Since commit 7361c36c52 (af_unix: Allow credentials to work across
user and pid namespaces) af_unix performance dropped a lot.
This is because we now take a reference on pid and cred in each write(),
and release them in read(), usually done from another process,
eventually from another cpu. This triggers false sharing.
# Events: 154K cycles
#
# Overhead Command Shared Object Symbol
# ........ ....... .................. .........................
#
10.40% hackbench [kernel.kallsyms] [k] put_pid
8.60% hackbench [kernel.kallsyms] [k] unix_stream_recvmsg
7.87% hackbench [kernel.kallsyms] [k] unix_stream_sendmsg
6.11% hackbench [kernel.kallsyms] [k] do_raw_spin_lock
4.95% hackbench [kernel.kallsyms] [k] unix_scm_to_skb
4.87% hackbench [kernel.kallsyms] [k] pid_nr_ns
4.34% hackbench [kernel.kallsyms] [k] cred_to_ucred
2.39% hackbench [kernel.kallsyms] [k] unix_destruct_scm
2.24% hackbench [kernel.kallsyms] [k] sub_preempt_count
1.75% hackbench [kernel.kallsyms] [k] fget_light
1.51% hackbench [kernel.kallsyms] [k]
__mutex_lock_interruptible_slowpath
1.42% hackbench [kernel.kallsyms] [k] sock_alloc_send_pskb
This patch includes SCM_CREDENTIALS information in a af_unix message/skb
only if requested by the sender, [man 7 unix for details how to include
ancillary data using sendmsg() system call]
Note: This might break buggy applications that expected SCM_CREDENTIAL
from an unaware write() system call, and receiver not using SO_PASSCRED
socket option.
If SOCK_PASSCRED is set on source or destination socket, we still
include credentials for mere write() syscalls.
Performance boost in hackbench : more than 50% gain on a 16 thread
machine (2 quad-core cpus, 2 threads per core)
hackbench 20 thread 2000
4.228 sec instead of 9.102 sec
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
RCU api had been completed and rcu_access_pointer() or
rcu_dereference_protected() are better than generic
rcu_dereference_raw()
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Consider the following situation:
* a dump that would show 8 entries, four in the first
round, and four in the second
* between the first and second rounds, 6 entries are
removed
* now the second round will not show any entry, and
even if there is a sequence/generation counter the
application will not know
To solve this problem, add a new flag NLM_F_DUMP_INTR
to the netlink header that indicates the dump wasn't
consistent, this flag can also be set on the MSG_DONE
message that terminates the dump, and as such above
situation can be detected.
To achieve this, add a sequence counter to the netlink
callback struct. Of course, netlink code still needs
to use this new functionality. The correct way to do
that is to always set cb->seq when a dumpit callback
is invoked and call nl_dump_check_consistent() for
each new message. The core code will also call this
function for the final MSG_DONE message.
To make it usable with generic netlink, a new function
genlmsg_nlhdr() is needed to obtain the netlink header
from the genetlink user header.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
In c7ac8679be "rtnetlink: Compute and store minimum ifinfo dump
size", we moved the allocation under the lock so we need to unlock
on error path.
Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@conan.davemloft.net>
The message size allocated for rtnl ifinfo dumps was limited to
a single page. This is not enough for additional interface info
available with devices that support SR-IOV and caused a bug in
which VF info would not be displayed if more than approximately
40 VFs were created per interface.
Implement a new function pointer for the rtnl_register service that will
calculate the amount of data required for the ifinfo dump and allocate
enough data to satisfy the request.
Signed-off-by: Greg Rose <gregory.v.rose@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
The %pK format specifier is designed to hide exposed kernel pointers,
specifically via /proc interfaces. Exposing these pointers provides an
easy target for kernel write vulnerabilities, since they reveal the
locations of writable structures containing easily triggerable function
pointers. The behavior of %pK depends on the kptr_restrict sysctl.
If kptr_restrict is set to 0, no deviation from the standard %p behavior
occurs. If kptr_restrict is set to 1, the default, if the current user
(intended to be a reader via seq_printf(), etc.) does not have CAP_SYSLOG
(currently in the LSM tree), kernel pointers using %pK are printed as 0's.
