Stephen Hemminger a écrit :
> On Thu, 12 Nov 2009 15:11:36 +0100
> Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
>> When handling large number of netdevices, inet_dump_ifaddr()
>> is very slow because it has O(N^2) complexity.
>>
>> Instead of scanning one single list, we can use the NETDEV_HASHENTRIES
>> sub lists of the dev_index hash table, and RCU lookups.
>>
>> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
>
> You might be able to make RCU critical section smaller by moving
> it into loop.
>
Indeed. But we dump at most one skb (<= 8192 bytes ?), so rcu_read_lock
holding time is small, unless we meet many netdevices without
addresses. I wonder if its really common...
Thanks
[PATCH net-next-2.6] ipv4: speedup inet_dump_ifaddr()
When handling large number of netdevices, inet_dump_ifaddr()
is very slow because it has O(N2) complexity.
Instead of scanning one single list, we can use the NETDEV_HASHENTRIES
sub lists of the dev_index hash table, and RCU lookups.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We need to use next_det_device_rcu() in RCU protected section.
We also can avoid in_dev_get()/in_dev_put() overhead (code size mainly)
in rcu_read_lock() sections.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
No longer need read_lock(&dev_base_lock), use RCU instead.
We also can avoid taking references on inet6_dev structs.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Define two symbols needed in both kernel and user space.
Remove old (somewhat incorrect) kernel variant that wasn't used in
most cases. Default should apply to both RMSS and SMSS (RFC2581).
Replace numeric constants with defined symbols.
Stand-alone patch, originally developed for TCPCT.
Signed-off-by: William.Allen.Simpson@gmail.com
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Both vlan and macvlan devices usually don't use a qdisc and immediately
queue packets to the underlying device. Propagate transmission state of
the underlying device to the upper layers so they can react on congestion
and/or inform the sending process.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently the ->ndo_hard_start_xmit() callbacks are only permitted to return
one of the NETDEV_TX codes. This prevents any kind of error propagation for
virtual devices, like queue congestion of the underlying device in case of
layered devices, or unreachability in case of tunnels.
This patches changes the NET_XMIT codes to avoid clashes with the NETDEV_TX
codes and changes the two callers of dev_hard_start_xmit() to expect either
errno codes, NET_XMIT codes or NETDEV_TX codes as return value.
In case of qdisc_restart(), all non NETDEV_TX codes are mapped to NETDEV_TX_OK
since no error propagation is possible when using qdiscs. In case of
dev_queue_xmit(), the error is propagated upwards.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
We have two implementations of the compat_ioctl handling for ATM, the
one that we have had for ages in fs/compat_ioctl.c and the one added to
net/atm/ioctl.c by David Woodhouse. Unfortunately, both versions are
incomplete, and in practice we use a very confusing combination of the
two.
For ioctl numbers that have the same identifier on 32 and 64 bit systems,
we go directly through the compat_ioctl socket operation, for those that
differ, we do a conversion in fs/compat_ioctl.c.
This patch moves both variants into the vcc_compat_ioctl() function,
while preserving the current behaviour. It also kills off the COMPATIBLE_IOCTL
definitions that we never use here.
Doing it this way is clearly not a good solution, but I hope it is a
step into the right direction, so that someone is able to clean up this
mess for real.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Handling for SIOCSHWTSTAMP is broken on architectures
with a split user/kernel address space like s390,
because it passes a real user pointer while using
set_fs(KERNEL_DS).
A similar problem might arise the next time somebody
adds code to dev_ifsioc.
Split up dev_ifsioc into three separate functions for
SIOCSHWTSTAMP, SIOC*IFMAP and all other numbers so
we can get rid of set_fs in all potentially affected
cases.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Patrick Ohly <patrick.ohly@intel.com>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no reason for this lock to be reader/writer since
the reader only has lock held for a very brief period.
The overhead of read_lock is more expensive than spinlock.
Compile tested only, I am not a decnet user.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add missing locking in the case of auto binding to the
default device. The address list might change while this code is looking
at the list.
Compile tested only, I am not a decnet user.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The full_name_hash function does not produce well distributed values in
the lower bits, so most code uses hash_32() to fold it. This is really
a bug introduced when name hashing was added, back in 2.5 when I added
name hashing.
hash_32 is all that is needed since full_name_hash returns unsigned int
which is only 32 bits on 64 bit platforms.
Also, there is no point in using hash_32 on ifindex, because the is naturally
sequential and usually well distributed.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
NAPI drivers try to recycle SKBs in their polling routine, but we
generally don't know the context in which the polling will be called,
and the skb recycling itself may require IRQs to be enabled.
This patch adds irqs_disabled() test to the skb_recycle_check()
routine, so that we'll not let the drivers hit the skb recycling
path with IRQs disabled.
