I have benchmarked this on an x86_64 NUMA system and see no significant
performance difference on kernbench. Tested on both x86_64 and powerpc.
The way we do file struct accounting is not very suitable for batched
freeing. For scalability reasons, file accounting was
constructor/destructor based. This meant that nr_files was decremented
only when the object was removed from the slab cache. This is susceptible
to slab fragmentation. With RCU based file structure, consequent batched
freeing and a test program like Serge's, we just speed this up and end up
with a very fragmented slab -
llm22:~ # cat /proc/sys/fs/file-nr
587730 0 758844
At the same time, I see only a 2000+ objects in filp cache. The following
patch I fixes this problem.
This patch changes the file counting by removing the filp_count_lock.
Instead we use a separate percpu counter, nr_files, for now and all
accesses to it are through get_nr_files() api. In the sysctl handler for
nr_files, we populate files_stat.nr_files before returning to user.
Counting files as an when they are created and destroyed (as opposed to
inside slab) allows us to correctly count open files with RCU.
Signed-off-by: Dipankar Sarma <dipankar@in.ibm.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
The size of the skb carrying the netlink message is not
equivalent to the length of the actual netlink message
due to padding. ip_queue matches the length of the payload
against the original packet size to determine if packet
mangling is desired, due to the above wrong assumption
arbitary packets may not be mangled depening on their
original size.
Signed-off-by: Thomas Graf <tgraf@suug.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
In rare circumstances 0 is returned by dccp_li_hist_calc_i_mean which
leads to a divide by zero in ccid3_hc_rx_packet_recv. Explicitly check
for zero return now. Update copyright notice at same time.
Found by Arnaldo.
Signed-off-by: Ian McDonald <imcdnzl@gmail.com>
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The earlier round of kobject/sysfs changes to bridge caused
it not to generate a uevent on removal. Don't think any application
cares (not sure about Xen) but since it generates add uevent
it should generate remove as well.
Signed-off-by: Stephen Hemminger <shemmigner@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Initialize the STP timers for a port when it is created,
rather than when it is enabled. This will prevent future race conditions
where timer gets started before port is enabled.
Signed-off-by: Stephen Hemminger <shemmigner@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Bridge would crash because of uninitailized timer if STP is used and
device was inserted into a bridge before bridge was up. This got
introduced when the delayed port checking was added. Fix is to not
enable STP on port unless bridge is up.
Bugzilla: http://bugzilla.kernel.org/show_bug.cgi?id=6140
Dup: http://bugzilla.kernel.org/show_bug.cgi?id=6156
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
In 295f7324ff I moved defer_accept from
tcp_sock to request_queue and mistakingly reset it at reqsl_queue_alloc, causing
calls to setsockopt(TCP_DEFER_ACCEPT ) to be lost after bind, the fix is to
remove the zeroing of rskq_defer_accept from reqsl_queue_alloc.
Thanks to Alexandra N. Kossovsky <Alexandra.Kossovsky@oktetlabs.ru> for
reporting and testing the suggested fix.
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The nfnetlink_log infrastructure changes broke compatiblity of the LOG
targets. They currently use whatever log backend was registered first,
which means that if ipt_ULOG was loaded first, no messages will be printed
to the ring buffer anymore.
Restore compatiblity by using the old log functions by default and only use
the nf_log backend if the user explicitly said so.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
The comparison wants to find out if the last list iteration reached the
end of the list. It needs to compare the iterator with the list head to
do this, not the element it is looking for.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
The only point of registering a queue handler is to provide an outfn,
so there is no need to check for it.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Packets should be rerouted when they come back from userspace, not before.
Also move the queue_rerouters to RCU to avoid taking the queue_handler_lock
for each reinjected packet.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Every rerouter needs to provide a save and a reroute function, we don't
need to check for them. But we do need to check if a rerouter is registered
at all for the current family, with bridging for example packets of
unregistered families can hit nf_queue.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use the registered data structure instead of copying it.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
The only reason post_input exists at all is that it gives us the
potential to adjust the checksums incrementally in future which
we ought to do.
However, after thinking about it for a bit we can adjust the
checksums without using this post_input stuff at all. The crucial
point is that only the inner-most NAT-T SA needs to be considered
when adjusting checksums. What's more, the checksum adjustment
comes down to a single u32 due to the linearity of IP checksums.
We just happen to have a spare u32 lying around in our skb structure :)
When ip_summed is set to CHECKSUM_NONE on input, the value of skb->csum
is currently unused. All we have to do is to make that the checksum
adjustment and voila, there goes all the post_input and decap structures!
I've left in the decap data structures for now since it's intricately
woven into the sec_path stuff. We can kill them later too.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
We used to keep sg on the stack which is why the extra block was useful.
