net: sched: add Flow Queue PIE packet scheduler
authorMohit P. Tahiliani <tahiliani@nitk.edu.in>
Wed, 22 Jan 2020 18:22:33 +0000 (23:52 +0530)
committerDavid S. Miller <davem@davemloft.net>
Thu, 23 Jan 2020 10:38:31 +0000 (11:38 +0100)
Principles:
  - Packets are classified on flows.
  - This is a Stochastic model (as we use a hash, several flows might
                                be hashed to the same slot)
  - Each flow has a PIE managed queue.
  - Flows are linked onto two (Round Robin) lists,
    so that new flows have priority on old ones.
  - For a given flow, packets are not reordered.
  - Drops during enqueue only.
  - ECN capability is off by default.
  - ECN threshold (if ECN is enabled) is at 10% by default.
  - Uses timestamps to calculate queue delay by default.

Usage:
tc qdisc ... fq_pie [ limit PACKETS ] [ flows NUMBER ]
                    [ target TIME ] [ tupdate TIME ]
                    [ alpha NUMBER ] [ beta NUMBER ]
                    [ quantum BYTES ] [ memory_limit BYTES ]
                    [ ecnprob PERCENTAGE ] [ [no]ecn ]
                    [ [no]bytemode ] [ [no_]dq_rate_estimator ]

defaults:
  limit: 10240 packets, flows: 1024
  target: 15 ms, tupdate: 15 ms (in jiffies)
  alpha: 1/8, beta : 5/4
  quantum: device MTU, memory_limit: 32 Mb
  ecnprob: 10%, ecn: off
  bytemode: off, dq_rate_estimator: off

Signed-off-by: Mohit P. Tahiliani <tahiliani@nitk.edu.in>
Signed-off-by: Sachin D. Patil <sdp.sachin@gmail.com>
Signed-off-by: V. Saicharan <vsaicharan1998@gmail.com>
Signed-off-by: Mohit Bhasi <mohitbhasi1998@gmail.com>
Signed-off-by: Leslie Monis <lesliemonis@gmail.com>
Signed-off-by: Gautam Ramakrishnan <gautamramk@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/pie.h
include/uapi/linux/pkt_sched.h
net/sched/Kconfig
net/sched/Makefile
net/sched/sch_fq_pie.c [new file with mode: 0644]

index 90f5db3d29e7cfe316d8af414de9e99ebe62ac61..fd5a37cb799325f658098811b74ba88c622e37ed 100644 (file)
@@ -81,9 +81,11 @@ struct pie_stats {
 /**
  * struct pie_skb_cb - contains private skb vars
  * @enqueue_time:      timestamp when the packet is enqueued
+ * @mem_usage:         size of the skb during enqueue
  */
 struct pie_skb_cb {
        psched_time_t enqueue_time;
+       u32 mem_usage;
 };
 
 static inline void pie_params_init(struct pie_params *params)
index bf5a5b1dfb0b03fcf410cef38b0e106ebdc90ad7..bbe791b241682fafddfac2d40aaad2bc789eebe8 100644 (file)
@@ -971,6 +971,37 @@ struct tc_pie_xstats {
        __u32 ecn_mark;                 /* packets marked with ecn*/
 };
 
+/* FQ PIE */
+enum {
+       TCA_FQ_PIE_UNSPEC,
+       TCA_FQ_PIE_LIMIT,
+       TCA_FQ_PIE_FLOWS,
+       TCA_FQ_PIE_TARGET,
+       TCA_FQ_PIE_TUPDATE,
+       TCA_FQ_PIE_ALPHA,
+       TCA_FQ_PIE_BETA,
+       TCA_FQ_PIE_QUANTUM,
+       TCA_FQ_PIE_MEMORY_LIMIT,
+       TCA_FQ_PIE_ECN_PROB,
+       TCA_FQ_PIE_ECN,
+       TCA_FQ_PIE_BYTEMODE,
+       TCA_FQ_PIE_DQ_RATE_ESTIMATOR,
+       __TCA_FQ_PIE_MAX
+};
+#define TCA_FQ_PIE_MAX   (__TCA_FQ_PIE_MAX - 1)
+
+struct tc_fq_pie_xstats {
+       __u32 packets_in;       /* total number of packets enqueued */
+       __u32 dropped;          /* packets dropped due to fq_pie_action */
+       __u32 overlimit;        /* dropped due to lack of space in queue */
+       __u32 overmemory;       /* dropped due to lack of memory in queue */
+       __u32 ecn_mark;         /* packets marked with ecn */
+       __u32 new_flow_count;   /* count of new flows created by packets */
+       __u32 new_flows_len;    /* count of flows in new list */
+       __u32 old_flows_len;    /* count of flows in old list */
+       __u32 memory_usage;     /* total memory across all queues */
+};
+
 /* CBS */
 struct tc_cbs_qopt {
        __u8 offload;
index b1e7ec7269582ce6930ff2cedc1bed7fa7867bc5..edde0e5194385a8e2ac5504d9f3ac9720ddcf1e8 100644 (file)
@@ -366,6 +366,19 @@ config NET_SCH_PIE
 
          If unsure, say N.