As a side effect, this patch actually disables skb recycling for some
[broken] drivers. E.g. gianfar driver grabs an irqsave spinlock during
TX ring processing, and then tries to recycle an skb, and that caused
the following badness:
nf_conntrack version 0.5.0 (1008 buckets, 4032 max)
------------[ cut here ]------------
Badness at kernel/softirq.c:143
NIP: c003e3c4 LR: c423a528 CTR: c003e344
...
NIP [c003e3c4] local_bh_enable+0x80/0xc4
LR [c423a528] destroy_conntrack+0xd4/0x13c [nf_conntrack]
Call Trace:
[c15d1b60] [c003e32c] local_bh_disable+0x1c/0x34 (unreliable)
[c15d1b70] [c423a528] destroy_conntrack+0xd4/0x13c [nf_conntrack]
[c15d1b80] [c02c6370] nf_conntrack_destroy+0x3c/0x70
Signed-off-by: David S. Miller <davem@davemloft.net>
Reported by Stephen Rothwell:
--------------------
Today's linux-next build (x86_64 allmodconfig) produced this warning:
net/ipv6/addrconf.c: In function 'inet6_dump_ifinfo':
net/ipv6/addrconf.c:3833: warning: unused variable 'err'
Introduced by commit 84d2697d96 ("ipv6:
speedup inet6_dump_ifinfo()").
--------------------
Signed-off-by: David S. Miller <davem@davemloft.net>
Use new function to avoid doing read_lock().
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Oliver Hartkopp <oliver@hartkopp.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This also needs to be optimized for large number of devices.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When showing device statistics use RCU rather than read_lock(&dev_base_lock)
Compile tested only.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use RCU to walk list of network devices in qdisc dump.
This could be optimized for large number of devices.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Do not need to use read_lock(&dev_base_lock), use RCU instead.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Change udp6_portaddr_hash() to use ipv6_addr_v4mapped()
inline instead of ipv6_addr_type().
Signed-off-by: Brian Haley <brian.haley@hp.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch rearranges the SIT DF bit handling using the new IPIP DF
code. The only externally visible effect should be the case where
PMTU is enabled and the MTU is exactly 1280 bytes. In this case the
previous code would send packets out with DF off while the new code
would set the DF bit. This is inline with RFC 4213.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Thanks,
Signed-off-by: David S. Miller <davem@davemloft.net>
Apparently, inet6_dump_addr() is not able to handle more than
64 ipv6 addresses per device. We must break from inner loops
in case skb is full, or else cursor is put at the end of list.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When handling large number of netdevice, inet6_dump_ifinfo()
is very slow because it has O(N^2) complexity.
Instead of scanning one single list, we can use the 256 sub lists
of the dev_index hash table, and RCU lookups.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use guard DECLARE_SOCKADDR in a few more places which allow
us to catch if the structure copied back is too big.
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
UDP bind() can be O(N^2) in some pathological cases.
Thanks to secondary hash tables, we can make it O(N)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This fixes the following bug in the current implementation of
net/xfrm: SAD entries timeouts do not count the time spent by the machine
in the suspended state. This leads to the connectivity problems because
after resuming local machine thinks that the SAD entry is still valid, while
it has already been expired on the remote server.
The cause of this is very simple: the timeouts in the net/xfrm are bound to
the old mod_timer() timers. This patch reassigns them to the
CLOCK_REALTIME hrtimer.
I have been using this version of the patch for a few months on my
machines without any problems. Also run a few stress tests w/o any
issues.
This version of the patch uses tasklet_hrtimer by Peter Zijlstra
(commit 9ba5f0).
This patch is against 2.6.31.4. Please CC me.
Signed-off-by: Yury Polyanskiy <polyanskiy@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This adds compat_ioctl support for SIOCWANDEV, which has
always been missing.
The definition of struct compat_ifreq was missing an
ifru_settings fields that is needed to support SIOCWANDEV,
so add that and clean up the whitespace damage in the
struct definition.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
SIOCGMIIPHY and SIOCGMIIREG return data through ifreq,
so it needs to be converted on the way out as well.
SIOCGIFPFLAGS is unused, but has the same problem in theory.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
When skb_clone() fails, we should increment sk_drops and SNMP counters.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
IPV6 UDP multicast rx path is a bit complex and can hold a spinlock
for a long time.
Using a small (32 or 64 entries) stack of socket pointers can help
to perform expensive operations (skb_clone(), udp_queue_rcv_skb())
outside of the lock, in most cases.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
UDP multicast rx path is a bit complex and can hold a spinlock
for a long time.
Using a small (32 or 64 entries) stack of socket pointers can help
to perform expensive operations (skb_clone(), udp_queue_rcv_skb())
outside of the lock, in most cases.
It's also a base for a future RCU conversion of multicast recption.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Lucian Adrian Grijincu <lgrijincu@ixiacom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We first locate the (local port) hash chain head
If few sockets are in this chain, we proceed with previous lookup algo.
If too many sockets are listed, we take a look at the secondary
(port, address) hash chain.