We've long since stopped doing that so let's kill the block and save
some indentation.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Based on patch by Hoerdt Mickael <hoerdt@clarinet.u-strasbg.fr>.
Signed-off-by: YOSHIFUJI Hideaki <yosufuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The included patch fixes ip6_tunnel to release the cached dst entry
when the tunnel parameters (such as tunnel endpoints) are changed so
they are used immediatly for the next encapsulated packets.
Signed-off-by: Hugo Santos <hsantos@av.it.pt>
Acked-by: Ville Nuorvala <vnuorval@tcs.hut.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
We should use the TOS because it's one of the routing keys. It also
means that we update the correct routing cache entry when PMTU occurs.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
When you turn off ARP on a netdevice then the first packet always goes
out with a dstMAC of all zeroes. This is because the first packet is
used to resolve ARP entries. Even though the ARP entry may be resolved
(I tried by setting a static ARP entry for a host i was pinging from),
it gets overwritten by virtue of having the netdevice disabling ARP.
Subsequent packets go out fine with correct dstMAC address (which may
be why people have ignored reporting this issue).
To cut the story short:
the culprit code is in net/ethernet/eth.c::eth_header()
----
/*
* Anyway, the loopback-device should never use this
function...
*/
if (dev->flags & (IFF_LOOPBACK|IFF_NOARP))
{
memset(eth->h_dest, 0, dev->addr_len);
return ETH_HLEN;
}
if(daddr)
{
memcpy(eth->h_dest,daddr,dev->addr_len);
return ETH_HLEN;
}
----
Note how the h_dest is being reset when device has IFF_NOARP.
As a note:
All devices including loopback pass a daddr. loopback in fact passes
a 0 all the time ;->
This means i can delete the check totaly or i can remove the IFF_NOARP
Alexey says:
--------------------
I think, it was me who did this crap. It was so long ago I do not remember
why it was made.
I remember some troubles with dummy device. It tried to resolve
addresses, apparently, without success and generated errors instead of
blackholing. I think the problem was eventually solved at neighbour
level.
After some thinking I suspect the deletion of this chunk could change
behaviour of some parts which do not use neighbour cache f.e. packet
socket.
I think safer approach would be to move this chunk after if (daddr).
And the possibility to remove this completely could be analyzed later.
--------------------
Patch updated with Alexey's safer suggestions.
Signed-off-by: Jamal Hadi Salim <hadi@cyberus.ca>
Acked-by: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
We often just do an atomic_dec(&x->refcnt) on an xfrm_state object
because we know there is more than 1 reference remaining and thus
we can elide the heavier xfrm_state_put() call.
Do this behind an inline function called __xfrm_state_put() so that is
more obvious and also to allow us to more cleanly add refcount
debugging later.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
When garbage collecting route cache entries of multipath routes
in rt_garbage_collect(), entries were deleted from the hash bucket
'i' while holding a spin lock on bucket 'k' resulting in a system
hang. Delete entries, if any, from bucket 'k' instead.
Signed-off-by: Suresh Bhogavilli <sbhogavilli@verisign.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The bridge-netfilter code attaches a fake dst_entry with dst->ops == NULL
to purely bridged packets. When these packets are SNATed and a policy
lookup is done, xfrm_lookup crashes because it tries to dereference
dst->ops.
Change xfrm_lookup not to dereference dst->ops before checking for the
DST_NOXFRM flag and set this flag in the fake dst_entry.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some of netfilter-related members are initalized / copied twice in
skb_clone(). Remove one.
Pointed out by Olivier MATZ <olivier.matz@6wind.com>.
And this patch also fixes order of copying / clearing members.
Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
When redirecting an outgoing packet to loopback, it keeps the original
conntrack reference and information from the outgoing path, which
falsely triggers the check for DNAT on input and the dst_entry is
released to trigger rerouting. ip_route_input refuses to route the
packet because it has a local source address and it is dropped.
Look at the packet itself to dermine if it was NATed. Also fix a
missing inversion that causes unneccesary xfrm lookups.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
ICMP errors are only SNATed when their source matches the source of the
connection they are related to, otherwise the source address is not
changed. This creates problems with ICMP frag. required messages
originating from a router behind the NAT, if private IPs are used the
packet has a good change of getting dropped on the path to its destination.
Always NAT ICMP errors similar to the original connection.
Based on report by Al Viro.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
If skb->ip_summed is CHECKSUM_HW here, skb->csum includes checksum
of actual IPv6 header and extension headers. Then such excess
checksum must be subtruct when nf_conntrack calculates TCP/UDP checksum
with pseudo IPv6 header. Spotted by Ben Skeggs.
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>
Locally generated ICMPv6 errors should be associated with the conntrack
of the original packet. Since the conntrack entry may not be in the hash
tables (for the first packet), it must be manually attached.
