commit 8efa885406 (sch_sfq: avoid giving spurious NET_XMIT_CN signals)
forgot to call qdisc_tree_decrease_qlen() to signal upper levels that a
packet (from another flow) was dropped, leading to various problems.
With help from Michal Soltys and Michal Pokrywka, who did a bisection.
Bugzilla ref: https://bugzilla.kernel.org/show_bug.cgi?id=39372
Debian ref: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=631945
Reported-by: Lucas Bocchi <lucas.bocchi@gmail.com>
Reported-and-bisected-by: Michal Pokrywka <wolfmoon@o2.pl>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Michal Soltys <soltys@ziu.info>
Acked-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove it, as it indirectly exposes netdev features. It's not used in
iproute2 (2.6.38) - is anything else using its interface?
Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
Signed-off-by: David S. Miller <davem@davemloft.net>
Results on dummy device can be seen in my netconf 2011
slides. These results are for a 10Gige IXGBE intel
nic - on another i5 machine, very similar specs to
the one used in the netconf2011 results.
It turns out - this is a hell lot worse than dummy
and so this patch is even more beneficial for 10G.
Test setup:
----------
System under test sending packets out.
Additional box connected directly dropping packets.
Installed prio qdisc on the eth device and default
netdev default length of 1000 used as is.
The 3 prio bands each were set to 100 (didnt factor in
the results).
5 packet runs were made and the middle 3 picked.
results
-------
The "cpu" column indicates the which cpu the sample
was taken on,
The "Pkt runx" carries the number of packets a cpu
dequeued when forced to be in the "dequeuer" role.
The "avg" for each run is the number of times each
cpu should be a "dequeuer" if the system was fair.
3.0-rc4 (plain)
cpu Pkt run1 Pkt run2 Pkt run3
================================================
cpu0 21853354 21598183 22199900
cpu1 431058 473476 393159
cpu2 481975 477529 458466
cpu3 23261406 23412299 22894315
avg 11506948 11490372 11486460
3.0-rc4 with patch and default weight 64
cpu Pkt run1 Pkt run2 Pkt run3
================================================
cpu0 13205312 13109359 13132333
cpu1 10189914 10159127 10122270
cpu2 10213871 10124367 10168722
cpu3 13165760 13164767 13096705
avg 11693714 11639405 11630008
As you can see the system is still not perfect but
is a lot better than what it was before...
At the moment we use the old backlog weight, weight_p
which is 64 packets. It seems to be reasonably fine
with that value.
The system could be made more fair if we reduce the
weight_p (as per my presentation), but we are going
to affect the shared backlog weight. Unless deemed
necessary, I think the default value is fine. If not
we could add yet another knob.
Signed-off-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There are enough instances of this:
iph->frag_off & htons(IP_MF | IP_OFFSET)
that a helper function is probably warranted.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Remove linux/mm.h inclusion from netdevice.h -- it's unused (I've checked manually).
To prevent mm.h inclusion via other channels also extract "enum dma_data_direction"
definition into separate header. This tiny piece is what gluing netdevice.h with mm.h
via "netdevice.h => dmaengine.h => dma-mapping.h => scatterlist.h => mm.h".
Removal of mm.h from scatterlist.h was tried and was found not feasible
on most archs, so the link was cutoff earlier.
Hope people are OK with tiny include file.
Note, that mm_types.h is still dragged in, but it is a separate story.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: David S. Miller <davem@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>
* remove interrupt.g inclusion from netdevice.h -- not needed
* fixup fallout, add interrupt.h and hardirq.h back where needed.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This interface uses a temporary buffer, but for no real reason.
And now can generate warnings like:
net/sched/sch_generic.c: In function dev_watchdog
net/sched/sch_generic.c:254:10: warning: unused variable drivername
Just return driver->name directly or "".
Reported-by: Connor Hansen <cmdkhh@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since commit eeaeb068f1 (sch_sfq: allow big packets and be fair),
sfq_peek() can return a different skb that would be normally dequeued by
sfq_dequeue() [ if current slot->allot is negative ]
Use generic qdisc_peek_dequeued() instead of custom implementation, to
get consistent result.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Jarek Poplawski <jarkao2@gmail.com>
CC: Patrick McHardy <kaber@trash.net>
CC: Jesper Dangaard Brouer <hawk@diku.dk>
Signed-off-by: David S. Miller <davem@davemloft.net>
While chasing a possible net_sched bug, I found that IP fragments have
litle chance to pass a congestioned SFQ qdisc :
- Say SFQ qdisc is full because one flow is non responsive.
