Release 4.11 net/sched/sch_fq.c

/*
 * net/sched/sch_fq.c Fair Queue Packet Scheduler (per flow pacing)
 *
 *  Copyright (C) 2013-2015 Eric Dumazet <edumazet@google.com>
 *
 *      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.
 *
 *  Meant to be mostly used for locally generated traffic :
 *  Fast classification depends on skb->sk being set before reaching us.
 *  If not, (router workload), we use rxhash as fallback, with 32 bits wide hash.
 *  All packets belonging to a socket are considered as a 'flow'.
 *
 *  Flows are dynamically allocated and stored in a hash table of RB trees
 *  They are also part of one Round Robin 'queues' (new or old flows)
 *
 *  Burst avoidance (aka pacing) capability :
 *
 *  Transport (eg TCP) can set in sk->sk_pacing_rate a rate, enqueue a
 *  bunch of packets, and this packet scheduler adds delay between
 *  packets to respect rate limitation.
 *
 *  enqueue() :
 *   - lookup one RB tree (out of 1024 or more) to find the flow.
 *     If non existent flow, create it, add it to the tree.
 *     Add skb to the per flow list of skb (fifo).
 *   - Use a special fifo for high prio packets
 *
 *  dequeue() : serves flows in Round Robin
 *  Note : When a flow becomes empty, we do not immediately remove it from
 *  rb trees, for performance reasons (its expected to send additional packets,
 *  or SLAB cache will reuse socket for another flow)
 */

#include <linux/module.h>
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/jiffies.h>
#include <linux/string.h>
#include <linux/in.h>
#include <linux/errno.h>
#include <linux/init.h>
#include <linux/skbuff.h>
#include <linux/slab.h>
#include <linux/rbtree.h>
#include <linux/hash.h>
#include <linux/prefetch.h>
#include <linux/vmalloc.h>
#include <net/netlink.h>
#include <net/pkt_sched.h>
#include <net/sock.h>
#include <net/tcp_states.h>
#include <net/tcp.h>

/*
 * Per flow structure, dynamically allocated
 */

struct fq_flow {
	
struct sk_buff	*head;		/* list of skbs for this flow : first skb */
	union {
		
struct sk_buff *tail;	/* last skb in the list */
		
unsigned long  age;	/* jiffies when flow was emptied, for gc */
	};
	
struct rb_node	fq_node;	/* anchor in fq_root[] trees */
	
struct sock	*sk;
	
int		qlen;		/* number of packets in flow queue */
	
int		credit;
	
u32		socket_hash;	/* sk_hash */
	
struct fq_flow *next;		/* next pointer in RR lists, or &detached */

	
struct rb_node  rate_node;	/* anchor in q->delayed tree */
	
u64		time_next_packet;
};


struct fq_flow_head {
	
struct fq_flow *first;
	
struct fq_flow *last;
};


struct fq_sched_data {
	
struct fq_flow_head new_flows;

	
struct fq_flow_head old_flows;

	
struct rb_root	delayed;	/* for rate limited flows */
	
u64		time_next_delayed_flow;
	
unsigned long	unthrottle_latency_ns;

	
struct fq_flow	internal;	/* for non classified or high prio packets */
	
u32		quantum;
	
u32		initial_quantum;
	
u32		flow_refill_delay;
	
u32		flow_max_rate;	/* optional max rate per flow */
	
u32		flow_plimit;	/* max packets per flow */
	
u32		orphan_mask;	/* mask for orphaned skb */
	
u32		low_rate_threshold;
	
struct rb_root	*fq_root;
	
u8		rate_enable;
	
u8		fq_trees_log;

	
u32		flows;
	
u32		inactive_flows;
	
u32		throttled_flows;

	
u64		stat_gc_flows;
	
u64		stat_internal_packets;
	
u64		stat_tcp_retrans;
	
u64		stat_throttled;
	
u64		stat_flows_plimit;
	
u64		stat_pkts_too_long;
	
u64		stat_allocation_errors;
	
struct qdisc_watchdog watchdog;
};