If kptr_restrict is set to 2, kernel pointers using %pK are printed as
0's regardless of privileges. Replacing with 0's was chosen over the
default "(null)", which cannot be parsed by userland %p, which expects
"(nil)".
The supporting code for kptr_restrict and %pK are currently in the -mm
tree. This patch converts users of %p in net/ to %pK. Cases of printing
pointers to the syslog are not covered, since this would eliminate useful
information for postmortem debugging and the reading of the syslog is
already optionally protected by the dmesg_restrict sysctl.
Signed-off-by: Dan Rosenberg <drosenberg@vsecurity.com>
Cc: James Morris <jmorris@namei.org>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Thomas Graf <tgraf@infradead.org>
Cc: Eugene Teo <eugeneteo@kernel.org>
Cc: Kees Cook <kees.cook@canonical.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: David S. Miller <davem@davemloft.net>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Eric Paris <eparis@parisplace.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The rcu callback listeners_free_rcu() just calls a kfree(),
so we use kfree_rcu() instead of the call_rcu(listeners_free_rcu).
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Reviewed-by: Josh Triplett <josh@joshtriplett.org>
Netlink message processing in the kernel is synchronous these days,
capabilities can be checked directly in security_netlink_recv() from
the current process.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Reviewed-by: James Morris <jmorris@namei.org>
[chrisw: update to include pohmelfs and uvesafb]
Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Netlink message processing in the kernel is synchronous these days, the
session information can be collected when needed.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
netlink_dump() may failed, but nobody handle its error.
It generates output data, when a previous portion has been returned to
user space. This mechanism works when all data isn't go in skb. If we
enter in netlink_recvmsg() and skb is absent in the recv queue, the
netlink_dump() will not been executed. So if netlink_dump() is failed
one time, the new data never appear and the reader will sleep forever.
netlink_dump() is called from two places:
1. from netlink_sendmsg->...->netlink_dump_start().
In this place we can report error directly and it will be returned
by sendmsg().
2. from netlink_recvmsg
There we can't report error directly, because we have a portion of
valid output data and call netlink_dump() for prepare the next portion.
If netlink_dump() is failed, the socket will be mark as error and the
next recvmsg will be failed.
Signed-off-by: Andrey Vagin <avagin@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
commit 6c04bb18dd (netlink: use call_rcu for netlink_change_ngroups)
used a somewhat convoluted and racy way to perform call_rcu().
The old block of memory is freed after a grace period, but the rcu_head
used to track it is located in new block.
This can clash if we call two times or more netlink_change_ngroups(),
and a block is freed before another. call_rcu() called on different cpus
makes no guarantee in order of callbacks.
Fix this using a more standard way of handling this : Each block of
memory contains its own rcu_head, so that no 'use after free' can
happens.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Johannes Berg <johannes@sipsolutions.net>
CC: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Once we started enforcing the a nl_table[] entry exist for
a protocol, NETLINK_USERSOCK stopped working. Add a dummy
table entry so that it works again.
Reported-by: Thomas Voegtle <tv@lio96.de>
Tested-by: Thomas Voegtle <tv@lio96.de>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since
commit 1dacc76d00
Author: Johannes Berg <johannes@sipsolutions.net>
Date: Wed Jul 1 11:26:02 2009 +0000
net/compat/wext: send different messages to compat tasks
we had a race condition when setting and then
restoring frag_list. Eric attempted to fix it,
but the fix created even worse problems.
However, the original motivation I had when I
added the code that turned out to be racy is
no longer clear to me, since we only copy up
to skb->len to userspace, which doesn't include
the frag_list length. As a result, not doing
any frag_list clearing and restoring avoids
the race condition, while not introducing any
other problems.
Additionally, while preparing this patch I found
that since none of the remaining netlink code is
really aware of the frag_list, we need to use the
original skb's information for packet information
and credentials. This fixes, for example, the
group information received by compat tasks.
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: stable@kernel.org [2.6.31+, for 2.6.35 revert 1235f504aa]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This reverts commit 1235f504aa.
