Since the mdb add/del code was introduced there have been 2 br_mdb_notify
calls when doing br_mdb_add() resulting in 2 notifications on each add.
Example:
Command: bridge mdb add dev br0 port eth1 grp 239.0.0.1 permanent
Before patch:
root@debian:~# bridge monitor all
[MDB]dev br0 port eth1 grp 239.0.0.1 permanent
[MDB]dev br0 port eth1 grp 239.0.0.1 permanent
After patch:
root@debian:~# bridge monitor all
[MDB]dev br0 port eth1 grp 239.0.0.1 permanent
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Fixes: cfd5675435 ("bridge: add support of adding and deleting mdb entries")
Signed-off-by: David S. Miller <davem@davemloft.net>
A report with INCLUDE/Change_to_include and empty source list should be
treated as a leave, specified by RFC 3376, section 3.1:
"If the requested filter mode is INCLUDE *and* the requested source
list is empty, then the entry corresponding to the requested
interface and multicast address is deleted if present. If no such
entry is present, the request is ignored."
Signed-off-by: Satish Ashok <sashok@cumulusnetworks.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull rdma fixes from Doug Ledford:
"Mainly fix-ups for the various 4.2 items"
* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dledford/rdma: (24 commits)
IB/core: Destroy ocrdma_dev_id IDR on module exit
IB/core: Destroy multcast_idr on module exit
IB/mlx4: Optimize do_slave_init
IB/mlx4: Fix memory leak in do_slave_init
IB/mlx4: Optimize freeing of items on error unwind
IB/mlx4: Fix use of flow-counters for process_mad
IB/ipath: Convert use of __constant_<foo> to <foo>
IB/ipoib: Set MTU to max allowed by mode when mode changes
IB/ipoib: Scatter-Gather support in connected mode
IB/ucm: Fix bitmap wrap when devnum > IB_UCM_MAX_DEVICES
IB/ipoib: Prevent lockdep warning in __ipoib_ib_dev_flush
IB/ucma: Fix lockdep warning in ucma_lock_files
rds: rds_ib_device.refcount overflow
RDMA/nes: Fix for incorrect recording of the MAC address
RDMA/nes: Fix for resolving the neigh
RDMA/core: Fixes for port mapper client registration
IB/IPoIB: Fix bad error flow in ipoib_add_port()
IB/mlx4: Do not attemp to report HCA clock offset on VFs
IB/cm: Do not queue work to a device that's going away
IB/srp: Avoid using uninitialized variable
...
When we calculate the checksum on the recv path, we store the
result in the skb as an optimisation in case we need the checksum
again down the line.
This is in fact bogus for the MSG_PEEK case as this is done without
any locking. So multiple threads can peek and then store the result
to the same skb, potentially resulting in bogus skb states.
This patch fixes this by only storing the result if the skb is not
shared. This preserves the optimisations for the few cases where
it can be done safely due to locking or other reasons, e.g., SIOCINQ.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Shared skbs must not be modified and this is crucial for broadcast
and/or multicast paths where we use it as an optimisation to avoid
unnecessary cloning.
The function skb_recv_datagram breaks this rule by setting peeked
without cloning the skb first. This causes funky races which leads
to double-free.
This patch fixes this by cloning the skb and replacing the skb
in the list when setting skb->peeked.
Fixes: a59322be07 ("[UDP]: Only increment counter on first peek/recv")
Reported-by: Konstantin Khlebnikov <khlebnikov@yandex-team.ru>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Similarly as in commit 4f7d2cdfdd ("rtnetlink: verify IFLA_VF_INFO
attributes before passing them to driver"), we have a double nesting
of netlink attributes, i.e. IFLA_VF_PORTS only contains IFLA_VF_PORT
that is nested itself. While IFLA_VF_PORTS is a verified attribute
from ifla_policy[], we only check if the IFLA_VF_PORTS container has
IFLA_VF_PORT attributes and then pass the attribute's content itself
via nla_parse_nested(). It would be more correct to reject inner types
other than IFLA_VF_PORT instead of continuing parsing and also similarly
as in commit 4f7d2cdfdd, to check for a minimum of NLA_HDRLEN.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Cc: Roopa Prabhu <roopa@cumulusnetworks.com>
Cc: Scott Feldman <sfeldma@gmail.com>
Cc: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Acked-by: Roopa Prabhu <roopa@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Don't bother testing if we need to switch to alternate stack
unless TEE target is used.
