linux/net/sched/sch_plug.c
Eric Dumazet 520ac30f45 net_sched: drop packets after root qdisc lock is released
Qdisc performance suffers when packets are dropped at enqueue()
time because drops (kfree_skb()) are done while qdisc lock is held,
delaying a dequeue() draining the queue.

Nominal throughput can be reduced by 50 % when this happens,
at a time we would like the dequeue() to proceed as fast as possible.

Even FQ is vulnerable to this problem, while one of FQ goals was
to provide some flow isolation.

This patch adds a 'struct sk_buff **to_free' parameter to all
qdisc->enqueue(), and in qdisc_drop() helper.

I measured a performance increase of up to 12 %, but this patch
is a prereq so that future batches in enqueue() can fly.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
2016-06-25 12:19:35 -04:00

234 lines
6.6 KiB
C

/*
* sch_plug.c Queue traffic until an explicit release command
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version
* 2 of the License, or (at your option) any later version.
*
* There are two ways to use this qdisc:
* 1. A simple "instantaneous" plug/unplug operation, by issuing an alternating
* sequence of TCQ_PLUG_BUFFER & TCQ_PLUG_RELEASE_INDEFINITE commands.
*
* 2. For network output buffering (a.k.a output commit) functionality.
* Output commit property is commonly used by applications using checkpoint
* based fault-tolerance to ensure that the checkpoint from which a system
* is being restored is consistent w.r.t outside world.
*
* Consider for e.g. Remus - a Virtual Machine checkpointing system,
* wherein a VM is checkpointed, say every 50ms. The checkpoint is replicated
* asynchronously to the backup host, while the VM continues executing the
* next epoch speculatively.
*
* The following is a typical sequence of output buffer operations:
* 1.At epoch i, start_buffer(i)
* 2. At end of epoch i (i.e. after 50ms):
* 2.1 Stop VM and take checkpoint(i).
* 2.2 start_buffer(i+1) and Resume VM
* 3. While speculatively executing epoch(i+1), asynchronously replicate
* checkpoint(i) to backup host.
* 4. When checkpoint_ack(i) is received from backup, release_buffer(i)
* Thus, this Qdisc would receive the following sequence of commands:
* TCQ_PLUG_BUFFER (epoch i)
* .. TCQ_PLUG_BUFFER (epoch i+1)
* ....TCQ_PLUG_RELEASE_ONE (epoch i)
* ......TCQ_PLUG_BUFFER (epoch i+2)
* ........
*/
#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/errno.h>
#include <linux/netdevice.h>
#include <linux/skbuff.h>
#include <net/pkt_sched.h>
/*
* State of the queue, when used for network output buffering:
*
* plug(i+1) plug(i) head
* ------------------+--------------------+---------------->
* | |
* | |
* pkts_current_epoch| pkts_last_epoch |pkts_to_release
* ----------------->|<--------+--------->|+--------------->
* v v
*
*/
struct plug_sched_data {
/* If true, the dequeue function releases all packets
* from head to end of the queue. The queue turns into
* a pass-through queue for newly arriving packets.
*/
bool unplug_indefinite;
bool throttled;
/* Queue Limit in bytes */
u32 limit;
/* Number of packets (output) from the current speculatively
* executing epoch.
*/
u32 pkts_current_epoch;
/* Number of packets corresponding to the recently finished
* epoch. These will be released when we receive a
* TCQ_PLUG_RELEASE_ONE command. This command is typically
