read_sock() process as many segments as possible, so
that tcp_rcv_space_adjust() and tcp_cleanup_rbuf() are called less
often.
With this change, splice() behaves like tcp_recvmsg(), being able
to consume many skbs in one system call. With typical 1460 bytes
of payload per frame, that means splice(SPLICE_F_NONBLOCK) can return
16*1460 = 23360 bytes.
Signed-off-by: Willy Tarreau <w@1wt.eu>
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
An old bug crept back into the ICMP/ICMPv6 conntrack protocols: the timeout
values are defined as unsigned longs, the sysctl's maxsize is set to
sizeof(unsigned int). Use unsigned int for the timeout values as in the
other conntrack protocols.
Reported-by: Jean-Mickael Guerin <jean-mickael.guerin@6wind.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