/* special value to mark a detached flow (not on old/new list) */


static struct fq_flow detached, throttled;



static void fq_flow_set_detached(struct fq_flow *f)
{
	f->next = &detached;
	f->age = jiffies;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
24
100.00%
2
100.00%
Total
24
100.00%
2
100.00%



static bool fq_flow_is_detached(const struct fq_flow *f)
{
	return f->next == &detached;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
20
100.00%
1
100.00%
Total
20
100.00%
1
100.00%



static void fq_flow_set_throttled(struct fq_sched_data *q, struct fq_flow *f)
{
	struct rb_node **p = &q->delayed.rb_node, *parent = NULL;

	while (*p) {
		struct fq_flow *aux;

		parent = *p;
		aux = rb_entry(parent, struct fq_flow, rate_node);
		if (f->time_next_packet >= aux->time_next_packet)
			p = &parent->rb_right;
		else
			p = &parent->rb_left;
	}
	rb_link_node(&f->rate_node, parent, p);
	rb_insert_color(&f->rate_node, &q->delayed);
	q->throttled_flows++;
	q->stat_throttled++;

	f->next = &throttled;
	if (q->time_next_delayed_flow > f->time_next_packet)
		q->time_next_delayed_flow = f->time_next_packet;
}


Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
147
99.32%
1
50.00%
Geliang Tang
1
0.68%
1
50.00%
Total
148
100.00%
2
100.00%


static struct kmem_cache *fq_flow_cachep __read_mostly;



static void fq_flow_add_tail(struct fq_flow_head *head, struct fq_flow *flow)
{
	if (head->first)
		head->last->next = flow;
	else
		head->first = flow;
	head->last = flow;
	flow->next = NULL;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
49
100.00%
1
100.00%
Total
49
100.00%
1
100.00%

/* limit number of collected flows per round */

#define FQ_GC_MAX 8

#define FQ_GC_AGE (3*HZ)



static bool fq_gc_candidate(const struct fq_flow *f)
{
	return fq_flow_is_detached(f) &&
	       time_after(jiffies, f->age + FQ_GC_AGE);
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
29
100.00%
1
100.00%
Total
29
100.00%
1
100.00%



static void fq_gc(struct fq_sched_data *q,
		  struct rb_root *root,
		  struct sock *sk)
{
	struct fq_flow *f, *tofree[FQ_GC_MAX];
	struct rb_node **p, *parent;
	int fcnt = 0;

	p = &root->rb_node;
	parent = NULL;
	while (*p) {
		parent = *p;

		f = rb_entry(parent, struct fq_flow, fq_node);
		if (f->sk == sk)
			break;

		if (fq_gc_candidate(f)) {
			tofree[fcnt++] = f;
			if (fcnt == FQ_GC_MAX)
				break;
		}

		if (f->sk > sk)
			p = &parent->rb_right;
		else
			p = &parent->rb_left;
	}

	q->flows -= fcnt;
	q->inactive_flows -= fcnt;
	q->stat_gc_flows += fcnt;
	while (fcnt) {
		struct fq_flow *f = tofree[--fcnt];

		rb_erase(&f->fq_node, root);
		kmem_cache_free(fq_flow_cachep, f);
	}
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
188
99.47%
1
50.00%
Geliang Tang
1
0.53%
1
50.00%
Total
189
100.00%
2
100.00%



static struct fq_flow *fq_classify(struct sk_buff *skb, struct fq_sched_data *q)
{
	struct rb_node **p, *parent;
	struct sock *sk = skb->sk;
	struct rb_root *root;
	struct fq_flow *f;

	/* warning: no starvation prevention... */
	if (unlikely((skb->priority & TC_PRIO_MAX) == TC_PRIO_CONTROL))
		return &q->internal;