- ip_fragment() wants to send two fragments belonging to an idle flow.
- sfq_enqueue() queues first packet, but see queue limit reached :
- sfq_enqueue() drops one packet from 'big consumer', and returns
NET_XMIT_CN.
- ip_fragment() cancel remaining fragments.
This patch restores fairness, making sure we return NET_XMIT_CN only if
we dropped a packet from the same flow.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Patrick McHardy <kaber@trash.net>
CC: Jarek Poplawski <jarkao2@gmail.com>
CC: Jamal Hadi Salim <hadi@cyberus.ca>
CC: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
dev_deactivate_many() issues one synchronize_rcu() call after qdiscs set
to noop_qdisc.
This call is here to make sure they are no outstanding qdisc-less
dev_queue_xmit calls before returning to caller.
But in dismantle phase, we dont have to wait, because we wont activate
again the device, and we are going to wait one rcu grace period later in
rollback_registered_many().
After this patch, device dismantle uses one synchronize_net() and one
rcu_barrier() call only, so we have a ~30% speedup and a smaller RTNL
latency.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Patrick McHardy <kaber@trash.net>,
CC: Ben Greear <greearb@candelatech.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
IP_ROUTE_CLASSID depends on INET and NET_CLS_ROUTE4 selects
IP_ROUTE_CLASSID, but when INET is not enabled, this kconfig warning
is produced, so fix it by making NET_CLS_ROUTE4 depend on INET.
warning: (NET_CLS_ROUTE4) selects IP_ROUTE_CLASSID which has unmet direct dependencies (NET && INET)
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is no callback of this module maybe queued
since we use kfree_rcu(), we can safely remove the rcu_barrier().
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>
[PATCH 05/17] net,rcu: convert call_rcu(tcf_police_free_rcu) to kfree_rcu()
The rcu callback tcf_police_free_rcu() just calls a kfree(),
so we use kfree_rcu() instead of the call_rcu(tcf_police_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>
The rcu callback tcf_common_free_rcu() just calls a kfree(),
so we use kfree_rcu() instead of the call_rcu(tcf_common_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>
Add const qualifiers to structs iphdr, ipv6hdr and in6_addr pointers
where possible, to make code intention more obvious.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is an implementation of the Quick Fair Queue scheduler developed
by Fabio Checconi. The same algorithm is already implemented in ipfw
in FreeBSD. Fabio had an earlier version developed on Linux, I just
cleaned it up. Thanks to Eric Dumazet for testing this under load.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The only necessary parts are the src/dst addresses, the
interface indexes, the TOS, and the mark.
The rest is unnecessary bloat, which amounts to nearly
50 bytes on 64-bit.
Signed-off-by: David S. Miller <davem@davemloft.net>
Because of various alignements [SLUB / qdisc], we use 512 bytes of
memory for one {p|b}fifo qdisc, instead of 256 bytes on 64bit arches and
192 bytes on 32bit ones.
Move the "u32 limit" inside "struct Qdisc" (no impact on other qdiscs)
Change qdisc_alloc(), first trying a regular allocation before an
oversized one.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Better document choke skb->cb[] use, like we did in netem and sfb
This adds a compile time check to make sure we dont exhaust skb->cb[]
space.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Get rid of debug message that are not useful, and enable
the log messages in case of error.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is a patch originated with Stefano Salsano and Fabio Ludovici.
It provides several alternative loss models for use with netem.
This patch adds two state machine based loss models.
See: http://netgroup.uniroma2.it/twiki/bin/view.cgi/Main/NetemCLG
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Many users have wanted the old functionality that was lost
to be able to use pfifo as inner qdisc for netem. The reason that
netem could not be classful with the older API was because of the
limitations of the old dequeue/requeue interface; now that qdisc API has
a peek function, there is no longer a problem with using any
inner qdisc's.
This reverts commit 0220146411.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Rather than magic constant in code, expose the maximum size of
packet distribution table in API. In iproute2, q_netem defines
MAX_DIST as 16K already.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The netem probability table can be large (up to 64K bytes)
which may be too large to allocate in one contiguous chunk.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Use nla_put_nested to update netlink attribute value.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
* make qdisc_ops local
* add sparse annotation about expected unlock/unlock in dump_class_stats
* fix indentation
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This is the Stochastic Fair Blue scheduler, based on work from :
W. Feng, D. Kandlur, D. Saha, K. Shin. Blue: A New Class of Active Queue
Management Algorithms. U. Michigan CSE-TR-387-99, April 1999.