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
In most cases there is no reentrancy into ip/ip6tables.
For skbs sent by REJECT or SYNPROXY targets, there is one level
of reentrancy, but its not relevant as those targets issue an absolute
verdict, i.e. the jumpstack can be clobbered since its not used
after the target issues absolute verdict (ACCEPT, DROP, STOLEN, etc).
So the only special case where it is relevant is the TEE target, which
returns XT_CONTINUE.
This patch changes ip(6)_do_table to always use the jump stack starting
from 0.
When we detect we're operating on an skb sent via TEE (percpu
nf_skb_duplicated is 1) we switch to an alternate stack to leave
the original one alone.
Since there is no TEE support for arptables, it doesn't need to
test if tee is active.
The jump stack overflow tests are no longer needed as well --
since ->stacksize is the largest call depth we cannot exceed it.
A much better alternative to the external jumpstack would be to just
declare a jumps[32] stack on the local stack frame, but that would mean
we'd have to reject iptables rulesets that used to work before.
Another alternative would be to start rejecting rulesets with a larger
call depth, e.g. 1000 -- in this case it would be feasible to allocate the
entire stack in the percpu area which would avoid one dereference.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
This prepares for a TEE like expression in nftables.
We want to ensure only one duplicate is sent, so both will
use the same percpu variable to detect duplication.
The other use case is detection of recursive call to xtables, but since
we don't want dependency from nft to xtables core its put into core.c
instead of the x_tables core.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The {arp,ip,ip6tables} jump stack is currently sized based
on the number of user chains.
However, its rather unlikely that every user defined chain jumps to the
next, so lets use the existing loop detection logic to also track the
chain depths.
The stacksize is then set to the largest chain depth seen.
Signed-off-by: Florian Westphal <fw@strlen.de>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
- Register the nftables chains in the network namespace that they need
to run in.
- Remove the hacks that stopped chains running in the wrong network
namespace.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
- Add a new set of functions for registering and unregistering per
network namespace hooks.
- Modify the old global namespace hook functions to use the per
network namespace hooks in their implementation, so their remains a
single list that needs to be walked for any hook (this is important
for keeping the hook priority working and for keeping the code
walking the hooks simple).
- Only allow registering the per netdevice hooks in the network
namespace where the network device lives.
- Dynamically allocate the structures in the per network namespace
hook list in nf_register_net_hook, and unregister them in
nf_unregister_net_hook.
Dynamic allocate is required somewhere as the number of network
namespaces are not fixed so we might as well allocate them in the
registration function.
The chain of registered hooks on any list is expected to be small so
the cost of walking that list to find the entry we are unregistering
should also be small.
Performing the management of the dynamically allocated list entries
in the registration and unregistration functions keeps the complexity
from spreading.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
- Add a new function find_nf_hook_list to select the nf_hook_list
- Fail nf_register_hook when asked for a per netdevice hook list when
support for per netdevice hook lists is not built into the kernel.
- Move the hook list head selection outside of nf_hook_mutex as
nothing in the selection requires the hook list, and error handling
is simpler if a mutex is not held.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Replace an overcomplicated switch statement with a simple if statement.
This also removes the ingress queue enable outside of nf_hook_mutex as
the protection provided by the mutex is not necessary and the code is
clearer having both of the static key increments together.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
The module_put() function tests whether its argument is NULL and then
returns immediately. Thus the test around the call is not needed.
This issue was detected by using the Coccinelle software.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Fixes: 3e0249f9c0 ("RDS/IB: add refcount tracking to struct rds_ib_device")
There lacks a dropping on rds_ib_device.refcount in case rds_ib_alloc_fmr
failed(mr pool running out). this lead to the refcount overflow.
A complain in line 117(see following) is seen. From vmcore:
s_ib_rdma_mr_pool_depleted is 2147485544 and rds_ibdev->refcount is -2147475448.