 
+config NET_SCH_FQ_PIE
+       depends on NET_SCH_PIE
+       tristate "Flow Queue Proportional Integral controller Enhanced (FQ-PIE)"
+       help
+         Say Y here if you want to use the Flow Queue Proportional Integral
+         controller Enhanced (FQ-PIE) packet scheduling algorithm.
+         For more information, please see https://tools.ietf.org/html/rfc8033
+
+         To compile this driver as a module, choose M here: the module
+         will be called sch_fq_pie.
+
+         If unsure, say N.
+
 config NET_SCH_INGRESS
        tristate "Ingress/classifier-action Qdisc"
        depends on NET_CLS_ACT
index bc8856b865ff5541c061c17adcf4d022c81c0a49..31c367a6cd093ba9e8fff379a80ac76d0248e0f8 100644 (file)
@@ -59,6 +59,7 @@ obj-$(CONFIG_NET_SCH_CAKE)    += sch_cake.o
 obj-$(CONFIG_NET_SCH_FQ)       += sch_fq.o
 obj-$(CONFIG_NET_SCH_HHF)      += sch_hhf.o
 obj-$(CONFIG_NET_SCH_PIE)      += sch_pie.o
+obj-$(CONFIG_NET_SCH_FQ_PIE)   += sch_fq_pie.o
 obj-$(CONFIG_NET_SCH_CBS)      += sch_cbs.o
 obj-$(CONFIG_NET_SCH_ETF)      += sch_etf.o
 obj-$(CONFIG_NET_SCH_TAPRIO)   += sch_taprio.o
diff --git a/net/sched/sch_fq_pie.c b/net/sched/sch_fq_pie.c
new file mode 100644 (file)
index 0000000..bbd0dea
--- /dev/null
@@ -0,0 +1,562 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Flow Queue PIE discipline
+ *
+ * Copyright (C) 2019 Mohit P. Tahiliani <tahiliani@nitk.edu.in>
+ * Copyright (C) 2019 Sachin D. Patil <sdp.sachin@gmail.com>
+ * Copyright (C) 2019 V. Saicharan <vsaicharan1998@gmail.com>
+ * Copyright (C) 2019 Mohit Bhasi <mohitbhasi1998@gmail.com>
+ * Copyright (C) 2019 Leslie Monis <lesliemonis@gmail.com>
+ * Copyright (C) 2019 Gautam Ramakrishnan <gautamramk@gmail.com>
+ */
+
+#include <linux/jhash.h>
+#include <linux/sizes.h>
+#include <linux/vmalloc.h>
+#include <net/pkt_cls.h>
+#include <net/pie.h>
+
+/* Flow Queue PIE
+ *
+ * Principles:
+ *   - Packets are classified on flows.
+ *   - This is a Stochastic model (as we use a hash, several flows might
+ *                                 be hashed to the same slot)
+ *   - Each flow has a PIE managed queue.
+ *   - Flows are linked onto two (Round Robin) lists,
+ *     so that new flows have priority on old ones.
+ *   - For a given flow, packets are not reordered.
+ *   - Drops during enqueue only.
+ *   - ECN capability is off by default.
+ *   - ECN threshold (if ECN is enabled) is at 10% by default.
+ *   - Uses timestamps to calculate queue delay by default.