We choose the shortest chain and proceed with a RCU lookup on the elected chain.
But, if we chose (port, address) chain, and fail to find a socket on given address,
we must try another lookup on (port, in6addr_any) chain to find sockets not bound
to a particular IP.
-> No extra cost for typical setups, where the first lookup will probabbly
be performed.
RCU lookups everywhere, we dont acquire spinlock.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
We first locate the (local port) hash chain head
If few sockets are in this chain, we proceed with previous lookup algo.
If too many sockets are listed, we take a look at the secondary
(port, address) hash chain we added in previous patch.
We choose the shortest chain and proceed with a RCU lookup on the elected chain.
But, if we chose (port, address) chain, and fail to find a socket on given address,
we must try another lookup on (port, INADDR_ANY) chain to find socket not bound
to a particular IP.
-> No extra cost for typical setups, where the first lookup will probabbly
be performed.
RCU lookups everywhere, we dont acquire spinlock.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Extends udp_table to contain a secondary hash table.
socket anchor for this second hash is free, because UDP
doesnt use skc_bind_node : We define an union to hold
both skc_bind_node & a new hlist_nulls_node udp_portaddr_node
udp_lib_get_port() inserts sockets into second hash chain
(additional cost of one atomic op)
udp_lib_unhash() deletes socket from second hash chain
(additional cost of one atomic op)
Note : No spinlock lockdep annotation is needed, because
lock for the secondary hash chain is always get after
lock for primary hash chain.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Union sk_hash with two u16 hashes for udp (no extra memory taken)
One 16 bits hash on (local port) value (the previous udp 'hash')
One 16 bits hash on (local address, local port) values, initialized
but not yet used. This second hash is using jenkin hash for better
distribution.
Because the 'port' is xored later, a partial hash is performed
on local address + net_hash_mix(net)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Adds a counter in udp_hslot to keep an accurate count
of sockets present in chain.
This will permit to upcoming UDP lookup algo to chose
the shortest chain when secondary hash is added.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no good reason to not support userspace specifying the
network namespace during device creation, and it makes it easier
to create a network device and pass it to a child network namespace
with a well known name.
We have to be careful to ensure that the target network namespace
for the new device exists through the life of the call. To keep
that logic clear I have factored out the network namespace grabbing
logic into rtnl_link_get_net.
In addtion we need to continue to pass the source network namespace
to the rtnl_link_ops.newlink method so that we can find the base
device source network namespace.
Signed-off-by: Eric W. Biederman <ebiederm@aristanetworks.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
atalk_sum_partial can now use the rol16 function in bitops.h
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Using RCU helps not touching device refcount in rawv6_bind()
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
bcm_proc_getifname() is called with RTNL and dev_base_lock
not held. It calls __dev_get_by_index() without locks, and
this is illegal (might crash)
Close the race by holding dev_base_lock and copying dev->name
in the protected section.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Oliver Hartkopp <oliver@hartkopp.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
The x25 driver uses lock_kernel() implicitly through
its proto_ops wrapper. The makes the usage explicit
in order to get rid of that wrapper and to better document
the usage of the BKL.
The next step should be to get rid of the usage of the BKL
in x25 entirely, which requires understanding what data
structures need serialized accesses.
Cc: Henner Eisen <eis@baty.hanse.de>
Cc: David S. Miller <davem@davemloft.net>
Cc: linux-x25@vger.kernel.org
Cc: netdev@vger.kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
The irda driver uses the BKL implicitly in its protocol
operations. Replace the wrapped proto_ops with explicit
lock_kernel() calls makes the usage more obvious and
shrinks the size of the object code.
The calls t lock_kernel() should eventually all be replaced
by other serialization methods, which requires finding out
The calls t lock_kernel() should eventually all be replaced
by other serialization methods, which requires finding out
which data actually needs protection.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Making the BKL usage explicit in ipx makes it more
obvious where it is used, reduces code size and helps
getting rid of the BKL in common code.
I did not analyse how to kill lock_kernel from ipx
entirely, this will involve either proving that it's not
needed, or replacing with a proper mutex or spinlock,
after finding out which data structures are protected
by the lock.
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: David S. Miller <davem@davemloft.net>
Cc: Stephen Hemminger <shemminger@vyatta.com>
Cc: netdev@vger.kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Making the BKL usage explicit in appletalk makes it more
obvious where it is used, reduces code size and helps
getting rid of the BKL in common code.
I did not analyse how to kill lock_kernel from appletalk
entirely, this will involve either proving that it's not
needed, or replacing with a proper mutex or spinlock,
after finding out which data structures are protected
by the lock.
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
Cc: David S. Miller <davem@davemloft.net>
Cc: Stephen Hemminger <shemminger@vyatta.com>
Cc: netdev@vger.kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
SPIN_LOCK_UNLOCKED is deprecated. Use DEFINE_SPINLOCK instead.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