It causes regressions worse than the problem it was trying
to fix. Eric will try to solve the problem another way.
Signed-off-by: David S. Miller <davem@davemloft.net>
commit 1dacc76d00
(net/compat/wext: send different messages to compat tasks)
introduced a race condition on netlink, in case MSG_PEEK is used.
An skb given by skb_recv_datagram() might be shared, we must copy it
before any modification, or risk fatal corruption.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Convert a few calls from kfree_skb to consume_skb
Noticed while I was working on dropwatch that I was detecting lots of internal
skb drops in several places. While some are legitimate, several were not,
freeing skbs that were at the end of their life, rather than being discarded due
to an error. This patch converts those calls sites from using kfree_skb to
consume_skb, which quiets the in-kernel drop_monitor code from detecting them as
drops. Tested successfully by myself
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
scm_send occasionally allocates state in the scm_cookie, so I have
modified netlink_sendmsg to guarantee that when scm_send succeeds
scm_destory will be called to free that state.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Reviewed-by: Daniel Lezcano <daniel.lezcano@free.fr>
Acked-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
When netlink sockets are used to convey data that is in a namespace
we need a way to select a subset of the listening sockets to deliver
the packet to. For the network namespace we have been doing this
by only transmitting packets in the correct network namespace.
For data belonging to other namespaces netlink_bradcast_filtered
provides a mechanism that allows us to examine the destination
socket and to decide if we should transmit the specified packet
to it.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
check the length of the socket address passed to connect(2).
Check the length of the socket address passed to connect(2). If the
length is invalid, -EINVAL will be returned.
Signed-off-by: Changli Gao <xiaosuo@gmail.com>
----
net/bluetooth/l2cap.c | 3 ++-
net/bluetooth/rfcomm/sock.c | 3 ++-
net/bluetooth/sco.c | 3 ++-
net/can/bcm.c | 3 +++
net/ieee802154/af_ieee802154.c | 3 +++
net/ipv4/af_inet.c | 5 +++++
net/netlink/af_netlink.c | 3 +++
7 files changed, 20 insertions(+), 3 deletions(-)
Signed-off-by: David S. Miller <davem@davemloft.net>
This was included in OpenVZ kernels but wasn't integrated upstream.
>From git://git.openvz.org/pub/linux-2.6.24-openvz:
commit 5c69402f18adf7276352e051ece2cf31feefab02
Author: Alexey Dobriyan <adobriyan@openvz.org>
Date: Mon Dec 24 14:37:45 2007 +0300
netlink: fixup ->tgid to work in multiple PID namespaces
Signed-off-by: Tom Goff <thomas.goff@boeing.com>
Acked-by: Alexey Dobriyan <adobriyan@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, ENOBUFS errors are reported to the socket via
netlink_set_err() even if NETLINK_RECV_NO_ENOBUFS is set. However,
that should not happen. This fixes this problem and it changes the
prototype of netlink_set_err() to return the number of sockets that
have set the NETLINK_RECV_NO_ENOBUFS socket option. This return
value is used in the next patch in these bugfix series.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
The Inode field in /proc/net/{tcp,udp,packet,raw,...} is useful to know the types of
file descriptors associated to a process. Actually lsof utility uses the field.
Unfortunately, unlike /proc/net/{tcp,udp,packet,raw,...}, /proc/net/netlink doesn't have the field.
This patch adds the field to /proc/net/netlink.
Signed-off-by: Masatake YAMATO <yamato@redhat.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Generated with the following semantic patch
@@
struct net *n1;
struct net *n2;
@@
- n1 == n2
+ net_eq(n1, n2)
@@
struct net *n1;
struct net *n2;
@@
- n1 != n2
+ !net_eq(n1, n2)
applied over {include,net,drivers/net}.
Signed-off-by: Octavian Purdila <opurdila@ixiacom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The netlink URELEASE notifier doesn't notify for
sockets that have been used to receive multicast
but it should be called for such sockets as well
since they might _also_ be used for sending and
not solely for receiving multicast. We will need
that for nl80211 (generic netlink sockets) in the
future.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Cc: Patrick McHardy <kaber@trash.net>
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>
The generic __sock_create function has a kern argument which allows the
security system to make decisions based on if a socket is being created by
the kernel or by userspace. This patch passes that flag to the
net_proto_family specific create function, so it can do the same thing.