+ */
+
+/**
+ * struct fq_pie_flow - contains data for each flow
+ * @vars:      pie vars associated with the flow
+ * @deficit:   number of remaining byte credits
+ * @backlog:   size of data in the flow
+ * @qlen:      number of packets in the flow
+ * @flowchain: flowchain for the flow
+ * @head:      first packet in the flow
+ * @tail:      last packet in the flow
+ */
+struct fq_pie_flow {
+       struct pie_vars vars;
+       s32 deficit;
+       u32 backlog;
+       u32 qlen;
+       struct list_head flowchain;
+       struct sk_buff *head;
+       struct sk_buff *tail;
+};
+
+struct fq_pie_sched_data {
+       struct tcf_proto __rcu *filter_list; /* optional external classifier */
+       struct tcf_block *block;
+       struct fq_pie_flow *flows;
+       struct Qdisc *sch;
+       struct list_head old_flows;
+       struct list_head new_flows;
+       struct pie_params p_params;
+       u32 ecn_prob;
+       u32 flows_cnt;
+       u32 quantum;
+       u32 memory_limit;
+       u32 new_flow_count;
+       u32 memory_usage;
+       u32 overmemory;
+       struct pie_stats stats;
+       struct timer_list adapt_timer;
+};
+
+static unsigned int fq_pie_hash(const struct fq_pie_sched_data *q,
+                               struct sk_buff *skb)
+{
+       return reciprocal_scale(skb_get_hash(skb), q->flows_cnt);
+}
+
+static unsigned int fq_pie_classify(struct sk_buff *skb, struct Qdisc *sch,
+                                   int *qerr)
+{
+       struct fq_pie_sched_data *q = qdisc_priv(sch);
+       struct tcf_proto *filter;
+       struct tcf_result res;
+       int result;
+
+       if (TC_H_MAJ(skb->priority) == sch->handle &&
+           TC_H_MIN(skb->priority) > 0 &&
+           TC_H_MIN(skb->priority) <= q->flows_cnt)
+               return TC_H_MIN(skb->priority);
+
+       filter = rcu_dereference_bh(q->filter_list);
+       if (!filter)
+               return fq_pie_hash(q, skb) + 1;
+
+       *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
+       result = tcf_classify(skb, filter, &res, false);
+       if (result >= 0) {
+#ifdef CONFIG_NET_CLS_ACT
+               switch (result) {
+               case TC_ACT_STOLEN:
+               case TC_ACT_QUEUED:
+               case TC_ACT_TRAP:
+                       *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
+                       /* fall through */
+               case TC_ACT_SHOT:
+                       return 0;
+               }
+#endif
+               if (TC_H_MIN(res.classid) <= q->flows_cnt)
+                       return TC_H_MIN(res.classid);
+       }
+       return 0;
+}
+
+/* add skb to flow queue (tail add) */
+static inline void flow_queue_add(struct fq_pie_flow *flow,
+                                 struct sk_buff *skb)
+{
+       if (!flow->head)
+               flow->head = skb;
+       else
+               flow->tail->next = skb;
+       flow->tail = skb;
+       skb->next = NULL;
+}
+
+static int fq_pie_qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch,
+                               struct sk_buff **to_free)
+{
+       struct fq_pie_sched_data *q = qdisc_priv(sch);
+       struct fq_pie_flow *sel_flow;
+       int uninitialized_var(ret);
+       u8 memory_limited = false;
+       u8 enqueue = false;
+       u32 pkt_len;
+       u32 idx;
+
+       /* Classifies packet into corresponding flow */
+       idx = fq_pie_classify(skb, sch, &ret);
+       sel_flow = &q->flows[idx];
+
+       /* Checks whether adding a new packet would exceed memory limit */
+       get_pie_cb(skb)->mem_usage = skb->truesize;
+       memory_limited = q->memory_usage > q->memory_limit + skb->truesize;
+
+       /* Checks if the qdisc is full */
+       if (unlikely(qdisc_qlen(sch) >= sch->limit)) {
+               q->stats.overlimit++;
+               goto out;
+       } else if (unlikely(memory_limited)) {
+               q->overmemory++;
+       }
+
+       if (!pie_drop_early(sch, &q->p_params, &sel_flow->vars,
+                           sel_flow->backlog, skb->len)) {
+               enqueue = true;
+       } else if (q->p_params.ecn &&
+                  sel_flow->vars.prob <= (MAX_PROB / 100) * q->ecn_prob &&
+                  INET_ECN_set_ce(skb)) {
+               /* If packet is ecn capable, mark it if drop probability
+                * is lower than the parameter ecn_prob, else drop it.