That is the evidence the mr pool is used up. so rds_ib_alloc_fmr is very likely
to return ERR_PTR(-EAGAIN).
115 void rds_ib_dev_put(struct rds_ib_device *rds_ibdev)
116 {
117 BUG_ON(atomic_read(&rds_ibdev->refcount) <= 0);
118 if (atomic_dec_and_test(&rds_ibdev->refcount))
119 queue_work(rds_wq, &rds_ibdev->free_work);
120 }
fix is to drop refcount when rds_ib_alloc_fmr failed.
Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Reviewed-by: Haggai Eran <haggaie@mellanox.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
Fix crash in 3.5+ if FTP is used after switching
sync_version to 0.
Fixes: 749c42b620 ("ipvs: reduce sync rate with time thresholds")
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
It is possible that we bind against a local socket in early_demux when we
are actually going to want to forward it. In this case, the socket serves
no purpose and only serves to confuse things (particularly functions which
implicitly expect sk_fullsock to be true, like ip_local_out).
Additionally, skb_set_owner_w is totally broken for non full-socks.
Signed-off-by: Alex Gartrell <agartrell@fb.com>
Fixes: 41063e9dd1 ("ipv4: Early TCP socket demux.")
Acked-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
I overlooked the svc->sched_data usage from schedulers
when the services were converted to RCU in 3.10. Now
the rare ipvsadm -E command can change the scheduler
but due to the reverse order of ip_vs_bind_scheduler
and ip_vs_unbind_scheduler we provide new sched_data
to the old scheduler resulting in a crash.
To fix it without changing the scheduler methods we
have to use synchronize_rcu() only for the editing case.
It means all svc->scheduler readers should expect a
NULL value. To avoid breakage for the service listing
and ipvsadm -R we can use the "none" name to indicate
that scheduler is not assigned, a state when we drop
new connections.
Reported-by: Alexander Vasiliev <a.vasylev@404-group.com>
Fixes: ceec4c3816 ("ipvs: convert services to rcu")
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
Michael Vallaly reports about wrong source address used
in rare cases for tunneled traffic. Looks like
__ip_vs_get_out_rt in 3.10+ is providing uninitialized
dest_dst->dst_saddr.ip because ip_vs_dest_dst_alloc uses
kmalloc. While we retry after seeing EINVAL from routing
for data that does not look like valid local address, it
still succeeded when this memory was previously used from
other dests and with different local addresses. As result,
we can use valid local address that is not suitable for
our real server.
Fix it by providing 0.0.0.0 every time our cache is refreshed.
By this way we will get preferred source address from routing.
Reported-by: Michael Vallaly <lvs@nolatency.com>
Fixes: 026ace060d ("ipvs: optimize dst usage for real server")
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
Previously there was a trivial panic
unshare -n /bin/bash <<EOF
ip addr add dev lo face::1/128
ipvsadm -A -t [face::1]:15213
ipvsadm -a -t [face::1]:15213 -r b00c::1
echo boom | nc face::1 15213
EOF
This patch allows us to replicate the net logic above and simply capture
the skb_dst(skb)->dev and use that for the purpose of the invocation.
Signed-off-by: Alex Gartrell <agartrell@fb.com>
Acked-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
Conflicts:
net/bridge/br_mdb.c
Minor conflict in br_mdb.c, in 'net' we added a memset of the
on-stack 'ip' variable whereas in 'net-next' we assign a new
member 'vid'.
Signed-off-by: David S. Miller <davem@davemloft.net>
Until now all user mdb entries were added in vlan 0, this patch adds
support to allow the user to specify the vlan for the entry.
About the uapi change a hole in struct br_mdb_entry is used so the size
and offsets are kept the same (verified with pahole and tested with older
iproute2).
Example:
$ bridge mdb
dev br0 port eth1 grp 239.0.0.1 permanent vlan 2000
dev br0 port eth1 grp 239.0.0.1 permanent vlan 200
dev br0 port eth1 grp 239.0.0.1 permanent
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch makes the default to build IPv6 into the kernel. IPv6
now has significant traction and any remaining vestiges of IPv6
not being provided parity with IPv4 should be swept away. IPv6 is now
core to the Internet and kernel.