Signed-off-by: Eric Paris <eparis@redhat.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
All usages of structure net_proto_ops should be declared const.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This provides safety against negative optlen at the type
level instead of depending upon (sometimes non-trivial)
checks against this sprinkled all over the the place, in
each and every implementation.
Based upon work done by Arjan van de Ven and feedback
from Linus Torvalds.
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, the nlmsg->len field is not set correctly in netlink_ack()
for ack messages that include the nlmsg of the error frame. This
corrects the length field passed to __nlmsg_put to use the correct
payload size.
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Similar to commit d136f1bd36,
there's a bug when unregistering a generic netlink family,
which is caught by the might_sleep() added in that commit:
BUG: sleeping function called from invalid context at net/netlink/af_netlink.c:183
in_atomic(): 1, irqs_disabled(): 0, pid: 1510, name: rmmod
2 locks held by rmmod/1510:
#0: (genl_mutex){+.+.+.}, at: [<ffffffff8138283b>] genl_unregister_family+0x2b/0x130
#1: (rcu_read_lock){.+.+..}, at: [<ffffffff8138270c>] __genl_unregister_mc_group+0x1c/0x120
Pid: 1510, comm: rmmod Not tainted 2.6.31-wl #444
Call Trace:
[<ffffffff81044ff9>] __might_sleep+0x119/0x150
[<ffffffff81380501>] netlink_table_grab+0x21/0x100
[<ffffffff813813a3>] netlink_clear_multicast_users+0x23/0x60
[<ffffffff81382761>] __genl_unregister_mc_group+0x71/0x120
[<ffffffff81382866>] genl_unregister_family+0x56/0x130
[<ffffffffa0007d85>] nl80211_exit+0x15/0x20 [cfg80211]
[<ffffffffa000005a>] cfg80211_exit+0x1a/0x40 [cfg80211]
Fix in the same way by grabbing the netlink table lock
before doing rcu_read_lock().
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sizing of memory allocations shouldn't depend on the number of physical
pages found in a system, as that generally includes (perhaps a huge amount
of) non-RAM pages. The amount of what actually is usable as storage
should instead be used as a basis here.
Some of the calculations (i.e. those not intending to use high memory)
should likely even use (totalram_pages - totalhigh_pages).
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Ingo Molnar <mingo@elte.hu>
Cc: Dave Airlie <airlied@linux.ie>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Patrick McHardy <kaber@trash.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Since my commits introducing netns awareness into
genetlink we can get this problem:
BUG: scheduling while atomic: modprobe/1178/0x00000002
2 locks held by modprobe/1178:
#0: (genl_mutex){+.+.+.}, at: [<ffffffff8135ee1a>] genl_register_mc_grou
#1: (rcu_read_lock){.+.+..}, at: [<ffffffff8135eeb5>] genl_register_mc_g
Pid: 1178, comm: modprobe Not tainted 2.6.31-rc8-wl-34789-g95cb731-dirty #
Call Trace:
[<ffffffff8103e285>] __schedule_bug+0x85/0x90
[<ffffffff81403138>] schedule+0x108/0x588
[<ffffffff8135b131>] netlink_table_grab+0xa1/0xf0
[<ffffffff8135c3a7>] netlink_change_ngroups+0x47/0x100
[<ffffffff8135ef0f>] genl_register_mc_group+0x12f/0x290
because I overlooked that netlink_table_grab() will
schedule, thinking it was just the rwlock. However,
in the contention case, that isn't actually true.
Fix this by letting the code grab the netlink table
lock first and then the RCU for netns protection.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Consitfy nlmsghdr arguments to a couple of functions as preparation
for the next patch, which will constify the netlink message data in
all nfnetlink users.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Wireless extensions have the unfortunate problem that events
are multicast netlink messages, and are not independent of
pointer size. Thus, currently 32-bit tasks on 64-bit platforms
cannot properly receive events and fail with all kinds of
strange problems, for instance wpa_supplicant never notices
disassociations, due to the way the 64-bit event looks (to a
32-bit process), the fact that the address is all zeroes is
lost, it thinks instead it is 00:00:00:00:01:00.