http://www.thefengs.com/wuchang/blue/CSE-TR-387-99.pdf
This implementation is based on work done by Juliusz Chroboczek
General SFB algorithm can be found in figure 14, page 15:
B[l][n] : L x N array of bins (L levels, N bins per level)
enqueue()
Calculate hash function values h{0}, h{1}, .. h{L-1}
Update bins at each level
for i = 0 to L - 1
if (B[i][h{i}].qlen > bin_size)
B[i][h{i}].p_mark += p_increment;
else if (B[i][h{i}].qlen == 0)
B[i][h{i}].p_mark -= p_decrement;
p_min = min(B[0][h{0}].p_mark ... B[L-1][h{L-1}].p_mark);
if (p_min == 1.0)
ratelimit();
else
mark/drop with probabilty p_min;
I did the adaptation of Juliusz code to meet current kernel standards,
and various changes to address previous comments :
http://thread.gmane.org/gmane.linux.network/90225http://thread.gmane.org/gmane.linux.network/90375
Default flow classifier is the rxhash introduced by RPS in 2.6.35, but
we can use an external flow classifier if wanted.
tc qdisc add dev $DEV parent 1:11 handle 11: \
est 0.5sec 2sec sfb limit 128
tc filter add dev $DEV protocol ip parent 11: handle 3 \
flow hash keys dst divisor 1024
Notes:
1) SFB default child qdisc is pfifo_fast. It can be changed by another
qdisc but a child qdisc MUST not drop a packet previously queued. This
is because SFB needs to handle a dequeued packet in order to maintain
its virtual queue states. pfifo_head_drop or CHOKe should not be used.
2) ECN is enabled by default, unlike RED/CHOKe/GRED
With help from Patrick McHardy & Andi Kleen
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Juliusz Chroboczek <Juliusz.Chroboczek@pps.jussieu.fr>
CC: Stephen Hemminger <shemminger@vyatta.com>
CC: Patrick McHardy <kaber@trash.net>
CC: Andi Kleen <andi@firstfloor.org>
CC: John W. Linville <linville@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The variable _data is used in asm-generic to define sections
which causes sparse warnings, so just rename the variable.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
From: Eric W. Biederman <ebiederm@xmission.com>
In the beginning with batching unreg_list was a list that was used only
once in the lifetime of a network device (I think). Now we have calls
using the unreg_list that can happen multiple times in the life of a
network device like dev_deactivate and dev_close that are also using the
unreg_list. In addition in unregister_netdevice_queue we also do a
list_move because for devices like veth pairs it is possible that
unregister_netdevice_queue will be called multiple times.
So I think the change below to fix dev_deactivate which Eric D. missed
will fix this problem. Now to go test that.
Signed-off-by: David S. Miller <davem@davemloft.net>
All the cleanup code in mqprio_destroy() is currently conditional on
priv->qdiscs being non-null, but that condition should only apply to
the per-queue qdisc cleanup. We should always set the number of
traffic classes back to 0 here.
Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
CHOKe ("CHOose and Kill" or "CHOose and Keep") is an alternative
packet scheduler based on the Random Exponential Drop (RED) algorithm.
The core idea is:
For every packet arrival:
Calculate Qave
if (Qave < minth)
Queue the new packet
else
Select randomly a packet from the queue
if (both packets from same flow)
then Drop both the packets
else if (Qave > maxth)
Drop packet
else
Admit packet with proability p (same as RED)
See also:
Rong Pan, Balaji Prabhakar, Konstantinos Psounis, "CHOKe: a stateless active
queue management scheme for approximating fair bandwidth allocation",
Proceeding of INFOCOM'2000, March 2000.
Help from:
Eric Dumazet <eric.dumazet@gmail.com>
Patrick McHardy <kaber@trash.net>
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The change to allow divisor to be a parameter (in 2.6.38-rc1)
commit 817fb15dfd
introduced a possible deadlock caught by sparse.
The scheduler tree lock was left locked in the case of an incorrect
divisor value. Simplest fix is to move test outside of lock
which also solves problem of partial update.
Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
mqprio_dump() should make sure all fields of struct tc_mqprio_qopt are
initialized.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now qdisc stab is handled before TCQ_F_CAN_BYPASS test in
__dev_xmit_skb(), we can generalize TCQ_F_CAN_BYPASS to other qdiscs
than pfifo_fast : pfifo, bfifo, pfifo_head_drop and sfq
SFQ is special because it can have external classifiers, and in these
cases, we cannot bypass queue discipline (packet could be dropped by
classifier) without admin asking it, or further changes.