Points on IPv6 adoption:
- Per Google statistics, IPv6 usage has reached 7% on the Internet
and continues to exhibit an exponential growth rate
https://www.google.com/intl/en/ipv6/statistics.html
- Just a few days ago ARIN officially depleted its IPv4 pool
- IPv6 only data centers are being successfully built
(e.g. at Facebook)
This patch changes the IPv6 Kconfig for IPV6. Default for CONFIG_IPV6
is set to "y" and the text has been updated to reflect the maturity of
IPv6.
Impact:
Under some circumstances building modules in to kernel might have a
performance advantage. In my testing, I did notice a very slight
improvement.
This will obviously increase the size of the kernel image. In my
configuration I see:
IPv6 as module:
text data bss dec hex filename
9703666 1899288 933888 12536842 bf4c0a vmlinux
IPv6 built into kernel
text data bss dec hex filename
9436490 1879600 913408 12229498 ba9b7a vmlinux
Which increases text size by ~270K (2.8% increase in size for me). If
image size is an issue, presumably for a device which does not do IP
networking (IMO we should be discouraging IPv4-only devices), IPV6 can
be disabled or still built as a module.
Acked-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull networking fixes from David Miller:
1) Missing list head init in bluetooth hidp session creation, from Tedd
Ho-Jeong An.
2) Don't leak SKB in bridge netfilter error paths, from Florian
Westphal.
3) ipv6 netdevice private leak in netfilter bridging, fixed by Julien
Grall.
4) Fix regression in IP over hamradio bpq encapsulation, from Ralf
Baechle.
5) Fix race between rhashtable resize events and table walks, from Phil
Sutter.
6) Missing validation of IFLA_VF_INFO netlink attributes, fix from
Daniel Borkmann.
7) Missing security layer socket state initialization in tipc code,
from Stephen Smalley.
8) Fix shared IRQ handling in boomerang 3c59x interrupt handler, from
Denys Vlasenko.
9) Missing minor_idr destroy on module unload on macvtap driver, from
Johannes Thumshirn.
10) Various pktgen kernel thread races, from Oleg Nesterov.
11) Fix races that can cause packets to be processed in the backlog even
after a device attached to that SKB has been fully unregistered.
From Julian Anastasov.
12) bcmgenet driver doesn't account packet drops vs. errors properly,
fix from Petri Gynther.
13) Array index validation and off by one fix in DSA layer from Florian
Fainelli
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (66 commits)
can: replace timestamp as unique skb attribute
ARM: dts: dra7x-evm: Prevent glitch on DCAN1 pinmux
can: c_can: Fix default pinmux glitch at init
can: rcar_can: unify error messages
can: rcar_can: print request_irq() error code
can: rcar_can: fix typo in error message
can: rcar_can: print signed IRQ #
can: rcar_can: fix IRQ check
net: dsa: Fix off-by-one in switch address parsing
net: dsa: Test array index before use
net: switchdev: don't abort unsupported operations
net: bcmgenet: fix accounting of packet drops vs errors
cdc_ncm: update specs URL
Doc: z8530book: Fix typo in API-z8530-sync-txdma-open.html
net: inet_diag: always export IPV6_V6ONLY sockopt for listening sockets
bridge: mdb: allow the user to delete mdb entry if there's a querier
net: call rcu_read_lock early in process_backlog
net: do not process device backlog during unregistration
bridge: fix potential crash in __netdev_pick_tx()
net: axienet: Fix devm_ioremap_resource return value check
...
On device shutdown/removal, virtio drivers need to trigger a reset on
the device; if this is neglected, the virtio core will complain about
non-zero device status.
This patch resets the status when the 9p virtio driver is removed
from the system by calling vdev->config->reset on the virtio_device
to send a reset to the host virtio device.
Signed-off-by: Pierre Morel <pmorel@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Marc Kleine-Budde says:
====================
pull-request: can 2015-07-12
this is a pull request of 8 patchs for net/master.
Sergei Shtylyov contributes 5 patches for the rcar_can driver, fixing the IRQ
check and several info and error messages. There are two patches by J.D.