+                */
+               q->stats.ecn_mark++;
+               enqueue = true;
+       }
+       if (enqueue) {
+               /* Set enqueue time only when dq_rate_estimator is disabled. */
+               if (!q->p_params.dq_rate_estimator)
+                       pie_set_enqueue_time(skb);
+
+               pkt_len = qdisc_pkt_len(skb);
+               q->stats.packets_in++;
+               q->memory_usage += skb->truesize;
+               sch->qstats.backlog += pkt_len;
+               sch->q.qlen++;
+               flow_queue_add(sel_flow, skb);
+               if (list_empty(&sel_flow->flowchain)) {
+                       list_add_tail(&sel_flow->flowchain, &q->new_flows);
+                       q->new_flow_count++;
+                       sel_flow->deficit = q->quantum;
+                       sel_flow->qlen = 0;
+                       sel_flow->backlog = 0;
+               }
+               sel_flow->qlen++;
+               sel_flow->backlog += pkt_len;
+               return NET_XMIT_SUCCESS;
+       }
+out:
+       q->stats.dropped++;
+       sel_flow->vars.accu_prob = 0;
+       sel_flow->vars.accu_prob_overflows = 0;
+       __qdisc_drop(skb, to_free);
+       qdisc_qstats_drop(sch);
+       return NET_XMIT_CN;
+}
+
+static const struct nla_policy fq_pie_policy[TCA_FQ_PIE_MAX + 1] = {
+       [TCA_FQ_PIE_LIMIT]              = {.type = NLA_U32},
+       [TCA_FQ_PIE_FLOWS]              = {.type = NLA_U32},
+       [TCA_FQ_PIE_TARGET]             = {.type = NLA_U32},
+       [TCA_FQ_PIE_TUPDATE]            = {.type = NLA_U32},
+       [TCA_FQ_PIE_ALPHA]              = {.type = NLA_U32},
+       [TCA_FQ_PIE_BETA]               = {.type = NLA_U32},
+       [TCA_FQ_PIE_QUANTUM]            = {.type = NLA_U32},
+       [TCA_FQ_PIE_MEMORY_LIMIT]       = {.type = NLA_U32},
+       [TCA_FQ_PIE_ECN_PROB]           = {.type = NLA_U32},
+       [TCA_FQ_PIE_ECN]                = {.type = NLA_U32},
+       [TCA_FQ_PIE_BYTEMODE]           = {.type = NLA_U32},
+       [TCA_FQ_PIE_DQ_RATE_ESTIMATOR]  = {.type = NLA_U32},
+};
+
+static inline struct sk_buff *dequeue_head(struct fq_pie_flow *flow)
+{
+       struct sk_buff *skb = flow->head;
+
+       flow->head = skb->next;
+       skb->next = NULL;
+       return skb;
+}
+
+static struct sk_buff *fq_pie_qdisc_dequeue(struct Qdisc *sch)
+{
+       struct fq_pie_sched_data *q = qdisc_priv(sch);
+       struct sk_buff *skb = NULL;
+       struct fq_pie_flow *flow;
+       struct list_head *head;
+       u32 pkt_len;
+
+begin:
+       head = &q->new_flows;
+       if (list_empty(head)) {
+               head = &q->old_flows;
+               if (list_empty(head))
+                       return NULL;
+       }
+
+       flow = list_first_entry(head, struct fq_pie_flow, flowchain);
+       /* Flow has exhausted all its credits */
+       if (flow->deficit <= 0) {
+               flow->deficit += q->quantum;
+               list_move_tail(&flow->flowchain, &q->old_flows);
+               goto begin;
+       }
+
+       if (flow->head) {
+               skb = dequeue_head(flow);
+               pkt_len = qdisc_pkt_len(skb);
+               sch->qstats.backlog -= pkt_len;
+               sch->q.qlen--;
+               qdisc_bstats_update(sch, skb);
+       }
+
+       if (!skb) {
+               /* force a pass through old_flows to prevent starvation */