	/* SYNACK messages are attached to a TCP_NEW_SYN_RECV request socket
         * or a listener (SYNCOOKIE mode)
         * 1) request sockets are not full blown,
         *    they do not contain sk_pacing_rate
         * 2) They are not part of a 'flow' yet
         * 3) We do not want to rate limit them (eg SYNFLOOD attack),
         *    especially if the listener set SO_MAX_PACING_RATE
         * 4) We pretend they are orphaned
         */
	if (!sk || sk_listener(sk)) {
		unsigned long hash = skb_get_hash(skb) & q->orphan_mask;

		/* By forcing low order bit to 1, we make sure to not
                 * collide with a local flow (socket pointers are word aligned)
                 */
		sk = (struct sock *)((hash << 1) | 1UL);
		skb_orphan(skb);
	}

	root = &q->fq_root[hash_ptr(sk, q->fq_trees_log)];

	if (q->flows >= (2U << q->fq_trees_log) &&
	    q->inactive_flows > q->flows/2)
		fq_gc(q, root, sk);

	p = &root->rb_node;
	parent = NULL;
	while (*p) {
		parent = *p;

		f = rb_entry(parent, struct fq_flow, fq_node);
		if (f->sk == sk) {
			/* socket might have been reallocated, so check
                         * if its sk_hash is the same.
                         * It not, we need to refill credit with
                         * initial quantum
                         */
			if (unlikely(skb->sk &&
				     f->socket_hash != sk->sk_hash)) {
				f->credit = q->initial_quantum;
				f->socket_hash = sk->sk_hash;
				f->time_next_packet = 0ULL;
			}
			return f;
		}
		if (f->sk > sk)
			p = &parent->rb_right;
		else
			p = &parent->rb_left;
	}

	f = kmem_cache_zalloc(fq_flow_cachep, GFP_ATOMIC | __GFP_NOWARN);
	if (unlikely(!f)) {
		q->stat_allocation_errors++;
		return &q->internal;
	}
	fq_flow_set_detached(f);
	f->sk = sk;
	if (skb->sk)
		f->socket_hash = sk->sk_hash;
	f->credit = q->initial_quantum;

	rb_link_node(&f->fq_node, parent, p);
	rb_insert_color(&f->fq_node, root);

	q->flows++;
	q->inactive_flows++;
	return f;
}


Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
372
97.89%
5
71.43%
Maciej Żenczykowski
7
1.84%
1
14.29%
Geliang Tang
1
0.26%
1
14.29%
Total
380
100.00%
7
100.00%

/* remove one skb from head of flow queue */


static struct sk_buff *fq_dequeue_head(struct Qdisc *sch, struct fq_flow *flow)
{
	struct sk_buff *skb = flow->head;

	if (skb) {
		flow->head = skb->next;
		skb->next = NULL;
		flow->qlen--;
		qdisc_qstats_backlog_dec(sch, skb);
		sch->q.qlen--;
	}
	return skb;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
66
95.65%
2
66.67%
John Fastabend
3
4.35%
1
33.33%
Total
69
100.00%
3
100.00%

/* We might add in the future detection of retransmits
 * For the time being, just return false
 */


static bool skb_is_retransmit(struct sk_buff *skb)
{
	return false;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
14
100.00%
1
100.00%
Total
14
100.00%
1
100.00%

/* add skb to flow queue
 * flow queue is a linked list, kind of FIFO, except for TCP retransmits
 * We special case tcp retransmits to be transmitted before other packets.
 * We rely on fact that TCP retransmits are unlikely, so we do not waste
 * a separate queue or a pointer.
 * head->  [retrans pkt 1]
 *         [retrans pkt 2]
 *         [ normal pkt 1]
 *         [ normal pkt 2]
 *         [ normal pkt 3]
 * tail->  [ normal pkt 4]
 */


static void flow_queue_add(struct fq_flow *flow, struct sk_buff *skb)
{
	struct sk_buff *prev, *head = flow->head;

	skb->next = NULL;
	if (!head) {
		flow->head = skb;
		flow->tail = skb;
		return;
	}
	if (likely(!skb_is_retransmit(skb))) {
		flow->tail->next = skb;
		flow->tail = skb;
		return;
	}