Schroeder and Roger Quadros for the c_can driver and dra7x-evm device tree,
which precent a glitch in the DCAN1 pinmux. Oliver Hartkopp provides a better
approach to make the CAN skbs unique, the timestamp is replaced by a counter.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 514ac99c64 "can: fix multiple delivery of a single CAN frame for
overlapping CAN filters" requires the skb->tstamp to be set to check for
identical CAN skbs.
Without timestamping to be required by user space applications this timestamp
was not generated which lead to commit 36c01245eb "can: fix loss of CAN frames
in raw_rcv" - which forces the timestamp to be set in all CAN related skbuffs
by introducing several __net_timestamp() calls.
This forces e.g. out of tree drivers which are not using alloc_can{,fd}_skb()
to add __net_timestamp() after skbuff creation to prevent the frame loss fixed
in mainline Linux.
This patch removes the timestamp dependency and uses an atomic counter to
create an unique identifier together with the skbuff pointer.
Btw: the new skbcnt element introduced in struct can_skb_priv has to be
initialized with zero in out-of-tree drivers which are not using
alloc_can{,fd}_skb() too.
Signed-off-by: Oliver Hartkopp <socketcan@hartkopp.net>
Cc: linux-stable <stable@vger.kernel.org>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
cd->sw_addr is used as a MDIO bus address, which cannot exceed
PHY_MAX_ADDR (32), our check was off-by-one.
Fixes: 5e95329b70 ("dsa: add device tree bindings to register DSA switches")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
port_index is used an index into an array, and this information comes
from Device Tree, make sure that port_index is not equal to the array
size before using it. Move the check against port_index earlier in the
loop.
Fixes: 5e95329b70: ("dsa: add device tree bindings to register DSA switches")
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no need to abort attribute setting or object addition, if the
prepare phase returned operation not supported.
Thus, abort these two transactions only if the error is not -EOPNOTSUPP.
Signed-off-by: Vivien Didelot <vivien.didelot@savoirfairelinux.com>
Acked-by: Jiri Pirko <jiri@resnulli.us>
Acked-by: Scott Feldman <sfeldma@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit 3cc4949269.
There is nothing wrong with coalescing during defragmentation, it
reduces truesize overhead and simplifies things for the receiving
socket (no fraglist walk needed).
However, it also destroys geometry of the original fragments.
While that doesn't cause any breakage (we make sure to not exceed largest
original size) ip_do_fragment contains a 'fastpath' that takes advantage
of a present frag list and results in fragments that (in most cases)
match what was received.
In case its needed the coalescing could be done later, when we're sure
the skb is not forwarded. But discussion during NFWS resulted in
'lets just remove this for now'.
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: Florian Westphal <fw@strlen.de>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reconsidering my commit 20462155 "net: inet_diag: export IPV6_V6ONLY
sockopt", I am not happy with the limitations it causes for socket
analysing code in userspace. Exporting the value only if it is set makes
it hard for userspace to decide whether the option is not set or the
kernel does not support exporting the option at all.
>From an auditor's perspective, the interesting question for listening
AF_INET6 sockets is: "Does it NOT have IPV6_V6ONLY set?" Because it is
the unexpected case. This patch allows to answer this question reliably.
Signed-off-by: Phil Sutter <phil@nwl.cc>
Cc: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If outgoing interface is specified and the candidate address is
restricted to the outgoing interface, it is enough to iterate
over that given interface only.
Signed-off-by: YOSHIFUJI Hideaki <hideaki.yoshifuji@miraclelinux.com>
Acked-by: Erik Kline <ek@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Until now when a querier was present static entries couldn't be deleted.
Fix this and allow the user to manipulate the mdb with or without a
querier.
