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/kernel.h>
  13 #include <linux/string.h>
  14 #include <linux/errno.h>
  15 #include <linux/skbuff.h>
  16 #include <net/netlink.h>
  17 #include <net/pkt_sched.h>
  18
  19 struct mq_sched {
  20         struct Qdisc            **qdiscs;
  21 };
  22
  23 static void mq_destroy(struct Qdisc *sch)
  24 {
  25         struct net_device *dev = qdisc_dev(sch);
  26         struct mq_sched *priv = qdisc_priv(sch);
  27         unsigned int ntx;
  28
  29         if (!priv->qdiscs)
  30                 return;
  31         for (ntx = 0; ntx < dev->num_tx_queues && priv->qdiscs[ntx]; ntx++)
  32                 qdisc_destroy(priv->qdiscs[ntx]);
  33         kfree(priv->qdiscs);
  34 }
  35
  36 static int mq_init(struct Qdisc *sch, struct nlattr *opt)
  37 {
  38         struct net_device *dev = qdisc_dev(sch);
  39         struct mq_sched *priv = qdisc_priv(sch);
  40         struct netdev_queue *dev_queue;
  41         struct Qdisc *qdisc;
  42         unsigned int ntx;
  43
  44         if (sch->parent != TC_H_ROOT)
  45                 return -EOPNOTSUPP;
  46
  47         if (!netif_is_multiqueue(dev))
  48                 return -EOPNOTSUPP;
  49
  50         /* pre-allocate qdiscs, attachment can't fail */
  51         priv->qdiscs = kcalloc(dev->num_tx_queues, sizeof(priv->qdiscs[0]),
  52                                GFP_KERNEL);
  53         if (priv->qdiscs == NULL)
  54                 return -ENOMEM;
  55
  56         for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
  57                 dev_queue = netdev_get_tx_queue(dev, ntx);
  58                 qdisc = qdisc_create_dflt(dev, dev_queue, &pfifo_fast_ops,
  59                                           TC_H_MAKE(TC_H_MAJ(sch->handle),
  60                                                     TC_H_MIN(ntx + 1)));
  61                 if (qdisc == NULL)
  62                         goto err;
  63                 qdisc->flags |= TCQ_F_CAN_BYPASS;
  64                 priv->qdiscs[ntx] = qdisc;
  65         }
  66
  67         sch->flags |= TCQ_F_MQROOT;
  68         return 0;
  69
  70 err:
  71         mq_destroy(sch);
  72         return -ENOMEM;
  73 }
  74
  75 static void mq_attach(struct Qdisc *sch)
  76 {
  77         struct net_device *dev = qdisc_dev(sch);
  78         struct mq_sched *priv = qdisc_priv(sch);
  79         struct Qdisc *qdisc;
  80         unsigned int ntx;
  81
  82         for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
  83                 qdisc = priv->qdiscs[ntx];
  84                 qdisc = dev_graft_qdisc(qdisc->dev_queue, qdisc);
  85                 if (qdisc)
  86                         qdisc_destroy(qdisc);
  87         }
  88         kfree(priv->qdiscs);
  89         priv->qdiscs = NULL;
  90 }
  91
  92 static int mq_dump(struct Qdisc *sch, struct sk_buff *skb)
  93 {
  94         struct net_device *dev = qdisc_dev(sch);
  95         struct Qdisc *qdisc;
  96         unsigned int ntx;
  97
  98         sch->q.qlen = 0;
  99         memset(&sch->bstats, 0, sizeof(sch->bstats));
 100         memset(&sch->qstats, 0, sizeof(sch->qstats));
 101
 102         for (ntx = 0; ntx < dev->num_tx_queues; ntx++) {
 103                 qdisc = netdev_get_tx_queue(dev, ntx)->qdisc_sleeping;
 104                 spin_lock_bh(qdisc_lock(qdisc));
 105                 sch->q.qlen             += qdisc->q.qlen;
 106                 sch->bstats.bytes       += qdisc->bstats.bytes;
 107                 sch->bstats.packets     += qdisc->bstats.packets;
 108                 sch->qstats.qlen        += qdisc->qstats.qlen;
 109                 sch->qstats.backlog     += qdisc->qstats.backlog;
 110                 sch->qstats.drops       += qdisc->qstats.drops;
 111                 sch->qstats.requeues    += qdisc->qstats.requeues;
 112                 sch->qstats.overlimits  += qdisc->qstats.overlimits;
 113                 spin_unlock_bh(qdisc_lock(qdisc));
 114         }
 115         return 0;
 116 }
 117
 118 static struct netdev_queue *mq_queue_get(struct Qdisc *sch, unsigned long cl)
 119 {
 120         struct net_device *dev = qdisc_dev(sch);
 121         unsigned long ntx = cl - 1;
 122
 123         if (ntx >= dev->num_tx_queues)
 124                 return NULL;
 125         return netdev_get_tx_queue(dev, ntx);
 126 }
 127
 128 static struct netdev_queue *mq_select_queue(struct Qdisc *sch,
 129                                             struct tcmsg *tcm)
 130 {
 131         unsigned int ntx = TC_H_MIN(tcm->tcm_parent);
 132         struct netdev_queue *dev_queue = mq_queue_get(sch, ntx);
 133
 134         if (!dev_queue) {
 135                 struct net_device *dev = qdisc_dev(sch);
 136
 137                 return netdev_get_tx_queue(dev, 0);
 138         }
 139         return dev_queue;
 140 }
 141
 142 static int mq_graft(struct Qdisc *sch, unsigned long cl, struct Qdisc *new,
 143                     struct Qdisc **old)
 144 {
 145         struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
 146         struct net_device *dev = qdisc_dev(sch);
 147
 148         if (dev->flags & IFF_UP)
 149                 dev_deactivate(dev);
 150
 151         *old = dev_graft_qdisc(dev_queue, new);
 152
 153         if (dev->flags & IFF_UP)
 154                 dev_activate(dev);
 155         return 0;
 156 }
 157
 158 static struct Qdisc *mq_leaf(struct Qdisc *sch, unsigned long cl)
 159 {
 160         struct netdev_queue *dev_queue = mq_queue_get(sch, cl);
 161
 162         return dev_queue->qdisc_sleeping;
 163 }
 164
 165 static unsigned long mq_get(struct Qdisc *sch, u32 classid)
 166 {
 167         unsigned int ntx = TC_H_MIN(classid);
 168
 169         if (!mq_queue_get(sch, ntx))
 170                 return 0;
 171         return ntx;
 172 }
 173
 174 static void mq_put(struct Qdisc *sch, unsigned long cl)
 175 {
 176         return;
 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 };