+               if (head == &q->new_flows && !list_empty(&q->old_flows))
+                       list_move_tail(&flow->flowchain, &q->old_flows);
+               else
+                       list_del_init(&flow->flowchain);
+               goto begin;
+       }
+
+       flow->qlen--;
+       flow->deficit -= pkt_len;
+       flow->backlog -= pkt_len;
+       q->memory_usage -= get_pie_cb(skb)->mem_usage;
+       pie_process_dequeue(skb, &q->p_params, &flow->vars, flow->backlog);
+       return skb;
+}
+
+static int fq_pie_change(struct Qdisc *sch, struct nlattr *opt,
+                        struct netlink_ext_ack *extack)
+{
+       struct fq_pie_sched_data *q = qdisc_priv(sch);
+       struct nlattr *tb[TCA_FQ_PIE_MAX + 1];
+       unsigned int len_dropped = 0;
+       unsigned int num_dropped = 0;
+       int err;
+
+       if (!opt)
+               return -EINVAL;
+
+       err = nla_parse_nested(tb, TCA_FQ_PIE_MAX, opt, fq_pie_policy, extack);
+       if (err < 0)
+               return err;
+
+       sch_tree_lock(sch);
+       if (tb[TCA_FQ_PIE_LIMIT]) {
+               u32 limit = nla_get_u32(tb[TCA_FQ_PIE_LIMIT]);
+
+               q->p_params.limit = limit;
+               sch->limit = limit;
+       }
+       if (tb[TCA_FQ_PIE_FLOWS]) {
+               if (q->flows) {
+                       NL_SET_ERR_MSG_MOD(extack,
+                                          "Number of flows cannot be changed");
+                       goto flow_error;
+               }
+               q->flows_cnt = nla_get_u32(tb[TCA_FQ_PIE_FLOWS]);
+               if (!q->flows_cnt || q->flows_cnt > 65536) {
+                       NL_SET_ERR_MSG_MOD(extack,
+                                          "Number of flows must be < 65536");
+                       goto flow_error;
+               }
+       }
+
+       /* convert from microseconds to pschedtime */
+       if (tb[TCA_FQ_PIE_TARGET]) {
+               /* target is in us */
+               u32 target = nla_get_u32(tb[TCA_FQ_PIE_TARGET]);
+
+               /* convert to pschedtime */
+               q->p_params.target =
+                       PSCHED_NS2TICKS((u64)target * NSEC_PER_USEC);
+       }
+
+       /* tupdate is in jiffies */
+       if (tb[TCA_FQ_PIE_TUPDATE])
+               q->p_params.tupdate =
+                       usecs_to_jiffies(nla_get_u32(tb[TCA_FQ_PIE_TUPDATE]));
+
+       if (tb[TCA_FQ_PIE_ALPHA])
+               q->p_params.alpha = nla_get_u32(tb[TCA_FQ_PIE_ALPHA]);
+
+       if (tb[TCA_FQ_PIE_BETA])
+               q->p_params.beta = nla_get_u32(tb[TCA_FQ_PIE_BETA]);
+
+       if (tb[TCA_FQ_PIE_QUANTUM])
+               q->quantum = nla_get_u32(tb[TCA_FQ_PIE_QUANTUM]);
+
+       if (tb[TCA_FQ_PIE_MEMORY_LIMIT])
+               q->memory_limit = nla_get_u32(tb[TCA_FQ_PIE_MEMORY_LIMIT]);
+
+       if (tb[TCA_FQ_PIE_ECN_PROB])
+               q->ecn_prob = nla_get_u32(tb[TCA_FQ_PIE_ECN_PROB]);
+
+       if (tb[TCA_FQ_PIE_ECN])
+               q->p_params.ecn = nla_get_u32(tb[TCA_FQ_PIE_ECN]);
+
+       if (tb[TCA_FQ_PIE_BYTEMODE])
+               q->p_params.bytemode = nla_get_u32(tb[TCA_FQ_PIE_BYTEMODE]);
+
+       if (tb[TCA_FQ_PIE_DQ_RATE_ESTIMATOR])
+               q->p_params.dq_rate_estimator =
+                       nla_get_u32(tb[TCA_FQ_PIE_DQ_RATE_ESTIMATOR]);
+
+       /* Drop excess packets if new limit is lower */
+       while (sch->q.qlen > sch->limit) {