Signed-off-by: Satish Ashok <sashok@cumulusnetworks.com>
Signed-off-by: Nikolay Aleksandrov <nikolay@cumulusnetworks.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Incoming packet should be either in backlog queue or
in RCU read-side section. Otherwise, the final sequence of
flush_backlog() and synchronize_net() may miss packets
that can run without device reference:
CPU 1 CPU 2
skb->dev: no reference
process_backlog:__skb_dequeue
process_backlog:local_irq_enable
on_each_cpu for
flush_backlog => IPI(hardirq): flush_backlog
- packet not found in backlog
CPU delayed ...
synchronize_net
- no ongoing RCU
read-side sections
netdev_run_todo,
rcu_barrier: no
ongoing callbacks
__netif_receive_skb_core:rcu_read_lock
- too late
free dev
process packet for freed dev
Fixes: 6e583ce524 ("net: eliminate refcounting in backlog queue")
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
commit 381c759d99 ("ipv4: Avoid crashing in ip_error")
fixes a problem where processed packet comes from device
with destroyed inetdev (dev->ip_ptr). This is not expected
because inetdev_destroy is called in NETDEV_UNREGISTER
phase and packets should not be processed after
dev_close_many() and synchronize_net(). Above fix is still
required because inetdev_destroy can be called for other
reasons. But it shows the real problem: backlog can keep
packets for long time and they do not hold reference to
device. Such packets are then delivered to upper levels
at the same time when device is unregistered.
Calling flush_backlog after NETDEV_UNREGISTER_FINAL still
accounts all packets from backlog but before that some packets
continue to be delivered to upper levels long after the
synchronize_net call which is supposed to wait the last
ones. Also, as Eric pointed out, processed packets, mostly
from other devices, can continue to add new packets to backlog.
Fix the problem by moving flush_backlog early, after the
device driver is stopped and before the synchronize_net() call.
Then use netif_running check to make sure we do not add more
packets to backlog. We have to do it in enqueue_to_backlog
context when the local IRQ is disabled. As result, after the
flush_backlog and synchronize_net sequence all packets
should be accounted.
Thanks to Eric W. Biederman for the test script and his
valuable feedback!
Reported-by: Vittorio Gambaletta <linuxbugs@vittgam.net>
Fixes: 6e583ce524 ("net: eliminate refcounting in backlog queue")
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: David S. Miller <davem@davemloft.net>
We have to put back the references to the master conntrack and the expectation
that we just created, otherwise we'll leak them.
Fixes: 0ef71ee1a5 ("netfilter: ctnetlink: refactor ctnetlink_create_expect")
Reported-by: Tim Wiess <Tim.Wiess@watchguard.com>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Commit c29390c6df ("xps: must clear sender_cpu before forwarding")
fixed an issue in normal forward path, caused by sender_cpu & napi_id
skb fields being an union.
Bridge is another point where skb can be forwarded, so we need
the same cure.
Bug triggers if packet was received on a NIC using skb_mark_napi_id()
Fixes: 2bd82484bb ("xps: fix xps for stacked devices")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Bob Liu <bob.liu@oracle.com>
Tested-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
flags local variable in __mkroute_input is not used as a variable.
Signed-off-by: Masatake YAMATO <yamato@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add support to allow non-local binds similar to how this was done for IPv4.
Non-local binds are very useful in emulating the Internet in a box, etc.
This add the ip_nonlocal_bind sysctl under ipv6.
Testing:
Set up nonlocal binding and receive routing on a host, e.g.:
ip -6 rule add from ::/0 iif eth0 lookup 200
ip -6 route add local 2001:0:0:1::/64 dev lo proto kernel scope host table 200
sysctl -w net.ipv6.ip_nonlocal_bind=1
Set up routing to 2001:0:0:1::/64 on peer to go to first host
ping6 -I 2001:0:0:1::1 peer-address -- to verify
Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
inet_twsk_deschedule() calls are followed by inet_twsk_put().
Only particular case is in inet_twsk_purge() but there is no point
to defer the inet_twsk_put() after re-enabling BH.
Lets rename inet_twsk_deschedule() to inet_twsk_deschedule_put()
and move the inet_twsk_put() inside.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
timewait sockets have a complex refcounting logic.
Once we realize it should be similar to established and
syn_recv sockets, we can use sk_nulls_del_node_init_rcu()
and remove inet_twsk_unhash()
In particular, deferred inet_twsk_put() added in commit
13475a30b6 ("tcp: connect() race with timewait reuse")
looks unecessary : When removing a timewait socket from
ehash or bhash, caller must own a reference on the socket
anyway.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