	/* This skb is a tcp retransmit,
         * find the last retrans packet in the queue
         */
	prev = NULL;
	while (skb_is_retransmit(head)) {
		prev = head;
		head = head->next;
		if (!head)
			break;
	}
	if (!prev) { /* no rtx packet in queue, become the new head */
		skb->next = flow->head;
		flow->head = skb;
	} else {
		if (prev == flow->tail)
			flow->tail = skb;
		else
			skb->next = prev->next;
		prev->next = skb;
	}
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
166
100.00%
1
100.00%
Total
166
100.00%
1
100.00%



static int fq_enqueue(struct sk_buff *skb, struct Qdisc *sch,
		      struct sk_buff **to_free)
{
	struct fq_sched_data *q = qdisc_priv(sch);
	struct fq_flow *f;

	if (unlikely(sch->q.qlen >= sch->limit))
		return qdisc_drop(skb, sch, to_free);

	f = fq_classify(skb, q);
	if (unlikely(f->qlen >= q->flow_plimit && f != &q->internal)) {
		q->stat_flows_plimit++;
		return qdisc_drop(skb, sch, to_free);
	}

	f->qlen++;
	if (skb_is_retransmit(skb))
		q->stat_tcp_retrans++;
	qdisc_qstats_backlog_inc(sch, skb);
	if (fq_flow_is_detached(f)) {
		fq_flow_add_tail(&q->new_flows, f);
		if (time_after(jiffies, f->age + q->flow_refill_delay))
			f->credit = max_t(u32, f->credit, q->quantum);
		q->inactive_flows--;
	}

	/* Note: this overwrites f->age */
	flow_queue_add(f, skb);

	if (unlikely(f == &q->internal)) {
		q->stat_internal_packets++;
	}
	sch->q.qlen++;

	return NET_XMIT_SUCCESS;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
222
98.67%
3
75.00%
John Fastabend
3
1.33%
1
25.00%
Total
225
100.00%
4
100.00%



static void fq_check_throttled(struct fq_sched_data *q, u64 now)
{
	unsigned long sample;
	struct rb_node *p;

	if (q->time_next_delayed_flow > now)
		return;

	/* Update unthrottle latency EWMA.
         * This is cheap and can help diagnosing timer/latency problems.
         */
	sample = (unsigned long)(now - q->time_next_delayed_flow);
	q->unthrottle_latency_ns -= q->unthrottle_latency_ns >> 3;
	q->unthrottle_latency_ns += sample >> 3;

	q->time_next_delayed_flow = ~0ULL;
	while ((p = rb_first(&q->delayed)) != NULL) {
		struct fq_flow *f = rb_entry(p, struct fq_flow, rate_node);

		if (f->time_next_packet > now) {
			q->time_next_delayed_flow = f->time_next_packet;
			break;
		}
		rb_erase(p, &q->delayed);
		q->throttled_flows--;
		fq_flow_add_tail(&q->old_flows, f);
	}
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
148
99.33%
2
66.67%
Geliang Tang
1
0.67%
1
33.33%
Total
149
100.00%
3
100.00%



static struct sk_buff *fq_dequeue(struct Qdisc *sch)
{
	struct fq_sched_data *q = qdisc_priv(sch);
	u64 now = ktime_get_ns();
	struct fq_flow_head *head;
	struct sk_buff *skb;
	struct fq_flow *f;
	u32 rate, plen;

	skb = fq_dequeue_head(sch, &q->internal);
	if (skb)
		goto out;
	fq_check_throttled(q, now);
begin:
	head = &q->new_flows;
	if (!head->first) {
		head = &q->old_flows;
		if (!head->first) {
			if (q->time_next_delayed_flow != ~0ULL)
				qdisc_watchdog_schedule_ns(&q->watchdog,
							   q->time_next_delayed_flow);
			return NULL;
		}
	}
	f = head->first;