+               struct sk_buff *skb = fq_pie_qdisc_dequeue(sch);
+
+               kfree_skb(skb);
+               len_dropped += qdisc_pkt_len(skb);
+               num_dropped += 1;
+       }
+       qdisc_tree_reduce_backlog(sch, num_dropped, len_dropped);
+
+       sch_tree_unlock(sch);
+       return 0;
+
+flow_error:
+       sch_tree_unlock(sch);
+       return -EINVAL;
+}
+
+static void fq_pie_timer(struct timer_list *t)
+{
+       struct fq_pie_sched_data *q = from_timer(q, t, adapt_timer);
+       struct Qdisc *sch = q->sch;
+       spinlock_t *root_lock; /* to lock qdisc for probability calculations */
+       u16 idx;
+
+       root_lock = qdisc_lock(qdisc_root_sleeping(sch));
+       spin_lock(root_lock);
+
+       for (idx = 0; idx < q->flows_cnt; idx++)
+               pie_calculate_probability(&q->p_params, &q->flows[idx].vars,
+                                         q->flows[idx].backlog);
+
+       /* reset the timer to fire after 'tupdate' jiffies. */
+       if (q->p_params.tupdate)
+               mod_timer(&q->adapt_timer, jiffies + q->p_params.tupdate);
+
+       spin_unlock(root_lock);
+}
+
+static int fq_pie_init(struct Qdisc *sch, struct nlattr *opt,
+                      struct netlink_ext_ack *extack)
+{
+       struct fq_pie_sched_data *q = qdisc_priv(sch);
+       int err;
+       u16 idx;
+
+       pie_params_init(&q->p_params);
+       sch->limit = 10 * 1024;
+       q->p_params.limit = sch->limit;
+       q->quantum = psched_mtu(qdisc_dev(sch));
+       q->sch = sch;
+       q->ecn_prob = 10;
+       q->flows_cnt = 1024;
+       q->memory_limit = SZ_32M;
+
+       INIT_LIST_HEAD(&q->new_flows);
+       INIT_LIST_HEAD(&q->old_flows);
+
+       if (opt) {
+               err = fq_pie_change(sch, opt, extack);
+
+               if (err)
+                       return err;
+       }
+
+       err = tcf_block_get(&q->block, &q->filter_list, sch, extack);
+       if (err)
+               goto init_failure;
+
+       q->flows = kvcalloc(q->flows_cnt, sizeof(struct fq_pie_flow),
+                           GFP_KERNEL);
+       if (!q->flows) {
+               err = -ENOMEM;
+               goto init_failure;
+       }
+       for (idx = 0; idx < q->flows_cnt; idx++) {
+               struct fq_pie_flow *flow = q->flows + idx;
+
+               INIT_LIST_HEAD(&flow->flowchain);
+               pie_vars_init(&flow->vars);
+       }
+
+       timer_setup(&q->adapt_timer, fq_pie_timer, 0);
+       mod_timer(&q->adapt_timer, jiffies + HZ / 2);
+
+       return 0;
+
+init_failure:
+       q->flows_cnt = 0;
+
+       return err;
+}
+
+static int fq_pie_dump(struct Qdisc *sch, struct sk_buff *skb)
+{
+       struct fq_pie_sched_data *q = qdisc_priv(sch);
+       struct nlattr *opts;
+
+       opts = nla_nest_start(skb, TCA_OPTIONS);
+       if (!opts)
+               return -EMSGSIZE;
+
+       /* convert target from pschedtime to us */
+       if (nla_put_u32(skb, TCA_FQ_PIE_LIMIT, sch->limit) ||
+           nla_put_u32(skb, TCA_FQ_PIE_FLOWS, q->flows_cnt) ||
+           nla_put_u32(skb, TCA_FQ_PIE_TARGET,
+                       ((u32)PSCHED_TICKS2NS(q->p_params.target)) /
+                       NSEC_PER_USEC) ||
+           nla_put_u32(skb, TCA_FQ_PIE_TUPDATE,
+                       jiffies_to_usecs(q->p_params.tupdate)) ||
+           nla_put_u32(skb, TCA_FQ_PIE_ALPHA, q->p_params.alpha) ||