The same problem existed with the ioctls, until David Miller
fixed those some time ago in an heroic effort.
A different problem caused by this is that we cannot send the
ASSOCREQIE/ASSOCRESPIE events because sending them causes a
32-bit wpa_supplicant on a 64-bit system to overwrite its
internal information, which is worse than it not getting the
information at all -- so we currently resort to sending a
custom string event that it then parses. This, however, has a
severe size limitation we are frequently hitting with modern
access points; this limitation would can be lifted after this
patch by sending the correct binary, not custom, event.
A similar problem apparently happens for some other netlink
users on x86_64 with 32-bit tasks due to the alignment for
64-bit quantities.
In order to fix these problems, I have implemented a way to
send compat messages to tasks. When sending an event, we send
the non-compat event data together with a compat event data in
skb_shinfo(main_skb)->frag_list. Then, when the event is read
from the socket, the netlink code makes sure to pass out only
the skb that is compatible with the task. This approach was
suggested by David Miller, my original approach required
always sending two skbs but that had various small problems.
To determine whether compat is needed or not, I have used the
MSG_CMSG_COMPAT flag, and adjusted the call path for recv and
recvfrom to include it, even if those calls do not have a cmsg
parameter.
I have not solved one small part of the problem, and I don't
think it is necessary to: if a 32-bit application uses read()
rather than any form of recvmsg() it will still get the wrong
(64-bit) event. However, neither do applications actually do
this, nor would it be a regression.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
For the network namespace work in generic netlink I need
to be able to call this function under rcu_read_lock(),
otherwise the locking becomes a nightmare and more locks
would be needed. Instead, just embed a struct rcu_head
(actually a struct listeners_rcu_head that also carries
the pointer to the memory block) into the listeners
memory so we can use call_rcu() instead of synchronising
and then freeing. No rcu_barrier() is needed since this
code cannot be modular.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
I added those myself in commits b4ff4f04 and 84659eb5,
but I see no reason now why they should be exported,
only generic netlink uses them which cannot be modular.
Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
commit 2b85a34e91
(net: No more expensive sock_hold()/sock_put() on each tx)
changed initial sk_wmem_alloc value.
We need to take into account this offset when reporting
sk_wmem_alloc to user, in PROC_FS files or various
ioctls (SIOCOUTQ/TIOCOUTQ)
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds the NETLINK_NO_ENOBUFS socket flag. This flag can
be used by unicast and broadcast listeners to avoid receiving
ENOBUFS errors.
Generally speaking, ENOBUFS errors are useful to notify two things
to the listener:
a) You may increase the receiver buffer size via setsockopt().
b) You have lost messages, you may be out of sync.
In some cases, ignoring ENOBUFS errors can be useful. For example:
a) nfnetlink_queue: this subsystem does not have any sort of resync
method and you can decide to ignore ENOBUFS once you have set a
given buffer size.
b) ctnetlink: you can use this together with the socket flag
NETLINK_BROADCAST_SEND_ERROR to stop getting ENOBUFS errors as
you do not need to resync (packets whose event are not delivered
are drop to provide reliable logging and state-synchronization).
Moreover, the use of NETLINK_NO_ENOBUFS also reduces a "go up, go down"
effect in terms of performance which is due to the netlink congestion
control when the listener cannot back off. The effect is the following:
1) throughput rate goes up and netlink messages are inserted in the
receiver buffer.
2) Then, netlink buffer fills and overruns (set on nlk->state bit 0).
3) While the listener empties the receiver buffer, netlink keeps
dropping messages. Thus, throughput goes dramatically down.
4) Then, once the listener has emptied the buffer (nlk->state
bit 0 is set off), goto step 1.
This effect is easy to trigger with netlink broadcast under heavy
load, and it is more noticeable when using a big receiver buffer.
