net/sched/sch_mq.c

   1 /*
   2  * net/sched/sch_mq.c           Classful multiqueue dummy scheduler
   3  *
   4  * Copyright (c) 2009 Patrick McHardy <kaber@trash.net>
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License
   8  * version 2 as published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/types.h>
  12 #include <linux/slab.h>
  13 #include <linux/kernel.h>
  14 #include <linux/string.h>
  15 #include <linux/errno.h>
  16 #include <linux/skbuff.h>
  17 #include <net/netlink.h>
  18 #include <net/pkt_sched.h>
  19
  20 struct mq_sched {
  21         struct Qdisc            **qdiscs;
  22 };
  23
  24 static void mq_destroy(struct Qdisc *sch)
  25 {
  26         struct net_device *dev = qdisc_dev(sch);
  27         struct mq_sched *priv = qdisc_priv(sch);
  28         unsigned int ntx;
  29
  30         if (!priv->qdiscs)
  31                 return;
  32         for (ntx = 0; ntx < dev->num_tx_queues && priv->qdiscs[ntx]; ntx++)
  33                 qdisc_destroy(priv->qdiscs[ntx]);
  34         kfree(priv->qdiscs);
  35 }
  36
  37 static int mq_init(struct Qdisc *sch, struct nlattr *opt)
  38 {
  39         struct net_device *dev = qdisc_dev(sch);
  40         struct mq_sched *priv = qdisc_priv(sch);
  41         struct netdev_queue *dev_queue;
  42         struct Qdisc *qdisc;
  43         unsigned int ntx;
  44
  45         if (sch->parent != TC_H_ROOT)
  46                 return -EOPNOTSUPP;
  47
  48         if (!netif_is_multiqueue(dev))
  49                 return -EOPNOTSUPP;
  50
  51         /* pre-allocate qdiscs, attachment can't fail */
  52         priv->qdiscs = kcalloc(dev->num_tx_queues, sizeof(priv->qdiscs[0]),
  53                                GFP_KERNEL);
  54         if (priv->qdiscs == NULL)
  55                 return -ENOMEM;
  56
  57         for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
  58                 dev_queue = netdev_get_tx_queue(dev, ntx);
  59                 qdisc = qdisc_create_dflt(dev_queue, &pfifo_fast_ops,
  60                                           TC_H_MAKE(TC_H_MAJ(sch->handle),
  61                                                     TC_H_MIN(ntx + 1)));
  62                 if (qdisc == NULL)
  63                         goto err;
  64                 qdisc->flags |= TCQ_F_CAN_BYPASS;
  65                 priv->qdiscs[ntx] = qdisc;
  66         }
  67
  68         sch->flags |= TCQ_F_MQROOT;
  69         return 0;
  70
  71 err:
  72         mq_destroy(sch);
  73         return -ENOMEM;
  74 }
  75
  76 static void mq_attach(struct Qdisc *sch)
  77 {
  78         struct net_device *dev = qdisc_dev(sch);
  79         struct mq_sched *priv = qdisc_priv(sch);
  80         struct Qdisc *qdisc;
  81         unsigned int ntx;
  82
  83         for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
  84                 qdisc = priv->qdiscs[ntx];
  85                 qdisc = dev_graft_qdisc(qdisc->dev_queue, qdisc);
  86                 if (qdisc)
  87                         qdisc_destroy(qdisc);
  88         }
  89         kfree(priv->qdiscs);
  90         priv->qdiscs = NULL;
  91 }
  92
  93 static int mq_dump(struct Qdisc *sch, struct sk_buff *skb)
  94 {
  95         struct net_device *dev = qdisc_dev(sch);
  96         struct Qdisc *qdisc;
  97         unsigned int ntx;
  98
  99         sch->q.qlen = 0;
 100         memset(&sch->bstats, 0, sizeof(sch->bstats));
 101         memset(&sch->qstats, 0, sizeof(sch->qstats));
 102
 103         for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
 104                 qdisc = netdev_get_tx_queue(dev, ntx)->qdisc_sleeping;
 105                 spin_lock_bh(qdisc_lock(qdisc));
 106                 sch->q.qlen             += qdisc->q.qlen;
 107                 sch->bstats.bytes       += qdisc->bstats.bytes;
 108                 sch->bstats.packets     += qdisc->bstats.packets;
 109                 sch->qstats.qlen        += qdisc->qstats.qlen;
 110                 sch->qstats.backlog     += qdisc->qstats.backlog;
 111                 sch->qstats.drops       += qdisc->qstats.drops;
 112                 sch->qstats.requeues    += qdisc->qstats.requeues;
 113                 sch->qstats.overlimits  += qdisc->qstats.overlimits;
