Alexander V. Lukyanov found a regression in 2.6.29 and made a complete
analysis found in http://bugzilla.kernel.org/show_bug.cgi?id=13339
Quoted here because its a perfect one :
begin_of_quotation
2.6.29 patch has introduced flexible route cache rebuilding. Unfortunately the
patch has at least one critical flaw, and another problem.
rt_intern_hash calculates rthi pointer, which is later used for new entry
insertion. The same loop calculates cand pointer which is used to clean the
list. If the pointers are the same, rtable leak occurs, as first the cand is
removed then the new entry is appended to it.
This leak leads to unregister_netdevice problem (usage count > 0).
Another problem of the patch is that it tries to insert the entries in certain
order, to facilitate counting of entries distinct by all but QoS parameters.
Unfortunately, referencing an existing rtable entry moves it to list beginning,
to speed up further lookups, so the carefully built order is destroyed.
For the first problem the simplest patch it to set rthi=0 when rthi==cand, but
it will also destroy the ordering.
end_of_quotation
Problematic commit is 1080d709fb
(net: implement emergency route cache rebulds when gc_elasticity is exceeded)
Trying to keep dst_entries ordered is too complex and breaks the fact that
order should depend on the frequency of use for garbage collection.
A possible fix is to make rt_intern_hash() simpler, and only makes
rt_check_expire() a litle bit smarter, being able to cope with an arbitrary
entries order. The added loop is running on cache hot data, while cpu
is prefetching next object, so should be unnoticied.
Reported-and-analyzed-by: Alexander V. Lukyanov <lav@yar.ru>
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rt_check_expire() computes average and standard deviation of chain lengths,
but not correclty reset length to 0 at beginning of each chain.
This probably gives overflows for sum2 (and sum) on loaded machines instead
of meaningful results.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Its possible for cfg80211 to have scheduled the work and for
the global workqueue to not have kicked in prior to a cfg80211
driver's regulatory hint or wiphy_apply_custom_regulatory().
Although this is very unlikely its possible and should fix
this race. When this race would happen you are expected to have
hit a null pointer dereference panic.
Cc: stable@kernel.org
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Tested-by: Alan Jenkins <alan-jenkins@tuffmail.co.uk>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Another design flaw in wireless extensions (is anybody
surprised?) in the way it handles the iw_encode_ext
structure: The structure is part of the 'extra' memory
but contains the key length explicitly, instead of it
just being the length of the extra buffer - size of
the struct and using the explicit key length only for
the get operation (which only writes it).
Therefore, we have this layout:
extra: +-------------------------+
| struct iw_encode_ext { |
| ... |
| u16 key_len; |
| u8 key[0]; |
| }; |
+-------------------------+
| key material |
+-------------------------+
Now, all drivers I checked use ext->key_len without
checking that both key_len and the struct fit into the
extra buffer that has been copied from userspace. This
leads to a buffer overrun while reading that buffer,
depending on the driver it may be possible to specify
arbitrary key_len or it may need to be a proper length
for the key algorithm specified.
Thankfully, this is only exploitable by root, but root
can actually cause a segfault or use kernel memory as
a key (which you can even get back with siocgiwencode
or siocgiwencodeext from the key buffer).
Fix this by verifying that key_len fits into the buffer
along with struct iw_encode_ext.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Commit e81963b1 ("ipv4: Make INET_LRO a bool instead of tristate.")
changed this config from tristate to bool. Add default so that it is
consistent with the help text.
Signed-off-by: Frans Pop <elendil@planet.nl>
Signed-off-by: David S. Miller <davem@davemloft.net>
When called with a consumed value that is less than skb_headlen(skb)
bytes into a page frag, skb_seq_read() incorrectly returns an
offset/length relative to skb->data. Ensure that data which should come
from a page frag does.
Signed-off-by: Thomas Chenault <thomas_chenault@dell.com>
Tested-by: Shyam Iyer <shyam_iyer@dell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
gen_estimator can overflow bps (bytes per second) with Gb links, while
it was designed with a u32 API, with a theorical limit of 34360Mbit
(2^32 bytes)
Using 64 bit intermediate avbps/brate counters can allow us to reach
this theorical limit.
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It is illegal to dereference a skb after a successful ndo_start_xmit()
call. We must store skb length in a local variable instead.
Bug was introduced in 2.6.27 by commit 0abf77e55a
(net_sched: Add accessor function for packet length for qdiscs)
Signed-off-by: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Commit 518a09ef11 (tcp: Fix recvmsg MSG_PEEK influence of
blocking behavior) lets the loop run longer than the race check
did previously expect, so we need to be more careful with this
check and consider the work we have been doing.