You can find some results in [1] that show this problem.
[1] http://1984.lsi.us.es/linux/netlink/
This patch also includes the use of sk_drop to account the number of
netlink messages drop due to overrun. This value is shown in
/proc/net/netlink.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds nfnetlink_set_err() to propagate the error to netlink
broadcast listener in case of memory allocation errors in the
message building.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Patrick McHardy <kaber@trash.net>
The callers of netlink_set_err() currently pass a negative value
as parameter for the error code. However, sk->sk_err wants a
positive error value. Without this patch, skb_recv_datagram() called
by netlink_recvmsg() may return a positive value to report an error.
Another choice to fix this is to change callers to pass a positive
error value, but this seems a bit inconsistent and error prone
to me. Indeed, the callers of netlink_set_err() assumed that the
(usual) negative value for error codes was fine before this patch :).
This patch also includes some documentation in docbook format
for netlink_set_err() to avoid this sort of confusion.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove some pointless conditionals before kfree_skb().
Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch changes the return value of nlmsg_notify() as follows:
If NETLINK_BROADCAST_ERROR is set by any of the listeners and
an error in the delivery happened, return the broadcast error;
else if there are no listeners apart from the socket that
requested a change with the echo flag, return the result of the
unicast notification. Thus, with this patch, the unicast
notification is handled in the same way of a broadcast listener
that has set the NETLINK_BROADCAST_ERROR socket flag.
This patch is useful in case that the caller of nlmsg_notify()
wants to know the result of the delivery of a netlink notification
(including the broadcast delivery) and take any action in case
that the delivery failed. For example, ctnetlink can drop packets
if the event delivery failed to provide reliable logging and
state-synchronization at the cost of dropping packets.
This patch also modifies the rtnetlink code to ignore the return
value of rtnl_notify() in all callers. The function rtnl_notify()
(before this patch) returned the error of the unicast notification
which makes rtnl_set_sk_err() reports errors to all listeners. This
is not of any help since the origin of the change (the socket that
requested the echoing) notices the ENOBUFS error if the notification
fails and should resync itself.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch adds NETLINK_BROADCAST_ERROR which is a netlink
socket option that the listener can set to make netlink_broadcast()
return errors in the delivery to the caller. This option is useful
if the caller of netlink_broadcast() do something with the result
of the message delivery, like in ctnetlink where it drops a network
packet if the event delivery failed, this is used to enable reliable
logging and state-synchronization. If this socket option is not set,
netlink_broadcast() only reports ESRCH errors and silently ignore
ENOBUFS errors, which is what most netlink_broadcast() callers
should do.
This socket option is based on a suggestion from Patrick McHardy.
Patrick McHardy can exchange this patch for a beer from me ;).
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, netlink_broadcast() reports errors to the caller if no
messages at all were delivered:
1) If, at least, one message has been delivered correctly, returns 0.
2) Otherwise, if no messages at all were delivered due to skb_clone()
failure, return -ENOBUFS.
3) Otherwise, if there are no listeners, return -ESRCH.
With this patch, the caller knows if the delivery of any of the
messages to the listeners have failed:
1) If it fails to deliver any message (for whatever reason), return
-ENOBUFS.
2) Otherwise, if all messages were delivered OK, returns 0.
3) Otherwise, if no listeners, return -ESRCH.
In the current ctnetlink code and in Netfilter in general, we can add
reliable logging and connection tracking event delivery by dropping the
packets whose events were not successfully delivered over Netlink. Of
course, this option would be settable via /proc as this approach reduces
performance (in terms of filtered connections per seconds by a stateful
firewall) but providing reliable logging and event delivery (for
conntrackd) in return.
This patch also changes some clients of netlink_broadcast() that
may report ENOBUFS errors via printk. This error handling is not
of any help. Instead, the userspace daemons that are listening to
those netlink messages should resync themselves with the kernel-side
if they hit ENOBUFS.
BTW, netlink_broadcast() clients include those that call
cn_netlink_send(), nlmsg_multicast() and genlmsg_multicast() since they
internally call netlink_broadcast() and return its error value.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
