net/sched/sch_skbprio.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * net/sched/sch_skbprio.c  SKB Priority Queue.
   4  *
   5  * Authors:     Nishanth Devarajan, <ndev2021@gmail.com>
   6  *              Cody Doucette, <doucette@bu.edu>
   7  *              original idea by Michel Machado, Cody Doucette, and Qiaobin Fu
   8  */
   9
  10 #include <linux/string.h>
  11 #include <linux/module.h>
  12 #include <linux/slab.h>
  13 #include <linux/types.h>
  14 #include <linux/kernel.h>
  15 #include <linux/errno.h>
  16 #include <linux/skbuff.h>
  17 #include <net/pkt_sched.h>
  18 #include <net/sch_generic.h>
  19 #include <net/inet_ecn.h>
  20
  21 /*              SKB Priority Queue
  22  *      =================================
  23  *
  24  * Skbprio (SKB Priority Queue) is a queueing discipline that prioritizes
  25  * packets according to their skb->priority field. Under congestion,
  26  * Skbprio drops already-enqueued lower priority packets to make space
  27  * available for higher priority packets; it was conceived as a solution
  28  * for denial-of-service defenses that need to route packets with different
  29  * priorities as a mean to overcome DoS attacks.
  30  */
  31
  32 struct skbprio_sched_data {
  33         /* Queue state. */
  34         struct sk_buff_head qdiscs[SKBPRIO_MAX_PRIORITY];
  35         struct gnet_stats_queue qstats[SKBPRIO_MAX_PRIORITY];
  36         u16 highest_prio;
  37         u16 lowest_prio;
  38 };
  39
  40 static u16 calc_new_high_prio(const struct skbprio_sched_data *q)
  41 {
  42         int prio;
  43
  44         for (prio = q->highest_prio - 1; prio >= q->lowest_prio; prio--) {
  45                 if (!skb_queue_empty(&q->qdiscs[prio]))
  46                         return prio;
  47         }
  48
  49         /* SKB queue is empty, return 0 (default highest priority setting). */
  50         return 0;
  51 }
  52
  53 static u16 calc_new_low_prio(const struct skbprio_sched_data *q)
  54 {
  55         int prio;
  56
  57         for (prio = q->lowest_prio + 1; prio <= q->highest_prio; prio++) {
  58                 if (!skb_queue_empty(&q->qdiscs[prio]))
  59                         return prio;
  60         }
  61
  62         /* SKB queue is empty, return SKBPRIO_MAX_PRIORITY - 1
  63          * (default lowest priority setting).
  64          */
  65         return SKBPRIO_MAX_PRIORITY - 1;
  66 }
  67
  68 static int skbprio_enqueue(struct sk_buff *skb, struct Qdisc *sch,
  69                           struct sk_buff **to_free)
  70 {
  71         const unsigned int max_priority = SKBPRIO_MAX_PRIORITY - 1;
  72         struct skbprio_sched_data *q = qdisc_priv(sch);
  73         struct sk_buff_head *qdisc;
  74         struct sk_buff_head *lp_qdisc;
  75         struct sk_buff *to_drop;
  76         u16 prio, lp;
  77
  78         /* Obtain the priority of @skb. */
  79         prio = min(skb->priority, max_priority);
  80
  81         qdisc = &q->qdiscs[prio];
  82         if (sch->q.qlen < sch->limit) {
  83                 __skb_queue_tail(qdisc, skb);
  84                 qdisc_qstats_backlog_inc(sch, skb);
  85                 q->qstats[prio].backlog += qdisc_pkt_len(skb);
  86
  87                 /* Check to update highest and lowest priorities. */
  88                 if (prio > q->highest_prio)
  89                         q->highest_prio = prio;
  90
  91                 if (prio < q->lowest_prio)
  92                         q->lowest_prio = prio;
  93
  94                 sch->q.qlen++;
  95                 return NET_XMIT_SUCCESS;
  96         }
  97
  98         /* If this packet has the lowest priority, drop it. */
  99         lp = q->lowest_prio;
 100         if (prio <= lp) {
 101                 q->qstats[prio].drops++;