 114                 spin_unlock_bh(qdisc_lock(qdisc));
 115         }
 116         return 0;
 117 }
 118
 119 static struct netdev_queue *mq_queue_get(struct Qdisc *sch, unsigned long cl)
 120 {
 121         struct net_device *dev = qdisc_dev(sch);
 122         unsigned long ntx = cl - 1;
 123
 124         if (ntx >= dev->num_tx_queues)
 125                 return NULL;
 126         return netdev_get_tx_queue(dev, ntx);
 127 }
 128
 129 static struct netdev_queue *mq_select_queue(struct Qdisc *sch,
 130                                             struct tcmsg *tcm)
 131 {
 132         unsigned int ntx = TC_H_MIN(tcm->tcm_parent);
 133         struct netdev_queue *dev_queue = mq_queue_get(sch, ntx);
 134
 135         if (!dev_queue) {
 136                 struct net_device *dev = qdisc_dev(sch);
 137
 138                 return netdev_get_tx_queue(dev, 0);
 139         }
 140         return dev_queue;
 141 }
 142
 143 static int mq_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
 144                     struct Qdisc **old)
 145 {
 146         struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
 147         struct net_device *dev = qdisc_dev(sch);
 148
 149         if (dev->flags & IFF_UP)
 150                 dev_deactivate(dev);
 151
 152         *old = dev_graft_qdisc(dev_queue, new);
 153
 154         if (dev->flags & IFF_UP)
 155                 dev_activate(dev);
 156         return 0;
 157 }
 158
 159 static struct Qdisc *mq_leaf(struct Qdisc *sch, unsigned long cl)
 160 {
 161         struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
 162
 163         return dev_queue->qdisc_sleeping;
 164 }
 165
 166 static unsigned long mq_get(struct Qdisc *sch, u32 classid)
 167 {
 168         unsigned int ntx = TC_H_MIN(classid);
 169
 170         if (!mq_queue_get(sch, ntx))
 171                 return 0;
 172         return ntx;
 173 }
 174
 175 static void mq_put(struct Qdisc *sch, unsigned long cl)
 176 {
 177 }
 178
 179 static int mq_dump_class(struct Qdisc *sch, unsigned long cl,
 180                          struct sk_buff *skb, struct tcmsg *tcm)
 181 {
 182         struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
 183
 184         tcm->tcm_parent = TC_H_ROOT;
 185         tcm->tcm_handle |= TC_H_MIN(cl);
 186         tcm->tcm_info = dev_queue->qdisc_sleeping->handle;
 187         return 0;
 188 }
 189
 190 static int mq_dump_class_stats(struct Qdisc *sch, unsigned long cl,
 191                                struct gnet_dump *d)
 192 {
 193         struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
 194
 195         sch = dev_queue->qdisc_sleeping;
 196         sch->qstats.qlen = sch->q.qlen;
 197         if (gnet_stats_copy_basic(d, &sch->bstats) < 0 ||
 198             gnet_stats_copy_queue(d, &sch->qstats) < 0)
 199                 return -1;
 200         return 0;
 201 }
 202
 203 static void mq_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 204 {
 205         struct net_device *dev = qdisc_dev(sch);
 206         unsigned int ntx;
 207
 208         if (arg->stop)
 209                 return;
 210
 211         arg->count = arg->skip;
 212         for (ntx = arg->skip; ntx < dev->num_tx_queues; ntx++) {
 213                 if (arg->fn(sch, ntx + 1, arg) < 0) {
 214                         arg->stop = 1;
 215                         break;
 216                 }
 217                 arg->count++;
 218         }
 219 }
 220
 221 static const struct Qdisc_class_ops mq_class_ops = {
 222         .select_queue   = mq_select_queue,
 223         .graft          = mq_graft,
 224         .leaf           = mq_leaf,
 225         .get            = mq_get,
 226         .put            = mq_put,
 227         .walk           = mq_walk,
 228         .dump           = mq_dump_class,
 229         .dump_stats     = mq_dump_class_stats,
 230 };
 231
 232 struct Qdisc_ops mq_qdisc_ops __read_mostly = {
 233         .cl_ops         = &mq_class_ops,
 234         .id             = "mq",
 235         .priv_size      = sizeof(struct mq_sched),
 236         .init           = mq_init,
 237         .destroy        = mq_destroy,
 238         .attach         = mq_attach,
 239         .dump           = mq_dump,
 240         .owner          = THIS_MODULE,
 241 };