	if (f->credit <= 0) {
		f->credit += q->quantum;
		head->first = f->next;
		fq_flow_add_tail(&q->old_flows, f);
		goto begin;
	}

	skb = f->head;
	if (unlikely(skb && now < f->time_next_packet &&
		     !skb_is_tcp_pure_ack(skb))) {
		head->first = f->next;
		fq_flow_set_throttled(q, f);
		goto begin;
	}

	skb = fq_dequeue_head(sch, f);
	if (!skb) {
		head->first = f->next;
		/* force a pass through old_flows to prevent starvation */
		if ((head == &q->new_flows) && q->old_flows.first) {
			fq_flow_add_tail(&q->old_flows, f);
		} else {
			fq_flow_set_detached(f);
			q->inactive_flows++;
		}
		goto begin;
	}
	prefetch(&skb->end);
	f->credit -= qdisc_pkt_len(skb);

	if (!q->rate_enable)
		goto out;

	/* Do not pace locally generated ack packets */
	if (skb_is_tcp_pure_ack(skb))
		goto out;

	rate = q->flow_max_rate;
	if (skb->sk)
		rate = min(skb->sk->sk_pacing_rate, rate);

	if (rate <= q->low_rate_threshold) {
		f->credit = 0;
		plen = qdisc_pkt_len(skb);
	} else {
		plen = max(qdisc_pkt_len(skb), q->quantum);
		if (f->credit > 0)
			goto out;
	}
	if (rate != ~0U) {
		u64 len = (u64)plen * NSEC_PER_SEC;

		if (likely(rate))
			do_div(len, rate);
		/* Since socket rate can change later,
                 * clamp the delay to 1 second.
                 * Really, providers of too big packets should be fixed !
                 */
		if (unlikely(len > NSEC_PER_SEC)) {
			len = NSEC_PER_SEC;
			q->stat_pkts_too_long++;
		}
		/* Account for schedule/timers drifts.
                 * f->time_next_packet was set when prior packet was sent,
                 * and current time (@now) can be too late by tens of us.
                 */
		if (f->time_next_packet)
			len -= min(len/2, now - f->time_next_packet);
		f->time_next_packet = now + len;
	}
out:
	qdisc_bstats_update(sch, skb);
	return skb;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
503
100.00%
10
100.00%
Total
503
100.00%
10
100.00%



static void fq_flow_purge(struct fq_flow *flow)
{
	rtnl_kfree_skbs(flow->head, flow->tail);
	flow->head = NULL;
	flow->qlen = 0;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
34
100.00%
1
100.00%
Total
34
100.00%
1
100.00%



static void fq_reset(struct Qdisc *sch)
{
	struct fq_sched_data *q = qdisc_priv(sch);
	struct rb_root *root;
	struct rb_node *p;
	struct fq_flow *f;
	unsigned int idx;

	sch->q.qlen = 0;
	sch->qstats.backlog = 0;

	fq_flow_purge(&q->internal);

	if (!q->fq_root)
		return;

	for (idx = 0; idx < (1U << q->fq_trees_log); idx++) {
		root = &q->fq_root[idx];
		while ((p = rb_first(root)) != NULL) {
			f = rb_entry(p, struct fq_flow, fq_node);
			rb_erase(p, root);

			fq_flow_purge(f);

			kmem_cache_free(fq_flow_cachep, f);
		}
	}
	q->new_flows.first	= NULL;
	q->old_flows.first	= NULL;
	q->delayed		= RB_ROOT;
	q->flows		= 0;
	q->inactive_flows	= 0;
	q->throttled_flows	= 0;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
188
99.47%
3
75.00%
Geliang Tang
1
0.53%
1
25.00%
Total
189
100.00%
4
100.00%



static void fq_rehash(struct fq_sched_data *q,
		      struct rb_root *old_array, u32 old_log,
		      struct rb_root *new_array, u32 new_log)
{
	struct rb_node *op, **np, *parent;
	struct rb_root *oroot, *nroot;
	struct fq_flow *of, *nf;
	int fcnt = 0;
	u32 idx;