I tried my best to deal with urg hole madness too which happens
here:
if (!sock_flag(sk, SOCK_URGINLINE)) {
++*seq;
...
by using additional offset by one but I certainly have very
little interest in testing that part.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Tested-by: Frans Pop <elendil@planet.nl>
Tested-by: Ian Zimmermann <itz@buug.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
If bridge is configured with no STP and forwarding delay of 0 (which
is typical for virtualization) then when link starts it will flood all
packets for the first 20 seconds.
This bug was introduced by a combination of earlier changes:
* forwarding database uses hold time of zero to indicate
user wants to always flood packets
* optimzation of the case of forwarding delay of 0 avoids the initial
timer tick
The fix is to just skip all the topology change detection code if
kernel STP is not being used.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently the bridge catches all STP packets; even if STP is turned
off. This prevents other systems (which do have STP turned on)
from being able to detect loops in the network.
With this patch, if STP is off, then any packet sent to the STP
multicast group address is forwarded to all ports.
Based on earlier patch by Joakim Tjernlund with changes
to go through forwarding (not local chain), and optimization
that only last octet needs to be checked.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
If a DHCP server is delayed, it's possible for the client to receive the
DHCPOFFER after it has already sent out a new DHCPDISCOVER message from
a second interface. The client then sends out a DHCPREQUEST from the
second interface, but the server doesn't recognize the device and
rejects the request.
This patch simply tracks the current device being configured and throws
away the OFFER if it is not intended for the current device. A more
sophisticated approach would be to put the OFFER information into the
struct ic_device rather than storing it globally.
Signed-off-by: Chris Friesen <cfriesen@nortel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
It looks like the dev in netpoll_poll can be NULL - at lease it's
checked at the function beginning. Thus the dev->netde_ops dereference
looks dangerous.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
* git://git.kernel.org/pub/scm/linux/kernel/git/holtmann/bluetooth-2.6:
Bluetooth: Don't trigger disconnect timeout for security mode 3 pairing
Bluetooth: Don't use hci_acl_connect_cancel() for incoming connections
Bluetooth: Fix wrong module refcount when connection setup fails
Another case of me handling the fallout from Davem's unfortunate
addiction to shuffleboard.
Won't anybody think of the children? Join the anti-shuffleboard league
today!
* git://git.kernel.org/pub/scm/linux/kernel/git/linville/wireless-2.6:
iwlwifi: fix device id registration for 6000 series 2x2 devices
ath5k: update channel in sw state after stopping RX and TX
rtl8187: use DMA-aware buffers with usb_control_msg
mac80211: avoid NULL ptr deref when finding max_rates in PID and minstrel
airo: airo_get_encode{,ext} potential buffer overflow
Pulled directly by Linus because Davem is off playing shuffle-board at
some Alaskan cruise, and the NULL ptr deref issue hits people and should
get merged sooner rather than later.
David - make us proud on the shuffle-board tournament!
* 'for-2.6.30' of git://linux-nfs.org/~bfields/linux:
nfsd: silence lockdep warning
lockd: fix list corruption on lockd restart
nfsd4: check for negative dentry before use in nfsv4 readdir
nfsd41: slots are freed with session
svcrdma: clean up error paths.
svcrdma: Fix dma map direction for rdma read targets
"There is another problem with this piece of code. The sband will be NULL
after second iteration on single band device and cause null pointer
dereference. Everything is working with dual band card. Sorry, but i
don't know how to explain this clearly in English. I have looked on the
second patch for pid algorithm and found similar bug."
Reported-by: Karol Szuster <qflon@o2.pl>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6: (26 commits)
bonding: fix panic if initialization fails
IXP4xx: complete Ethernet netdev setup before calling register_netdev().
IXP4xx: use "ENODEV" instead of "ENOSYS" in module initialization.
ipvs: Fix IPv4 FWMARK virtual services
ipv4: Make INET_LRO a bool instead of tristate.
net: remove stale reference to fastroute from Kconfig help text
net: update skb_recycle_check() for hardware timestamping changes
bnx2: Fix panic in bnx2_poll_work().
net-sched: fix bfifo default limit
igb: resolve panic on shutdown when SR-IOV is enabled
wimax: oops: wimax_dev_add() is the only one that can initialize the state
wimax: fix oops if netlink fails to add attribute
Bluetooth: Move dev_set_name() to a context that can sleep
netfilter: ctnetlink: fix wrong message type in user updates
netfilter: xt_cluster: fix use of cluster match with 32 nodes
netfilter: ip6t_ipv6header: fix match on packets ending with NEXTHDR_NONE
netfilter: add missing linux/types.h include to xt_LED.h
mac80211: pid, fix memory corruption
mac80211: minstrel, fix memory corruption
cfg80211: fix comment on regulatory hint processing
...