Its worth doing this, especially for SFQ, avoiding dirtying memory in
case no packets are already waiting in queue.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 44b8288308 (net_sched: pfifo_head_drop problem), we fixed
a problem with pfifo_head drops that incorrectly decreased
sch->bstats.bytes and sch->bstats.packets
Several qdiscs (CHOKe, SFQ, pfifo_head, ...) are able to drop a
previously enqueued packet, and bstats cannot be changed, so
bstats/rates are not accurate (over estimated)
This patch changes the qdisc_bstats updates to be done at dequeue() time
instead of enqueue() time. bstats counters no longer account for dropped
frames, and rates are more correct, since enqueue() bursts dont have
effect on dequeue() rate.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch converts stab qdisc management to RCU, so that we can perform
the qdisc_calculate_pkt_len() call before getting qdisc lock.
This shortens the lock's held time in __dev_xmit_skb().
This permits more qdiscs to get TCQ_F_CAN_BYPASS status, avoiding lot of
cache misses and so reducing latencies.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Patrick McHardy <kaber@trash.net>
CC: Jesper Dangaard Brouer <hawk@diku.dk>
CC: Jarek Poplawski <jarkao2@gmail.com>
CC: Jamal Hadi Salim <hadi@cyberus.ca>
CC: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In commit 3711210576 (net: QDISC_STATE_RUNNING dont need atomic bit
ops) I moved QDISC_STATE_RUNNING flag to __state container, located in
the cache line containing qdisc lock and often dirtied fields.
I now move TCQ_F_THROTTLED bit too, so that we let first cache line read
mostly, and shared by all cpus. This should speedup HTB/CBQ for example.
Not using test_bit()/__clear_bit()/__test_and_set_bit allows to use an
"unsigned int" for __state container, reducing by 8 bytes Qdisc size.
Introduce helpers to hide implementation details.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Patrick McHardy <kaber@trash.net>
CC: Jesper Dangaard Brouer <hawk@diku.dk>
CC: Jarek Poplawski <jarkao2@gmail.com>
CC: Jamal Hadi Salim <hadi@cyberus.ca>
CC: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
SFQ currently uses a 1024 slots hash table, and its internal structure
(sfq_sched_data) allocation needs order-1 page on x86_64
Allow tc command to specify a divisor value (hash table size), between 1
and 65536.
If no value is provided, assume the 1024 default size.
This allows admins to setup smaller (or bigger) SFQ for specific needs.
This also brings back sfq_sched_data allocations to order-0 ones, saving
3KB per SFQ qdisc.
Jesper uses ~55.000 SFQ in one machine, this patch should free 165 MB of
memory.
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Patrick McHardy <kaber@trash.net>
CC: Jesper Dangaard Brouer <hawk@diku.dk>
CC: Jarek Poplawski <jarkao2@gmail.com>
CC: Jamal Hadi Salim <hadi@cyberus.ca>
CC: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Cleanup net/sched code to current CodingStyle and practices.
Reduce inline abuse
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This implements a mqprio queueing discipline that by default creates
a pfifo_fast qdisc per tx queue and provides the needed configuration
interface.
Using the mqprio qdisc the number of tcs currently in use along
with the range of queues alloted to each class can be configured. By
default skbs are mapped to traffic classes using the skb priority.
This mapping is configurable.
Configurable parameters,
struct tc_mqprio_qopt {
__u8 num_tc;
__u8 prio_tc_map[TC_BITMASK + 1];
__u8 hw;
__u16 count[TC_MAX_QUEUE];
__u16 offset[TC_MAX_QUEUE];
};
Here the count/offset pairing give the queue alignment and the
prio_tc_map gives the mapping from skb->priority to tc.
The hw bit determines if the hardware should configure the count
and offset values. If the hardware bit is set then the operation
will fail if the hardware does not implement the ndo_setup_tc
operation. This is to avoid undetermined states where the hardware
may or may not control the queue mapping. Also minimal bounds
checking is done on the count/offset to verify a queue does not
exceed num_tx_queues and that queue ranges do not overlap. Otherwise
it is left to user policy or hardware configuration to create
useful mappings.
It is expected that hardware QOS schemes can be implemented by
creating appropriate mappings of queues in ndo_tc_setup().
One expected use case is drivers will use the ndo_setup_tc to map
queue ranges onto 802.1Q traffic classes. This provides a generic
mechanism to map network traffic onto these traffic classes and
removes the need for lower layer drivers to know specifics about
traffic types.
Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet