This patch improves ctnetlink event reliability if one broadcast
listener has set the NETLINK_BROADCAST_ERROR socket option.
The logic is the following: if an event delivery fails, we keep
the undelivered events in the missed event cache. Once the next
packet arrives, we add the new events (if any) to the missed
events in the cache and we try a new delivery, and so on. Thus,
if ctnetlink fails to deliver an event, we try to deliver them
once we see a new packet. Therefore, we may lose state
transitions but the userspace process gets in sync at some point.
At worst case, if no events were delivered to userspace, we make
sure that destroy events are successfully delivered. Basically,
if ctnetlink fails to deliver the destroy event, we remove the
conntrack entry from the hashes and we insert them in the dying
list, which contains inactive entries. Then, the conntrack timer
is added with an extra grace timeout of random32() % 15 seconds
to trigger the event again (this grace timeout is tunable via
/proc). The use of a limited random timeout value allows
distributing the "destroy" resends, thus, avoiding accumulating
lots "destroy" events at the same time. Event delivery may
re-order but we can identify them by means of the tuple plus
the conntrack ID.
The maximum number of conntrack entries (active or inactive) is
still handled by nf_conntrack_max. Thus, we may start dropping
packets at some point if we accumulate a lot of inactive conntrack
entries that did not successfully report the destroy event to
userspace.
During my stress tests consisting of setting a very small buffer
of 2048 bytes for conntrackd and the NETLINK_BROADCAST_ERROR socket
flag, and generating lots of very small connections, I noticed
very few destroy entries on the fly waiting to be resend.
A simple way to test this patch consist of creating a lot of
entries, set a very small Netlink buffer in conntrackd (+ a patch
which is not in the git tree to set the BROADCAST_ERROR flag)
and invoke `conntrack -F'.
For expectations, no changes are introduced in this patch.
Currently, event delivery is only done for new expectations (no
events from expectation expiration, removal and confirmation).
In that case, they need a per-expectation event cache to implement
the same idea that is exposed in this patch.
This patch can be useful to provide reliable flow-accouting. We
still have to add a new conntrack extension to store the creation
and destroy time.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Patrick McHardy <kaber@trash.net>
This patch adds the hlist_nulls_add_head() function which is
based on hlist_nulls_add_head_rcu() but without the use of
rcu_assign_pointer(). It also adds hlist_nulls_del which is
exactly the same like hlist_nulls_del_rcu().
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
This patch moves the helper destruction to a function that lives
in nf_conntrack_helper.c. This new function is used in the patch
to add ctnetlink reliable event delivery.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Patrick McHardy <kaber@trash.net>
This patch reworks the per-cpu event caching to use the conntrack
extension infrastructure.
The main drawback is that we consume more memory per conntrack
if event delivery is enabled. This patch is required by the
reliable event delivery that follows to this patch.
BTW, this patch allows you to enable/disable event delivery via
/proc/sys/net/netfilter/nf_conntrack_events in runtime, although
you can still disable event caching as compilation option.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Use mod_timer_pending() instead of atomic sequence of del_timer()/
add_timer(). mod_timer_pending() does not rearm an inactive timer,
so we don't need the conntrack lock anymore to make sure we don't
accidentally rearm a timer of a conntrack which is in the process
of being destroyed.
With this change, we don't need to take the global lock anymore at all,
counter updates can be performed under the per-conntrack lock.
Signed-off-by: Patrick McHardy <kaber@trash.net>
We need to enforce the IP alignment on the non-mergeable RX path just
like the other RX path. Not doing so results in misaligned IP
headers.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some drivers incorrectly use ntohs() instead of htons()
A cleanup as htons() returns same result than ntohs(),
but better to use the proper one.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix build error introduced by commit bb70dfa5 (netfilter: xtables:
consolidate comefrom debug cast access):
net/ipv4/netfilter/ip_tables.c: In function 'ipt_do_table':
net/ipv4/netfilter/ip_tables.c:421: error: 'comefrom' undeclared (first use in this function)
net/ipv4/netfilter/ip_tables.c:421: error: (Each undeclared identifier is reported only once
net/ipv4/netfilter/ip_tables.c:421: error: for each function it appears in.)
Signed-off-by: Patrick McHardy <kaber@trash.net>
SH4's BUG() seems to confuse the compiler as it is considered to
return; thus, some functions would trigger usage of uninitialized
variables or non-void functions returning void.
Work around by initializing/returning.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
Caused by an API update. The return value can be safely ignored, as
there is notthing we can do with it.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
This fix triggering the WARN_ON_ONCE(in_irq() || irqs_disabled()); in
local_bh_enable().
Here is no need to grab this lock, this was wrong at all and may
cause a deadlock and access to freed memory, since on a TEI remove
the current listelement can be deleted under us. So this is clearly
a case for list_for_each_entry_safe.
Signed-off-by: Karsten Keil <keil@b1-systems.de>
The check for overindexing of dev->mdm.info[] has an off-by-one.
Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Karsten Keil <keil@b1-systems.de>
If we get no interrupts for after 3 resets we need to unregister
the interrupt function, which is already done outside the loop.
Signed-off-by: Andreas Mohr <andi@lisas.de>
Signed-off-by: Karsten Keil <keil@b1-systems.de>
Some users still load bond module multiple times to create bonding
devices. This accidentally was broken by a later patch about
the time sysfs was fixed. According to Jay, it was broken
by:
commit b8a9787edd
Author: Jay Vosburgh <fubar@us.ibm.com>
Date: Fri Jun 13 18:12:04 2008 -0700
bonding: Allow setting max_bonds to zero
Note: sysfs and procfs still produce WARN() messages when this is done
so the sysfs method is the recommended API.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The current code errors out the INCOMPLETE neigh entry skb queue only from
the timer if maximum probes have been attempted and there has been no reply.
This also causes the transtion to FAILED state.
However, the neigh entry can be also updated via Netlink to inform that the
address is unavailable. Currently, neigh_update() just stops the timers and
leaves the pending skb's unreleased. This results that the clean up code in
the timer callback is never called, preventing also proper garbage collection.
This fixes neigh_update() to process the pending skb queue immediately if
INCOMPLETE -> FAILED state transtion occurs due to a Netlink request.
Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
When the i2400m device resets, the driver code will force some
functions to return a -ERESTARTSYS error code, which can is used by
the caller to determine which recovery actions to take.
However, in certain situations the only thing that can be done is to
bubble up said error code to user space, for handling.
However, -ERESTARSYS was a poor choice, as it is supposed to be used
by the kernel only.
As such, replace -ERESTARTSYS with -EL3RST; as well, in
i2400m_msg_to_dev(), when the device is in boot mode (following a
recent reset), return -EL3RST instead of -ENODEV (meaning the device
is in bootrom mode after a reset, not that the device was
disconnected, and thus, normal commands cannot be executed).
Signed-off-by: Cindy H Kao <cindy.h.kao@intel.com>
When a device reset happens during firmware load [in
i2400m_dev_bootstrap()], __i2400m_dev_start() will retry a number of
times. However, for those retries to be able to accomplish anything,
the device's bootrom has to be reinitialized.
Thus, on the retry path, pass the I2400M_MAC_REINIT to the firmware
load code.
Signed-off-by: Cindy H Kao <cindy.h.kao@intel.com>
The current SDIO code was working in polling mode for boot-mode
(firmware load) mode. This was causing issues on some hardware.
Moved all the RX code to use a unified IRQ handler that based on the
type of data the device is sending can discriminate and decide which
is the right destination.
As well, all the reads from the device are made to be at least the
block size (256); the driver will ignore the rest when not needed.
Signed-off-by: Dirk Brandewie <dirk.j.brandewie@intel.com>
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
When i2400m_bootrom_init() fails to put the device into a state of
being ready to accept firmware, the driver was currently trying to
reset it if it failed to do so. This is not too useful; as part of
trying to put the device in the right state a few resets have already
been tried.
At this point, things are probably fried out and an extra reset might
do more harm than good (for example causing reseting of other
functions in the same composite device).
So it is left up to the callers to determine the error path to take
(at the end this is always i2400m_setup(), who depending on how many
retries are left, might give up on the device).
From a fix by Cindy H. Kao.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
This change moves the table of "pokes" performed on the device at boot
time to the bus specific portion of the driver.
Different models of the i2400m device supported by this driver require
different poke tables, thus having a single table that works for all
is impossible. For that, the table is moved to the bus-specific
driver, who can decide which table to use based on the specifics of
the device and point the generic driver to it.
Signed-off-by: Dirk Brandewie <dirk.j.brandewie@intel.com>
The code that sets up the i2400m (firmware load and general driver
setup after it) includes a couple of retry loops.
The SDIO device sometimes can get in more complicated corners than the
USB one (due to its interaction with other SDIO functions), that
require trying a few more times.
To solve that, without having a failing USB device taking longer to be
considered dead, allow the retry counts to be specified by the
bus-specific driver, which the general driver takes as a parameter.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
When a device reboot happens when we are under probe, with init_mutex
taken, make sure we can recover. Have dev_reset_handle set boot mode
and i2400m_msg_to_dev() will see it and fail gracefully instead of
timing out.
Found and diagnosed by Cindy H. Kao.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
When the TX FIFO filled up and i2400m_tx_new() failed to allocate a
new TX message header, a missing check for said condition was causing a
kernel oops when trying to dereference a NULL i2400m->tx_msg pointer.
Found and diagnosed by Cindy H. Kao.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
i2400m_dev_shutdown() tried to reset the device to put it in a known
state before shutting down.
But that turned out to be pointless. We reach this case in two paths:
1 - when the device resets, to clean up state
2 - when the driver is unloaded, for the same
however, in both cases it is pointless; in (1) the device is already
reset, why do it again? in (2) we can't -- the USB stack, for example,
doesn't allow communicating with the device when the driver is being
unbound and if the device is disconnected, the device is gone already.
So just remove it. Leave the function as a placeholder for future
cleanups that will be done from data allocated by the driver during
device operation.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
i2400m_tx_skip_tail() needs to handle the special case of being called
when the tail room that is left over in the FIFO is zero.
This happens when a TX message header was opened at the very end of
the FIFO (without payloads). The i2400m_tx_close() code already marked
said TX message (header) to be skipped and this function should be
doing nothing.
It is called anyway because it is part of a common "corner case" path
handling which takes care of more cases than only this one.
The tail room computation was also improved to take care of the case
when tx_in is at the end of the buffer boundary; tail_room has to be
modded (%) to the buffer size. To do that in a single well-documented
place, __i2400m_tx_tail_room() is introduced and used.
Treat i2400m->tx_in == 0 as a corner case and handle it accordingly.
Found and diagnosed by Cindy H. Kao.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
In some situations, when a new TX message header is started, there
might be no space for data payloads. In this case the message is left
with zero payloads and the i2400m_tx_close() function has just to mark
it as "to skip". If it tries to go ahead it will overwrite things
because there is no space to add padding as defined by the
bus-specific layer. This can cause buffer overruns and in some stress
cases, panics.
Found and diagnosed by Cindy H. Kao.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
The constant is being use as an alignment factor, not as a padding
factor; made reading/reviewing the code quite confusing.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
This reset type causes the WiMAX function to be disabled and
re-enabled, which will force the WiMAX device to reset and enter boot
mode.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
Signed-off-by: Dirk Brandewie <dirk.j.brandewie@intel.com>
By mistake, the BUG_ON() check was left in there and it will fail when
called if i2400m->work_queue is still not setup.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
RX support is the only user of the work-queue, to process
reports/notifications from the device. Thus, it needs the work queue
to be initialized first.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
Reported and fixed by Cindy H Kao.
When the device is stopped __i2400m_dev_stop() stops the network
queue.
However, when this is done in the middle of heavy network operation,
when the bus-specific subdriver is still wrapping up and it reports a
sent TX transaction with _tx_msg_sent() right after the device was
stopped, the queue was being started again, which was causing a stream
of oopsen and finally a panic.
In any case, said call has no place there. It's a left over from an
early implementation that was discarded later on.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
The i2400m driver waits for the device to report being ready for
entering power save before asking it to do so. This module parameter
allows control of said operation; if disabled, the driver won't ask
the device to enter power save mode.
This is useful in setups where power saving is not so important or
when the overhead imposed by network reentry after power save is not
acceptable; by combining this with parameter 'idle_mode_disabled', the
driver will always maintain both the connection and the device in
active state.
Signed-off-by: Inaky Perez-Gonzalez <inaky@linux.intel.com>
One of the problem with sock memory accounting is it uses
a pair of sock_hold()/sock_put() for each transmitted packet.
This slows down bidirectional flows because the receive path
also needs to take a refcount on socket and might use a different
cpu than transmit path or transmit completion path. So these
two atomic operations also trigger cache line bounces.
We can see this in tx or tx/rx workloads (media gateways for example),
where sock_wfree() can be in top five functions in profiles.
We use this sock_hold()/sock_put() so that sock freeing
is delayed until all tx packets are completed.
As we also update sk_wmem_alloc, we could offset sk_wmem_alloc
by one unit at init time, until sk_free() is called.
Once sk_free() is called, we atomic_dec_and_test(sk_wmem_alloc)
to decrement initial offset and atomicaly check if any packets
are in flight.
skb_set_owner_w() doesnt call sock_hold() anymore
sock_wfree() doesnt call sock_put() anymore, but check if sk_wmem_alloc
reached 0 to perform the final freeing.
Drawback is that a skb->truesize error could lead to unfreeable sockets, or
even worse, prematurely calling __sk_free() on a live socket.
Nice speedups on SMP. tbench for example, going from 2691 MB/s to 2711 MB/s
on my 8 cpu dev machine, even if tbench was not really hitting sk_refcnt
contention point. 5 % speedup on a UDP transmit workload (depends
on number of flows), lowering TX completion cpu usage.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
vfree() does its own 'NULL' check, so no need for check before
calling it.
Signed-off-by: Figo.zhang <figo1802@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the tx queue destroy path, be_tx_q_clean() is currently called after the tx queues are freed; it must be called before.
Signed-off-by: Sathya Perla <sathyap@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
be_rx_compl_get() must not reset(via the valid word) the rx_compl as the rx_compl is not processed yet; it must be reset after it is processed.
Signed-off-by: Sathya Perla <sathyap@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rx stats are not getting updated when an rx_compl with only one frag is rcvd in non-lro path.
Signed-off-by: Sathya Perla <sathyap@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fix netdev stat rx_errors to cover length related errors and checksum errors and rx_dropped to the pkts dropped due to lack of buffers
Signed-off-by: Sathya Perla <sathyap@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use cancel_delayed_work_sycn instead of cancel_delayed_work() to reliably kill be_worker() as it rearms itself.
Signed-off-by: Sathya Perla <sathyap@serverengines.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