 102                 q->qstats[prio].overlimits++;
 103                 return qdisc_drop(skb, sch, to_free);
 104         }
 105
 106         __skb_queue_tail(qdisc, skb);
 107         qdisc_qstats_backlog_inc(sch, skb);
 108         q->qstats[prio].backlog += qdisc_pkt_len(skb);
 109
 110         /* Drop the packet at the tail of the lowest priority qdisc. */
 111         lp_qdisc = &q->qdiscs[lp];
 112         to_drop = __skb_dequeue_tail(lp_qdisc);
 113         BUG_ON(!to_drop);
 114         qdisc_qstats_backlog_dec(sch, to_drop);
 115         qdisc_drop(to_drop, sch, to_free);
 116
 117         q->qstats[lp].backlog -= qdisc_pkt_len(to_drop);
 118         q->qstats[lp].drops++;
 119         q->qstats[lp].overlimits++;
 120
 121         /* Check to update highest and lowest priorities. */
 122         if (skb_queue_empty(lp_qdisc)) {
 123                 if (q->lowest_prio == q->highest_prio) {
 124                         /* The incoming packet is the only packet in queue. */
 125                         BUG_ON(sch->q.qlen != 1);
 126                         q->lowest_prio = prio;
 127                         q->highest_prio = prio;
 128                 } else {
 129                         q->lowest_prio = calc_new_low_prio(q);
 130                 }
 131         }
 132
 133         if (prio > q->highest_prio)
 134                 q->highest_prio = prio;
 135
 136         return NET_XMIT_CN;
 137 }
 138
 139 static struct sk_buff *skbprio_dequeue(struct Qdisc *sch)
 140 {
 141         struct skbprio_sched_data *q = qdisc_priv(sch);
 142         struct sk_buff_head *hpq = &q->qdiscs[q->highest_prio];
 143         struct sk_buff *skb = __skb_dequeue(hpq);
 144
 145         if (unlikely(!skb))
 146                 return NULL;
 147
 148         sch->q.qlen--;
 149         qdisc_qstats_backlog_dec(sch, skb);
 150         qdisc_bstats_update(sch, skb);
 151
 152         q->qstats[q->highest_prio].backlog -= qdisc_pkt_len(skb);
 153
 154         /* Update highest priority field. */
 155         if (skb_queue_empty(hpq)) {
 156                 if (q->lowest_prio == q->highest_prio) {
 157                         BUG_ON(sch->q.qlen);
 158                         q->highest_prio = 0;
 159                         q->lowest_prio = SKBPRIO_MAX_PRIORITY - 1;
 160                 } else {
 161                         q->highest_prio = calc_new_high_prio(q);
 162                 }
 163         }
 164         return skb;
 165 }
 166
 167 static int skbprio_change(struct Qdisc *sch, struct nlattr *opt,
 168                         struct netlink_ext_ack *extack)
 169 {
 170         struct tc_skbprio_qopt *ctl = nla_data(opt);
 171
 172         if (opt->nla_len != nla_attr_size(sizeof(*ctl)))
 173                 return -EINVAL;
 174
 175         sch->limit = ctl->limit;
 176         return 0;
 177 }
 178
 179 static int skbprio_init(struct Qdisc *sch, struct nlattr *opt,
 180                         struct netlink_ext_ack *extack)
 181 {
 182         struct skbprio_sched_data *q = qdisc_priv(sch);
 183         int prio;
 184
 185         /* Initialise all queues, one for each possible priority. */
 186         for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++)
 187                 __skb_queue_head_init(&q->qdiscs[prio]);
 188
 189         memset(&q->qstats, 0, sizeof(q->qstats));
 190         q->highest_prio = 0;
 191         q->lowest_prio = SKBPRIO_MAX_PRIORITY - 1;
 192         sch->limit = 64;
 193         if (!opt)
 194                 return 0;
 195
 196         return skbprio_change(sch, opt, extack);
 197 }
 198
 199 static int skbprio_dump(struct Qdisc *sch, struct sk_buff *skb)
 200 {
 201         struct tc_skbprio_qopt opt;
 202
 203         opt.limit = sch->limit;
 204
 205         if (nla_put(skb, TCA_OPTIONS, sizeof(opt), &opt))
 206                 return -1;
 207
 208         return skb->len;
 209 }
 210
 211 static void skbprio_reset(struct Qdisc *sch)
 212 {
 213         struct skbprio_sched_data *q = qdisc_priv(sch);
 214         int prio;
 215
 216         sch->qstats.backlog = 0;
 217         sch->q.qlen = 0;
 218
 219         for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++)
 220                 __skb_queue_purge(&q->qdiscs[prio]);
 221
 222         memset(&q->qstats, 0, sizeof(q->qstats));
 223         q->highest_prio = 0;
 224         q->lowest_prio = SKBPRIO_MAX_PRIORITY - 1;
 225 }
 226
 227 static void skbprio_destroy(struct Qdisc *sch)
 228 {
 229         struct skbprio_sched_data *q = qdisc_priv(sch);
 230         int prio;
 231
 232         for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++)
 233                 __skb_queue_purge(&q->qdiscs[prio]);
 234 }
 235
 236 static struct Qdisc *skbprio_leaf(struct Qdisc *sch, unsigned long arg)
 237 {
 238         return NULL;
 239 }
 240
 241 static unsigned long skbprio_find(struct Qdisc *sch, u32 classid)
 242 {
 243         return 0;
 244 }
 245
 246 static int skbprio_dump_class(struct Qdisc *sch, unsigned long cl,
 247                              struct sk_buff *skb, struct tcmsg *tcm)
 248 {
 249         tcm->tcm_handle |= TC_H_MIN(cl);
 250         return 0;
 251 }
 252
 253 static int skbprio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
 254                                    struct gnet_dump *d)
 255 {
 256         struct skbprio_sched_data *q = qdisc_priv(sch);
 257         if (gnet_stats_copy_queue(d, NULL, &q->qstats[cl - 1],
 258                 q->qstats[cl - 1].qlen) < 0)
 259                 return -1;
 260         return 0;
 261 }
 262
 263 static void skbprio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
 264 {
 265         unsigned int i;
 266
 267         if (arg->stop)
 268                 return;
 269
 270         for (i = 0; i < SKBPRIO_MAX_PRIORITY; i++) {
 271                 if (arg->count < arg->skip) {
 272                         arg->count++;
 273                         continue;
 274                 }
 275                 if (arg->fn(sch, i + 1, arg) < 0) {
 276                         arg->stop = 1;
 277                         break;
 278                 }
 279                 arg->count++;
 280         }
 281 }
 282
 283 static const struct Qdisc_class_ops skbprio_class_ops = {
 284         .leaf           =       skbprio_leaf,
 285         .find           =       skbprio_find,
 286         .dump           =       skbprio_dump_class,
 287         .dump_stats     =       skbprio_dump_class_stats,
 288         .walk           =       skbprio_walk,
 289 };
 290
 291 static struct Qdisc_ops skbprio_qdisc_ops __read_mostly = {
 292         .cl_ops         =       &skbprio_class_ops,
 293         .id             =       "skbprio",
 294         .priv_size      =       sizeof(struct skbprio_sched_data),
 295         .enqueue        =       skbprio_enqueue,
 296         .dequeue        =       skbprio_dequeue,
 297         .peek           =       qdisc_peek_dequeued,
 298         .init           =       skbprio_init,
 299         .reset          =       skbprio_reset,
 300         .change         =       skbprio_change,
 301         .dump           =       skbprio_dump,
 302         .destroy        =       skbprio_destroy,
 303         .owner          =       THIS_MODULE,
 304 };
 305
 306 static int __init skbprio_module_init(void)
 307 {
 308         return register_qdisc(&skbprio_qdisc_ops);
 309 }
 310
 311 static void __exit skbprio_module_exit(void)
 312 {
 313         unregister_qdisc(&skbprio_qdisc_ops);
 314 }
 315
 316 module_init(skbprio_module_init)
 317 module_exit(skbprio_module_exit)
 318
 319 MODULE_LICENSE("GPL");