	for (idx = 0; idx < (1U << old_log); idx++) {
		oroot = &old_array[idx];
		while ((op = rb_first(oroot)) != NULL) {
			rb_erase(op, oroot);
			of = rb_entry(op, struct fq_flow, fq_node);
			if (fq_gc_candidate(of)) {
				fcnt++;
				kmem_cache_free(fq_flow_cachep, of);
				continue;
			}
			nroot = &new_array[hash_ptr(of->sk, new_log)];

			np = &nroot->rb_node;
			parent = NULL;
			while (*np) {
				parent = *np;

				nf = rb_entry(parent, struct fq_flow, fq_node);
				BUG_ON(nf->sk == of->sk);

				if (nf->sk > of->sk)
					np = &parent->rb_right;
				else
					np = &parent->rb_left;
			}

			rb_link_node(&of->fq_node, parent, np);
			rb_insert_color(&of->fq_node, nroot);
		}
	}
	q->flows -= fcnt;
	q->inactive_flows -= fcnt;
	q->stat_gc_flows += fcnt;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
268
99.26%
2
66.67%
Geliang Tang
2
0.74%
1
33.33%
Total
270
100.00%
3
100.00%



static void *fq_alloc_node(size_t sz, int node)
{
	void *ptr;

	ptr = kmalloc_node(sz, GFP_KERNEL | __GFP_REPEAT | __GFP_NOWARN, node);
	if (!ptr)
		ptr = vmalloc_node(sz, node);
	return ptr;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
49
100.00%
1
100.00%
Total
49
100.00%
1
100.00%



static void fq_free(void *addr)
{
	kvfree(addr);
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
14
93.33%
1
50.00%
Américo Wang
1
6.67%
1
50.00%
Total
15
100.00%
2
100.00%



static int fq_resize(struct Qdisc *sch, u32 log)
{
	struct fq_sched_data *q = qdisc_priv(sch);
	struct rb_root *array;
	void *old_fq_root;
	u32 idx;

	if (q->fq_root && log == q->fq_trees_log)
		return 0;

	/* If XPS was setup, we can allocate memory on right NUMA node */
	array = fq_alloc_node(sizeof(struct rb_root) << log,
			      netdev_queue_numa_node_read(sch->dev_queue));
	if (!array)
		return -ENOMEM;

	for (idx = 0; idx < (1U << log); idx++)
		array[idx] = RB_ROOT;

	sch_tree_lock(sch);

	old_fq_root = q->fq_root;
	if (old_fq_root)
		fq_rehash(q, old_fq_root, q->fq_trees_log, array, log);

	q->fq_root = array;
	q->fq_trees_log = log;

	sch_tree_unlock(sch);

	fq_free(old_fq_root);

	return 0;
}

Contributors
Person
Tokens
Prop
Commits
CommitProp
Eric Dumazet
160
100.00%
3
100.00%
Total
160
100.00%
3
100.00%


static const struct nla_policy fq_policy[TCA_FQ_MAX + 1] = {
	[TCA_FQ_PLIMIT]			= { .type = NLA_U32 },
	[TCA_FQ_FLOW_PLIMIT]		= { .type = NLA_U32 },
	[TCA_FQ_QUANTUM]		= { .type = NLA_U32 },
	[TCA_FQ_INITIAL_QUANTUM]	= { .type = NLA_U32 },
	[TCA_FQ_RATE_ENABLE]		= { .type = NLA_U32 },
	[TCA_FQ_FLOW_DEFAULT_RATE]	= { .type = NLA_U32 },
	[TCA_FQ_FLOW_MAX_RATE]		= { .type = NLA_U32 },
	[TCA_FQ_BUCKETS_LOG]		= { .type = NLA_U32 },
	[TCA_FQ_FLOW_REFILL_DELAY]	= { .type = NLA_U32