+           nla_put_u32(skb, TCA_FQ_PIE_BETA, q->p_params.beta) ||
+           nla_put_u32(skb, TCA_FQ_PIE_QUANTUM, q->quantum) ||
+           nla_put_u32(skb, TCA_FQ_PIE_MEMORY_LIMIT, q->memory_limit) ||
+           nla_put_u32(skb, TCA_FQ_PIE_ECN_PROB, q->ecn_prob) ||
+           nla_put_u32(skb, TCA_FQ_PIE_ECN, q->p_params.ecn) ||
+           nla_put_u32(skb, TCA_FQ_PIE_BYTEMODE, q->p_params.bytemode) ||
+           nla_put_u32(skb, TCA_FQ_PIE_DQ_RATE_ESTIMATOR,
+                       q->p_params.dq_rate_estimator))
+               goto nla_put_failure;
+
+       return nla_nest_end(skb, opts);
+
+nla_put_failure:
+       nla_nest_cancel(skb, opts);
+       return -EMSGSIZE;
+}
+
+static int fq_pie_dump_stats(struct Qdisc *sch, struct gnet_dump *d)
+{
+       struct fq_pie_sched_data *q = qdisc_priv(sch);
+       struct tc_fq_pie_xstats st = {
+               .packets_in     = q->stats.packets_in,
+               .overlimit      = q->stats.overlimit,
+               .overmemory     = q->overmemory,
+               .dropped        = q->stats.dropped,
+               .ecn_mark       = q->stats.ecn_mark,
+               .new_flow_count = q->new_flow_count,
+               .memory_usage   = q->memory_usage,
+       };
+       struct list_head *pos;
+
+       sch_tree_lock(sch);
+       list_for_each(pos, &q->new_flows)
+               st.new_flows_len++;
+
+       list_for_each(pos, &q->old_flows)
+               st.old_flows_len++;
+       sch_tree_unlock(sch);
+
+       return gnet_stats_copy_app(d, &st, sizeof(st));
+}
+
+static void fq_pie_reset(struct Qdisc *sch)
+{
+       struct fq_pie_sched_data *q = qdisc_priv(sch);
+       u16 idx;
+
+       INIT_LIST_HEAD(&q->new_flows);
+       INIT_LIST_HEAD(&q->old_flows);
+       for (idx = 0; idx < q->flows_cnt; idx++) {
+               struct fq_pie_flow *flow = q->flows + idx;
+
+               /* Removes all packets from flow */
+               rtnl_kfree_skbs(flow->head, flow->tail);
+               flow->head = NULL;
+
+               INIT_LIST_HEAD(&flow->flowchain);
+               pie_vars_init(&flow->vars);
+       }
+
+       sch->q.qlen = 0;
+       sch->qstats.backlog = 0;
+}
+
+static void fq_pie_destroy(struct Qdisc *sch)
+{
+       struct fq_pie_sched_data *q = qdisc_priv(sch);
+
+       tcf_block_put(q->block);
+       del_timer_sync(&q->adapt_timer);
+       kvfree(q->flows);
+}
+
+static struct Qdisc_ops fq_pie_qdisc_ops __read_mostly = {
+       .id             = "fq_pie",
+       .priv_size      = sizeof(struct fq_pie_sched_data),
+       .enqueue        = fq_pie_qdisc_enqueue,
+       .dequeue        = fq_pie_qdisc_dequeue,
+       .peek           = qdisc_peek_dequeued,
+       .init           = fq_pie_init,
+       .destroy        = fq_pie_destroy,
+       .reset          = fq_pie_reset,
+       .change         = fq_pie_change,
+       .dump           = fq_pie_dump,
+       .dump_stats     = fq_pie_dump_stats,
+       .owner          = THIS_MODULE,
+};
+
+static int __init fq_pie_module_init(void)
+{
+       return register_qdisc(&fq_pie_qdisc_ops);
+}
+
+static void __exit fq_pie_module_exit(void)
+{
+       unregister_qdisc(&fq_pie_qdisc_ops);
+}
+
+module_init(fq_pie_module_init);
+module_exit(fq_pie_module_exit);
+
+MODULE_DESCRIPTION("Flow Queue Proportional Integral controller Enhanced (FQ-PIE)");
+MODULE_AUTHOR("Mohit P. Tahiliani");
+MODULE_LICENSE("GPL");