A remote device in security mode 3 that tries to connect will require
the pairing during the connection setup phase. The disconnect timeout
is now triggered within 10 milliseconds and causes the pairing to fail.
If a connection is not fully established and a PIN code request is
received, don't trigger the disconnect timeout. The either successful
or failing connection complete event will make sure that the timeout
is triggered at the right time.
The biggest problem with security mode 3 is that many Bluetooth 2.0
device and before use a temporary security mode 3 for dedicated
bonding.
Based on a report by Johan Hedberg <johan.hedberg@nokia.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Tested-by: Johan Hedberg <johan.hedberg@nokia.com>
The connection setup phase takes around 2 seconds or longer and in
that time it is possible that the need for an ACL connection is no
longer present. If that happens then, the connection attempt will
be canceled.
This only applies to outgoing connections, but currently it can also
be triggered by incoming connection. Don't call hci_acl_connect_cancel()
on incoming connection since these have to be either accepted or rejected
in this state. Once they are successfully connected they need to be
fully disconnected anyway.
Also remove the wrong hci_acl_disconn() call for SCO and eSCO links
since at this stage they can't be disconnected either, because the
connection handle is still unknown.
Based on a report by Johan Hedberg <johan.hedberg@nokia.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Tested-by: Johan Hedberg <johan.hedberg@nokia.com>
The module refcount is increased by hci_dev_hold() call in hci_conn_add()
and decreased by hci_dev_put() call in del_conn(). In case the connection
setup fails, hci_dev_put() is never called.
Procedure to reproduce the issue:
# hciconfig hci0 up
# lsmod | grep btusb -> "used by" refcount = 1
# hcitool cc <non-exisiting bdaddr> -> will get timeout
# lsmod | grep btusb -> "used by" refcount = 2
# hciconfig hci0 down
# lsmod | grep btusb -> "used by" refcount = 1
# rmmod btusb -> ERROR: Module btusb is in use
The hci_dev_put() call got moved into del_conn() with the 2.6.25 kernel
to fix an issue with hci_dev going away before hci_conn. However that
change was wrong and introduced this problem.
When calling hci_conn_del() it has to call hci_dev_put() after freeing
the connection details. This handling should be fully symmetric. The
execution of del_conn() is done in a work queue and needs it own calls
to hci_dev_hold() and hci_dev_put() to ensure that the hci_dev stays
until the connection cleanup has been finished.
Based on a report by Bing Zhao <bzhao@marvell.com>
Signed-off-by: Marcel Holtmann <marcel@holtmann.org>
Tested-by: Bing Zhao <bzhao@marvell.com>
This fixes the use of fwmarks to denote IPv4 virtual services
which was unfortunately broken as a result of the integration
of IPv6 support into IPVS, which was included in 2.6.28.
The problem arises because fwmarks are stored in the 4th octet
of a union nf_inet_addr .all, however in the case of IPv4 only
the first octet, corresponding to .ip, is assigned and compared.
In other words, using .all = { 0, 0, 0, htonl(svc->fwmark) always
results in a value of 0 (32bits) being stored for IPv4. This means
that one fwmark can be used, as it ends up being mapped to 0, but things
break down when multiple fwmarks are used, as they all end up being mapped
to 0.
As fwmarks are 32bits a reasonable fix seems to be to just store the fwmark
in .ip, and comparing and storing .ip when fwmarks are used.
This patch makes the assumption that in calls to ip_vs_ct_in_get()
and ip_vs_sched_persist() if the proto parameter is IPPROTO_IP then
we are dealing with an fwmark. I believe this is valid as ip_vs_in()
does fairly strict filtering on the protocol and IPPROTO_IP should
not be used in these calls unless explicitly passed when making
these calls for fwmarks in ip_vs_sched_persist().
Tested-by: Fabien Duchêne <fabien.duchene@student.uclouvain.be>
Cc: Joseph Mack NA3T <jmack@wm7d.net>
Cc: Julius Volz <julius.volz@gmail.com>
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
This code is used as a library by several device drivers,
which select INET_LRO.
If some are modules and some are statically built into the
kernel, we get build failures if INET_LRO is modular.
Signed-off-by: David S. Miller <davem@davemloft.net>
David S. Miller
Eric Dumazet
Luis R. Rodriguez
Johannes Berg
Frans Pop
Thomas Chenault
Ilpo Järvinen
Stephen Hemminger
Chris Friesen
Pavel Emelyanov
Linus Torvalds
John W. Linville
Marcel Holtmann
Simon Horman
Ashish Karkare