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 for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++) 217 __skb_queue_purge(&q->qdiscs[prio]); 218 219 memset(&q->qstats, 0, sizeof(q->qstats)); 220 q->highest_prio = 0; 221 q->lowest_prio = SKBPRIO_MAX_PRIORITY - 1; 222 } 223 224 static void skbprio_destroy(struct Qdisc *sch) 225 { 226 struct skbprio_sched_data *q = qdisc_priv(sch); 227 int prio; 228 229 for (prio = 0; prio < SKBPRIO_MAX_PRIORITY; prio++) 230 __skb_queue_purge(&q->qdiscs[prio]); 231 } 232 233 static struct Qdisc *skbprio_leaf(struct Qdisc *sch, unsigned long arg) 234 { 235 return NULL; 236 } 237 238 static unsigned long skbprio_find(struct Qdisc *sch, u32 classid) 239 { 240 return 0; 241 } 242 243 static int skbprio_dump_class(struct Qdisc *sch, unsigned long cl, 244 struct sk_buff *skb, struct tcmsg *tcm) 245 { 246 tcm->tcm_handle |= TC_H_MIN(cl); 247 return 0; 248 } 249 250 static int skbprio_dump_class_stats(struct Qdisc *sch, unsigned long cl, 251 struct gnet_dump *d) 252 { 253 struct skbprio_sched_data *q = qdisc_priv(sch); 254 if (gnet_stats_copy_queue(d, NULL, &q->qstats[cl - 1], 255 q->qstats[cl - 1].qlen) < 0) 256 return -1; 257 return 0; 258 } 259 260 static void skbprio_walk(struct Qdisc *sch, struct qdisc_walker *arg) 261 { 262 unsigned int i; 263 264 if (arg->stop) 265 return; 266 267 for (i = 0; i < SKBPRIO_MAX_PRIORITY; i++) { 268 if (!tc_qdisc_stats_dump(sch, i + 1, arg)) 269 break; 270 } 271 } 272 273 static const struct Qdisc_class_ops skbprio_class_ops = { 274 .leaf = skbprio_leaf, 275 .find = skbprio_find, 276 .dump = skbprio_dump_class, 277 .dump_stats = skbprio_dump_class_stats, 278 .walk = skbprio_walk, 279 }; 280 281 static struct Qdisc_ops skbprio_qdisc_ops __read_mostly = { 282 .cl_ops = &skbprio_class_ops, 283 .id = "skbprio", 284 .priv_size = sizeof(struct skbprio_sched_data), 285 .enqueue = skbprio_enqueue, 286 .dequeue = skbprio_dequeue, 287 .peek = qdisc_peek_dequeued, 288 .init = skbprio_init, 289 .reset = skbprio_reset, 290 .change = skbprio_change, 291 .dump = skbprio_dump, 292 .destroy = skbprio_destroy, 293 .owner = THIS_MODULE, 294 }; 295 296 static int __init skbprio_module_init(void) 297 { 298 return register_qdisc(&skbprio_qdisc_ops); 299 } 300 301 static void __exit skbprio_module_exit(void) 302 { 303 unregister_qdisc(&skbprio_qdisc_ops); 304 } 305 306 module_init(skbprio_module_init) 307 module_exit(skbprio_module_exit) 308 309 MODULE_LICENSE("GPL"); 310