* issued after committing a checkpoint at the target.
*/
u32 pkts_last_epoch;
/*
* Number of packets from the head of the queue, that can
* be released (committed checkpoint).
*/
u32 pkts_to_release;
};
static int plug_enqueue(struct sk_buff *skb, struct Qdisc *sch,
struct sk_buff **to_free)
{
struct plug_sched_data *q = qdisc_priv(sch);
if (likely(sch->qstats.backlog + skb->len <= q->limit)) {
if (!q->unplug_indefinite)
q->pkts_current_epoch++;
return qdisc_enqueue_tail(skb, sch);
}
return qdisc_drop(skb, sch, to_free);
}
static struct sk_buff *plug_dequeue(struct Qdisc *sch)
{
struct plug_sched_data *q = qdisc_priv(sch);
if (q->throttled)
return NULL;
if (!q->unplug_indefinite) {
if (!q->pkts_to_release) {
/* No more packets to dequeue. Block the queue
* and wait for the next release command.
*/
q->throttled = true;
return NULL;
}
q->pkts_to_release--;
}
return qdisc_dequeue_head(sch);
}
static int plug_init(struct Qdisc *sch, struct nlattr *opt)
{
struct plug_sched_data *q = qdisc_priv(sch);
q->pkts_current_epoch = 0;
q->pkts_last_epoch = 0;
q->pkts_to_release = 0;
q->unplug_indefinite = false;
if (opt == NULL) {
q->limit = qdisc_dev(sch)->tx_queue_len
* psched_mtu(qdisc_dev(sch));
} else {
struct tc_plug_qopt *ctl = nla_data(opt);
if (nla_len(opt) < sizeof(*ctl))
return -EINVAL;
q->limit = ctl->limit;
}
q->throttled = true;
return 0;
}
/* Receives 4 types of messages:
* TCQ_PLUG_BUFFER: Inset a plug into the queue and
* buffer any incoming packets
* TCQ_PLUG_RELEASE_ONE: Dequeue packets from queue head
* to beginning of the next plug.
* TCQ_PLUG_RELEASE_INDEFINITE: Dequeue all packets from queue.
* Stop buffering packets until the next TCQ_PLUG_BUFFER
* command is received (just act as a pass-thru queue).
* TCQ_PLUG_LIMIT: Increase/decrease queue size
*/
static int plug_change(struct Qdisc *sch, struct nlattr *opt)
{
struct plug_sched_data *q = qdisc_priv(sch);
struct tc_plug_qopt *msg;
if (opt == NULL)
return -EINVAL;
msg = nla_data(opt);
if (nla_len(opt) < sizeof(*msg))
return -EINVAL;
switch (msg->action) {
case TCQ_PLUG_BUFFER:
/* Save size of the current buffer */
q->pkts_last_epoch = q->pkts_current_epoch;
q->pkts_current_epoch = 0;
if (q->unplug_indefinite)
q->throttled = true;
q->unplug_indefinite = false;
break;
case TCQ_PLUG_RELEASE_ONE:
/* Add packets from the last complete buffer to the
* packets to be released set.
*/
q->pkts_to_release += q->pkts_last_epoch;
q->pkts_last_epoch = 0;
q->throttled = false;
netif_schedule_queue(sch->dev_queue);
break;
case TCQ_PLUG_RELEASE_INDEFINITE:
q->unplug_indefinite = true;
q->pkts_to_release = 0;
q->pkts_last_epoch = 0;
q->pkts_current_epoch = 0;
q->throttled = false;
netif_schedule_queue(sch->dev_queue);
break;
case TCQ_PLUG_LIMIT:
/* Limit is supplied in bytes */
q->limit = msg->limit;
break;
default:
return -EINVAL;
}
return 0;
}
static struct Qdisc_ops plug_qdisc_ops __read_mostly = {
.id = "plug",
.priv_size = sizeof(struct plug_sched_data),
.enqueue = plug_enqueue,
.dequeue = plug_dequeue,
.peek = qdisc_peek_head,
.init = plug_init,
.change = plug_change,
.reset = qdisc_reset_queue,
.owner = THIS_MODULE,
};
static int __init plug_module_init(void)
{
return register_qdisc(&plug_qdisc_ops);
}
static void __exit plug_module_exit(void)
{
unregister_qdisc(&plug_qdisc_ops);
}
module_init(plug_module_init)
module_exit(plug_module_exit)
MODULE_LICENSE("GPL");