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>
TCP RSTs generated by the REJECT target should be associated with the
conntrack of the original TCP packet. Since the conntrack entry is
usually not is the hash tables, it must be manually attached.
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>
Move registration of __nf_ct_attach to nf_conntrack_core to make it usable
for IPv6 connection tracking as well.
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>
NF_CONNTRACK_MARK is bool and depends on NF_CONNTRACK which is
tristate. If a variable depends on NF_CONNTRACK_MARK and doesn't take
care about NF_CONNTRACK, it can be y even if NF_CONNTRACK isn't y.
NF_CT_ACCT have same issue, too.
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>
When a packet matching an IPsec policy is SNATed so it doesn't match any
policy anymore it looses its xfrm bundle, which makes xfrm4_output_finish
crash because of a NULL pointer dereference.
This patch directs these packets to the original output path instead. Since
the packets have already passed the POST_ROUTING hook, but need to start at
the beginning of the original output path which includes another
POST_ROUTING invocation, a flag is added to the IPCB to indicate that the
packet was rerouted and doesn't need to pass the POST_ROUTING hook again.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Looks like somebody forgot to use the _bh spin_lock variant. We ran into a
deadlock where br->hello_timer expired while br_stp_disable_br() walked
br->port_list.
Signed-off-by: Adrian Drzewiecki <z@drze.net>
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
To find out if a packet needs to be handled by IPsec after SNAT, packets
are currently rerouted in POST_ROUTING and a new xfrm lookup is done. This
breaks SNAT of non-unicast packets to non-local addresses because the
packet is routed as incoming packet and no neighbour entry is bound to the
dst_entry. In general, it seems to be a bad idea to replace the dst_entry
after the packet was already sent to the output routine because its state
might not match what's expected.
This patch changes the xfrm lookup in POST_ROUTING to re-use the original
dst_entry without routing the packet again. This means no policy routing
can be used for transport mode transforms (which keep the original route)
when packets are SNATed to match the policy, but it looks like the best
we can do for now.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Revert the following changeset:
bc8dfcb939
Recursive SKB frag lists are really possible and disallowing
them breaks things.
Noticed by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Problem discovered and initial patch by Olaf Kirch:
there's a problem with IPsec that has been bugging some of our users
for the last couple of kernel revs. Every now and then, IPsec will
freeze the machine completely. This is with openswan user land,
and with kernels up to and including 2.6.16-rc2.
I managed to debug this a little, and what happens is that we end
up looping in xfrm_lookup, and never get out. With a bit of debug
printks added, I can this happening:
ip_route_output_flow calls xfrm_lookup
xfrm_find_bundle returns NULL (apparently we're in the
middle of negotiating a new SA or something)
We therefore call xfrm_tmpl_resolve. This returns EAGAIN
We go to sleep, waiting for a policy update.
Then we loop back to the top
Apparently, the dst_orig that was passed into xfrm_lookup
has been dropped from the routing table (obsolete=2)
This leads to the endless loop, because we now create
a new bundle, check the new bundle and find it's stale
(stale_bundle -> xfrm_bundle_ok -> dst_check() return 0)
People have been testing with the patch below, which seems to fix the
problem partially. They still see connection hangs however (things
only clear up when they start a new ping or new ssh). So the patch
is obvsiouly not sufficient, and something else seems to go wrong.
I'm grateful for any hints you may have...
I suggest that we simply bail out always. If the dst decides to die
on us later on, the packet will be dropped anyway. So there is no
great urgency to retry here. Once we have the proper resolution
queueing, we can then do the retry again.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Olaf Kirch <okir@suse.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
- panic() doesn't return.
- Don't forget to unlock on genl_register_family() error path
- genl_rcv_msg() is called via pointer so there's no point in declaring it
`inline'.
Notes:
genl_ctrl_event() ignores the genlmsg_multicast() return value.
lots of things ignore the genl_ctrl_event() return value.
Signed-off-by: Jamal Hadi Salim <hadi@cyberus.ca>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Horms patch was the best of the three fixes. Dave, already applied
Harald's version, so this patch converts that to the better one.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The new x_tables related Kconfig options appear at the wrong menu level
without this patch.
Signed-off-by: Harald Welte <laforge@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Missing license tag.
I've assumed this is GPL. (It could also use a MODULE_AUTHOR)
Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
isic can trigger these msgs to be spewed at a very high rate.
There's already a sysctl to turn them off. Given these messages
aren't useful for most people, this patch disables them by
default.
Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch fixes the two NULL pointer dereferences found by the sfuzz
tool from Ilja van Sprundel. The first one was a call of getsockname()
for an unbound socket and the second was calling accept() while this
operation isn't implemented for the HCI socket interface.
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Dipankar Sarma